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         __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 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         /*
 347          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 348          * if the vcpu is no longer assigned to a cpu.  This is used for the
 349          * optimized make_all_cpus_request path.
 350          */
 351         vcpu->cpu = -1;
 352
 353         kvm_arm_set_running_vcpu(NULL);
 354 }
 355
 356 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 357                                         struct kvm_guest_debug *dbg)
 358 {
 359         return -EINVAL;
 360 }
 361
 362
 363 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 364                                     struct kvm_mp_state *mp_state)
 365 {
 366         return -EINVAL;
 367 }
 368
 369 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 370                                     struct kvm_mp_state *mp_state)
 371 {
 372         return -EINVAL;
 373 }
 374
 375 /**
 376  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 377  * @v:          The VCPU pointer
 378  *
 379  * If the guest CPU is not waiting for interrupts or an interrupt line is
 380  * asserted, the CPU is by definition runnable.
 381  */
 382 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 383 {
 384         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 385 }
 386
 387 /* Just ensure a guest exit from a particular CPU */
 388 static void exit_vm_noop(void *info)
 389 {
 390 }
 391
 392 void force_vm_exit(const cpumask_t *mask)
 393 {
 394         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 395 }
 396
 397 /**
 398  * need_new_vmid_gen - check that the VMID is still valid
 399  * @kvm: The VM's VMID to checkt
 400  *
 401  * return true if there is a new generation of VMIDs being used
 402  *
 403  * The hardware supports only 256 values with the value zero reserved for the
 404  * host, so we check if an assigned value belongs to a previous generation,
 405  * which which requires us to assign a new value. If we're the first to use a
 406  * VMID for the new generation, we must flush necessary caches and TLBs on all
 407  * CPUs.
 408  */
 409 static bool need_new_vmid_gen(struct kvm *kvm)
 410 {
 411         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 412 }
 413
 414 /**
 415  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 416  * @kvm The guest that we are about to run
 417  *
 418  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 419  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 420  * caches and TLBs.
 421  */
 422 static void update_vttbr(struct kvm *kvm)
 423 {
 424         phys_addr_t pgd_phys;
 425         u64 vmid;
 426
 427         if (!need_new_vmid_gen(kvm))
 428                 return;
 429
 430         spin_lock(&kvm_vmid_lock);
 431
 432         /*
 433          * We need to re-check the vmid_gen here to ensure that if another vcpu
 434          * already allocated a valid vmid for this vm, then this vcpu should
 435          * use the same vmid.
 436          */
 437         if (!need_new_vmid_gen(kvm)) {
 438                 spin_unlock(&kvm_vmid_lock);
 439                 return;
 440         }
 441
 442         /* First user of a new VMID generation? */
 443         if (unlikely(kvm_next_vmid == 0)) {
 444                 atomic64_inc(&kvm_vmid_gen);
 445                 kvm_next_vmid = 1;
 446
 447                 /*
 448                  * On SMP we know no other CPUs can use this CPU's or each
 449                  * other's VMID after force_vm_exit returns since the
 450                  * kvm_vmid_lock blocks them from reentry to the guest.
 451                  */
 452                 force_vm_exit(cpu_all_mask);
 453                 /*
 454                  * Now broadcast TLB + ICACHE invalidation over the inner
 455                  * shareable domain to make sure all data structures are
 456                  * clean.
 457                  */
 458                 kvm_call_hyp(__kvm_flush_vm_context);
 459         }
 460
 461         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 462         kvm->arch.vmid = kvm_next_vmid;
 463         kvm_next_vmid++;
 464
 465         /* update vttbr to be used with the new vmid */
 466         pgd_phys = virt_to_phys(kvm->arch.pgd);
 467         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 468         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 469         kvm->arch.vttbr |= vmid;
 470
 471         spin_unlock(&kvm_vmid_lock);
 472 }
 473
 474 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 475 {
 476         int ret;
 477
 478         if (likely(vcpu->arch.has_run_once))
 479                 return 0;
 480
 481         vcpu->arch.has_run_once = true;
 482
 483         /*
 484          * Initialize the VGIC before running a vcpu the first time on
 485          * this VM.
 486          */
 487         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 488                 ret = kvm_vgic_init(vcpu->kvm);
 489                 if (ret)
 490                         return ret;
 491         }
 492
 493         return 0;
 494 }
 495
 496 static void vcpu_pause(struct kvm_vcpu *vcpu)
 497 {
 498         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 499
 500         wait_event_interruptible(*wq, !vcpu->arch.pause);
 501 }
 502
 503 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 504 {
 505         return vcpu->arch.target >= 0;
 506 }
 507
 508 /**
 509  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 510  * @vcpu:       The VCPU pointer
 511  * @run:        The kvm_run structure pointer used for userspace state exchange
 512  *
 513  * This function is called through the VCPU_RUN ioctl called from user space. It
 514  * will execute VM code in a loop until the time slice for the process is used
 515  * or some emulation is needed from user space in which case the function will
 516  * return with return value 0 and with the kvm_run structure filled in with the
 517  * required data for the requested emulation.
 518  */
 519 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 520 {
 521         int ret;
 522         sigset_t sigsaved;
 523
 524         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 525                 return -ENOEXEC;
 526
 527         ret = kvm_vcpu_first_run_init(vcpu);
 528         if (ret)
 529                 return ret;
 530
 531         if (run->exit_reason == KVM_EXIT_MMIO) {
 532                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 533                 if (ret)
 534                         return ret;
 535         }
 536
 537         if (vcpu->sigset_active)
 538                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 539
 540         ret = 1;
 541         run->exit_reason = KVM_EXIT_UNKNOWN;
 542         while (ret > 0) {
 543                 /*
 544                  * Check conditions before entering the guest
 545                  */
 546                 cond_resched();
 547
 548                 update_vttbr(vcpu->kvm);
 549
 550                 if (vcpu->arch.pause)
 551                         vcpu_pause(vcpu);
 552
 553                 kvm_vgic_flush_hwstate(vcpu);
 554                 kvm_timer_flush_hwstate(vcpu);
 555
 556                 local_irq_disable();
 557
 558                 /*
 559                  * Re-check atomic conditions
 560                  */
 561                 if (signal_pending(current)) {
 562                         ret = -EINTR;
 563                         run->exit_reason = KVM_EXIT_INTR;
 564                 }
 565
 566                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 567                         local_irq_enable();
 568                         kvm_timer_sync_hwstate(vcpu);
 569                         kvm_vgic_sync_hwstate(vcpu);
 570                         continue;
 571                 }
 572
 573                 /**************************************************************
 574                  * Enter the guest
 575                  */
 576                 trace_kvm_entry(*vcpu_pc(vcpu));
 577                 kvm_guest_enter();
 578                 vcpu->mode = IN_GUEST_MODE;
 579
 580                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 581
 582                 vcpu->mode = OUTSIDE_GUEST_MODE;
 583                 vcpu->arch.last_pcpu = smp_processor_id();
 584                 kvm_guest_exit();
 585                 trace_kvm_exit(*vcpu_pc(vcpu));
 586                 /*
 587                  * We may have taken a host interrupt in HYP mode (ie
 588                  * while executing the guest). This interrupt is still
 589                  * pending, as we haven't serviced it yet!
 590                  *
 591                  * We're now back in SVC mode, with interrupts
 592                  * disabled.  Enabling the interrupts now will have
 593                  * the effect of taking the interrupt again, in SVC
 594                  * mode this time.
 595                  */
 596                 local_irq_enable();
 597
 598                 /*
 599                  * Back from guest
 600                  *************************************************************/
 601
 602                 kvm_timer_sync_hwstate(vcpu);
 603                 kvm_vgic_sync_hwstate(vcpu);
 604
 605                 ret = handle_exit(vcpu, run, ret);
 606         }
 607
 608         if (vcpu->sigset_active)
 609                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 610         return ret;
 611 }
 612
 613 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 614 {
 615         int bit_index;
 616         bool set;
 617         unsigned long *ptr;
 618
 619         if (number == KVM_ARM_IRQ_CPU_IRQ)
 620                 bit_index = __ffs(HCR_VI);
 621         else /* KVM_ARM_IRQ_CPU_FIQ */
 622                 bit_index = __ffs(HCR_VF);
 623
 624         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 625         if (level)
 626                 set = test_and_set_bit(bit_index, ptr);
 627         else
 628                 set = test_and_clear_bit(bit_index, ptr);
 629
 630         /*
 631          * If we didn't change anything, no need to wake up or kick other CPUs
 632          */
 633         if (set == level)
 634                 return 0;
 635
 636         /*
 637          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 638          * trigger a world-switch round on the running physical CPU to set the
 639          * virtual IRQ/FIQ fields in the HCR appropriately.
 640          */
 641         kvm_vcpu_kick(vcpu);
 642
 643         return 0;
 644 }
 645
 646 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 647                           bool line_status)
 648 {
 649         u32 irq = irq_level->irq;
 650         unsigned int irq_type, vcpu_idx, irq_num;
 651         int nrcpus = atomic_read(&kvm->online_vcpus);
 652         struct kvm_vcpu *vcpu = NULL;
 653         bool level = irq_level->level;
 654
 655         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 656         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 657         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 658
 659         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 660
 661         switch (irq_type) {
 662         case KVM_ARM_IRQ_TYPE_CPU:
 663                 if (irqchip_in_kernel(kvm))
 664                         return -ENXIO;
 665
 666                 if (vcpu_idx >= nrcpus)
 667                         return -EINVAL;
 668
 669                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 670                 if (!vcpu)
 671                         return -EINVAL;
 672
 673                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 674                         return -EINVAL;
 675
 676                 return vcpu_interrupt_line(vcpu, irq_num, level);
 677         case KVM_ARM_IRQ_TYPE_PPI:
 678                 if (!irqchip_in_kernel(kvm))
 679                         return -ENXIO;
 680
 681                 if (vcpu_idx >= nrcpus)
 682                         return -EINVAL;
 683
 684                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 685                 if (!vcpu)
 686                         return -EINVAL;
 687
 688                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 689                         return -EINVAL;
 690
 691                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 692         case KVM_ARM_IRQ_TYPE_SPI:
 693                 if (!irqchip_in_kernel(kvm))
 694                         return -ENXIO;
 695
 696                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 697                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 698                         return -EINVAL;
 699
 700                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 701         }
 702
 703         return -EINVAL;
 704 }
 705
 706 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 707                                          struct kvm_vcpu_init *init)
 708 {
 709         int ret;
 710
 711         ret = kvm_vcpu_set_target(vcpu, init);
 712         if (ret)
 713                 return ret;
 714
 715         /*
 716          * Handle the "start in power-off" case by marking the VCPU as paused.
 717          */
 718         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 719                 vcpu->arch.pause = true;
 720
 721         return 0;
 722 }
 723
 724 long kvm_arch_vcpu_ioctl(struct file *filp,
 725                          unsigned int ioctl, unsigned long arg)
 726 {
 727         struct kvm_vcpu *vcpu = filp->private_data;
 728         void __user *argp = (void __user *)arg;
 729
 730         switch (ioctl) {
 731         case KVM_ARM_VCPU_INIT: {
 732                 struct kvm_vcpu_init init;
 733
 734                 if (copy_from_user(&init, argp, sizeof(init)))
 735                         return -EFAULT;
 736
 737                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 738         }
 739         case KVM_SET_ONE_REG:
 740         case KVM_GET_ONE_REG: {
 741                 struct kvm_one_reg reg;
 742
 743                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 744                         return -ENOEXEC;
 745
 746                 if (copy_from_user(&reg, argp, sizeof(reg)))
 747                         return -EFAULT;
 748                 if (ioctl == KVM_SET_ONE_REG)
 749                         return kvm_arm_set_reg(vcpu, &reg);
 750                 else
 751                         return kvm_arm_get_reg(vcpu, &reg);
 752         }
 753         case KVM_GET_REG_LIST: {
 754                 struct kvm_reg_list __user *user_list = argp;
 755                 struct kvm_reg_list reg_list;
 756                 unsigned n;
 757
 758                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 759                         return -ENOEXEC;
 760
 761                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 762                         return -EFAULT;
 763                 n = reg_list.n;
 764                 reg_list.n = kvm_arm_num_regs(vcpu);
 765                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 766                         return -EFAULT;
 767                 if (n < reg_list.n)
 768                         return -E2BIG;
 769                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 770         }
 771         default:
 772                 return -EINVAL;
 773         }
 774 }
 775
 776 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 777 {
 778         return -EINVAL;
 779 }
 780
 781 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 782                                         struct kvm_arm_device_addr *dev_addr)
 783 {
 784         unsigned long dev_id, type;
 785
 786         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 787                 KVM_ARM_DEVICE_ID_SHIFT;
 788         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 789                 KVM_ARM_DEVICE_TYPE_SHIFT;
 790
 791         switch (dev_id) {
 792         case KVM_ARM_DEVICE_VGIC_V2:
 793                 if (!vgic_present)
 794                         return -ENXIO;
 795                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 796         default:
 797                 return -ENODEV;
 798         }
 799 }
 800
 801 long kvm_arch_vm_ioctl(struct file *filp,
 802                        unsigned int ioctl, unsigned long arg)
 803 {
 804         struct kvm *kvm = filp->private_data;
 805         void __user *argp = (void __user *)arg;
 806
 807         switch (ioctl) {
 808         case KVM_CREATE_IRQCHIP: {
 809                 if (vgic_present)
 810                         return kvm_vgic_create(kvm);
 811                 else
 812                         return -ENXIO;
 813         }
 814         case KVM_ARM_SET_DEVICE_ADDR: {
 815                 struct kvm_arm_device_addr dev_addr;
 816
 817                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 818                         return -EFAULT;
 819                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 820         }
 821         case KVM_ARM_PREFERRED_TARGET: {
 822                 int err;
 823                 struct kvm_vcpu_init init;
 824
 825                 err = kvm_vcpu_preferred_target(&init);
 826                 if (err)
 827                         return err;
 828
 829                 if (copy_to_user(argp, &init, sizeof(init)))
 830                         return -EFAULT;
 831
 832                 return 0;
 833         }
 834         default:
 835                 return -EINVAL;
 836         }
 837 }
 838
 839 static void cpu_init_hyp_mode(void *dummy)
 840 {
 841         phys_addr_t boot_pgd_ptr;
 842         phys_addr_t pgd_ptr;
 843         unsigned long hyp_stack_ptr;
 844         unsigned long stack_page;
 845         unsigned long vector_ptr;
 846
 847         /* Switch from the HYP stub to our own HYP init vector */
 848         __hyp_set_vectors(kvm_get_idmap_vector());
 849
 850         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 851         pgd_ptr = kvm_mmu_get_httbr();
 852         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 853         hyp_stack_ptr = stack_page + PAGE_SIZE;
 854         vector_ptr = (unsigned long)__kvm_hyp_vector;
 855
 856         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 857 }
 858
 859 static int hyp_init_cpu_notify(struct notifier_block *self,
 860                                unsigned long action, void *cpu)
 861 {
 862         switch (action) {
 863         case CPU_STARTING:
 864         case CPU_STARTING_FROZEN:
 865                 cpu_init_hyp_mode(NULL);
 866                 break;
 867         }
 868
 869         return NOTIFY_OK;
 870 }
 871
 872 static struct notifier_block hyp_init_cpu_nb = {
 873         .notifier_call = hyp_init_cpu_notify,
 874 };
 875
 876 #ifdef CONFIG_CPU_PM
 877 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 878                                     unsigned long cmd,
 879                                     void *v)
 880 {
 881         if (cmd == CPU_PM_EXIT &&
 882             __hyp_get_vectors() == hyp_default_vectors) {
 883                 cpu_init_hyp_mode(NULL);
 884                 return NOTIFY_OK;
 885         }
 886
 887         return NOTIFY_DONE;
 888 }
 889
 890 static struct notifier_block hyp_init_cpu_pm_nb = {
 891         .notifier_call = hyp_init_cpu_pm_notifier,
 892 };
 893
 894 static void __init hyp_cpu_pm_init(void)
 895 {
 896         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 897 }
 898 #else
 899 static inline void hyp_cpu_pm_init(void)
 900 {
 901 }
 902 #endif
 903
 904 /**
 905  * Inits Hyp-mode on all online CPUs
 906  */
 907 static int init_hyp_mode(void)
 908 {
 909         int cpu;
 910         int err = 0;
 911
 912         /*
 913          * Allocate Hyp PGD and setup Hyp identity mapping
 914          */
 915         err = kvm_mmu_init();
 916         if (err)
 917                 goto out_err;
 918
 919         /*
 920          * It is probably enough to obtain the default on one
 921          * CPU. It's unlikely to be different on the others.
 922          */
 923         hyp_default_vectors = __hyp_get_vectors();
 924
 925         /*
 926          * Allocate stack pages for Hypervisor-mode
 927          */
 928         for_each_possible_cpu(cpu) {
 929                 unsigned long stack_page;
 930
 931                 stack_page = __get_free_page(GFP_KERNEL);
 932                 if (!stack_page) {
 933                         err = -ENOMEM;
 934                         goto out_free_stack_pages;
 935                 }
 936
 937                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 938         }
 939
 940         /*
 941          * Map the Hyp-code called directly from the host
 942          */
 943         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 944         if (err) {
 945                 kvm_err("Cannot map world-switch code\n");
 946                 goto out_free_mappings;
 947         }
 948
 949         /*
 950          * Map the Hyp stack pages
 951          */
 952         for_each_possible_cpu(cpu) {
 953                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 954                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 955
 956                 if (err) {
 957                         kvm_err("Cannot map hyp stack\n");
 958                         goto out_free_mappings;
 959                 }
 960         }
 961
 962         /*
 963          * Map the host CPU structures
 964          */
 965         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 966         if (!kvm_host_cpu_state) {
 967                 err = -ENOMEM;
 968                 kvm_err("Cannot allocate host CPU state\n");
 969                 goto out_free_mappings;
 970         }
 971
 972         for_each_possible_cpu(cpu) {
 973                 kvm_cpu_context_t *cpu_ctxt;
 974
 975                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 976                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 977
 978                 if (err) {
 979                         kvm_err("Cannot map host CPU state: %d\n", err);
 980                         goto out_free_context;
 981                 }
 982         }
 983
 984         /*
 985          * Execute the init code on each CPU.
 986          */
 987         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 988
 989         /*
 990          * Init HYP view of VGIC
 991          */
 992         err = kvm_vgic_hyp_init();
 993         if (err)
 994                 goto out_free_context;
 995
 996 #ifdef CONFIG_KVM_ARM_VGIC
 997                 vgic_present = true;
 998 #endif
 999
1000         /*
1001          * Init HYP architected timer support
1002          */
1003         err = kvm_timer_hyp_init();
1004         if (err)
1005                 goto out_free_mappings;
1006
1007 #ifndef CONFIG_HOTPLUG_CPU
1008         free_boot_hyp_pgd();
1009 #endif
1010
1011         kvm_perf_init();
1012
1013         kvm_info("Hyp mode initialized successfully\n");
1014
1015         return 0;
1016 out_free_context:
1017         free_percpu(kvm_host_cpu_state);
1018 out_free_mappings:
1019         free_hyp_pgds();
1020 out_free_stack_pages:
1021         for_each_possible_cpu(cpu)
1022                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1023 out_err:
1024         kvm_err("error initializing Hyp mode: %d\n", err);
1025         return err;
1026 }
1027
1028 static void check_kvm_target_cpu(void *ret)
1029 {
1030         *(int *)ret = kvm_target_cpu();
1031 }
1032
1033 /**
1034  * Initialize Hyp-mode and memory mappings on all CPUs.
1035  */
1036 int kvm_arch_init(void *opaque)
1037 {
1038         int err;
1039         int ret, cpu;
1040
1041         if (!is_hyp_mode_available()) {
1042                 kvm_err("HYP mode not available\n");
1043                 return -ENODEV;
1044         }
1045
1046         for_each_online_cpu(cpu) {
1047                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1048                 if (ret < 0) {
1049                         kvm_err("Error, CPU %d not supported!\n", cpu);
1050                         return -ENODEV;
1051                 }
1052         }
1053
1054         cpu_notifier_register_begin();
1055
1056         err = init_hyp_mode();
1057         if (err)
1058                 goto out_err;
1059
1060         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1061         if (err) {
1062                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1063                 goto out_err;
1064         }
1065
1066         cpu_notifier_register_done();
1067
1068         hyp_cpu_pm_init();
1069
1070         kvm_coproc_table_init();
1071         return 0;
1072 out_err:
1073         cpu_notifier_register_done();
1074         return err;
1075 }
1076
1077 /* NOP: Compiling as a module not supported */
1078 void kvm_arch_exit(void)
1079 {
1080         kvm_perf_teardown();
1081 }
1082
1083 static int arm_init(void)
1084 {
1085         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1086         return rc;
1087 }
1088
1089 module_init(arm_init);