virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43
  44 #include <asm/processor.h>
  45 #include <asm/io.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/pgtable.h>
  48
  49 MODULE_AUTHOR("Qumranet");
  50 MODULE_LICENSE("GPL");
  51
  52 DEFINE_SPINLOCK(kvm_lock);
  53 LIST_HEAD(vm_list);
  54
  55 static cpumask_t cpus_hardware_enabled;
  56
  57 struct kmem_cache *kvm_vcpu_cache;
  58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  59
  60 static __read_mostly struct preempt_ops kvm_preempt_ops;
  61
  62 static struct dentry *debugfs_dir;
  63
  64 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  65                            unsigned long arg);
  66
  67 static inline int valid_vcpu(int n)
  68 {
  69         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  70 }
  71
  72 /*
  73  * Switches to specified vcpu, until a matching vcpu_put()
  74  */
  75 void vcpu_load(struct kvm_vcpu *vcpu)
  76 {
  77         int cpu;
  78
  79         mutex_lock(&vcpu->mutex);
  80         cpu = get_cpu();
  81         preempt_notifier_register(&vcpu->preempt_notifier);
  82         kvm_arch_vcpu_load(vcpu, cpu);
  83         put_cpu();
  84 }
  85
  86 void vcpu_put(struct kvm_vcpu *vcpu)
  87 {
  88         preempt_disable();
  89         kvm_arch_vcpu_put(vcpu);
  90         preempt_notifier_unregister(&vcpu->preempt_notifier);
  91         preempt_enable();
  92         mutex_unlock(&vcpu->mutex);
  93 }
  94
  95 static void ack_flush(void *_completed)
  96 {
  97 }
  98
  99 void kvm_flush_remote_tlbs(struct kvm *kvm)
 100 {
 101         int i, cpu;
 102         cpumask_t cpus;
 103         struct kvm_vcpu *vcpu;
 104
 105         cpus_clear(cpus);
 106         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 107                 vcpu = kvm->vcpus[i];
 108                 if (!vcpu)
 109                         continue;
 110                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 111                         continue;
 112                 cpu = vcpu->cpu;
 113                 if (cpu != -1 && cpu != raw_smp_processor_id())
 114                         cpu_set(cpu, cpus);
 115         }
 116         if (cpus_empty(cpus))
 117                 return;
 118         ++kvm->stat.remote_tlb_flush;
 119         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 120 }
 121
 122 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 123 {
 124         struct page *page;
 125         int r;
 126
 127         mutex_init(&vcpu->mutex);
 128         vcpu->cpu = -1;
 129         vcpu->kvm = kvm;
 130         vcpu->vcpu_id = id;
 131         init_waitqueue_head(&vcpu->wq);
 132
 133         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 134         if (!page) {
 135                 r = -ENOMEM;
 136                 goto fail;
 137         }
 138         vcpu->run = page_address(page);
 139
 140         r = kvm_arch_vcpu_init(vcpu);
 141         if (r < 0)
 142                 goto fail_free_run;
 143         return 0;
 144
 145 fail_free_run:
 146         free_page((unsigned long)vcpu->run);
 147 fail:
 148         return r;
 149 }
 150 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 151
 152 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 153 {
 154         kvm_arch_vcpu_uninit(vcpu);
 155         free_page((unsigned long)vcpu->run);
 156 }
 157 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 158
 159 static struct kvm *kvm_create_vm(void)
 160 {
 161         struct kvm *kvm = kvm_arch_create_vm();
 162
 163         if (IS_ERR(kvm))
 164                 goto out;
 165
 166         kvm->mm = current->mm;
 167         atomic_inc(&kvm->mm->mm_count);
 168         kvm_io_bus_init(&kvm->pio_bus);
 169         mutex_init(&kvm->lock);
 170         kvm_io_bus_init(&kvm->mmio_bus);
 171         spin_lock(&kvm_lock);
 172         list_add(&kvm->vm_list, &vm_list);
 173         spin_unlock(&kvm_lock);
 174 out:
 175         return kvm;
 176 }
 177
 178 /*
 179  * Free any memory in @free but not in @dont.
 180  */
 181 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 182                                   struct kvm_memory_slot *dont)
 183 {
 184         if (!dont || free->rmap != dont->rmap)
 185                 vfree(free->rmap);
 186
 187         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 188                 vfree(free->dirty_bitmap);
 189
 190         free->npages = 0;
 191         free->dirty_bitmap = NULL;
 192         free->rmap = NULL;
 193 }
 194
 195 void kvm_free_physmem(struct kvm *kvm)
 196 {
 197         int i;
 198
 199         for (i = 0; i < kvm->nmemslots; ++i)
 200                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 201 }
 202
 203 static void kvm_destroy_vm(struct kvm *kvm)
 204 {
 205         struct mm_struct *mm = kvm->mm;
 206
 207         spin_lock(&kvm_lock);
 208         list_del(&kvm->vm_list);
 209         spin_unlock(&kvm_lock);
 210         kvm_io_bus_destroy(&kvm->pio_bus);
 211         kvm_io_bus_destroy(&kvm->mmio_bus);
 212         kvm_arch_destroy_vm(kvm);
 213         mmdrop(mm);
 214 }
 215
 216 static int kvm_vm_release(struct inode *inode, struct file *filp)
 217 {
 218         struct kvm *kvm = filp->private_data;
 219
 220         kvm_destroy_vm(kvm);
 221         return 0;
 222 }
 223
 224 /*
 225  * Allocate some memory and give it an address in the guest physical address
 226  * space.
 227  *
 228  * Discontiguous memory is allowed, mostly for framebuffers.
 229  *
 230  * Must be called holding kvm->lock.
 231  */
 232 int __kvm_set_memory_region(struct kvm *kvm,
 233                             struct kvm_userspace_memory_region *mem,
 234                             int user_alloc)
 235 {
 236         int r;
 237         gfn_t base_gfn;
 238         unsigned long npages;
 239         unsigned long i;
 240         struct kvm_memory_slot *memslot;
 241         struct kvm_memory_slot old, new;
 242
 243         r = -EINVAL;
 244         /* General sanity checks */
 245         if (mem->memory_size & (PAGE_SIZE - 1))
 246                 goto out;
 247         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 248                 goto out;
 249         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 250                 goto out;
 251         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 252                 goto out;
 253
 254         memslot = &kvm->memslots[mem->slot];
 255         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 256         npages = mem->memory_size >> PAGE_SHIFT;
 257
 258         if (!npages)
 259                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 260
 261         new = old = *memslot;
 262
 263         new.base_gfn = base_gfn;
 264         new.npages = npages;
 265         new.flags = mem->flags;
 266
 267         /* Disallow changing a memory slot's size. */
 268         r = -EINVAL;
 269         if (npages && old.npages && npages != old.npages)
 270                 goto out_free;
 271
 272         /* Check for overlaps */
 273         r = -EEXIST;
 274         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 275                 struct kvm_memory_slot *s = &kvm->memslots[i];
 276
 277                 if (s == memslot)
 278                         continue;
 279                 if (!((base_gfn + npages <= s->base_gfn) ||
 280                       (base_gfn >= s->base_gfn + s->npages)))
 281                         goto out_free;
 282         }
 283
 284         /* Free page dirty bitmap if unneeded */
 285         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 286                 new.dirty_bitmap = NULL;
 287
 288         r = -ENOMEM;
 289
 290         /* Allocate if a slot is being created */
 291         if (npages && !new.rmap) {
 292                 new.rmap = vmalloc(npages * sizeof(struct page *));
 293
 294                 if (!new.rmap)
 295                         goto out_free;
 296
 297                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 298
 299                 new.user_alloc = user_alloc;
 300                 new.userspace_addr = mem->userspace_addr;
 301         }
 302
 303         /* Allocate page dirty bitmap if needed */
 304         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 305                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 306
 307                 new.dirty_bitmap = vmalloc(dirty_bytes);
 308                 if (!new.dirty_bitmap)
 309                         goto out_free;
 310                 memset(new.dirty_bitmap, 0, dirty_bytes);
 311         }
 312
 313         if (mem->slot >= kvm->nmemslots)
 314                 kvm->nmemslots = mem->slot + 1;
 315
 316         *memslot = new;
 317
 318         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 319         if (r) {
 320                 *memslot = old;
 321                 goto out_free;
 322         }
 323
 324         kvm_free_physmem_slot(&old, &new);
 325         return 0;
 326
 327 out_free:
 328         kvm_free_physmem_slot(&new, &old);
 329 out:
 330         return r;
 331
 332 }
 333 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 334
 335 int kvm_set_memory_region(struct kvm *kvm,
 336                           struct kvm_userspace_memory_region *mem,
 337                           int user_alloc)
 338 {
 339         int r;
 340
 341         mutex_lock(&kvm->lock);
 342         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 343         mutex_unlock(&kvm->lock);
 344         return r;
 345 }
 346 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 347
 348 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 349                                    struct
 350                                    kvm_userspace_memory_region *mem,
 351                                    int user_alloc)
 352 {
 353         if (mem->slot >= KVM_MEMORY_SLOTS)
 354                 return -EINVAL;
 355         return kvm_set_memory_region(kvm, mem, user_alloc);
 356 }
 357
 358 int kvm_get_dirty_log(struct kvm *kvm,
 359                         struct kvm_dirty_log *log, int *is_dirty)
 360 {
 361         struct kvm_memory_slot *memslot;
 362         int r, i;
 363         int n;
 364         unsigned long any = 0;
 365
 366         r = -EINVAL;
 367         if (log->slot >= KVM_MEMORY_SLOTS)
 368                 goto out;
 369
 370         memslot = &kvm->memslots[log->slot];
 371         r = -ENOENT;
 372         if (!memslot->dirty_bitmap)
 373                 goto out;
 374
 375         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 376
 377         for (i = 0; !any && i < n/sizeof(long); ++i)
 378                 any = memslot->dirty_bitmap[i];
 379
 380         r = -EFAULT;
 381         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 382                 goto out;
 383
 384         if (any)
 385                 *is_dirty = 1;
 386
 387         r = 0;
 388 out:
 389         return r;
 390 }
 391
 392 int is_error_page(struct page *page)
 393 {
 394         return page == bad_page;
 395 }
 396 EXPORT_SYMBOL_GPL(is_error_page);
 397
 398 static inline unsigned long bad_hva(void)
 399 {
 400         return PAGE_OFFSET;
 401 }
 402
 403 int kvm_is_error_hva(unsigned long addr)
 404 {
 405         return addr == bad_hva();
 406 }
 407 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 408
 409 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 410 {
 411         int i;
 412
 413         for (i = 0; i < kvm->nmemslots; ++i) {
 414                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 415
 416                 if (gfn >= memslot->base_gfn
 417                     && gfn < memslot->base_gfn + memslot->npages)
 418                         return memslot;
 419         }
 420         return NULL;
 421 }
 422
 423 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 424 {
 425         gfn = unalias_gfn(kvm, gfn);
 426         return __gfn_to_memslot(kvm, gfn);
 427 }
 428
 429 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 430 {
 431         int i;
 432
 433         gfn = unalias_gfn(kvm, gfn);
 434         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 435                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 436
 437                 if (gfn >= memslot->base_gfn
 438                     && gfn < memslot->base_gfn + memslot->npages)
 439                         return 1;
 440         }
 441         return 0;
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 444
 445 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 446 {
 447         struct kvm_memory_slot *slot;
 448
 449         gfn = unalias_gfn(kvm, gfn);
 450         slot = __gfn_to_memslot(kvm, gfn);
 451         if (!slot)
 452                 return bad_hva();
 453         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 454 }
 455
 456 /*
 457  * Requires current->mm->mmap_sem to be held
 458  */
 459 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 460 {
 461         struct page *page[1];
 462         unsigned long addr;
 463         int npages;
 464
 465         might_sleep();
 466
 467         addr = gfn_to_hva(kvm, gfn);
 468         if (kvm_is_error_hva(addr)) {
 469                 get_page(bad_page);
 470                 return bad_page;
 471         }
 472
 473         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 474                                 NULL);
 475
 476         if (npages != 1) {
 477                 get_page(bad_page);
 478                 return bad_page;
 479         }
 480
 481         return page[0];
 482 }
 483
 484 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 485 {
 486         struct page *page;
 487
 488         down_read(&current->mm->mmap_sem);
 489         page = __gfn_to_page(kvm, gfn);
 490         up_read(&current->mm->mmap_sem);
 491
 492         return page;
 493 }
 494
 495 EXPORT_SYMBOL_GPL(gfn_to_page);
 496
 497 void kvm_release_page_clean(struct page *page)
 498 {
 499         put_page(page);
 500 }
 501 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 502
 503 void kvm_release_page_dirty(struct page *page)
 504 {
 505         if (!PageReserved(page))
 506                 SetPageDirty(page);
 507         put_page(page);
 508 }
 509 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 510
 511 static int next_segment(unsigned long len, int offset)
 512 {
 513         if (len > PAGE_SIZE - offset)
 514                 return PAGE_SIZE - offset;
 515         else
 516                 return len;
 517 }
 518
 519 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 520                         int len)
 521 {
 522         int r;
 523         unsigned long addr;
 524
 525         addr = gfn_to_hva(kvm, gfn);
 526         if (kvm_is_error_hva(addr))
 527                 return -EFAULT;
 528         r = copy_from_user(data, (void __user *)addr + offset, len);
 529         if (r)
 530                 return -EFAULT;
 531         return 0;
 532 }
 533 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 534
 535 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 536 {
 537         gfn_t gfn = gpa >> PAGE_SHIFT;
 538         int seg;
 539         int offset = offset_in_page(gpa);
 540         int ret;
 541
 542         while ((seg = next_segment(len, offset)) != 0) {
 543                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 544                 if (ret < 0)
 545                         return ret;
 546                 offset = 0;
 547                 len -= seg;
 548                 data += seg;
 549                 ++gfn;
 550         }
 551         return 0;
 552 }
 553 EXPORT_SYMBOL_GPL(kvm_read_guest);
 554
 555 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 556                          int offset, int len)
 557 {
 558         int r;
 559         unsigned long addr;
 560
 561         addr = gfn_to_hva(kvm, gfn);
 562         if (kvm_is_error_hva(addr))
 563                 return -EFAULT;
 564         r = copy_to_user((void __user *)addr + offset, data, len);
 565         if (r)
 566                 return -EFAULT;
 567         mark_page_dirty(kvm, gfn);
 568         return 0;
 569 }
 570 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 571
 572 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 573                     unsigned long len)
 574 {
 575         gfn_t gfn = gpa >> PAGE_SHIFT;
 576         int seg;
 577         int offset = offset_in_page(gpa);
 578         int ret;
 579
 580         while ((seg = next_segment(len, offset)) != 0) {
 581                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 582                 if (ret < 0)
 583                         return ret;
 584                 offset = 0;
 585                 len -= seg;
 586                 data += seg;
 587                 ++gfn;
 588         }
 589         return 0;
 590 }
 591
 592 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 593 {
 594         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 595 }
 596 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 597
 598 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 599 {
 600         gfn_t gfn = gpa >> PAGE_SHIFT;
 601         int seg;
 602         int offset = offset_in_page(gpa);
 603         int ret;
 604
 605         while ((seg = next_segment(len, offset)) != 0) {
 606                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 607                 if (ret < 0)
 608                         return ret;
 609                 offset = 0;
 610                 len -= seg;
 611                 ++gfn;
 612         }
 613         return 0;
 614 }
 615 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 616
 617 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 618 {
 619         struct kvm_memory_slot *memslot;
 620
 621         gfn = unalias_gfn(kvm, gfn);
 622         memslot = __gfn_to_memslot(kvm, gfn);
 623         if (memslot && memslot->dirty_bitmap) {
 624                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 625
 626                 /* avoid RMW */
 627                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 628                         set_bit(rel_gfn, memslot->dirty_bitmap);
 629         }
 630 }
 631
 632 /*
 633  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 634  */
 635 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 636 {
 637         DECLARE_WAITQUEUE(wait, current);
 638
 639         add_wait_queue(&vcpu->wq, &wait);
 640
 641         /*
 642          * We will block until either an interrupt or a signal wakes us up
 643          */
 644         while (!kvm_cpu_has_interrupt(vcpu)
 645                && !signal_pending(current)
 646                && !kvm_arch_vcpu_runnable(vcpu)) {
 647                 set_current_state(TASK_INTERRUPTIBLE);
 648                 vcpu_put(vcpu);
 649                 schedule();
 650                 vcpu_load(vcpu);
 651         }
 652
 653         __set_current_state(TASK_RUNNING);
 654         remove_wait_queue(&vcpu->wq, &wait);
 655 }
 656
 657 void kvm_resched(struct kvm_vcpu *vcpu)
 658 {
 659         if (!need_resched())
 660                 return;
 661         cond_resched();
 662 }
 663 EXPORT_SYMBOL_GPL(kvm_resched);
 664
 665 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 666 {
 667         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 668         struct page *page;
 669
 670         if (vmf->pgoff == 0)
 671                 page = virt_to_page(vcpu->run);
 672         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 673                 page = virt_to_page(vcpu->arch.pio_data);
 674         else
 675                 return VM_FAULT_SIGBUS;
 676         get_page(page);
 677         vmf->page = page;
 678         return 0;
 679 }
 680
 681 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 682         .fault = kvm_vcpu_fault,
 683 };
 684
 685 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 686 {
 687         vma->vm_ops = &kvm_vcpu_vm_ops;
 688         return 0;
 689 }
 690
 691 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 692 {
 693         struct kvm_vcpu *vcpu = filp->private_data;
 694
 695         fput(vcpu->kvm->filp);
 696         return 0;
 697 }
 698
 699 static struct file_operations kvm_vcpu_fops = {
 700         .release        = kvm_vcpu_release,
 701         .unlocked_ioctl = kvm_vcpu_ioctl,
 702         .compat_ioctl   = kvm_vcpu_ioctl,
 703         .mmap           = kvm_vcpu_mmap,
 704 };
 705
 706 /*
 707  * Allocates an inode for the vcpu.
 708  */
 709 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 710 {
 711         int fd, r;
 712         struct inode *inode;
 713         struct file *file;
 714
 715         r = anon_inode_getfd(&fd, &inode, &file,
 716                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 717         if (r)
 718                 return r;
 719         atomic_inc(&vcpu->kvm->filp->f_count);
 720         return fd;
 721 }
 722
 723 /*
 724  * Creates some virtual cpus.  Good luck creating more than one.
 725  */
 726 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 727 {
 728         int r;
 729         struct kvm_vcpu *vcpu;
 730
 731         if (!valid_vcpu(n))
 732                 return -EINVAL;
 733
 734         vcpu = kvm_arch_vcpu_create(kvm, n);
 735         if (IS_ERR(vcpu))
 736                 return PTR_ERR(vcpu);
 737
 738         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 739
 740         r = kvm_arch_vcpu_setup(vcpu);
 741         if (r)
 742                 goto vcpu_destroy;
 743
 744         mutex_lock(&kvm->lock);
 745         if (kvm->vcpus[n]) {
 746                 r = -EEXIST;
 747                 mutex_unlock(&kvm->lock);
 748                 goto vcpu_destroy;
 749         }
 750         kvm->vcpus[n] = vcpu;
 751         mutex_unlock(&kvm->lock);
 752
 753         /* Now it's all set up, let userspace reach it */
 754         r = create_vcpu_fd(vcpu);
 755         if (r < 0)
 756                 goto unlink;
 757         return r;
 758
 759 unlink:
 760         mutex_lock(&kvm->lock);
 761         kvm->vcpus[n] = NULL;
 762         mutex_unlock(&kvm->lock);
 763 vcpu_destroy:
 764         kvm_arch_vcpu_destroy(vcpu);
 765         return r;
 766 }
 767
 768 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 769 {
 770         if (sigset) {
 771                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 772                 vcpu->sigset_active = 1;
 773                 vcpu->sigset = *sigset;
 774         } else
 775                 vcpu->sigset_active = 0;
 776         return 0;
 777 }
 778
 779 static long kvm_vcpu_ioctl(struct file *filp,
 780                            unsigned int ioctl, unsigned long arg)
 781 {
 782         struct kvm_vcpu *vcpu = filp->private_data;
 783         void __user *argp = (void __user *)arg;
 784         int r;
 785
 786         if (vcpu->kvm->mm != current->mm)
 787                 return -EIO;
 788         switch (ioctl) {
 789         case KVM_RUN:
 790                 r = -EINVAL;
 791                 if (arg)
 792                         goto out;
 793                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 794                 break;
 795         case KVM_GET_REGS: {
 796                 struct kvm_regs kvm_regs;
 797
 798                 memset(&kvm_regs, 0, sizeof kvm_regs);
 799                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 800                 if (r)
 801                         goto out;
 802                 r = -EFAULT;
 803                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 804                         goto out;
 805                 r = 0;
 806                 break;
 807         }
 808         case KVM_SET_REGS: {
 809                 struct kvm_regs kvm_regs;
 810
 811                 r = -EFAULT;
 812                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 813                         goto out;
 814                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 815                 if (r)
 816                         goto out;
 817                 r = 0;
 818                 break;
 819         }
 820         case KVM_GET_SREGS: {
 821                 struct kvm_sregs kvm_sregs;
 822
 823                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 824                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 825                 if (r)
 826                         goto out;
 827                 r = -EFAULT;
 828                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 829                         goto out;
 830                 r = 0;
 831                 break;
 832         }
 833         case KVM_SET_SREGS: {
 834                 struct kvm_sregs kvm_sregs;
 835
 836                 r = -EFAULT;
 837                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 838                         goto out;
 839                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 840                 if (r)
 841                         goto out;
 842                 r = 0;
 843                 break;
 844         }
 845         case KVM_TRANSLATE: {
 846                 struct kvm_translation tr;
 847
 848                 r = -EFAULT;
 849                 if (copy_from_user(&tr, argp, sizeof tr))
 850                         goto out;
 851                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 852                 if (r)
 853                         goto out;
 854                 r = -EFAULT;
 855                 if (copy_to_user(argp, &tr, sizeof tr))
 856                         goto out;
 857                 r = 0;
 858                 break;
 859         }
 860         case KVM_DEBUG_GUEST: {
 861                 struct kvm_debug_guest dbg;
 862
 863                 r = -EFAULT;
 864                 if (copy_from_user(&dbg, argp, sizeof dbg))
 865                         goto out;
 866                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 867                 if (r)
 868                         goto out;
 869                 r = 0;
 870                 break;
 871         }
 872         case KVM_SET_SIGNAL_MASK: {
 873                 struct kvm_signal_mask __user *sigmask_arg = argp;
 874                 struct kvm_signal_mask kvm_sigmask;
 875                 sigset_t sigset, *p;
 876
 877                 p = NULL;
 878                 if (argp) {
 879                         r = -EFAULT;
 880                         if (copy_from_user(&kvm_sigmask, argp,
 881                                            sizeof kvm_sigmask))
 882                                 goto out;
 883                         r = -EINVAL;
 884                         if (kvm_sigmask.len != sizeof sigset)
 885                                 goto out;
 886                         r = -EFAULT;
 887                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 888                                            sizeof sigset))
 889                                 goto out;
 890                         p = &sigset;
 891                 }
 892                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 893                 break;
 894         }
 895         case KVM_GET_FPU: {
 896                 struct kvm_fpu fpu;
 897
 898                 memset(&fpu, 0, sizeof fpu);
 899                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 900                 if (r)
 901                         goto out;
 902                 r = -EFAULT;
 903                 if (copy_to_user(argp, &fpu, sizeof fpu))
 904                         goto out;
 905                 r = 0;
 906                 break;
 907         }
 908         case KVM_SET_FPU: {
 909                 struct kvm_fpu fpu;
 910
 911                 r = -EFAULT;
 912                 if (copy_from_user(&fpu, argp, sizeof fpu))
 913                         goto out;
 914                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 915                 if (r)
 916                         goto out;
 917                 r = 0;
 918                 break;
 919         }
 920         default:
 921                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 922         }
 923 out:
 924         return r;
 925 }
 926
 927 static long kvm_vm_ioctl(struct file *filp,
 928                            unsigned int ioctl, unsigned long arg)
 929 {
 930         struct kvm *kvm = filp->private_data;
 931         void __user *argp = (void __user *)arg;
 932         int r;
 933
 934         if (kvm->mm != current->mm)
 935                 return -EIO;
 936         switch (ioctl) {
 937         case KVM_CREATE_VCPU:
 938                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 939                 if (r < 0)
 940                         goto out;
 941                 break;
 942         case KVM_SET_USER_MEMORY_REGION: {
 943                 struct kvm_userspace_memory_region kvm_userspace_mem;
 944
 945                 r = -EFAULT;
 946                 if (copy_from_user(&kvm_userspace_mem, argp,
 947                                                 sizeof kvm_userspace_mem))
 948                         goto out;
 949
 950                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
 951                 if (r)
 952                         goto out;
 953                 break;
 954         }
 955         case KVM_GET_DIRTY_LOG: {
 956                 struct kvm_dirty_log log;
 957
 958                 r = -EFAULT;
 959                 if (copy_from_user(&log, argp, sizeof log))
 960                         goto out;
 961                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
 962                 if (r)
 963                         goto out;
 964                 break;
 965         }
 966         default:
 967                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 968         }
 969 out:
 970         return r;
 971 }
 972
 973 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 974 {
 975         struct kvm *kvm = vma->vm_file->private_data;
 976         struct page *page;
 977
 978         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
 979                 return VM_FAULT_SIGBUS;
 980         /* current->mm->mmap_sem is already held so call lockless version */
 981         page = __gfn_to_page(kvm, vmf->pgoff);
 982         if (is_error_page(page)) {
 983                 kvm_release_page_clean(page);
 984                 return VM_FAULT_SIGBUS;
 985         }
 986         vmf->page = page;
 987         return 0;
 988 }
 989
 990 static struct vm_operations_struct kvm_vm_vm_ops = {
 991         .fault = kvm_vm_fault,
 992 };
 993
 994 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
 995 {
 996         vma->vm_ops = &kvm_vm_vm_ops;
 997         return 0;
 998 }
 999
1000 static struct file_operations kvm_vm_fops = {
1001         .release        = kvm_vm_release,
1002         .unlocked_ioctl = kvm_vm_ioctl,
1003         .compat_ioctl   = kvm_vm_ioctl,
1004         .mmap           = kvm_vm_mmap,
1005 };
1006
1007 static int kvm_dev_ioctl_create_vm(void)
1008 {
1009         int fd, r;
1010         struct inode *inode;
1011         struct file *file;
1012         struct kvm *kvm;
1013
1014         kvm = kvm_create_vm();
1015         if (IS_ERR(kvm))
1016                 return PTR_ERR(kvm);
1017         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1018         if (r) {
1019                 kvm_destroy_vm(kvm);
1020                 return r;
1021         }
1022
1023         kvm->filp = file;
1024
1025         return fd;
1026 }
1027
1028 static long kvm_dev_ioctl(struct file *filp,
1029                           unsigned int ioctl, unsigned long arg)
1030 {
1031         void __user *argp = (void __user *)arg;
1032         long r = -EINVAL;
1033
1034         switch (ioctl) {
1035         case KVM_GET_API_VERSION:
1036                 r = -EINVAL;
1037                 if (arg)
1038                         goto out;
1039                 r = KVM_API_VERSION;
1040                 break;
1041         case KVM_CREATE_VM:
1042                 r = -EINVAL;
1043                 if (arg)
1044                         goto out;
1045                 r = kvm_dev_ioctl_create_vm();
1046                 break;
1047         case KVM_CHECK_EXTENSION:
1048                 r = kvm_dev_ioctl_check_extension((long)argp);
1049                 break;
1050         case KVM_GET_VCPU_MMAP_SIZE:
1051                 r = -EINVAL;
1052                 if (arg)
1053                         goto out;
1054                 r = 2 * PAGE_SIZE;
1055                 break;
1056         default:
1057                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1058         }
1059 out:
1060         return r;
1061 }
1062
1063 static struct file_operations kvm_chardev_ops = {
1064         .unlocked_ioctl = kvm_dev_ioctl,
1065         .compat_ioctl   = kvm_dev_ioctl,
1066 };
1067
1068 static struct miscdevice kvm_dev = {
1069         KVM_MINOR,
1070         "kvm",
1071         &kvm_chardev_ops,
1072 };
1073
1074 static void hardware_enable(void *junk)
1075 {
1076         int cpu = raw_smp_processor_id();
1077
1078         if (cpu_isset(cpu, cpus_hardware_enabled))
1079                 return;
1080         cpu_set(cpu, cpus_hardware_enabled);
1081         kvm_arch_hardware_enable(NULL);
1082 }
1083
1084 static void hardware_disable(void *junk)
1085 {
1086         int cpu = raw_smp_processor_id();
1087
1088         if (!cpu_isset(cpu, cpus_hardware_enabled))
1089                 return;
1090         cpu_clear(cpu, cpus_hardware_enabled);
1091         decache_vcpus_on_cpu(cpu);
1092         kvm_arch_hardware_disable(NULL);
1093 }
1094
1095 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1096                            void *v)
1097 {
1098         int cpu = (long)v;
1099
1100         val &= ~CPU_TASKS_FROZEN;
1101         switch (val) {
1102         case CPU_DYING:
1103                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1104                        cpu);
1105                 hardware_disable(NULL);
1106                 break;
1107         case CPU_UP_CANCELED:
1108                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1109                        cpu);
1110                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1111                 break;
1112         case CPU_ONLINE:
1113                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1114                        cpu);
1115                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1116                 break;
1117         }
1118         return NOTIFY_OK;
1119 }
1120
1121 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1122                       void *v)
1123 {
1124         if (val == SYS_RESTART) {
1125                 /*
1126                  * Some (well, at least mine) BIOSes hang on reboot if
1127                  * in vmx root mode.
1128                  */
1129                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1130                 on_each_cpu(hardware_disable, NULL, 0, 1);
1131         }
1132         return NOTIFY_OK;
1133 }
1134
1135 static struct notifier_block kvm_reboot_notifier = {
1136         .notifier_call = kvm_reboot,
1137         .priority = 0,
1138 };
1139
1140 void kvm_io_bus_init(struct kvm_io_bus *bus)
1141 {
1142         memset(bus, 0, sizeof(*bus));
1143 }
1144
1145 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1146 {
1147         int i;
1148
1149         for (i = 0; i < bus->dev_count; i++) {
1150                 struct kvm_io_device *pos = bus->devs[i];
1151
1152                 kvm_iodevice_destructor(pos);
1153         }
1154 }
1155
1156 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1157 {
1158         int i;
1159
1160         for (i = 0; i < bus->dev_count; i++) {
1161                 struct kvm_io_device *pos = bus->devs[i];
1162
1163                 if (pos->in_range(pos, addr))
1164                         return pos;
1165         }
1166
1167         return NULL;
1168 }
1169
1170 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1171 {
1172         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1173
1174         bus->devs[bus->dev_count++] = dev;
1175 }
1176
1177 static struct notifier_block kvm_cpu_notifier = {
1178         .notifier_call = kvm_cpu_hotplug,
1179         .priority = 20, /* must be > scheduler priority */
1180 };
1181
1182 static u64 vm_stat_get(void *_offset)
1183 {
1184         unsigned offset = (long)_offset;
1185         u64 total = 0;
1186         struct kvm *kvm;
1187
1188         spin_lock(&kvm_lock);
1189         list_for_each_entry(kvm, &vm_list, vm_list)
1190                 total += *(u32 *)((void *)kvm + offset);
1191         spin_unlock(&kvm_lock);
1192         return total;
1193 }
1194
1195 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1196
1197 static u64 vcpu_stat_get(void *_offset)
1198 {
1199         unsigned offset = (long)_offset;
1200         u64 total = 0;
1201         struct kvm *kvm;
1202         struct kvm_vcpu *vcpu;
1203         int i;
1204
1205         spin_lock(&kvm_lock);
1206         list_for_each_entry(kvm, &vm_list, vm_list)
1207                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1208                         vcpu = kvm->vcpus[i];
1209                         if (vcpu)
1210                                 total += *(u32 *)((void *)vcpu + offset);
1211                 }
1212         spin_unlock(&kvm_lock);
1213         return total;
1214 }
1215
1216 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1217
1218 static struct file_operations *stat_fops[] = {
1219         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1220         [KVM_STAT_VM]   = &vm_stat_fops,
1221 };
1222
1223 static void kvm_init_debug(void)
1224 {
1225         struct kvm_stats_debugfs_item *p;
1226
1227         debugfs_dir = debugfs_create_dir("kvm", NULL);
1228         for (p = debugfs_entries; p->name; ++p)
1229                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1230                                                 (void *)(long)p->offset,
1231                                                 stat_fops[p->kind]);
1232 }
1233
1234 static void kvm_exit_debug(void)
1235 {
1236         struct kvm_stats_debugfs_item *p;
1237
1238         for (p = debugfs_entries; p->name; ++p)
1239                 debugfs_remove(p->dentry);
1240         debugfs_remove(debugfs_dir);
1241 }
1242
1243 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1244 {
1245         hardware_disable(NULL);
1246         return 0;
1247 }
1248
1249 static int kvm_resume(struct sys_device *dev)
1250 {
1251         hardware_enable(NULL);
1252         return 0;
1253 }
1254
1255 static struct sysdev_class kvm_sysdev_class = {
1256         .name = "kvm",
1257         .suspend = kvm_suspend,
1258         .resume = kvm_resume,
1259 };
1260
1261 static struct sys_device kvm_sysdev = {
1262         .id = 0,
1263         .cls = &kvm_sysdev_class,
1264 };
1265
1266 struct page *bad_page;
1267
1268 static inline
1269 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1270 {
1271         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1272 }
1273
1274 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1275 {
1276         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1277
1278         kvm_arch_vcpu_load(vcpu, cpu);
1279 }
1280
1281 static void kvm_sched_out(struct preempt_notifier *pn,
1282                           struct task_struct *next)
1283 {
1284         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1285
1286         kvm_arch_vcpu_put(vcpu);
1287 }
1288
1289 int kvm_init(void *opaque, unsigned int vcpu_size,
1290                   struct module *module)
1291 {
1292         int r;
1293         int cpu;
1294
1295         kvm_init_debug();
1296
1297         r = kvm_arch_init(opaque);
1298         if (r)
1299                 goto out_fail;
1300
1301         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1302
1303         if (bad_page == NULL) {
1304                 r = -ENOMEM;
1305                 goto out;
1306         }
1307
1308         r = kvm_arch_hardware_setup();
1309         if (r < 0)
1310                 goto out_free_0;
1311
1312         for_each_online_cpu(cpu) {
1313                 smp_call_function_single(cpu,
1314                                 kvm_arch_check_processor_compat,
1315                                 &r, 0, 1);
1316                 if (r < 0)
1317                         goto out_free_1;
1318         }
1319
1320         on_each_cpu(hardware_enable, NULL, 0, 1);
1321         r = register_cpu_notifier(&kvm_cpu_notifier);
1322         if (r)
1323                 goto out_free_2;
1324         register_reboot_notifier(&kvm_reboot_notifier);
1325
1326         r = sysdev_class_register(&kvm_sysdev_class);
1327         if (r)
1328                 goto out_free_3;
1329
1330         r = sysdev_register(&kvm_sysdev);
1331         if (r)
1332                 goto out_free_4;
1333
1334         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1335         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1336                                            __alignof__(struct kvm_vcpu),
1337                                            0, NULL);
1338         if (!kvm_vcpu_cache) {
1339                 r = -ENOMEM;
1340                 goto out_free_5;
1341         }
1342
1343         kvm_chardev_ops.owner = module;
1344
1345         r = misc_register(&kvm_dev);
1346         if (r) {
1347                 printk(KERN_ERR "kvm: misc device register failed\n");
1348                 goto out_free;
1349         }
1350
1351         kvm_preempt_ops.sched_in = kvm_sched_in;
1352         kvm_preempt_ops.sched_out = kvm_sched_out;
1353
1354         return 0;
1355
1356 out_free:
1357         kmem_cache_destroy(kvm_vcpu_cache);
1358 out_free_5:
1359         sysdev_unregister(&kvm_sysdev);
1360 out_free_4:
1361         sysdev_class_unregister(&kvm_sysdev_class);
1362 out_free_3:
1363         unregister_reboot_notifier(&kvm_reboot_notifier);
1364         unregister_cpu_notifier(&kvm_cpu_notifier);
1365 out_free_2:
1366         on_each_cpu(hardware_disable, NULL, 0, 1);
1367 out_free_1:
1368         kvm_arch_hardware_unsetup();
1369 out_free_0:
1370         __free_page(bad_page);
1371 out:
1372         kvm_arch_exit();
1373         kvm_exit_debug();
1374 out_fail:
1375         return r;
1376 }
1377 EXPORT_SYMBOL_GPL(kvm_init);
1378
1379 void kvm_exit(void)
1380 {
1381         misc_deregister(&kvm_dev);
1382         kmem_cache_destroy(kvm_vcpu_cache);
1383         sysdev_unregister(&kvm_sysdev);
1384         sysdev_class_unregister(&kvm_sysdev_class);
1385         unregister_reboot_notifier(&kvm_reboot_notifier);
1386         unregister_cpu_notifier(&kvm_cpu_notifier);
1387         on_each_cpu(hardware_disable, NULL, 0, 1);
1388         kvm_arch_hardware_unsetup();
1389         kvm_arch_exit();
1390         kvm_exit_debug();
1391         __free_page(bad_page);
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_exit);