arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <[email protected]>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __this_cpu_read(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *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
 159 /**
 160  * kvm_arch_destroy_vm - destroy the VM data structure
 161  * @kvm:        pointer to the KVM struct
 162  */
 163 void kvm_arch_destroy_vm(struct kvm *kvm)
 164 {
 165         int i;
 166
 167         kvm_free_stage2_pgd(kvm);
 168
 169         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 170                 if (kvm->vcpus[i]) {
 171                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 172                         kvm->vcpus[i] = NULL;
 173                 }
 174         }
 175 }
 176
 177 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 178 {
 179         int r;
 180         switch (ext) {
 181         case KVM_CAP_IRQCHIP:
 182                 r = vgic_present;
 183                 break;
 184         case KVM_CAP_DEVICE_CTRL:
 185         case KVM_CAP_USER_MEMORY:
 186         case KVM_CAP_SYNC_MMU:
 187         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 188         case KVM_CAP_ONE_REG:
 189         case KVM_CAP_ARM_PSCI:
 190         case KVM_CAP_ARM_PSCI_0_2:
 191                 r = 1;
 192                 break;
 193         case KVM_CAP_COALESCED_MMIO:
 194                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 195                 break;
 196         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 197                 r = 1;
 198                 break;
 199         case KVM_CAP_NR_VCPUS:
 200                 r = num_online_cpus();
 201                 break;
 202         case KVM_CAP_MAX_VCPUS:
 203                 r = KVM_MAX_VCPUS;
 204                 break;
 205         default:
 206                 r = kvm_arch_dev_ioctl_check_extension(ext);
 207                 break;
 208         }
 209         return r;
 210 }
 211
 212 long kvm_arch_dev_ioctl(struct file *filp,
 213                         unsigned int ioctl, unsigned long arg)
 214 {
 215         return -EINVAL;
 216 }
 217
 218
 219 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 220 {
 221         int err;
 222         struct kvm_vcpu *vcpu;
 223
 224         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 225         if (!vcpu) {
 226                 err = -ENOMEM;
 227                 goto out;
 228         }
 229
 230         err = kvm_vcpu_init(vcpu, kvm, id);
 231         if (err)
 232                 goto free_vcpu;
 233
 234         err = create_hyp_mappings(vcpu, vcpu + 1);
 235         if (err)
 236                 goto vcpu_uninit;
 237
 238         return vcpu;
 239 vcpu_uninit:
 240         kvm_vcpu_uninit(vcpu);
 241 free_vcpu:
 242         kmem_cache_free(kvm_vcpu_cache, vcpu);
 243 out:
 244         return ERR_PTR(err);
 245 }
 246
 247 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 248 {
 249         return 0;
 250 }
 251
 252 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 253 {
 254         kvm_mmu_free_memory_caches(vcpu);
 255         kvm_timer_vcpu_terminate(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 kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 270 {
 271         int ret;
 272
 273         /* Force users to call KVM_ARM_VCPU_INIT */
 274         vcpu->arch.target = -1;
 275
 276         /* Set up VGIC */
 277         ret = kvm_vgic_vcpu_init(vcpu);
 278         if (ret)
 279                 return ret;
 280
 281         /* Set up the timer */
 282         kvm_timer_vcpu_init(vcpu);
 283
 284         return 0;
 285 }
 286
 287 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 288 {
 289 }
 290
 291 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 292 {
 293         vcpu->cpu = cpu;
 294         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 295
 296         /*
 297          * Check whether this vcpu requires the cache to be flushed on
 298          * this physical CPU. This is a consequence of doing dcache
 299          * operations by set/way on this vcpu. We do it here to be in
 300          * a non-preemptible section.
 301          */
 302         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 303                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 304
 305         kvm_arm_set_running_vcpu(vcpu);
 306 }
 307
 308 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 309 {
 310         /*
 311          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 312          * if the vcpu is no longer assigned to a cpu.  This is used for the
 313          * optimized make_all_cpus_request path.
 314          */
 315         vcpu->cpu = -1;
 316
 317         kvm_arm_set_running_vcpu(NULL);
 318 }
 319
 320 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 321                                         struct kvm_guest_debug *dbg)
 322 {
 323         return -EINVAL;
 324 }
 325
 326
 327 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 328                                     struct kvm_mp_state *mp_state)
 329 {
 330         return -EINVAL;
 331 }
 332
 333 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 334                                     struct kvm_mp_state *mp_state)
 335 {
 336         return -EINVAL;
 337 }
 338
 339 /**
 340  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 341  * @v:          The VCPU pointer
 342  *
 343  * If the guest CPU is not waiting for interrupts or an interrupt line is
 344  * asserted, the CPU is by definition runnable.
 345  */
 346 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 347 {
 348         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 349 }
 350
 351 /* Just ensure a guest exit from a particular CPU */
 352 static void exit_vm_noop(void *info)
 353 {
 354 }
 355
 356 void force_vm_exit(const cpumask_t *mask)
 357 {
 358         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 359 }
 360
 361 /**
 362  * need_new_vmid_gen - check that the VMID is still valid
 363  * @kvm: The VM's VMID to checkt
 364  *
 365  * return true if there is a new generation of VMIDs being used
 366  *
 367  * The hardware supports only 256 values with the value zero reserved for the
 368  * host, so we check if an assigned value belongs to a previous generation,
 369  * which which requires us to assign a new value. If we're the first to use a
 370  * VMID for the new generation, we must flush necessary caches and TLBs on all
 371  * CPUs.
 372  */
 373 static bool need_new_vmid_gen(struct kvm *kvm)
 374 {
 375         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 376 }
 377
 378 /**
 379  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 380  * @kvm The guest that we are about to run
 381  *
 382  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 383  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 384  * caches and TLBs.
 385  */
 386 static void update_vttbr(struct kvm *kvm)
 387 {
 388         phys_addr_t pgd_phys;
 389         u64 vmid;
 390
 391         if (!need_new_vmid_gen(kvm))
 392                 return;
 393
 394         spin_lock(&kvm_vmid_lock);
 395
 396         /*
 397          * We need to re-check the vmid_gen here to ensure that if another vcpu
 398          * already allocated a valid vmid for this vm, then this vcpu should
 399          * use the same vmid.
 400          */
 401         if (!need_new_vmid_gen(kvm)) {
 402                 spin_unlock(&kvm_vmid_lock);
 403                 return;
 404         }
 405
 406         /* First user of a new VMID generation? */
 407         if (unlikely(kvm_next_vmid == 0)) {
 408                 atomic64_inc(&kvm_vmid_gen);
 409                 kvm_next_vmid = 1;
 410
 411                 /*
 412                  * On SMP we know no other CPUs can use this CPU's or each
 413                  * other's VMID after force_vm_exit returns since the
 414                  * kvm_vmid_lock blocks them from reentry to the guest.
 415                  */
 416                 force_vm_exit(cpu_all_mask);
 417                 /*
 418                  * Now broadcast TLB + ICACHE invalidation over the inner
 419                  * shareable domain to make sure all data structures are
 420                  * clean.
 421                  */
 422                 kvm_call_hyp(__kvm_flush_vm_context);
 423         }
 424
 425         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 426         kvm->arch.vmid = kvm_next_vmid;
 427         kvm_next_vmid++;
 428
 429         /* update vttbr to be used with the new vmid */
 430         pgd_phys = virt_to_phys(kvm->arch.pgd);
 431         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 432         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 433         kvm->arch.vttbr |= vmid;
 434
 435         spin_unlock(&kvm_vmid_lock);
 436 }
 437
 438 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 439 {
 440         int ret;
 441
 442         if (likely(vcpu->arch.has_run_once))
 443                 return 0;
 444
 445         vcpu->arch.has_run_once = true;
 446
 447         /*
 448          * Initialize the VGIC before running a vcpu the first time on
 449          * this VM.
 450          */
 451         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 452                 ret = kvm_vgic_init(vcpu->kvm);
 453                 if (ret)
 454                         return ret;
 455         }
 456
 457         return 0;
 458 }
 459
 460 static void vcpu_pause(struct kvm_vcpu *vcpu)
 461 {
 462         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 463
 464         wait_event_interruptible(*wq, !vcpu->arch.pause);
 465 }
 466
 467 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 468 {
 469         return vcpu->arch.target >= 0;
 470 }
 471
 472 /**
 473  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 474  * @vcpu:       The VCPU pointer
 475  * @run:        The kvm_run structure pointer used for userspace state exchange
 476  *
 477  * This function is called through the VCPU_RUN ioctl called from user space. It
 478  * will execute VM code in a loop until the time slice for the process is used
 479  * or some emulation is needed from user space in which case the function will
 480  * return with return value 0 and with the kvm_run structure filled in with the
 481  * required data for the requested emulation.
 482  */
 483 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 484 {
 485         int ret;
 486         sigset_t sigsaved;
 487
 488         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 489                 return -ENOEXEC;
 490
 491         ret = kvm_vcpu_first_run_init(vcpu);
 492         if (ret)
 493                 return ret;
 494
 495         if (run->exit_reason == KVM_EXIT_MMIO) {
 496                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 497                 if (ret)
 498                         return ret;
 499         }
 500
 501         if (vcpu->sigset_active)
 502                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 503
 504         ret = 1;
 505         run->exit_reason = KVM_EXIT_UNKNOWN;
 506         while (ret > 0) {
 507                 /*
 508                  * Check conditions before entering the guest
 509                  */
 510                 cond_resched();
 511
 512                 update_vttbr(vcpu->kvm);
 513
 514                 if (vcpu->arch.pause)
 515                         vcpu_pause(vcpu);
 516
 517                 kvm_vgic_flush_hwstate(vcpu);
 518                 kvm_timer_flush_hwstate(vcpu);
 519
 520                 local_irq_disable();
 521
 522                 /*
 523                  * Re-check atomic conditions
 524                  */
 525                 if (signal_pending(current)) {
 526                         ret = -EINTR;
 527                         run->exit_reason = KVM_EXIT_INTR;
 528                 }
 529
 530                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 531                         local_irq_enable();
 532                         kvm_timer_sync_hwstate(vcpu);
 533                         kvm_vgic_sync_hwstate(vcpu);
 534                         continue;
 535                 }
 536
 537                 /**************************************************************
 538                  * Enter the guest
 539                  */
 540                 trace_kvm_entry(*vcpu_pc(vcpu));
 541                 kvm_guest_enter();
 542                 vcpu->mode = IN_GUEST_MODE;
 543
 544                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 545
 546                 vcpu->mode = OUTSIDE_GUEST_MODE;
 547                 vcpu->arch.last_pcpu = smp_processor_id();
 548                 kvm_guest_exit();
 549                 trace_kvm_exit(*vcpu_pc(vcpu));
 550                 /*
 551                  * We may have taken a host interrupt in HYP mode (ie
 552                  * while executing the guest). This interrupt is still
 553                  * pending, as we haven't serviced it yet!
 554                  *
 555                  * We're now back in SVC mode, with interrupts
 556                  * disabled.  Enabling the interrupts now will have
 557                  * the effect of taking the interrupt again, in SVC
 558                  * mode this time.
 559                  */
 560                 local_irq_enable();
 561
 562                 /*
 563                  * Back from guest
 564                  *************************************************************/
 565
 566                 kvm_timer_sync_hwstate(vcpu);
 567                 kvm_vgic_sync_hwstate(vcpu);
 568
 569                 ret = handle_exit(vcpu, run, ret);
 570         }
 571
 572         if (vcpu->sigset_active)
 573                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 574         return ret;
 575 }
 576
 577 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 578 {
 579         int bit_index;
 580         bool set;
 581         unsigned long *ptr;
 582
 583         if (number == KVM_ARM_IRQ_CPU_IRQ)
 584                 bit_index = __ffs(HCR_VI);
 585         else /* KVM_ARM_IRQ_CPU_FIQ */
 586                 bit_index = __ffs(HCR_VF);
 587
 588         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 589         if (level)
 590                 set = test_and_set_bit(bit_index, ptr);
 591         else
 592                 set = test_and_clear_bit(bit_index, ptr);
 593
 594         /*
 595          * If we didn't change anything, no need to wake up or kick other CPUs
 596          */
 597         if (set == level)
 598                 return 0;
 599
 600         /*
 601          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 602          * trigger a world-switch round on the running physical CPU to set the
 603          * virtual IRQ/FIQ fields in the HCR appropriately.
 604          */
 605         kvm_vcpu_kick(vcpu);
 606
 607         return 0;
 608 }
 609
 610 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 611                           bool line_status)
 612 {
 613         u32 irq = irq_level->irq;
 614         unsigned int irq_type, vcpu_idx, irq_num;
 615         int nrcpus = atomic_read(&kvm->online_vcpus);
 616         struct kvm_vcpu *vcpu = NULL;
 617         bool level = irq_level->level;
 618
 619         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 620         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 621         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 622
 623         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 624
 625         switch (irq_type) {
 626         case KVM_ARM_IRQ_TYPE_CPU:
 627                 if (irqchip_in_kernel(kvm))
 628                         return -ENXIO;
 629
 630                 if (vcpu_idx >= nrcpus)
 631                         return -EINVAL;
 632
 633                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 634                 if (!vcpu)
 635                         return -EINVAL;
 636
 637                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 638                         return -EINVAL;
 639
 640                 return vcpu_interrupt_line(vcpu, irq_num, level);
 641         case KVM_ARM_IRQ_TYPE_PPI:
 642                 if (!irqchip_in_kernel(kvm))
 643                         return -ENXIO;
 644
 645                 if (vcpu_idx >= nrcpus)
 646                         return -EINVAL;
 647
 648                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 649                 if (!vcpu)
 650                         return -EINVAL;
 651
 652                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 653                         return -EINVAL;
 654
 655                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 656         case KVM_ARM_IRQ_TYPE_SPI:
 657                 if (!irqchip_in_kernel(kvm))
 658                         return -ENXIO;
 659
 660                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 661                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 662                         return -EINVAL;
 663
 664                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 665         }
 666
 667         return -EINVAL;
 668 }
 669
 670 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 671                                          struct kvm_vcpu_init *init)
 672 {
 673         int ret;
 674
 675         ret = kvm_vcpu_set_target(vcpu, init);
 676         if (ret)
 677                 return ret;
 678
 679         /*
 680          * Handle the "start in power-off" case by marking the VCPU as paused.
 681          */
 682         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 683                 vcpu->arch.pause = true;
 684
 685         return 0;
 686 }
 687
 688 long kvm_arch_vcpu_ioctl(struct file *filp,
 689                          unsigned int ioctl, unsigned long arg)
 690 {
 691         struct kvm_vcpu *vcpu = filp->private_data;
 692         void __user *argp = (void __user *)arg;
 693
 694         switch (ioctl) {
 695         case KVM_ARM_VCPU_INIT: {
 696                 struct kvm_vcpu_init init;
 697
 698                 if (copy_from_user(&init, argp, sizeof(init)))
 699                         return -EFAULT;
 700
 701                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 702         }
 703         case KVM_SET_ONE_REG:
 704         case KVM_GET_ONE_REG: {
 705                 struct kvm_one_reg reg;
 706
 707                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 708                         return -ENOEXEC;
 709
 710                 if (copy_from_user(&reg, argp, sizeof(reg)))
 711                         return -EFAULT;
 712                 if (ioctl == KVM_SET_ONE_REG)
 713                         return kvm_arm_set_reg(vcpu, &reg);
 714                 else
 715                         return kvm_arm_get_reg(vcpu, &reg);
 716         }
 717         case KVM_GET_REG_LIST: {
 718                 struct kvm_reg_list __user *user_list = argp;
 719                 struct kvm_reg_list reg_list;
 720                 unsigned n;
 721
 722                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 723                         return -ENOEXEC;
 724
 725                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 726                         return -EFAULT;
 727                 n = reg_list.n;
 728                 reg_list.n = kvm_arm_num_regs(vcpu);
 729                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 730                         return -EFAULT;
 731                 if (n < reg_list.n)
 732                         return -E2BIG;
 733                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 734         }
 735         default:
 736                 return -EINVAL;
 737         }
 738 }
 739
 740 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 741 {
 742         return -EINVAL;
 743 }
 744
 745 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 746                                         struct kvm_arm_device_addr *dev_addr)
 747 {
 748         unsigned long dev_id, type;
 749
 750         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 751                 KVM_ARM_DEVICE_ID_SHIFT;
 752         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 753                 KVM_ARM_DEVICE_TYPE_SHIFT;
 754
 755         switch (dev_id) {
 756         case KVM_ARM_DEVICE_VGIC_V2:
 757                 if (!vgic_present)
 758                         return -ENXIO;
 759                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 760         default:
 761                 return -ENODEV;
 762         }
 763 }
 764
 765 long kvm_arch_vm_ioctl(struct file *filp,
 766                        unsigned int ioctl, unsigned long arg)
 767 {
 768         struct kvm *kvm = filp->private_data;
 769         void __user *argp = (void __user *)arg;
 770
 771         switch (ioctl) {
 772         case KVM_CREATE_IRQCHIP: {
 773                 if (vgic_present)
 774                         return kvm_vgic_create(kvm);
 775                 else
 776                         return -ENXIO;
 777         }
 778         case KVM_ARM_SET_DEVICE_ADDR: {
 779                 struct kvm_arm_device_addr dev_addr;
 780
 781                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 782                         return -EFAULT;
 783                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 784         }
 785         case KVM_ARM_PREFERRED_TARGET: {
 786                 int err;
 787                 struct kvm_vcpu_init init;
 788
 789                 err = kvm_vcpu_preferred_target(&init);
 790                 if (err)
 791                         return err;
 792
 793                 if (copy_to_user(argp, &init, sizeof(init)))
 794                         return -EFAULT;
 795
 796                 return 0;
 797         }
 798         default:
 799                 return -EINVAL;
 800         }
 801 }
 802
 803 static void cpu_init_hyp_mode(void *dummy)
 804 {
 805         phys_addr_t boot_pgd_ptr;
 806         phys_addr_t pgd_ptr;
 807         unsigned long hyp_stack_ptr;
 808         unsigned long stack_page;
 809         unsigned long vector_ptr;
 810
 811         /* Switch from the HYP stub to our own HYP init vector */
 812         __hyp_set_vectors(kvm_get_idmap_vector());
 813
 814         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 815         pgd_ptr = kvm_mmu_get_httbr();
 816         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 817         hyp_stack_ptr = stack_page + PAGE_SIZE;
 818         vector_ptr = (unsigned long)__kvm_hyp_vector;
 819
 820         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 821 }
 822
 823 static int hyp_init_cpu_notify(struct notifier_block *self,
 824                                unsigned long action, void *cpu)
 825 {
 826         switch (action) {
 827         case CPU_STARTING:
 828         case CPU_STARTING_FROZEN:
 829                 cpu_init_hyp_mode(NULL);
 830                 break;
 831         }
 832
 833         return NOTIFY_OK;
 834 }
 835
 836 static struct notifier_block hyp_init_cpu_nb = {
 837         .notifier_call = hyp_init_cpu_notify,
 838 };
 839
 840 #ifdef CONFIG_CPU_PM
 841 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 842                                     unsigned long cmd,
 843                                     void *v)
 844 {
 845         if (cmd == CPU_PM_EXIT &&
 846             __hyp_get_vectors() == hyp_default_vectors) {
 847                 cpu_init_hyp_mode(NULL);
 848                 return NOTIFY_OK;
 849         }
 850
 851         return NOTIFY_DONE;
 852 }
 853
 854 static struct notifier_block hyp_init_cpu_pm_nb = {
 855         .notifier_call = hyp_init_cpu_pm_notifier,
 856 };
 857
 858 static void __init hyp_cpu_pm_init(void)
 859 {
 860         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 861 }
 862 #else
 863 static inline void hyp_cpu_pm_init(void)
 864 {
 865 }
 866 #endif
 867
 868 /**
 869  * Inits Hyp-mode on all online CPUs
 870  */
 871 static int init_hyp_mode(void)
 872 {
 873         int cpu;
 874         int err = 0;
 875
 876         /*
 877          * Allocate Hyp PGD and setup Hyp identity mapping
 878          */
 879         err = kvm_mmu_init();
 880         if (err)
 881                 goto out_err;
 882
 883         /*
 884          * It is probably enough to obtain the default on one
 885          * CPU. It's unlikely to be different on the others.
 886          */
 887         hyp_default_vectors = __hyp_get_vectors();
 888
 889         /*
 890          * Allocate stack pages for Hypervisor-mode
 891          */
 892         for_each_possible_cpu(cpu) {
 893                 unsigned long stack_page;
 894
 895                 stack_page = __get_free_page(GFP_KERNEL);
 896                 if (!stack_page) {
 897                         err = -ENOMEM;
 898                         goto out_free_stack_pages;
 899                 }
 900
 901                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 902         }
 903
 904         /*
 905          * Map the Hyp-code called directly from the host
 906          */
 907         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 908         if (err) {
 909                 kvm_err("Cannot map world-switch code\n");
 910                 goto out_free_mappings;
 911         }
 912
 913         /*
 914          * Map the Hyp stack pages
 915          */
 916         for_each_possible_cpu(cpu) {
 917                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 918                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 919
 920                 if (err) {
 921                         kvm_err("Cannot map hyp stack\n");
 922                         goto out_free_mappings;
 923                 }
 924         }
 925
 926         /*
 927          * Map the host CPU structures
 928          */
 929         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 930         if (!kvm_host_cpu_state) {
 931                 err = -ENOMEM;
 932                 kvm_err("Cannot allocate host CPU state\n");
 933                 goto out_free_mappings;
 934         }
 935
 936         for_each_possible_cpu(cpu) {
 937                 kvm_cpu_context_t *cpu_ctxt;
 938
 939                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 940                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 941
 942                 if (err) {
 943                         kvm_err("Cannot map host CPU state: %d\n", err);
 944                         goto out_free_context;
 945                 }
 946         }
 947
 948         /*
 949          * Execute the init code on each CPU.
 950          */
 951         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 952
 953         /*
 954          * Init HYP view of VGIC
 955          */
 956         err = kvm_vgic_hyp_init();
 957         if (err)
 958                 goto out_free_context;
 959
 960 #ifdef CONFIG_KVM_ARM_VGIC
 961                 vgic_present = true;
 962 #endif
 963
 964         /*
 965          * Init HYP architected timer support
 966          */
 967         err = kvm_timer_hyp_init();
 968         if (err)
 969                 goto out_free_mappings;
 970
 971 #ifndef CONFIG_HOTPLUG_CPU
 972         free_boot_hyp_pgd();
 973 #endif
 974
 975         kvm_perf_init();
 976
 977         kvm_info("Hyp mode initialized successfully\n");
 978
 979         return 0;
 980 out_free_context:
 981         free_percpu(kvm_host_cpu_state);
 982 out_free_mappings:
 983         free_hyp_pgds();
 984 out_free_stack_pages:
 985         for_each_possible_cpu(cpu)
 986                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 987 out_err:
 988         kvm_err("error initializing Hyp mode: %d\n", err);
 989         return err;
 990 }
 991
 992 static void check_kvm_target_cpu(void *ret)
 993 {
 994         *(int *)ret = kvm_target_cpu();
 995 }
 996
 997 /**
 998  * Initialize Hyp-mode and memory mappings on all CPUs.
 999  */
1000 int kvm_arch_init(void *opaque)
1001 {
1002         int err;
1003         int ret, cpu;
1004
1005         if (!is_hyp_mode_available()) {
1006                 kvm_err("HYP mode not available\n");
1007                 return -ENODEV;
1008         }
1009
1010         for_each_online_cpu(cpu) {
1011                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1012                 if (ret < 0) {
1013                         kvm_err("Error, CPU %d not supported!\n", cpu);
1014                         return -ENODEV;
1015                 }
1016         }
1017
1018         cpu_notifier_register_begin();
1019
1020         err = init_hyp_mode();
1021         if (err)
1022                 goto out_err;
1023
1024         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1025         if (err) {
1026                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1027                 goto out_err;
1028         }
1029
1030         cpu_notifier_register_done();
1031
1032         hyp_cpu_pm_init();
1033
1034         kvm_coproc_table_init();
1035         return 0;
1036 out_err:
1037         cpu_notifier_register_done();
1038         return err;
1039 }
1040
1041 /* NOP: Compiling as a module not supported */
1042 void kvm_arch_exit(void)
1043 {
1044         kvm_perf_teardown();
1045 }
1046
1047 static int arm_init(void)
1048 {
1049         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1050         return rc;
1051 }
1052
1053 module_init(arm_init);