SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 mmap_sem for write.
 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         down_write(&current->mm->mmap_sem);
 342         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 343         up_write(&current->mm->mmap_sem);
 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 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 EXPORT_SYMBOL_GPL(gfn_to_page);
 485
 486 void kvm_release_page_clean(struct page *page)
 487 {
 488         put_page(page);
 489 }
 490 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 491
 492 void kvm_release_page_dirty(struct page *page)
 493 {
 494         if (!PageReserved(page))
 495                 SetPageDirty(page);
 496         put_page(page);
 497 }
 498 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 499
 500 static int next_segment(unsigned long len, int offset)
 501 {
 502         if (len > PAGE_SIZE - offset)
 503                 return PAGE_SIZE - offset;
 504         else
 505                 return len;
 506 }
 507
 508 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 509                         int len)
 510 {
 511         int r;
 512         unsigned long addr;
 513
 514         addr = gfn_to_hva(kvm, gfn);
 515         if (kvm_is_error_hva(addr))
 516                 return -EFAULT;
 517         r = copy_from_user(data, (void __user *)addr + offset, len);
 518         if (r)
 519                 return -EFAULT;
 520         return 0;
 521 }
 522 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 523
 524 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 525 {
 526         gfn_t gfn = gpa >> PAGE_SHIFT;
 527         int seg;
 528         int offset = offset_in_page(gpa);
 529         int ret;
 530
 531         while ((seg = next_segment(len, offset)) != 0) {
 532                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 533                 if (ret < 0)
 534                         return ret;
 535                 offset = 0;
 536                 len -= seg;
 537                 data += seg;
 538                 ++gfn;
 539         }
 540         return 0;
 541 }
 542 EXPORT_SYMBOL_GPL(kvm_read_guest);
 543
 544 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 545                          int offset, int len)
 546 {
 547         int r;
 548         unsigned long addr;
 549
 550         addr = gfn_to_hva(kvm, gfn);
 551         if (kvm_is_error_hva(addr))
 552                 return -EFAULT;
 553         r = copy_to_user((void __user *)addr + offset, data, len);
 554         if (r)
 555                 return -EFAULT;
 556         mark_page_dirty(kvm, gfn);
 557         return 0;
 558 }
 559 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 560
 561 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 562                     unsigned long len)
 563 {
 564         gfn_t gfn = gpa >> PAGE_SHIFT;
 565         int seg;
 566         int offset = offset_in_page(gpa);
 567         int ret;
 568
 569         while ((seg = next_segment(len, offset)) != 0) {
 570                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 571                 if (ret < 0)
 572                         return ret;
 573                 offset = 0;
 574                 len -= seg;
 575                 data += seg;
 576                 ++gfn;
 577         }
 578         return 0;
 579 }
 580
 581 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 582 {
 583         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 584 }
 585 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 586
 587 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 588 {
 589         gfn_t gfn = gpa >> PAGE_SHIFT;
 590         int seg;
 591         int offset = offset_in_page(gpa);
 592         int ret;
 593
 594         while ((seg = next_segment(len, offset)) != 0) {
 595                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 596                 if (ret < 0)
 597                         return ret;
 598                 offset = 0;
 599                 len -= seg;
 600                 ++gfn;
 601         }
 602         return 0;
 603 }
 604 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 605
 606 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 607 {
 608         struct kvm_memory_slot *memslot;
 609
 610         gfn = unalias_gfn(kvm, gfn);
 611         memslot = __gfn_to_memslot(kvm, gfn);
 612         if (memslot && memslot->dirty_bitmap) {
 613                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 614
 615                 /* avoid RMW */
 616                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 617                         set_bit(rel_gfn, memslot->dirty_bitmap);
 618         }
 619 }
 620
 621 /*
 622  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 623  */
 624 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 625 {
 626         DECLARE_WAITQUEUE(wait, current);
 627
 628         add_wait_queue(&vcpu->wq, &wait);
 629
 630         /*
 631          * We will block until either an interrupt or a signal wakes us up
 632          */
 633         while (!kvm_cpu_has_interrupt(vcpu)
 634                && !signal_pending(current)
 635                && !kvm_arch_vcpu_runnable(vcpu)) {
 636                 set_current_state(TASK_INTERRUPTIBLE);
 637                 vcpu_put(vcpu);
 638                 schedule();
 639                 vcpu_load(vcpu);
 640         }
 641
 642         __set_current_state(TASK_RUNNING);
 643         remove_wait_queue(&vcpu->wq, &wait);
 644 }
 645
 646 void kvm_resched(struct kvm_vcpu *vcpu)
 647 {
 648         if (!need_resched())
 649                 return;
 650         cond_resched();
 651 }
 652 EXPORT_SYMBOL_GPL(kvm_resched);
 653
 654 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 655 {
 656         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 657         struct page *page;
 658
 659         if (vmf->pgoff == 0)
 660                 page = virt_to_page(vcpu->run);
 661         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 662                 page = virt_to_page(vcpu->arch.pio_data);
 663         else
 664                 return VM_FAULT_SIGBUS;
 665         get_page(page);
 666         vmf->page = page;
 667         return 0;
 668 }
 669
 670 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 671         .fault = kvm_vcpu_fault,
 672 };
 673
 674 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 675 {
 676         vma->vm_ops = &kvm_vcpu_vm_ops;
 677         return 0;
 678 }
 679
 680 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 681 {
 682         struct kvm_vcpu *vcpu = filp->private_data;
 683
 684         fput(vcpu->kvm->filp);
 685         return 0;
 686 }
 687
 688 static struct file_operations kvm_vcpu_fops = {
 689         .release        = kvm_vcpu_release,
 690         .unlocked_ioctl = kvm_vcpu_ioctl,
 691         .compat_ioctl   = kvm_vcpu_ioctl,
 692         .mmap           = kvm_vcpu_mmap,
 693 };
 694
 695 /*
 696  * Allocates an inode for the vcpu.
 697  */
 698 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 699 {
 700         int fd, r;
 701         struct inode *inode;
 702         struct file *file;
 703
 704         r = anon_inode_getfd(&fd, &inode, &file,
 705                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 706         if (r)
 707                 return r;
 708         atomic_inc(&vcpu->kvm->filp->f_count);
 709         return fd;
 710 }
 711
 712 /*
 713  * Creates some virtual cpus.  Good luck creating more than one.
 714  */
 715 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 716 {
 717         int r;
 718         struct kvm_vcpu *vcpu;
 719
 720         if (!valid_vcpu(n))
 721                 return -EINVAL;
 722
 723         vcpu = kvm_arch_vcpu_create(kvm, n);
 724         if (IS_ERR(vcpu))
 725                 return PTR_ERR(vcpu);
 726
 727         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 728
 729         r = kvm_arch_vcpu_setup(vcpu);
 730         if (r)
 731                 goto vcpu_destroy;
 732
 733         mutex_lock(&kvm->lock);
 734         if (kvm->vcpus[n]) {
 735                 r = -EEXIST;
 736                 mutex_unlock(&kvm->lock);
 737                 goto vcpu_destroy;
 738         }
 739         kvm->vcpus[n] = vcpu;
 740         mutex_unlock(&kvm->lock);
 741
 742         /* Now it's all set up, let userspace reach it */
 743         r = create_vcpu_fd(vcpu);
 744         if (r < 0)
 745                 goto unlink;
 746         return r;
 747
 748 unlink:
 749         mutex_lock(&kvm->lock);
 750         kvm->vcpus[n] = NULL;
 751         mutex_unlock(&kvm->lock);
 752 vcpu_destroy:
 753         kvm_arch_vcpu_destroy(vcpu);
 754         return r;
 755 }
 756
 757 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 758 {
 759         if (sigset) {
 760                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 761                 vcpu->sigset_active = 1;
 762                 vcpu->sigset = *sigset;
 763         } else
 764                 vcpu->sigset_active = 0;
 765         return 0;
 766 }
 767
 768 static long kvm_vcpu_ioctl(struct file *filp,
 769                            unsigned int ioctl, unsigned long arg)
 770 {
 771         struct kvm_vcpu *vcpu = filp->private_data;
 772         void __user *argp = (void __user *)arg;
 773         int r;
 774
 775         if (vcpu->kvm->mm != current->mm)
 776                 return -EIO;
 777         switch (ioctl) {
 778         case KVM_RUN:
 779                 r = -EINVAL;
 780                 if (arg)
 781                         goto out;
 782                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 783                 break;
 784         case KVM_GET_REGS: {
 785                 struct kvm_regs kvm_regs;
 786
 787                 memset(&kvm_regs, 0, sizeof kvm_regs);
 788                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 789                 if (r)
 790                         goto out;
 791                 r = -EFAULT;
 792                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 793                         goto out;
 794                 r = 0;
 795                 break;
 796         }
 797         case KVM_SET_REGS: {
 798                 struct kvm_regs kvm_regs;
 799
 800                 r = -EFAULT;
 801                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 802                         goto out;
 803                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 804                 if (r)
 805                         goto out;
 806                 r = 0;
 807                 break;
 808         }
 809         case KVM_GET_SREGS: {
 810                 struct kvm_sregs kvm_sregs;
 811
 812                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 813                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 814                 if (r)
 815                         goto out;
 816                 r = -EFAULT;
 817                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 818                         goto out;
 819                 r = 0;
 820                 break;
 821         }
 822         case KVM_SET_SREGS: {
 823                 struct kvm_sregs kvm_sregs;
 824
 825                 r = -EFAULT;
 826                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 827                         goto out;
 828                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 829                 if (r)
 830                         goto out;
 831                 r = 0;
 832                 break;
 833         }
 834         case KVM_TRANSLATE: {
 835                 struct kvm_translation tr;
 836
 837                 r = -EFAULT;
 838                 if (copy_from_user(&tr, argp, sizeof tr))
 839                         goto out;
 840                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 841                 if (r)
 842                         goto out;
 843                 r = -EFAULT;
 844                 if (copy_to_user(argp, &tr, sizeof tr))
 845                         goto out;
 846                 r = 0;
 847                 break;
 848         }
 849         case KVM_DEBUG_GUEST: {
 850                 struct kvm_debug_guest dbg;
 851
 852                 r = -EFAULT;
 853                 if (copy_from_user(&dbg, argp, sizeof dbg))
 854                         goto out;
 855                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 856                 if (r)
 857                         goto out;
 858                 r = 0;
 859                 break;
 860         }
 861         case KVM_SET_SIGNAL_MASK: {
 862                 struct kvm_signal_mask __user *sigmask_arg = argp;
 863                 struct kvm_signal_mask kvm_sigmask;
 864                 sigset_t sigset, *p;
 865
 866                 p = NULL;
 867                 if (argp) {
 868                         r = -EFAULT;
 869                         if (copy_from_user(&kvm_sigmask, argp,
 870                                            sizeof kvm_sigmask))
 871                                 goto out;
 872                         r = -EINVAL;
 873                         if (kvm_sigmask.len != sizeof sigset)
 874                                 goto out;
 875                         r = -EFAULT;
 876                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 877                                            sizeof sigset))
 878                                 goto out;
 879                         p = &sigset;
 880                 }
 881                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 882                 break;
 883         }
 884         case KVM_GET_FPU: {
 885                 struct kvm_fpu fpu;
 886
 887                 memset(&fpu, 0, sizeof fpu);
 888                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 889                 if (r)
 890                         goto out;
 891                 r = -EFAULT;
 892                 if (copy_to_user(argp, &fpu, sizeof fpu))
 893                         goto out;
 894                 r = 0;
 895                 break;
 896         }
 897         case KVM_SET_FPU: {
 898                 struct kvm_fpu fpu;
 899
 900                 r = -EFAULT;
 901                 if (copy_from_user(&fpu, argp, sizeof fpu))
 902                         goto out;
 903                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 904                 if (r)
 905                         goto out;
 906                 r = 0;
 907                 break;
 908         }
 909         default:
 910                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 911         }
 912 out:
 913         return r;
 914 }
 915
 916 static long kvm_vm_ioctl(struct file *filp,
 917                            unsigned int ioctl, unsigned long arg)
 918 {
 919         struct kvm *kvm = filp->private_data;
 920         void __user *argp = (void __user *)arg;
 921         int r;
 922
 923         if (kvm->mm != current->mm)
 924                 return -EIO;
 925         switch (ioctl) {
 926         case KVM_CREATE_VCPU:
 927                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 928                 if (r < 0)
 929                         goto out;
 930                 break;
 931         case KVM_SET_USER_MEMORY_REGION: {
 932                 struct kvm_userspace_memory_region kvm_userspace_mem;
 933
 934                 r = -EFAULT;
 935                 if (copy_from_user(&kvm_userspace_mem, argp,
 936                                                 sizeof kvm_userspace_mem))
 937                         goto out;
 938
 939                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
 940                 if (r)
 941                         goto out;
 942                 break;
 943         }
 944         case KVM_GET_DIRTY_LOG: {
 945                 struct kvm_dirty_log log;
 946
 947                 r = -EFAULT;
 948                 if (copy_from_user(&log, argp, sizeof log))
 949                         goto out;
 950                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
 951                 if (r)
 952                         goto out;
 953                 break;
 954         }
 955         default:
 956                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 957         }
 958 out:
 959         return r;
 960 }
 961
 962 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 963 {
 964         struct kvm *kvm = vma->vm_file->private_data;
 965         struct page *page;
 966
 967         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
 968                 return VM_FAULT_SIGBUS;
 969         page = gfn_to_page(kvm, vmf->pgoff);
 970         if (is_error_page(page)) {
 971                 kvm_release_page_clean(page);
 972                 return VM_FAULT_SIGBUS;
 973         }
 974         vmf->page = page;
 975         return 0;
 976 }
 977
 978 static struct vm_operations_struct kvm_vm_vm_ops = {
 979         .fault = kvm_vm_fault,
 980 };
 981
 982 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
 983 {
 984         vma->vm_ops = &kvm_vm_vm_ops;
 985         return 0;
 986 }
 987
 988 static struct file_operations kvm_vm_fops = {
 989         .release        = kvm_vm_release,
 990         .unlocked_ioctl = kvm_vm_ioctl,
 991         .compat_ioctl   = kvm_vm_ioctl,
 992         .mmap           = kvm_vm_mmap,
 993 };
 994
 995 static int kvm_dev_ioctl_create_vm(void)
 996 {
 997         int fd, r;
 998         struct inode *inode;
 999         struct file *file;
1000         struct kvm *kvm;
1001
1002         kvm = kvm_create_vm();
1003         if (IS_ERR(kvm))
1004                 return PTR_ERR(kvm);
1005         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1006         if (r) {
1007                 kvm_destroy_vm(kvm);
1008                 return r;
1009         }
1010
1011         kvm->filp = file;
1012
1013         return fd;
1014 }
1015
1016 static long kvm_dev_ioctl(struct file *filp,
1017                           unsigned int ioctl, unsigned long arg)
1018 {
1019         void __user *argp = (void __user *)arg;
1020         long r = -EINVAL;
1021
1022         switch (ioctl) {
1023         case KVM_GET_API_VERSION:
1024                 r = -EINVAL;
1025                 if (arg)
1026                         goto out;
1027                 r = KVM_API_VERSION;
1028                 break;
1029         case KVM_CREATE_VM:
1030                 r = -EINVAL;
1031                 if (arg)
1032                         goto out;
1033                 r = kvm_dev_ioctl_create_vm();
1034                 break;
1035         case KVM_CHECK_EXTENSION:
1036                 r = kvm_dev_ioctl_check_extension((long)argp);
1037                 break;
1038         case KVM_GET_VCPU_MMAP_SIZE:
1039                 r = -EINVAL;
1040                 if (arg)
1041                         goto out;
1042                 r = 2 * PAGE_SIZE;
1043                 break;
1044         default:
1045                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1046         }
1047 out:
1048         return r;
1049 }
1050
1051 static struct file_operations kvm_chardev_ops = {
1052         .unlocked_ioctl = kvm_dev_ioctl,
1053         .compat_ioctl   = kvm_dev_ioctl,
1054 };
1055
1056 static struct miscdevice kvm_dev = {
1057         KVM_MINOR,
1058         "kvm",
1059         &kvm_chardev_ops,
1060 };
1061
1062 static void hardware_enable(void *junk)
1063 {
1064         int cpu = raw_smp_processor_id();
1065
1066         if (cpu_isset(cpu, cpus_hardware_enabled))
1067                 return;
1068         cpu_set(cpu, cpus_hardware_enabled);
1069         kvm_arch_hardware_enable(NULL);
1070 }
1071
1072 static void hardware_disable(void *junk)
1073 {
1074         int cpu = raw_smp_processor_id();
1075
1076         if (!cpu_isset(cpu, cpus_hardware_enabled))
1077                 return;
1078         cpu_clear(cpu, cpus_hardware_enabled);
1079         decache_vcpus_on_cpu(cpu);
1080         kvm_arch_hardware_disable(NULL);
1081 }
1082
1083 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1084                            void *v)
1085 {
1086         int cpu = (long)v;
1087
1088         val &= ~CPU_TASKS_FROZEN;
1089         switch (val) {
1090         case CPU_DYING:
1091                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1092                        cpu);
1093                 hardware_disable(NULL);
1094                 break;
1095         case CPU_UP_CANCELED:
1096                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1097                        cpu);
1098                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1099                 break;
1100         case CPU_ONLINE:
1101                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1102                        cpu);
1103                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1104                 break;
1105         }
1106         return NOTIFY_OK;
1107 }
1108
1109 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1110                       void *v)
1111 {
1112         if (val == SYS_RESTART) {
1113                 /*
1114                  * Some (well, at least mine) BIOSes hang on reboot if
1115                  * in vmx root mode.
1116                  */
1117                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1118                 on_each_cpu(hardware_disable, NULL, 0, 1);
1119         }
1120         return NOTIFY_OK;
1121 }
1122
1123 static struct notifier_block kvm_reboot_notifier = {
1124         .notifier_call = kvm_reboot,
1125         .priority = 0,
1126 };
1127
1128 void kvm_io_bus_init(struct kvm_io_bus *bus)
1129 {
1130         memset(bus, 0, sizeof(*bus));
1131 }
1132
1133 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1134 {
1135         int i;
1136
1137         for (i = 0; i < bus->dev_count; i++) {
1138                 struct kvm_io_device *pos = bus->devs[i];
1139
1140                 kvm_iodevice_destructor(pos);
1141         }
1142 }
1143
1144 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1145 {
1146         int i;
1147
1148         for (i = 0; i < bus->dev_count; i++) {
1149                 struct kvm_io_device *pos = bus->devs[i];
1150
1151                 if (pos->in_range(pos, addr))
1152                         return pos;
1153         }
1154
1155         return NULL;
1156 }
1157
1158 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1159 {
1160         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1161
1162         bus->devs[bus->dev_count++] = dev;
1163 }
1164
1165 static struct notifier_block kvm_cpu_notifier = {
1166         .notifier_call = kvm_cpu_hotplug,
1167         .priority = 20, /* must be > scheduler priority */
1168 };
1169
1170 static u64 vm_stat_get(void *_offset)
1171 {
1172         unsigned offset = (long)_offset;
1173         u64 total = 0;
1174         struct kvm *kvm;
1175
1176         spin_lock(&kvm_lock);
1177         list_for_each_entry(kvm, &vm_list, vm_list)
1178                 total += *(u32 *)((void *)kvm + offset);
1179         spin_unlock(&kvm_lock);
1180         return total;
1181 }
1182
1183 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1184
1185 static u64 vcpu_stat_get(void *_offset)
1186 {
1187         unsigned offset = (long)_offset;
1188         u64 total = 0;
1189         struct kvm *kvm;
1190         struct kvm_vcpu *vcpu;
1191         int i;
1192
1193         spin_lock(&kvm_lock);
1194         list_for_each_entry(kvm, &vm_list, vm_list)
1195                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1196                         vcpu = kvm->vcpus[i];
1197                         if (vcpu)
1198                                 total += *(u32 *)((void *)vcpu + offset);
1199                 }
1200         spin_unlock(&kvm_lock);
1201         return total;
1202 }
1203
1204 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1205
1206 static struct file_operations *stat_fops[] = {
1207         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1208         [KVM_STAT_VM]   = &vm_stat_fops,
1209 };
1210
1211 static void kvm_init_debug(void)
1212 {
1213         struct kvm_stats_debugfs_item *p;
1214
1215         debugfs_dir = debugfs_create_dir("kvm", NULL);
1216         for (p = debugfs_entries; p->name; ++p)
1217                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1218                                                 (void *)(long)p->offset,
1219                                                 stat_fops[p->kind]);
1220 }
1221
1222 static void kvm_exit_debug(void)
1223 {
1224         struct kvm_stats_debugfs_item *p;
1225
1226         for (p = debugfs_entries; p->name; ++p)
1227                 debugfs_remove(p->dentry);
1228         debugfs_remove(debugfs_dir);
1229 }
1230
1231 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1232 {
1233         hardware_disable(NULL);
1234         return 0;
1235 }
1236
1237 static int kvm_resume(struct sys_device *dev)
1238 {
1239         hardware_enable(NULL);
1240         return 0;
1241 }
1242
1243 static struct sysdev_class kvm_sysdev_class = {
1244         .name = "kvm",
1245         .suspend = kvm_suspend,
1246         .resume = kvm_resume,
1247 };
1248
1249 static struct sys_device kvm_sysdev = {
1250         .id = 0,
1251         .cls = &kvm_sysdev_class,
1252 };
1253
1254 struct page *bad_page;
1255
1256 static inline
1257 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1258 {
1259         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1260 }
1261
1262 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1263 {
1264         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1265
1266         kvm_arch_vcpu_load(vcpu, cpu);
1267 }
1268
1269 static void kvm_sched_out(struct preempt_notifier *pn,
1270                           struct task_struct *next)
1271 {
1272         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1273
1274         kvm_arch_vcpu_put(vcpu);
1275 }
1276
1277 int kvm_init(void *opaque, unsigned int vcpu_size,
1278                   struct module *module)
1279 {
1280         int r;
1281         int cpu;
1282
1283         kvm_init_debug();
1284
1285         r = kvm_arch_init(opaque);
1286         if (r)
1287                 goto out_fail;
1288
1289         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1290
1291         if (bad_page == NULL) {
1292                 r = -ENOMEM;
1293                 goto out;
1294         }
1295
1296         r = kvm_arch_hardware_setup();
1297         if (r < 0)
1298                 goto out_free_0;
1299
1300         for_each_online_cpu(cpu) {
1301                 smp_call_function_single(cpu,
1302                                 kvm_arch_check_processor_compat,
1303                                 &r, 0, 1);
1304                 if (r < 0)
1305                         goto out_free_1;
1306         }
1307
1308         on_each_cpu(hardware_enable, NULL, 0, 1);
1309         r = register_cpu_notifier(&kvm_cpu_notifier);
1310         if (r)
1311                 goto out_free_2;
1312         register_reboot_notifier(&kvm_reboot_notifier);
1313
1314         r = sysdev_class_register(&kvm_sysdev_class);
1315         if (r)
1316                 goto out_free_3;
1317
1318         r = sysdev_register(&kvm_sysdev);
1319         if (r)
1320                 goto out_free_4;
1321
1322         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1323         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1324                                            __alignof__(struct kvm_vcpu),
1325                                            0, NULL);
1326         if (!kvm_vcpu_cache) {
1327                 r = -ENOMEM;
1328                 goto out_free_5;
1329         }
1330
1331         kvm_chardev_ops.owner = module;
1332
1333         r = misc_register(&kvm_dev);
1334         if (r) {
1335                 printk(KERN_ERR "kvm: misc device register failed\n");
1336                 goto out_free;
1337         }
1338
1339         kvm_preempt_ops.sched_in = kvm_sched_in;
1340         kvm_preempt_ops.sched_out = kvm_sched_out;
1341
1342         return 0;
1343
1344 out_free:
1345         kmem_cache_destroy(kvm_vcpu_cache);
1346 out_free_5:
1347         sysdev_unregister(&kvm_sysdev);
1348 out_free_4:
1349         sysdev_class_unregister(&kvm_sysdev_class);
1350 out_free_3:
1351         unregister_reboot_notifier(&kvm_reboot_notifier);
1352         unregister_cpu_notifier(&kvm_cpu_notifier);
1353 out_free_2:
1354         on_each_cpu(hardware_disable, NULL, 0, 1);
1355 out_free_1:
1356         kvm_arch_hardware_unsetup();
1357 out_free_0:
1358         __free_page(bad_page);
1359 out:
1360         kvm_arch_exit();
1361         kvm_exit_debug();
1362 out_fail:
1363         return r;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_init);
1366
1367 void kvm_exit(void)
1368 {
1369         misc_deregister(&kvm_dev);
1370         kmem_cache_destroy(kvm_vcpu_cache);
1371         sysdev_unregister(&kvm_sysdev);
1372         sysdev_class_unregister(&kvm_sysdev_class);
1373         unregister_reboot_notifier(&kvm_reboot_notifier);
1374         unregister_cpu_notifier(&kvm_cpu_notifier);
1375         on_each_cpu(hardware_disable, NULL, 0, 1);
1376         kvm_arch_hardware_unsetup();
1377         kvm_arch_exit();
1378         kvm_exit_debug();
1379         __free_page(bad_page);
1380 }
1381 EXPORT_SYMBOL_GPL(kvm_exit);