drivers/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 "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <linux/magic.h>
  24 #include <asm/processor.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <asm/msr.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/reboot.h>
  33 #include <asm/io.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/highmem.h>
  36 #include <linux/file.h>
  37 #include <asm/desc.h>
  38 #include <linux/sysdev.h>
  39 #include <linux/cpu.h>
  40 #include <linux/file.h>
  41 #include <linux/fs.h>
  42 #include <linux/mount.h>
  43
  44 #include "x86_emulate.h"
  45 #include "segment_descriptor.h"
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 struct kvm_arch_ops *kvm_arch_ops;
  54 struct kvm_stat kvm_stat;
  55 EXPORT_SYMBOL_GPL(kvm_stat);
  56
  57 static struct kvm_stats_debugfs_item {
  58         const char *name;
  59         u32 *data;
  60         struct dentry *dentry;
  61 } debugfs_entries[] = {
  62         { "pf_fixed", &kvm_stat.pf_fixed },
  63         { "pf_guest", &kvm_stat.pf_guest },
  64         { "tlb_flush", &kvm_stat.tlb_flush },
  65         { "invlpg", &kvm_stat.invlpg },
  66         { "exits", &kvm_stat.exits },
  67         { "io_exits", &kvm_stat.io_exits },
  68         { "mmio_exits", &kvm_stat.mmio_exits },
  69         { "signal_exits", &kvm_stat.signal_exits },
  70         { "irq_window", &kvm_stat.irq_window_exits },
  71         { "halt_exits", &kvm_stat.halt_exits },
  72         { "request_irq", &kvm_stat.request_irq_exits },
  73         { "irq_exits", &kvm_stat.irq_exits },
  74         { NULL, NULL }
  75 };
  76
  77 static struct dentry *debugfs_dir;
  78
  79 struct vfsmount *kvmfs_mnt;
  80
  81 #define MAX_IO_MSRS 256
  82
  83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  84 #define LMSW_GUEST_MASK 0x0eULL
  85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  86 #define CR8_RESEVED_BITS (~0x0fULL)
  87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  88
  89 #ifdef CONFIG_X86_64
  90 // LDT or TSS descriptor in the GDT. 16 bytes.
  91 struct segment_descriptor_64 {
  92         struct segment_descriptor s;
  93         u32 base_higher;
  94         u32 pad_zero;
  95 };
  96
  97 #endif
  98
  99 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 100                            unsigned long arg);
 101
 102 static struct inode *kvmfs_inode(struct file_operations *fops)
 103 {
 104         int error = -ENOMEM;
 105         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 106
 107         if (!inode)
 108                 goto eexit_1;
 109
 110         inode->i_fop = fops;
 111
 112         /*
 113          * Mark the inode dirty from the very beginning,
 114          * that way it will never be moved to the dirty
 115          * list because mark_inode_dirty() will think
 116          * that it already _is_ on the dirty list.
 117          */
 118         inode->i_state = I_DIRTY;
 119         inode->i_mode = S_IRUSR | S_IWUSR;
 120         inode->i_uid = current->fsuid;
 121         inode->i_gid = current->fsgid;
 122         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 123         return inode;
 124
 125 eexit_1:
 126         return ERR_PTR(error);
 127 }
 128
 129 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 130 {
 131         struct file *file = get_empty_filp();
 132
 133         if (!file)
 134                 return ERR_PTR(-ENFILE);
 135
 136         file->f_path.mnt = mntget(kvmfs_mnt);
 137         file->f_path.dentry = d_alloc_anon(inode);
 138         if (!file->f_path.dentry)
 139                 return ERR_PTR(-ENOMEM);
 140         file->f_mapping = inode->i_mapping;
 141
 142         file->f_pos = 0;
 143         file->f_flags = O_RDWR;
 144         file->f_op = inode->i_fop;
 145         file->f_mode = FMODE_READ | FMODE_WRITE;
 146         file->f_version = 0;
 147         file->private_data = private_data;
 148         return file;
 149 }
 150
 151 unsigned long segment_base(u16 selector)
 152 {
 153         struct descriptor_table gdt;
 154         struct segment_descriptor *d;
 155         unsigned long table_base;
 156         typedef unsigned long ul;
 157         unsigned long v;
 158
 159         if (selector == 0)
 160                 return 0;
 161
 162         asm ("sgdt %0" : "=m"(gdt));
 163         table_base = gdt.base;
 164
 165         if (selector & 4) {           /* from ldt */
 166                 u16 ldt_selector;
 167
 168                 asm ("sldt %0" : "=g"(ldt_selector));
 169                 table_base = segment_base(ldt_selector);
 170         }
 171         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 172         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 173 #ifdef CONFIG_X86_64
 174         if (d->system == 0
 175             && (d->type == 2 || d->type == 9 || d->type == 11))
 176                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 177 #endif
 178         return v;
 179 }
 180 EXPORT_SYMBOL_GPL(segment_base);
 181
 182 static inline int valid_vcpu(int n)
 183 {
 184         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 185 }
 186
 187 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 188                    void *dest)
 189 {
 190         unsigned char *host_buf = dest;
 191         unsigned long req_size = size;
 192
 193         while (size) {
 194                 hpa_t paddr;
 195                 unsigned now;
 196                 unsigned offset;
 197                 hva_t guest_buf;
 198
 199                 paddr = gva_to_hpa(vcpu, addr);
 200
 201                 if (is_error_hpa(paddr))
 202                         break;
 203
 204                 guest_buf = (hva_t)kmap_atomic(
 205                                         pfn_to_page(paddr >> PAGE_SHIFT),
 206                                         KM_USER0);
 207                 offset = addr & ~PAGE_MASK;
 208                 guest_buf |= offset;
 209                 now = min(size, PAGE_SIZE - offset);
 210                 memcpy(host_buf, (void*)guest_buf, now);
 211                 host_buf += now;
 212                 addr += now;
 213                 size -= now;
 214                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 215         }
 216         return req_size - size;
 217 }
 218 EXPORT_SYMBOL_GPL(kvm_read_guest);
 219
 220 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 221                     void *data)
 222 {
 223         unsigned char *host_buf = data;
 224         unsigned long req_size = size;
 225
 226         while (size) {
 227                 hpa_t paddr;
 228                 unsigned now;
 229                 unsigned offset;
 230                 hva_t guest_buf;
 231                 gfn_t gfn;
 232
 233                 paddr = gva_to_hpa(vcpu, addr);
 234
 235                 if (is_error_hpa(paddr))
 236                         break;
 237
 238                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 239                 mark_page_dirty(vcpu->kvm, gfn);
 240                 guest_buf = (hva_t)kmap_atomic(
 241                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 242                 offset = addr & ~PAGE_MASK;
 243                 guest_buf |= offset;
 244                 now = min(size, PAGE_SIZE - offset);
 245                 memcpy((void*)guest_buf, host_buf, now);
 246                 host_buf += now;
 247                 addr += now;
 248                 size -= now;
 249                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 250         }
 251         return req_size - size;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_write_guest);
 254
 255 /*
 256  * Switches to specified vcpu, until a matching vcpu_put()
 257  */
 258 static void vcpu_load(struct kvm_vcpu *vcpu)
 259 {
 260         mutex_lock(&vcpu->mutex);
 261         kvm_arch_ops->vcpu_load(vcpu);
 262 }
 263
 264 /*
 265  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 266  * if the slot is not populated.
 267  */
 268 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 269 {
 270         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 271
 272         mutex_lock(&vcpu->mutex);
 273         if (!vcpu->vmcs) {
 274                 mutex_unlock(&vcpu->mutex);
 275                 return NULL;
 276         }
 277         kvm_arch_ops->vcpu_load(vcpu);
 278         return vcpu;
 279 }
 280
 281 static void vcpu_put(struct kvm_vcpu *vcpu)
 282 {
 283         kvm_arch_ops->vcpu_put(vcpu);
 284         mutex_unlock(&vcpu->mutex);
 285 }
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290         int i;
 291
 292         if (!kvm)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         spin_lock_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 298                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 299
 300                 mutex_init(&vcpu->mutex);
 301                 vcpu->cpu = -1;
 302                 vcpu->kvm = kvm;
 303                 vcpu->mmu.root_hpa = INVALID_PAGE;
 304                 INIT_LIST_HEAD(&vcpu->free_pages);
 305                 spin_lock(&kvm_lock);
 306                 list_add(&kvm->vm_list, &vm_list);
 307                 spin_unlock(&kvm_lock);
 308         }
 309         return kvm;
 310 }
 311
 312 static int kvm_dev_open(struct inode *inode, struct file *filp)
 313 {
 314         return 0;
 315 }
 316
 317 /*
 318  * Free any memory in @free but not in @dont.
 319  */
 320 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 321                                   struct kvm_memory_slot *dont)
 322 {
 323         int i;
 324
 325         if (!dont || free->phys_mem != dont->phys_mem)
 326                 if (free->phys_mem) {
 327                         for (i = 0; i < free->npages; ++i)
 328                                 if (free->phys_mem[i])
 329                                         __free_page(free->phys_mem[i]);
 330                         vfree(free->phys_mem);
 331                 }
 332
 333         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 334                 vfree(free->dirty_bitmap);
 335
 336         free->phys_mem = NULL;
 337         free->npages = 0;
 338         free->dirty_bitmap = NULL;
 339 }
 340
 341 static void kvm_free_physmem(struct kvm *kvm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < kvm->nmemslots; ++i)
 346                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 347 }
 348
 349 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 350 {
 351         if (!vcpu->vmcs)
 352                 return;
 353
 354         vcpu_load(vcpu);
 355         kvm_mmu_destroy(vcpu);
 356         vcpu_put(vcpu);
 357         kvm_arch_ops->vcpu_free(vcpu);
 358         free_page((unsigned long)vcpu->run);
 359         vcpu->run = NULL;
 360 }
 361
 362 static void kvm_free_vcpus(struct kvm *kvm)
 363 {
 364         unsigned int i;
 365
 366         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 367                 kvm_free_vcpu(&kvm->vcpus[i]);
 368 }
 369
 370 static int kvm_dev_release(struct inode *inode, struct file *filp)
 371 {
 372         return 0;
 373 }
 374
 375 static void kvm_destroy_vm(struct kvm *kvm)
 376 {
 377         spin_lock(&kvm_lock);
 378         list_del(&kvm->vm_list);
 379         spin_unlock(&kvm_lock);
 380         kvm_free_vcpus(kvm);
 381         kvm_free_physmem(kvm);
 382         kfree(kvm);
 383 }
 384
 385 static int kvm_vm_release(struct inode *inode, struct file *filp)
 386 {
 387         struct kvm *kvm = filp->private_data;
 388
 389         kvm_destroy_vm(kvm);
 390         return 0;
 391 }
 392
 393 static void inject_gp(struct kvm_vcpu *vcpu)
 394 {
 395         kvm_arch_ops->inject_gp(vcpu, 0);
 396 }
 397
 398 /*
 399  * Load the pae pdptrs.  Return true is they are all valid.
 400  */
 401 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 402 {
 403         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 404         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 405         int i;
 406         u64 pdpte;
 407         u64 *pdpt;
 408         int ret;
 409         struct kvm_memory_slot *memslot;
 410
 411         spin_lock(&vcpu->kvm->lock);
 412         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 413         /* FIXME: !memslot - emulate? 0xff? */
 414         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 415
 416         ret = 1;
 417         for (i = 0; i < 4; ++i) {
 418                 pdpte = pdpt[offset + i];
 419                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 420                         ret = 0;
 421                         goto out;
 422                 }
 423         }
 424
 425         for (i = 0; i < 4; ++i)
 426                 vcpu->pdptrs[i] = pdpt[offset + i];
 427
 428 out:
 429         kunmap_atomic(pdpt, KM_USER0);
 430         spin_unlock(&vcpu->kvm->lock);
 431
 432         return ret;
 433 }
 434
 435 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 436 {
 437         if (cr0 & CR0_RESEVED_BITS) {
 438                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 439                        cr0, vcpu->cr0);
 440                 inject_gp(vcpu);
 441                 return;
 442         }
 443
 444         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 445                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 446                 inject_gp(vcpu);
 447                 return;
 448         }
 449
 450         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 451                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 452                        "and a clear PE flag\n");
 453                 inject_gp(vcpu);
 454                 return;
 455         }
 456
 457         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 458 #ifdef CONFIG_X86_64
 459                 if ((vcpu->shadow_efer & EFER_LME)) {
 460                         int cs_db, cs_l;
 461
 462                         if (!is_pae(vcpu)) {
 463                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 464                                        "in long mode while PAE is disabled\n");
 465                                 inject_gp(vcpu);
 466                                 return;
 467                         }
 468                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 469                         if (cs_l) {
 470                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 471                                        "in long mode while CS.L == 1\n");
 472                                 inject_gp(vcpu);
 473                                 return;
 474
 475                         }
 476                 } else
 477 #endif
 478                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 479                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 480                                "reserved bits\n");
 481                         inject_gp(vcpu);
 482                         return;
 483                 }
 484
 485         }
 486
 487         kvm_arch_ops->set_cr0(vcpu, cr0);
 488         vcpu->cr0 = cr0;
 489
 490         spin_lock(&vcpu->kvm->lock);
 491         kvm_mmu_reset_context(vcpu);
 492         spin_unlock(&vcpu->kvm->lock);
 493         return;
 494 }
 495 EXPORT_SYMBOL_GPL(set_cr0);
 496
 497 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 498 {
 499         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 500         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 501 }
 502 EXPORT_SYMBOL_GPL(lmsw);
 503
 504 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 505 {
 506         if (cr4 & CR4_RESEVED_BITS) {
 507                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 508                 inject_gp(vcpu);
 509                 return;
 510         }
 511
 512         if (is_long_mode(vcpu)) {
 513                 if (!(cr4 & CR4_PAE_MASK)) {
 514                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 515                                "in long mode\n");
 516                         inject_gp(vcpu);
 517                         return;
 518                 }
 519         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 520                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 522                 inject_gp(vcpu);
 523         }
 524
 525         if (cr4 & CR4_VMXE_MASK) {
 526                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 527                 inject_gp(vcpu);
 528                 return;
 529         }
 530         kvm_arch_ops->set_cr4(vcpu, cr4);
 531         spin_lock(&vcpu->kvm->lock);
 532         kvm_mmu_reset_context(vcpu);
 533         spin_unlock(&vcpu->kvm->lock);
 534 }
 535 EXPORT_SYMBOL_GPL(set_cr4);
 536
 537 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 538 {
 539         if (is_long_mode(vcpu)) {
 540                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 541                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 542                         inject_gp(vcpu);
 543                         return;
 544                 }
 545         } else {
 546                 if (cr3 & CR3_RESEVED_BITS) {
 547                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 548                         inject_gp(vcpu);
 549                         return;
 550                 }
 551                 if (is_paging(vcpu) && is_pae(vcpu) &&
 552                     !load_pdptrs(vcpu, cr3)) {
 553                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 554                                "reserved bits\n");
 555                         inject_gp(vcpu);
 556                         return;
 557                 }
 558         }
 559
 560         vcpu->cr3 = cr3;
 561         spin_lock(&vcpu->kvm->lock);
 562         /*
 563          * Does the new cr3 value map to physical memory? (Note, we
 564          * catch an invalid cr3 even in real-mode, because it would
 565          * cause trouble later on when we turn on paging anyway.)
 566          *
 567          * A real CPU would silently accept an invalid cr3 and would
 568          * attempt to use it - with largely undefined (and often hard
 569          * to debug) behavior on the guest side.
 570          */
 571         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 572                 inject_gp(vcpu);
 573         else
 574                 vcpu->mmu.new_cr3(vcpu);
 575         spin_unlock(&vcpu->kvm->lock);
 576 }
 577 EXPORT_SYMBOL_GPL(set_cr3);
 578
 579 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 580 {
 581         if ( cr8 & CR8_RESEVED_BITS) {
 582                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 583                 inject_gp(vcpu);
 584                 return;
 585         }
 586         vcpu->cr8 = cr8;
 587 }
 588 EXPORT_SYMBOL_GPL(set_cr8);
 589
 590 void fx_init(struct kvm_vcpu *vcpu)
 591 {
 592         struct __attribute__ ((__packed__)) fx_image_s {
 593                 u16 control; //fcw
 594                 u16 status; //fsw
 595                 u16 tag; // ftw
 596                 u16 opcode; //fop
 597                 u64 ip; // fpu ip
 598                 u64 operand;// fpu dp
 599                 u32 mxcsr;
 600                 u32 mxcsr_mask;
 601
 602         } *fx_image;
 603
 604         fx_save(vcpu->host_fx_image);
 605         fpu_init();
 606         fx_save(vcpu->guest_fx_image);
 607         fx_restore(vcpu->host_fx_image);
 608
 609         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 610         fx_image->mxcsr = 0x1f80;
 611         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 612                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 613 }
 614 EXPORT_SYMBOL_GPL(fx_init);
 615
 616 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 617 {
 618         spin_lock(&vcpu->kvm->lock);
 619         kvm_mmu_slot_remove_write_access(vcpu, slot);
 620         spin_unlock(&vcpu->kvm->lock);
 621 }
 622
 623 /*
 624  * Allocate some memory and give it an address in the guest physical address
 625  * space.
 626  *
 627  * Discontiguous memory is allowed, mostly for framebuffers.
 628  */
 629 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 630                                           struct kvm_memory_region *mem)
 631 {
 632         int r;
 633         gfn_t base_gfn;
 634         unsigned long npages;
 635         unsigned long i;
 636         struct kvm_memory_slot *memslot;
 637         struct kvm_memory_slot old, new;
 638         int memory_config_version;
 639
 640         r = -EINVAL;
 641         /* General sanity checks */
 642         if (mem->memory_size & (PAGE_SIZE - 1))
 643                 goto out;
 644         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 645                 goto out;
 646         if (mem->slot >= KVM_MEMORY_SLOTS)
 647                 goto out;
 648         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 649                 goto out;
 650
 651         memslot = &kvm->memslots[mem->slot];
 652         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 653         npages = mem->memory_size >> PAGE_SHIFT;
 654
 655         if (!npages)
 656                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 657
 658 raced:
 659         spin_lock(&kvm->lock);
 660
 661         memory_config_version = kvm->memory_config_version;
 662         new = old = *memslot;
 663
 664         new.base_gfn = base_gfn;
 665         new.npages = npages;
 666         new.flags = mem->flags;
 667
 668         /* Disallow changing a memory slot's size. */
 669         r = -EINVAL;
 670         if (npages && old.npages && npages != old.npages)
 671                 goto out_unlock;
 672
 673         /* Check for overlaps */
 674         r = -EEXIST;
 675         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 676                 struct kvm_memory_slot *s = &kvm->memslots[i];
 677
 678                 if (s == memslot)
 679                         continue;
 680                 if (!((base_gfn + npages <= s->base_gfn) ||
 681                       (base_gfn >= s->base_gfn + s->npages)))
 682                         goto out_unlock;
 683         }
 684         /*
 685          * Do memory allocations outside lock.  memory_config_version will
 686          * detect any races.
 687          */
 688         spin_unlock(&kvm->lock);
 689
 690         /* Deallocate if slot is being removed */
 691         if (!npages)
 692                 new.phys_mem = NULL;
 693
 694         /* Free page dirty bitmap if unneeded */
 695         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 696                 new.dirty_bitmap = NULL;
 697
 698         r = -ENOMEM;
 699
 700         /* Allocate if a slot is being created */
 701         if (npages && !new.phys_mem) {
 702                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 703
 704                 if (!new.phys_mem)
 705                         goto out_free;
 706
 707                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 708                 for (i = 0; i < npages; ++i) {
 709                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 710                                                      | __GFP_ZERO);
 711                         if (!new.phys_mem[i])
 712                                 goto out_free;
 713                         set_page_private(new.phys_mem[i],0);
 714                 }
 715         }
 716
 717         /* Allocate page dirty bitmap if needed */
 718         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 719                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 720
 721                 new.dirty_bitmap = vmalloc(dirty_bytes);
 722                 if (!new.dirty_bitmap)
 723                         goto out_free;
 724                 memset(new.dirty_bitmap, 0, dirty_bytes);
 725         }
 726
 727         spin_lock(&kvm->lock);
 728
 729         if (memory_config_version != kvm->memory_config_version) {
 730                 spin_unlock(&kvm->lock);
 731                 kvm_free_physmem_slot(&new, &old);
 732                 goto raced;
 733         }
 734
 735         r = -EAGAIN;
 736         if (kvm->busy)
 737                 goto out_unlock;
 738
 739         if (mem->slot >= kvm->nmemslots)
 740                 kvm->nmemslots = mem->slot + 1;
 741
 742         *memslot = new;
 743         ++kvm->memory_config_version;
 744
 745         spin_unlock(&kvm->lock);
 746
 747         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 748                 struct kvm_vcpu *vcpu;
 749
 750                 vcpu = vcpu_load_slot(kvm, i);
 751                 if (!vcpu)
 752                         continue;
 753                 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
 754                         do_remove_write_access(vcpu, mem->slot);
 755                 kvm_mmu_reset_context(vcpu);
 756                 vcpu_put(vcpu);
 757         }
 758
 759         kvm_free_physmem_slot(&old, &new);
 760         return 0;
 761
 762 out_unlock:
 763         spin_unlock(&kvm->lock);
 764 out_free:
 765         kvm_free_physmem_slot(&new, &old);
 766 out:
 767         return r;
 768 }
 769
 770 /*
 771  * Get (and clear) the dirty memory log for a memory slot.
 772  */
 773 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 774                                       struct kvm_dirty_log *log)
 775 {
 776         struct kvm_memory_slot *memslot;
 777         int r, i;
 778         int n;
 779         int cleared;
 780         unsigned long any = 0;
 781
 782         spin_lock(&kvm->lock);
 783
 784         /*
 785          * Prevent changes to guest memory configuration even while the lock
 786          * is not taken.
 787          */
 788         ++kvm->busy;
 789         spin_unlock(&kvm->lock);
 790         r = -EINVAL;
 791         if (log->slot >= KVM_MEMORY_SLOTS)
 792                 goto out;
 793
 794         memslot = &kvm->memslots[log->slot];
 795         r = -ENOENT;
 796         if (!memslot->dirty_bitmap)
 797                 goto out;
 798
 799         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 800
 801         for (i = 0; !any && i < n/sizeof(long); ++i)
 802                 any = memslot->dirty_bitmap[i];
 803
 804         r = -EFAULT;
 805         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 806                 goto out;
 807
 808         if (any) {
 809                 cleared = 0;
 810                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 811                         struct kvm_vcpu *vcpu;
 812
 813                         vcpu = vcpu_load_slot(kvm, i);
 814                         if (!vcpu)
 815                                 continue;
 816                         if (!cleared) {
 817                                 do_remove_write_access(vcpu, log->slot);
 818                                 memset(memslot->dirty_bitmap, 0, n);
 819                                 cleared = 1;
 820                         }
 821                         kvm_arch_ops->tlb_flush(vcpu);
 822                         vcpu_put(vcpu);
 823                 }
 824         }
 825
 826         r = 0;
 827
 828 out:
 829         spin_lock(&kvm->lock);
 830         --kvm->busy;
 831         spin_unlock(&kvm->lock);
 832         return r;
 833 }
 834
 835 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 836 {
 837         int i;
 838
 839         for (i = 0; i < kvm->nmemslots; ++i) {
 840                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 841
 842                 if (gfn >= memslot->base_gfn
 843                     && gfn < memslot->base_gfn + memslot->npages)
 844                         return memslot;
 845         }
 846         return NULL;
 847 }
 848 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 849
 850 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 851 {
 852         int i;
 853         struct kvm_memory_slot *memslot = NULL;
 854         unsigned long rel_gfn;
 855
 856         for (i = 0; i < kvm->nmemslots; ++i) {
 857                 memslot = &kvm->memslots[i];
 858
 859                 if (gfn >= memslot->base_gfn
 860                     && gfn < memslot->base_gfn + memslot->npages) {
 861
 862                         if (!memslot || !memslot->dirty_bitmap)
 863                                 return;
 864
 865                         rel_gfn = gfn - memslot->base_gfn;
 866
 867                         /* avoid RMW */
 868                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 869                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 870                         return;
 871                 }
 872         }
 873 }
 874
 875 static int emulator_read_std(unsigned long addr,
 876                              unsigned long *val,
 877                              unsigned int bytes,
 878                              struct x86_emulate_ctxt *ctxt)
 879 {
 880         struct kvm_vcpu *vcpu = ctxt->vcpu;
 881         void *data = val;
 882
 883         while (bytes) {
 884                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 885                 unsigned offset = addr & (PAGE_SIZE-1);
 886                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 887                 unsigned long pfn;
 888                 struct kvm_memory_slot *memslot;
 889                 void *page;
 890
 891                 if (gpa == UNMAPPED_GVA)
 892                         return X86EMUL_PROPAGATE_FAULT;
 893                 pfn = gpa >> PAGE_SHIFT;
 894                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 895                 if (!memslot)
 896                         return X86EMUL_UNHANDLEABLE;
 897                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 898
 899                 memcpy(data, page + offset, tocopy);
 900
 901                 kunmap_atomic(page, KM_USER0);
 902
 903                 bytes -= tocopy;
 904                 data += tocopy;
 905                 addr += tocopy;
 906         }
 907
 908         return X86EMUL_CONTINUE;
 909 }
 910
 911 static int emulator_write_std(unsigned long addr,
 912                               unsigned long val,
 913                               unsigned int bytes,
 914                               struct x86_emulate_ctxt *ctxt)
 915 {
 916         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 917                addr, bytes);
 918         return X86EMUL_UNHANDLEABLE;
 919 }
 920
 921 static int emulator_read_emulated(unsigned long addr,
 922                                   unsigned long *val,
 923                                   unsigned int bytes,
 924                                   struct x86_emulate_ctxt *ctxt)
 925 {
 926         struct kvm_vcpu *vcpu = ctxt->vcpu;
 927
 928         if (vcpu->mmio_read_completed) {
 929                 memcpy(val, vcpu->mmio_data, bytes);
 930                 vcpu->mmio_read_completed = 0;
 931                 return X86EMUL_CONTINUE;
 932         } else if (emulator_read_std(addr, val, bytes, ctxt)
 933                    == X86EMUL_CONTINUE)
 934                 return X86EMUL_CONTINUE;
 935         else {
 936                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 937
 938                 if (gpa == UNMAPPED_GVA)
 939                         return X86EMUL_PROPAGATE_FAULT;
 940                 vcpu->mmio_needed = 1;
 941                 vcpu->mmio_phys_addr = gpa;
 942                 vcpu->mmio_size = bytes;
 943                 vcpu->mmio_is_write = 0;
 944
 945                 return X86EMUL_UNHANDLEABLE;
 946         }
 947 }
 948
 949 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 950                                unsigned long val, int bytes)
 951 {
 952         struct kvm_memory_slot *m;
 953         struct page *page;
 954         void *virt;
 955
 956         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 957                 return 0;
 958         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 959         if (!m)
 960                 return 0;
 961         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 962         kvm_mmu_pre_write(vcpu, gpa, bytes);
 963         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
 964         virt = kmap_atomic(page, KM_USER0);
 965         memcpy(virt + offset_in_page(gpa), &val, bytes);
 966         kunmap_atomic(virt, KM_USER0);
 967         kvm_mmu_post_write(vcpu, gpa, bytes);
 968         return 1;
 969 }
 970
 971 static int emulator_write_emulated(unsigned long addr,
 972                                    unsigned long val,
 973                                    unsigned int bytes,
 974                                    struct x86_emulate_ctxt *ctxt)
 975 {
 976         struct kvm_vcpu *vcpu = ctxt->vcpu;
 977         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 978
 979         if (gpa == UNMAPPED_GVA)
 980                 return X86EMUL_PROPAGATE_FAULT;
 981
 982         if (emulator_write_phys(vcpu, gpa, val, bytes))
 983                 return X86EMUL_CONTINUE;
 984
 985         vcpu->mmio_needed = 1;
 986         vcpu->mmio_phys_addr = gpa;
 987         vcpu->mmio_size = bytes;
 988         vcpu->mmio_is_write = 1;
 989         memcpy(vcpu->mmio_data, &val, bytes);
 990
 991         return X86EMUL_CONTINUE;
 992 }
 993
 994 static int emulator_cmpxchg_emulated(unsigned long addr,
 995                                      unsigned long old,
 996                                      unsigned long new,
 997                                      unsigned int bytes,
 998                                      struct x86_emulate_ctxt *ctxt)
 999 {
1000         static int reported;
1001
1002         if (!reported) {
1003                 reported = 1;
1004                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1005         }
1006         return emulator_write_emulated(addr, new, bytes, ctxt);
1007 }
1008
1009 #ifdef CONFIG_X86_32
1010
1011 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1012                                        unsigned long old_lo,
1013                                        unsigned long old_hi,
1014                                        unsigned long new_lo,
1015                                        unsigned long new_hi,
1016                                        struct x86_emulate_ctxt *ctxt)
1017 {
1018         static int reported;
1019         int r;
1020
1021         if (!reported) {
1022                 reported = 1;
1023                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1024         }
1025         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1026         if (r != X86EMUL_CONTINUE)
1027                 return r;
1028         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1029 }
1030
1031 #endif
1032
1033 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1034 {
1035         return kvm_arch_ops->get_segment_base(vcpu, seg);
1036 }
1037
1038 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1039 {
1040         return X86EMUL_CONTINUE;
1041 }
1042
1043 int emulate_clts(struct kvm_vcpu *vcpu)
1044 {
1045         unsigned long cr0;
1046
1047         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1048         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1049         kvm_arch_ops->set_cr0(vcpu, cr0);
1050         return X86EMUL_CONTINUE;
1051 }
1052
1053 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1054 {
1055         struct kvm_vcpu *vcpu = ctxt->vcpu;
1056
1057         switch (dr) {
1058         case 0 ... 3:
1059                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1060                 return X86EMUL_CONTINUE;
1061         default:
1062                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1063                        __FUNCTION__, dr);
1064                 return X86EMUL_UNHANDLEABLE;
1065         }
1066 }
1067
1068 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1069 {
1070         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1071         int exception;
1072
1073         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1074         if (exception) {
1075                 /* FIXME: better handling */
1076                 return X86EMUL_UNHANDLEABLE;
1077         }
1078         return X86EMUL_CONTINUE;
1079 }
1080
1081 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1082 {
1083         static int reported;
1084         u8 opcodes[4];
1085         unsigned long rip = ctxt->vcpu->rip;
1086         unsigned long rip_linear;
1087
1088         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1089
1090         if (reported)
1091                 return;
1092
1093         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1094
1095         printk(KERN_ERR "emulation failed but !mmio_needed?"
1096                " rip %lx %02x %02x %02x %02x\n",
1097                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1098         reported = 1;
1099 }
1100
1101 struct x86_emulate_ops emulate_ops = {
1102         .read_std            = emulator_read_std,
1103         .write_std           = emulator_write_std,
1104         .read_emulated       = emulator_read_emulated,
1105         .write_emulated      = emulator_write_emulated,
1106         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1107 #ifdef CONFIG_X86_32
1108         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1109 #endif
1110 };
1111
1112 int emulate_instruction(struct kvm_vcpu *vcpu,
1113                         struct kvm_run *run,
1114                         unsigned long cr2,
1115                         u16 error_code)
1116 {
1117         struct x86_emulate_ctxt emulate_ctxt;
1118         int r;
1119         int cs_db, cs_l;
1120
1121         kvm_arch_ops->cache_regs(vcpu);
1122
1123         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1124
1125         emulate_ctxt.vcpu = vcpu;
1126         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1127         emulate_ctxt.cr2 = cr2;
1128         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1129                 ? X86EMUL_MODE_REAL : cs_l
1130                 ? X86EMUL_MODE_PROT64 : cs_db
1131                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1132
1133         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1134                 emulate_ctxt.cs_base = 0;
1135                 emulate_ctxt.ds_base = 0;
1136                 emulate_ctxt.es_base = 0;
1137                 emulate_ctxt.ss_base = 0;
1138         } else {
1139                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1140                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1141                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1142                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1143         }
1144
1145         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1146         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1147
1148         vcpu->mmio_is_write = 0;
1149         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1150
1151         if ((r || vcpu->mmio_is_write) && run) {
1152                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1153                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1154                 run->mmio.len = vcpu->mmio_size;
1155                 run->mmio.is_write = vcpu->mmio_is_write;
1156         }
1157
1158         if (r) {
1159                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1160                         return EMULATE_DONE;
1161                 if (!vcpu->mmio_needed) {
1162                         report_emulation_failure(&emulate_ctxt);
1163                         return EMULATE_FAIL;
1164                 }
1165                 return EMULATE_DO_MMIO;
1166         }
1167
1168         kvm_arch_ops->decache_regs(vcpu);
1169         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1170
1171         if (vcpu->mmio_is_write)
1172                 return EMULATE_DO_MMIO;
1173
1174         return EMULATE_DONE;
1175 }
1176 EXPORT_SYMBOL_GPL(emulate_instruction);
1177
1178 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1179 {
1180         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1181
1182         kvm_arch_ops->cache_regs(vcpu);
1183         ret = -KVM_EINVAL;
1184 #ifdef CONFIG_X86_64
1185         if (is_long_mode(vcpu)) {
1186                 nr = vcpu->regs[VCPU_REGS_RAX];
1187                 a0 = vcpu->regs[VCPU_REGS_RDI];
1188                 a1 = vcpu->regs[VCPU_REGS_RSI];
1189                 a2 = vcpu->regs[VCPU_REGS_RDX];
1190                 a3 = vcpu->regs[VCPU_REGS_RCX];
1191                 a4 = vcpu->regs[VCPU_REGS_R8];
1192                 a5 = vcpu->regs[VCPU_REGS_R9];
1193         } else
1194 #endif
1195         {
1196                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1197                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1198                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1199                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1200                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1201                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1202                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1203         }
1204         switch (nr) {
1205         default:
1206                 ;
1207         }
1208         vcpu->regs[VCPU_REGS_RAX] = ret;
1209         kvm_arch_ops->decache_regs(vcpu);
1210         return 1;
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_hypercall);
1213
1214 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1215 {
1216         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1217 }
1218
1219 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1220 {
1221         struct descriptor_table dt = { limit, base };
1222
1223         kvm_arch_ops->set_gdt(vcpu, &dt);
1224 }
1225
1226 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1227 {
1228         struct descriptor_table dt = { limit, base };
1229
1230         kvm_arch_ops->set_idt(vcpu, &dt);
1231 }
1232
1233 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1234                    unsigned long *rflags)
1235 {
1236         lmsw(vcpu, msw);
1237         *rflags = kvm_arch_ops->get_rflags(vcpu);
1238 }
1239
1240 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1241 {
1242         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1243         switch (cr) {
1244         case 0:
1245                 return vcpu->cr0;
1246         case 2:
1247                 return vcpu->cr2;
1248         case 3:
1249                 return vcpu->cr3;
1250         case 4:
1251                 return vcpu->cr4;
1252         default:
1253                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1254                 return 0;
1255         }
1256 }
1257
1258 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1259                      unsigned long *rflags)
1260 {
1261         switch (cr) {
1262         case 0:
1263                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1264                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1265                 break;
1266         case 2:
1267                 vcpu->cr2 = val;
1268                 break;
1269         case 3:
1270                 set_cr3(vcpu, val);
1271                 break;
1272         case 4:
1273                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1274                 break;
1275         default:
1276                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1277         }
1278 }
1279
1280 /*
1281  * Register the para guest with the host:
1282  */
1283 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1284 {
1285         struct kvm_vcpu_para_state *para_state;
1286         hpa_t para_state_hpa, hypercall_hpa;
1287         struct page *para_state_page;
1288         unsigned char *hypercall;
1289         gpa_t hypercall_gpa;
1290
1291         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1292         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1293
1294         /*
1295          * Needs to be page aligned:
1296          */
1297         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1298                 goto err_gp;
1299
1300         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1301         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1302         if (is_error_hpa(para_state_hpa))
1303                 goto err_gp;
1304
1305         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1306         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1307         para_state = kmap_atomic(para_state_page, KM_USER0);
1308
1309         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1310         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1311
1312         para_state->host_version = KVM_PARA_API_VERSION;
1313         /*
1314          * We cannot support guests that try to register themselves
1315          * with a newer API version than the host supports:
1316          */
1317         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1318                 para_state->ret = -KVM_EINVAL;
1319                 goto err_kunmap_skip;
1320         }
1321
1322         hypercall_gpa = para_state->hypercall_gpa;
1323         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1324         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1325         if (is_error_hpa(hypercall_hpa)) {
1326                 para_state->ret = -KVM_EINVAL;
1327                 goto err_kunmap_skip;
1328         }
1329
1330         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1331         vcpu->para_state_page = para_state_page;
1332         vcpu->para_state_gpa = para_state_gpa;
1333         vcpu->hypercall_gpa = hypercall_gpa;
1334
1335         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1336         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1337                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1338         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1339         kunmap_atomic(hypercall, KM_USER1);
1340
1341         para_state->ret = 0;
1342 err_kunmap_skip:
1343         kunmap_atomic(para_state, KM_USER0);
1344         return 0;
1345 err_gp:
1346         return 1;
1347 }
1348
1349 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1350 {
1351         u64 data;
1352
1353         switch (msr) {
1354         case 0xc0010010: /* SYSCFG */
1355         case 0xc0010015: /* HWCR */
1356         case MSR_IA32_PLATFORM_ID:
1357         case MSR_IA32_P5_MC_ADDR:
1358         case MSR_IA32_P5_MC_TYPE:
1359         case MSR_IA32_MC0_CTL:
1360         case MSR_IA32_MCG_STATUS:
1361         case MSR_IA32_MCG_CAP:
1362         case MSR_IA32_MC0_MISC:
1363         case MSR_IA32_MC0_MISC+4:
1364         case MSR_IA32_MC0_MISC+8:
1365         case MSR_IA32_MC0_MISC+12:
1366         case MSR_IA32_MC0_MISC+16:
1367         case MSR_IA32_UCODE_REV:
1368         case MSR_IA32_PERF_STATUS:
1369                 /* MTRR registers */
1370         case 0xfe:
1371         case 0x200 ... 0x2ff:
1372                 data = 0;
1373                 break;
1374         case 0xcd: /* fsb frequency */
1375                 data = 3;
1376                 break;
1377         case MSR_IA32_APICBASE:
1378                 data = vcpu->apic_base;
1379                 break;
1380         case MSR_IA32_MISC_ENABLE:
1381                 data = vcpu->ia32_misc_enable_msr;
1382                 break;
1383 #ifdef CONFIG_X86_64
1384         case MSR_EFER:
1385                 data = vcpu->shadow_efer;
1386                 break;
1387 #endif
1388         default:
1389                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1390                 return 1;
1391         }
1392         *pdata = data;
1393         return 0;
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1396
1397 /*
1398  * Reads an msr value (of 'msr_index') into 'pdata'.
1399  * Returns 0 on success, non-0 otherwise.
1400  * Assumes vcpu_load() was already called.
1401  */
1402 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1403 {
1404         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1405 }
1406
1407 #ifdef CONFIG_X86_64
1408
1409 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1410 {
1411         if (efer & EFER_RESERVED_BITS) {
1412                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1413                        efer);
1414                 inject_gp(vcpu);
1415                 return;
1416         }
1417
1418         if (is_paging(vcpu)
1419             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1420                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1421                 inject_gp(vcpu);
1422                 return;
1423         }
1424
1425         kvm_arch_ops->set_efer(vcpu, efer);
1426
1427         efer &= ~EFER_LMA;
1428         efer |= vcpu->shadow_efer & EFER_LMA;
1429
1430         vcpu->shadow_efer = efer;
1431 }
1432
1433 #endif
1434
1435 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1436 {
1437         switch (msr) {
1438 #ifdef CONFIG_X86_64
1439         case MSR_EFER:
1440                 set_efer(vcpu, data);
1441                 break;
1442 #endif
1443         case MSR_IA32_MC0_STATUS:
1444                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1445                        __FUNCTION__, data);
1446                 break;
1447         case MSR_IA32_UCODE_REV:
1448         case MSR_IA32_UCODE_WRITE:
1449         case 0x200 ... 0x2ff: /* MTRRs */
1450                 break;
1451         case MSR_IA32_APICBASE:
1452                 vcpu->apic_base = data;
1453                 break;
1454         case MSR_IA32_MISC_ENABLE:
1455                 vcpu->ia32_misc_enable_msr = data;
1456                 break;
1457         /*
1458          * This is the 'probe whether the host is KVM' logic:
1459          */
1460         case MSR_KVM_API_MAGIC:
1461                 return vcpu_register_para(vcpu, data);
1462
1463         default:
1464                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1465                 return 1;
1466         }
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1470
1471 /*
1472  * Writes msr value into into the appropriate "register".
1473  * Returns 0 on success, non-0 otherwise.
1474  * Assumes vcpu_load() was already called.
1475  */
1476 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1477 {
1478         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1479 }
1480
1481 void kvm_resched(struct kvm_vcpu *vcpu)
1482 {
1483         vcpu_put(vcpu);
1484         cond_resched();
1485         vcpu_load(vcpu);
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_resched);
1488
1489 void load_msrs(struct vmx_msr_entry *e, int n)
1490 {
1491         int i;
1492
1493         for (i = 0; i < n; ++i)
1494                 wrmsrl(e[i].index, e[i].data);
1495 }
1496 EXPORT_SYMBOL_GPL(load_msrs);
1497
1498 void save_msrs(struct vmx_msr_entry *e, int n)
1499 {
1500         int i;
1501
1502         for (i = 0; i < n; ++i)
1503                 rdmsrl(e[i].index, e[i].data);
1504 }
1505 EXPORT_SYMBOL_GPL(save_msrs);
1506
1507 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1508 {
1509         int i;
1510         u32 function;
1511         struct kvm_cpuid_entry *e, *best;
1512
1513         kvm_arch_ops->cache_regs(vcpu);
1514         function = vcpu->regs[VCPU_REGS_RAX];
1515         vcpu->regs[VCPU_REGS_RAX] = 0;
1516         vcpu->regs[VCPU_REGS_RBX] = 0;
1517         vcpu->regs[VCPU_REGS_RCX] = 0;
1518         vcpu->regs[VCPU_REGS_RDX] = 0;
1519         best = NULL;
1520         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1521                 e = &vcpu->cpuid_entries[i];
1522                 if (e->function == function) {
1523                         best = e;
1524                         break;
1525                 }
1526                 /*
1527                  * Both basic or both extended?
1528                  */
1529                 if (((e->function ^ function) & 0x80000000) == 0)
1530                         if (!best || e->function > best->function)
1531                                 best = e;
1532         }
1533         if (best) {
1534                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1535                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1536                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1537                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1538         }
1539         kvm_arch_ops->decache_regs(vcpu);
1540         kvm_arch_ops->skip_emulated_instruction(vcpu);
1541 }
1542 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1543
1544 static void complete_pio(struct kvm_vcpu *vcpu)
1545 {
1546         struct kvm_io *io = &vcpu->run->io;
1547         long delta;
1548
1549         kvm_arch_ops->cache_regs(vcpu);
1550
1551         if (!io->string) {
1552                 if (io->direction == KVM_EXIT_IO_IN)
1553                         memcpy(&vcpu->regs[VCPU_REGS_RAX], &io->value,
1554                                io->size);
1555         } else {
1556                 delta = 1;
1557                 if (io->rep) {
1558                         delta *= io->count;
1559                         /*
1560                          * The size of the register should really depend on
1561                          * current address size.
1562                          */
1563                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1564                 }
1565                 if (io->string_down)
1566                         delta = -delta;
1567                 delta *= io->size;
1568                 if (io->direction == KVM_EXIT_IO_IN)
1569                         vcpu->regs[VCPU_REGS_RDI] += delta;
1570                 else
1571                         vcpu->regs[VCPU_REGS_RSI] += delta;
1572         }
1573
1574         vcpu->pio_pending = 0;
1575         vcpu->run->io_completed = 0;
1576
1577         kvm_arch_ops->decache_regs(vcpu);
1578
1579         kvm_arch_ops->skip_emulated_instruction(vcpu);
1580 }
1581
1582 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1583 {
1584         int r;
1585
1586         vcpu_load(vcpu);
1587
1588         /* re-sync apic's tpr */
1589         vcpu->cr8 = kvm_run->cr8;
1590
1591         if (kvm_run->io_completed) {
1592                 if (vcpu->pio_pending)
1593                         complete_pio(vcpu);
1594                 else {
1595                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1596                         vcpu->mmio_read_completed = 1;
1597                 }
1598         }
1599
1600         vcpu->mmio_needed = 0;
1601
1602         r = kvm_arch_ops->run(vcpu, kvm_run);
1603
1604         vcpu_put(vcpu);
1605         return r;
1606 }
1607
1608 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1609                                    struct kvm_regs *regs)
1610 {
1611         vcpu_load(vcpu);
1612
1613         kvm_arch_ops->cache_regs(vcpu);
1614
1615         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1616         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1617         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1618         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1619         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1620         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1621         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1622         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1623 #ifdef CONFIG_X86_64
1624         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1625         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1626         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1627         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1628         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1629         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1630         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1631         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1632 #endif
1633
1634         regs->rip = vcpu->rip;
1635         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1636
1637         /*
1638          * Don't leak debug flags in case they were set for guest debugging
1639          */
1640         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1641                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1642
1643         vcpu_put(vcpu);
1644
1645         return 0;
1646 }
1647
1648 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1649                                    struct kvm_regs *regs)
1650 {
1651         vcpu_load(vcpu);
1652
1653         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1654         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1655         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1656         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1657         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1658         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1659         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1660         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1661 #ifdef CONFIG_X86_64
1662         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1663         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1664         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1665         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1666         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1667         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1668         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1669         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1670 #endif
1671
1672         vcpu->rip = regs->rip;
1673         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1674
1675         kvm_arch_ops->decache_regs(vcpu);
1676
1677         vcpu_put(vcpu);
1678
1679         return 0;
1680 }
1681
1682 static void get_segment(struct kvm_vcpu *vcpu,
1683                         struct kvm_segment *var, int seg)
1684 {
1685         return kvm_arch_ops->get_segment(vcpu, var, seg);
1686 }
1687
1688 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1689                                     struct kvm_sregs *sregs)
1690 {
1691         struct descriptor_table dt;
1692
1693         vcpu_load(vcpu);
1694
1695         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1696         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1697         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1698         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1699         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1700         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1701
1702         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1703         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1704
1705         kvm_arch_ops->get_idt(vcpu, &dt);
1706         sregs->idt.limit = dt.limit;
1707         sregs->idt.base = dt.base;
1708         kvm_arch_ops->get_gdt(vcpu, &dt);
1709         sregs->gdt.limit = dt.limit;
1710         sregs->gdt.base = dt.base;
1711
1712         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1713         sregs->cr0 = vcpu->cr0;
1714         sregs->cr2 = vcpu->cr2;
1715         sregs->cr3 = vcpu->cr3;
1716         sregs->cr4 = vcpu->cr4;
1717         sregs->cr8 = vcpu->cr8;
1718         sregs->efer = vcpu->shadow_efer;
1719         sregs->apic_base = vcpu->apic_base;
1720
1721         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1722                sizeof sregs->interrupt_bitmap);
1723
1724         vcpu_put(vcpu);
1725
1726         return 0;
1727 }
1728
1729 static void set_segment(struct kvm_vcpu *vcpu,
1730                         struct kvm_segment *var, int seg)
1731 {
1732         return kvm_arch_ops->set_segment(vcpu, var, seg);
1733 }
1734
1735 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1736                                     struct kvm_sregs *sregs)
1737 {
1738         int mmu_reset_needed = 0;
1739         int i;
1740         struct descriptor_table dt;
1741
1742         vcpu_load(vcpu);
1743
1744         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1745         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1746         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1747         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1748         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1749         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1750
1751         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1752         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1753
1754         dt.limit = sregs->idt.limit;
1755         dt.base = sregs->idt.base;
1756         kvm_arch_ops->set_idt(vcpu, &dt);
1757         dt.limit = sregs->gdt.limit;
1758         dt.base = sregs->gdt.base;
1759         kvm_arch_ops->set_gdt(vcpu, &dt);
1760
1761         vcpu->cr2 = sregs->cr2;
1762         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1763         vcpu->cr3 = sregs->cr3;
1764
1765         vcpu->cr8 = sregs->cr8;
1766
1767         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1768 #ifdef CONFIG_X86_64
1769         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1770 #endif
1771         vcpu->apic_base = sregs->apic_base;
1772
1773         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1774
1775         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1776         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1777
1778         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1779         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1780         if (!is_long_mode(vcpu) && is_pae(vcpu))
1781                 load_pdptrs(vcpu, vcpu->cr3);
1782
1783         if (mmu_reset_needed)
1784                 kvm_mmu_reset_context(vcpu);
1785
1786         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1787                sizeof vcpu->irq_pending);
1788         vcpu->irq_summary = 0;
1789         for (i = 0; i < NR_IRQ_WORDS; ++i)
1790                 if (vcpu->irq_pending[i])
1791                         __set_bit(i, &vcpu->irq_summary);
1792
1793         vcpu_put(vcpu);
1794
1795         return 0;
1796 }
1797
1798 /*
1799  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1800  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1801  *
1802  * This list is modified at module load time to reflect the
1803  * capabilities of the host cpu.
1804  */
1805 static u32 msrs_to_save[] = {
1806         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1807         MSR_K6_STAR,
1808 #ifdef CONFIG_X86_64
1809         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1810 #endif
1811         MSR_IA32_TIME_STAMP_COUNTER,
1812 };
1813
1814 static unsigned num_msrs_to_save;
1815
1816 static u32 emulated_msrs[] = {
1817         MSR_IA32_MISC_ENABLE,
1818 };
1819
1820 static __init void kvm_init_msr_list(void)
1821 {
1822         u32 dummy[2];
1823         unsigned i, j;
1824
1825         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1826                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1827                         continue;
1828                 if (j < i)
1829                         msrs_to_save[j] = msrs_to_save[i];
1830                 j++;
1831         }
1832         num_msrs_to_save = j;
1833 }
1834
1835 /*
1836  * Adapt set_msr() to msr_io()'s calling convention
1837  */
1838 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1839 {
1840         return set_msr(vcpu, index, *data);
1841 }
1842
1843 /*
1844  * Read or write a bunch of msrs. All parameters are kernel addresses.
1845  *
1846  * @return number of msrs set successfully.
1847  */
1848 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1849                     struct kvm_msr_entry *entries,
1850                     int (*do_msr)(struct kvm_vcpu *vcpu,
1851                                   unsigned index, u64 *data))
1852 {
1853         int i;
1854
1855         vcpu_load(vcpu);
1856
1857         for (i = 0; i < msrs->nmsrs; ++i)
1858                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1859                         break;
1860
1861         vcpu_put(vcpu);
1862
1863         return i;
1864 }
1865
1866 /*
1867  * Read or write a bunch of msrs. Parameters are user addresses.
1868  *
1869  * @return number of msrs set successfully.
1870  */
1871 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1872                   int (*do_msr)(struct kvm_vcpu *vcpu,
1873                                 unsigned index, u64 *data),
1874                   int writeback)
1875 {
1876         struct kvm_msrs msrs;
1877         struct kvm_msr_entry *entries;
1878         int r, n;
1879         unsigned size;
1880
1881         r = -EFAULT;
1882         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1883                 goto out;
1884
1885         r = -E2BIG;
1886         if (msrs.nmsrs >= MAX_IO_MSRS)
1887                 goto out;
1888
1889         r = -ENOMEM;
1890         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1891         entries = vmalloc(size);
1892         if (!entries)
1893                 goto out;
1894
1895         r = -EFAULT;
1896         if (copy_from_user(entries, user_msrs->entries, size))
1897                 goto out_free;
1898
1899         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1900         if (r < 0)
1901                 goto out_free;
1902
1903         r = -EFAULT;
1904         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1905                 goto out_free;
1906
1907         r = n;
1908
1909 out_free:
1910         vfree(entries);
1911 out:
1912         return r;
1913 }
1914
1915 /*
1916  * Translate a guest virtual address to a guest physical address.
1917  */
1918 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1919                                     struct kvm_translation *tr)
1920 {
1921         unsigned long vaddr = tr->linear_address;
1922         gpa_t gpa;
1923
1924         vcpu_load(vcpu);
1925         spin_lock(&vcpu->kvm->lock);
1926         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1927         tr->physical_address = gpa;
1928         tr->valid = gpa != UNMAPPED_GVA;
1929         tr->writeable = 1;
1930         tr->usermode = 0;
1931         spin_unlock(&vcpu->kvm->lock);
1932         vcpu_put(vcpu);
1933
1934         return 0;
1935 }
1936
1937 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1938                                     struct kvm_interrupt *irq)
1939 {
1940         if (irq->irq < 0 || irq->irq >= 256)
1941                 return -EINVAL;
1942         vcpu_load(vcpu);
1943
1944         set_bit(irq->irq, vcpu->irq_pending);
1945         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1946
1947         vcpu_put(vcpu);
1948
1949         return 0;
1950 }
1951
1952 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1953                                       struct kvm_debug_guest *dbg)
1954 {
1955         int r;
1956
1957         vcpu_load(vcpu);
1958
1959         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1960
1961         vcpu_put(vcpu);
1962
1963         return r;
1964 }
1965
1966 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1967                                     unsigned long address,
1968                                     int *type)
1969 {
1970         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1971         unsigned long pgoff;
1972         struct page *page;
1973
1974         *type = VM_FAULT_MINOR;
1975         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1976         if (pgoff != 0)
1977                 return NOPAGE_SIGBUS;
1978         page = virt_to_page(vcpu->run);
1979         get_page(page);
1980         return page;
1981 }
1982
1983 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1984         .nopage = kvm_vcpu_nopage,
1985 };
1986
1987 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1988 {
1989         vma->vm_ops = &kvm_vcpu_vm_ops;
1990         return 0;
1991 }
1992
1993 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1994 {
1995         struct kvm_vcpu *vcpu = filp->private_data;
1996
1997         fput(vcpu->kvm->filp);
1998         return 0;
1999 }
2000
2001 static struct file_operations kvm_vcpu_fops = {
2002         .release        = kvm_vcpu_release,
2003         .unlocked_ioctl = kvm_vcpu_ioctl,
2004         .compat_ioctl   = kvm_vcpu_ioctl,
2005         .mmap           = kvm_vcpu_mmap,
2006 };
2007
2008 /*
2009  * Allocates an inode for the vcpu.
2010  */
2011 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2012 {
2013         int fd, r;
2014         struct inode *inode;
2015         struct file *file;
2016
2017         atomic_inc(&vcpu->kvm->filp->f_count);
2018         inode = kvmfs_inode(&kvm_vcpu_fops);
2019         if (IS_ERR(inode)) {
2020                 r = PTR_ERR(inode);
2021                 goto out1;
2022         }
2023
2024         file = kvmfs_file(inode, vcpu);
2025         if (IS_ERR(file)) {
2026                 r = PTR_ERR(file);
2027                 goto out2;
2028         }
2029
2030         r = get_unused_fd();
2031         if (r < 0)
2032                 goto out3;
2033         fd = r;
2034         fd_install(fd, file);
2035
2036         return fd;
2037
2038 out3:
2039         fput(file);
2040 out2:
2041         iput(inode);
2042 out1:
2043         fput(vcpu->kvm->filp);
2044         return r;
2045 }
2046
2047 /*
2048  * Creates some virtual cpus.  Good luck creating more than one.
2049  */
2050 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2051 {
2052         int r;
2053         struct kvm_vcpu *vcpu;
2054         struct page *page;
2055
2056         r = -EINVAL;
2057         if (!valid_vcpu(n))
2058                 goto out;
2059
2060         vcpu = &kvm->vcpus[n];
2061
2062         mutex_lock(&vcpu->mutex);
2063
2064         if (vcpu->vmcs) {
2065                 mutex_unlock(&vcpu->mutex);
2066                 return -EEXIST;
2067         }
2068
2069         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2070         r = -ENOMEM;
2071         if (!page)
2072                 goto out_unlock;
2073         vcpu->run = page_address(page);
2074
2075         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2076                                            FX_IMAGE_ALIGN);
2077         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2078
2079         r = kvm_arch_ops->vcpu_create(vcpu);
2080         if (r < 0)
2081                 goto out_free_vcpus;
2082
2083         r = kvm_mmu_create(vcpu);
2084         if (r < 0)
2085                 goto out_free_vcpus;
2086
2087         kvm_arch_ops->vcpu_load(vcpu);
2088         r = kvm_mmu_setup(vcpu);
2089         if (r >= 0)
2090                 r = kvm_arch_ops->vcpu_setup(vcpu);
2091         vcpu_put(vcpu);
2092
2093         if (r < 0)
2094                 goto out_free_vcpus;
2095
2096         r = create_vcpu_fd(vcpu);
2097         if (r < 0)
2098                 goto out_free_vcpus;
2099
2100         return r;
2101
2102 out_free_vcpus:
2103         kvm_free_vcpu(vcpu);
2104 out_unlock:
2105         mutex_unlock(&vcpu->mutex);
2106 out:
2107         return r;
2108 }
2109
2110 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2111                                     struct kvm_cpuid *cpuid,
2112                                     struct kvm_cpuid_entry __user *entries)
2113 {
2114         int r;
2115
2116         r = -E2BIG;
2117         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2118                 goto out;
2119         r = -EFAULT;
2120         if (copy_from_user(&vcpu->cpuid_entries, entries,
2121                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2122                 goto out;
2123         vcpu->cpuid_nent = cpuid->nent;
2124         return 0;
2125
2126 out:
2127         return r;
2128 }
2129
2130 static long kvm_vcpu_ioctl(struct file *filp,
2131                            unsigned int ioctl, unsigned long arg)
2132 {
2133         struct kvm_vcpu *vcpu = filp->private_data;
2134         void __user *argp = (void __user *)arg;
2135         int r = -EINVAL;
2136
2137         switch (ioctl) {
2138         case KVM_RUN:
2139                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2140                 break;
2141         case KVM_GET_REGS: {
2142                 struct kvm_regs kvm_regs;
2143
2144                 memset(&kvm_regs, 0, sizeof kvm_regs);
2145                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2146                 if (r)
2147                         goto out;
2148                 r = -EFAULT;
2149                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2150                         goto out;
2151                 r = 0;
2152                 break;
2153         }
2154         case KVM_SET_REGS: {
2155                 struct kvm_regs kvm_regs;
2156
2157                 r = -EFAULT;
2158                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2159                         goto out;
2160                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2161                 if (r)
2162                         goto out;
2163                 r = 0;
2164                 break;
2165         }
2166         case KVM_GET_SREGS: {
2167                 struct kvm_sregs kvm_sregs;
2168
2169                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2170                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2171                 if (r)
2172                         goto out;
2173                 r = -EFAULT;
2174                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2175                         goto out;
2176                 r = 0;
2177                 break;
2178         }
2179         case KVM_SET_SREGS: {
2180                 struct kvm_sregs kvm_sregs;
2181
2182                 r = -EFAULT;
2183                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2184                         goto out;
2185                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2186                 if (r)
2187                         goto out;
2188                 r = 0;
2189                 break;
2190         }
2191         case KVM_TRANSLATE: {
2192                 struct kvm_translation tr;
2193
2194                 r = -EFAULT;
2195                 if (copy_from_user(&tr, argp, sizeof tr))
2196                         goto out;
2197                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2198                 if (r)
2199                         goto out;
2200                 r = -EFAULT;
2201                 if (copy_to_user(argp, &tr, sizeof tr))
2202                         goto out;
2203                 r = 0;
2204                 break;
2205         }
2206         case KVM_INTERRUPT: {
2207                 struct kvm_interrupt irq;
2208
2209                 r = -EFAULT;
2210                 if (copy_from_user(&irq, argp, sizeof irq))
2211                         goto out;
2212                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2213                 if (r)
2214                         goto out;
2215                 r = 0;
2216                 break;
2217         }
2218         case KVM_DEBUG_GUEST: {
2219                 struct kvm_debug_guest dbg;
2220
2221                 r = -EFAULT;
2222                 if (copy_from_user(&dbg, argp, sizeof dbg))
2223                         goto out;
2224                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2225                 if (r)
2226                         goto out;
2227                 r = 0;
2228                 break;
2229         }
2230         case KVM_GET_MSRS:
2231                 r = msr_io(vcpu, argp, get_msr, 1);
2232                 break;
2233         case KVM_SET_MSRS:
2234                 r = msr_io(vcpu, argp, do_set_msr, 0);
2235                 break;
2236         case KVM_SET_CPUID: {
2237                 struct kvm_cpuid __user *cpuid_arg = argp;
2238                 struct kvm_cpuid cpuid;
2239
2240                 r = -EFAULT;
2241                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2242                         goto out;
2243                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2244                 if (r)
2245                         goto out;
2246                 break;
2247         }
2248         default:
2249                 ;
2250         }
2251 out:
2252         return r;
2253 }
2254
2255 static long kvm_vm_ioctl(struct file *filp,
2256                            unsigned int ioctl, unsigned long arg)
2257 {
2258         struct kvm *kvm = filp->private_data;
2259         void __user *argp = (void __user *)arg;
2260         int r = -EINVAL;
2261
2262         switch (ioctl) {
2263         case KVM_CREATE_VCPU:
2264                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2265                 if (r < 0)
2266                         goto out;
2267                 break;
2268         case KVM_SET_MEMORY_REGION: {
2269                 struct kvm_memory_region kvm_mem;
2270
2271                 r = -EFAULT;
2272                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2273                         goto out;
2274                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2275                 if (r)
2276                         goto out;
2277                 break;
2278         }
2279         case KVM_GET_DIRTY_LOG: {
2280                 struct kvm_dirty_log log;
2281
2282                 r = -EFAULT;
2283                 if (copy_from_user(&log, argp, sizeof log))
2284                         goto out;
2285                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2286                 if (r)
2287                         goto out;
2288                 break;
2289         }
2290         default:
2291                 ;
2292         }
2293 out:
2294         return r;
2295 }
2296
2297 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2298                                   unsigned long address,
2299                                   int *type)
2300 {
2301         struct kvm *kvm = vma->vm_file->private_data;
2302         unsigned long pgoff;
2303         struct kvm_memory_slot *slot;
2304         struct page *page;
2305
2306         *type = VM_FAULT_MINOR;
2307         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2308         slot = gfn_to_memslot(kvm, pgoff);
2309         if (!slot)
2310                 return NOPAGE_SIGBUS;
2311         page = gfn_to_page(slot, pgoff);
2312         if (!page)
2313                 return NOPAGE_SIGBUS;
2314         get_page(page);
2315         return page;
2316 }
2317
2318 static struct vm_operations_struct kvm_vm_vm_ops = {
2319         .nopage = kvm_vm_nopage,
2320 };
2321
2322 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2323 {
2324         vma->vm_ops = &kvm_vm_vm_ops;
2325         return 0;
2326 }
2327
2328 static struct file_operations kvm_vm_fops = {
2329         .release        = kvm_vm_release,
2330         .unlocked_ioctl = kvm_vm_ioctl,
2331         .compat_ioctl   = kvm_vm_ioctl,
2332         .mmap           = kvm_vm_mmap,
2333 };
2334
2335 static int kvm_dev_ioctl_create_vm(void)
2336 {
2337         int fd, r;
2338         struct inode *inode;
2339         struct file *file;
2340         struct kvm *kvm;
2341
2342         inode = kvmfs_inode(&kvm_vm_fops);
2343         if (IS_ERR(inode)) {
2344                 r = PTR_ERR(inode);
2345                 goto out1;
2346         }
2347
2348         kvm = kvm_create_vm();
2349         if (IS_ERR(kvm)) {
2350                 r = PTR_ERR(kvm);
2351                 goto out2;
2352         }
2353
2354         file = kvmfs_file(inode, kvm);
2355         if (IS_ERR(file)) {
2356                 r = PTR_ERR(file);
2357                 goto out3;
2358         }
2359         kvm->filp = file;
2360
2361         r = get_unused_fd();
2362         if (r < 0)
2363                 goto out4;
2364         fd = r;
2365         fd_install(fd, file);
2366
2367         return fd;
2368
2369 out4:
2370         fput(file);
2371 out3:
2372         kvm_destroy_vm(kvm);
2373 out2:
2374         iput(inode);
2375 out1:
2376         return r;
2377 }
2378
2379 static long kvm_dev_ioctl(struct file *filp,
2380                           unsigned int ioctl, unsigned long arg)
2381 {
2382         void __user *argp = (void __user *)arg;
2383         int r = -EINVAL;
2384
2385         switch (ioctl) {
2386         case KVM_GET_API_VERSION:
2387                 r = KVM_API_VERSION;
2388                 break;
2389         case KVM_CREATE_VM:
2390                 r = kvm_dev_ioctl_create_vm();
2391                 break;
2392         case KVM_GET_MSR_INDEX_LIST: {
2393                 struct kvm_msr_list __user *user_msr_list = argp;
2394                 struct kvm_msr_list msr_list;
2395                 unsigned n;
2396
2397                 r = -EFAULT;
2398                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2399                         goto out;
2400                 n = msr_list.nmsrs;
2401                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2402                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2403                         goto out;
2404                 r = -E2BIG;
2405                 if (n < num_msrs_to_save)
2406                         goto out;
2407                 r = -EFAULT;
2408                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2409                                  num_msrs_to_save * sizeof(u32)))
2410                         goto out;
2411                 if (copy_to_user(user_msr_list->indices
2412                                  + num_msrs_to_save * sizeof(u32),
2413                                  &emulated_msrs,
2414                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2415                         goto out;
2416                 r = 0;
2417                 break;
2418         }
2419         case KVM_CHECK_EXTENSION:
2420                 /*
2421                  * No extensions defined at present.
2422                  */
2423                 r = 0;
2424                 break;
2425         default:
2426                 ;
2427         }
2428 out:
2429         return r;
2430 }
2431
2432 static struct file_operations kvm_chardev_ops = {
2433         .open           = kvm_dev_open,
2434         .release        = kvm_dev_release,
2435         .unlocked_ioctl = kvm_dev_ioctl,
2436         .compat_ioctl   = kvm_dev_ioctl,
2437 };
2438
2439 static struct miscdevice kvm_dev = {
2440         KVM_MINOR,
2441         "kvm",
2442         &kvm_chardev_ops,
2443 };
2444
2445 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2446                        void *v)
2447 {
2448         if (val == SYS_RESTART) {
2449                 /*
2450                  * Some (well, at least mine) BIOSes hang on reboot if
2451                  * in vmx root mode.
2452                  */
2453                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2454                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2455         }
2456         return NOTIFY_OK;
2457 }
2458
2459 static struct notifier_block kvm_reboot_notifier = {
2460         .notifier_call = kvm_reboot,
2461         .priority = 0,
2462 };
2463
2464 /*
2465  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2466  * cached on it.
2467  */
2468 static void decache_vcpus_on_cpu(int cpu)
2469 {
2470         struct kvm *vm;
2471         struct kvm_vcpu *vcpu;
2472         int i;
2473
2474         spin_lock(&kvm_lock);
2475         list_for_each_entry(vm, &vm_list, vm_list)
2476                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2477                         vcpu = &vm->vcpus[i];
2478                         /*
2479                          * If the vcpu is locked, then it is running on some
2480                          * other cpu and therefore it is not cached on the
2481                          * cpu in question.
2482                          *
2483                          * If it's not locked, check the last cpu it executed
2484                          * on.
2485                          */
2486                         if (mutex_trylock(&vcpu->mutex)) {
2487                                 if (vcpu->cpu == cpu) {
2488                                         kvm_arch_ops->vcpu_decache(vcpu);
2489                                         vcpu->cpu = -1;
2490                                 }
2491                                 mutex_unlock(&vcpu->mutex);
2492                         }
2493                 }
2494         spin_unlock(&kvm_lock);
2495 }
2496
2497 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2498                            void *v)
2499 {
2500         int cpu = (long)v;
2501
2502         switch (val) {
2503         case CPU_DOWN_PREPARE:
2504         case CPU_UP_CANCELED:
2505                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2506                        cpu);
2507                 decache_vcpus_on_cpu(cpu);
2508                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2509                                          NULL, 0, 1);
2510                 break;
2511         case CPU_ONLINE:
2512                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2513                        cpu);
2514                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2515                                          NULL, 0, 1);
2516                 break;
2517         }
2518         return NOTIFY_OK;
2519 }
2520
2521 static struct notifier_block kvm_cpu_notifier = {
2522         .notifier_call = kvm_cpu_hotplug,
2523         .priority = 20, /* must be > scheduler priority */
2524 };
2525
2526 static __init void kvm_init_debug(void)
2527 {
2528         struct kvm_stats_debugfs_item *p;
2529
2530         debugfs_dir = debugfs_create_dir("kvm", NULL);
2531         for (p = debugfs_entries; p->name; ++p)
2532                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2533                                                p->data);
2534 }
2535
2536 static void kvm_exit_debug(void)
2537 {
2538         struct kvm_stats_debugfs_item *p;
2539
2540         for (p = debugfs_entries; p->name; ++p)
2541                 debugfs_remove(p->dentry);
2542         debugfs_remove(debugfs_dir);
2543 }
2544
2545 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2546 {
2547         decache_vcpus_on_cpu(raw_smp_processor_id());
2548         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2549         return 0;
2550 }
2551
2552 static int kvm_resume(struct sys_device *dev)
2553 {
2554         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2555         return 0;
2556 }
2557
2558 static struct sysdev_class kvm_sysdev_class = {
2559         set_kset_name("kvm"),
2560         .suspend = kvm_suspend,
2561         .resume = kvm_resume,
2562 };
2563
2564 static struct sys_device kvm_sysdev = {
2565         .id = 0,
2566         .cls = &kvm_sysdev_class,
2567 };
2568
2569 hpa_t bad_page_address;
2570
2571 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2572                         const char *dev_name, void *data, struct vfsmount *mnt)
2573 {
2574         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2575 }
2576
2577 static struct file_system_type kvm_fs_type = {
2578         .name           = "kvmfs",
2579         .get_sb         = kvmfs_get_sb,
2580         .kill_sb        = kill_anon_super,
2581 };
2582
2583 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2584 {
2585         int r;
2586
2587         if (kvm_arch_ops) {
2588                 printk(KERN_ERR "kvm: already loaded the other module\n");
2589                 return -EEXIST;
2590         }
2591
2592         if (!ops->cpu_has_kvm_support()) {
2593                 printk(KERN_ERR "kvm: no hardware support\n");
2594                 return -EOPNOTSUPP;
2595         }
2596         if (ops->disabled_by_bios()) {
2597                 printk(KERN_ERR "kvm: disabled by bios\n");
2598                 return -EOPNOTSUPP;
2599         }
2600
2601         kvm_arch_ops = ops;
2602
2603         r = kvm_arch_ops->hardware_setup();
2604         if (r < 0)
2605                 goto out;
2606
2607         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2608         r = register_cpu_notifier(&kvm_cpu_notifier);
2609         if (r)
2610                 goto out_free_1;
2611         register_reboot_notifier(&kvm_reboot_notifier);
2612
2613         r = sysdev_class_register(&kvm_sysdev_class);
2614         if (r)
2615                 goto out_free_2;
2616
2617         r = sysdev_register(&kvm_sysdev);
2618         if (r)
2619                 goto out_free_3;
2620
2621         kvm_chardev_ops.owner = module;
2622
2623         r = misc_register(&kvm_dev);
2624         if (r) {
2625                 printk (KERN_ERR "kvm: misc device register failed\n");
2626                 goto out_free;
2627         }
2628
2629         return r;
2630
2631 out_free:
2632         sysdev_unregister(&kvm_sysdev);
2633 out_free_3:
2634         sysdev_class_unregister(&kvm_sysdev_class);
2635 out_free_2:
2636         unregister_reboot_notifier(&kvm_reboot_notifier);
2637         unregister_cpu_notifier(&kvm_cpu_notifier);
2638 out_free_1:
2639         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2640         kvm_arch_ops->hardware_unsetup();
2641 out:
2642         kvm_arch_ops = NULL;
2643         return r;
2644 }
2645
2646 void kvm_exit_arch(void)
2647 {
2648         misc_deregister(&kvm_dev);
2649         sysdev_unregister(&kvm_sysdev);
2650         sysdev_class_unregister(&kvm_sysdev_class);
2651         unregister_reboot_notifier(&kvm_reboot_notifier);
2652         unregister_cpu_notifier(&kvm_cpu_notifier);
2653         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2654         kvm_arch_ops->hardware_unsetup();
2655         kvm_arch_ops = NULL;
2656 }
2657
2658 static __init int kvm_init(void)
2659 {
2660         static struct page *bad_page;
2661         int r;
2662
2663         r = register_filesystem(&kvm_fs_type);
2664         if (r)
2665                 goto out3;
2666
2667         kvmfs_mnt = kern_mount(&kvm_fs_type);
2668         r = PTR_ERR(kvmfs_mnt);
2669         if (IS_ERR(kvmfs_mnt))
2670                 goto out2;
2671         kvm_init_debug();
2672
2673         kvm_init_msr_list();
2674
2675         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2676                 r = -ENOMEM;
2677                 goto out;
2678         }
2679
2680         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2681         memset(__va(bad_page_address), 0, PAGE_SIZE);
2682
2683         return 0;
2684
2685 out:
2686         kvm_exit_debug();
2687         mntput(kvmfs_mnt);
2688 out2:
2689         unregister_filesystem(&kvm_fs_type);
2690 out3:
2691         return r;
2692 }
2693
2694 static __exit void kvm_exit(void)
2695 {
2696         kvm_exit_debug();
2697         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2698         mntput(kvmfs_mnt);
2699         unregister_filesystem(&kvm_fs_type);
2700 }
2701
2702 module_init(kvm_init)
2703 module_exit(kvm_exit)
2704
2705 EXPORT_SYMBOL_GPL(kvm_init_arch);
2706 EXPORT_SYMBOL_GPL(kvm_exit_arch);