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 <asm/processor.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <asm/msr.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/reboot.h>
  32 #include <asm/io.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <asm/desc.h>
  37
  38 #include "x86_emulate.h"
  39 #include "segment_descriptor.h"
  40
  41 MODULE_AUTHOR("Qumranet");
  42 MODULE_LICENSE("GPL");
  43
  44 struct kvm_arch_ops *kvm_arch_ops;
  45 struct kvm_stat kvm_stat;
  46 EXPORT_SYMBOL_GPL(kvm_stat);
  47
  48 static struct kvm_stats_debugfs_item {
  49         const char *name;
  50         u32 *data;
  51         struct dentry *dentry;
  52 } debugfs_entries[] = {
  53         { "pf_fixed", &kvm_stat.pf_fixed },
  54         { "pf_guest", &kvm_stat.pf_guest },
  55         { "tlb_flush", &kvm_stat.tlb_flush },
  56         { "invlpg", &kvm_stat.invlpg },
  57         { "exits", &kvm_stat.exits },
  58         { "io_exits", &kvm_stat.io_exits },
  59         { "mmio_exits", &kvm_stat.mmio_exits },
  60         { "signal_exits", &kvm_stat.signal_exits },
  61         { "irq_window", &kvm_stat.irq_window_exits },
  62         { "halt_exits", &kvm_stat.halt_exits },
  63         { "request_irq", &kvm_stat.request_irq_exits },
  64         { "irq_exits", &kvm_stat.irq_exits },
  65         { 0, 0 }
  66 };
  67
  68 static struct dentry *debugfs_dir;
  69
  70 #define MAX_IO_MSRS 256
  71
  72 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  73 #define LMSW_GUEST_MASK 0x0eULL
  74 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  75 #define CR8_RESEVED_BITS (~0x0fULL)
  76 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  77
  78 #ifdef CONFIG_X86_64
  79 // LDT or TSS descriptor in the GDT. 16 bytes.
  80 struct segment_descriptor_64 {
  81         struct segment_descriptor s;
  82         u32 base_higher;
  83         u32 pad_zero;
  84 };
  85
  86 #endif
  87
  88 unsigned long segment_base(u16 selector)
  89 {
  90         struct descriptor_table gdt;
  91         struct segment_descriptor *d;
  92         unsigned long table_base;
  93         typedef unsigned long ul;
  94         unsigned long v;
  95
  96         if (selector == 0)
  97                 return 0;
  98
  99         asm ("sgdt %0" : "=m"(gdt));
 100         table_base = gdt.base;
 101
 102         if (selector & 4) {           /* from ldt */
 103                 u16 ldt_selector;
 104
 105                 asm ("sldt %0" : "=g"(ldt_selector));
 106                 table_base = segment_base(ldt_selector);
 107         }
 108         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 109         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 110 #ifdef CONFIG_X86_64
 111         if (d->system == 0
 112             && (d->type == 2 || d->type == 9 || d->type == 11))
 113                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 114 #endif
 115         return v;
 116 }
 117 EXPORT_SYMBOL_GPL(segment_base);
 118
 119 static inline int valid_vcpu(int n)
 120 {
 121         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 122 }
 123
 124 int kvm_read_guest(struct kvm_vcpu *vcpu,
 125                              gva_t addr,
 126                              unsigned long size,
 127                              void *dest)
 128 {
 129         unsigned char *host_buf = dest;
 130         unsigned long req_size = size;
 131
 132         while (size) {
 133                 hpa_t paddr;
 134                 unsigned now;
 135                 unsigned offset;
 136                 hva_t guest_buf;
 137
 138                 paddr = gva_to_hpa(vcpu, addr);
 139
 140                 if (is_error_hpa(paddr))
 141                         break;
 142
 143                 guest_buf = (hva_t)kmap_atomic(
 144                                         pfn_to_page(paddr >> PAGE_SHIFT),
 145                                         KM_USER0);
 146                 offset = addr & ~PAGE_MASK;
 147                 guest_buf |= offset;
 148                 now = min(size, PAGE_SIZE - offset);
 149                 memcpy(host_buf, (void*)guest_buf, now);
 150                 host_buf += now;
 151                 addr += now;
 152                 size -= now;
 153                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 154         }
 155         return req_size - size;
 156 }
 157 EXPORT_SYMBOL_GPL(kvm_read_guest);
 158
 159 int kvm_write_guest(struct kvm_vcpu *vcpu,
 160                              gva_t addr,
 161                              unsigned long size,
 162                              void *data)
 163 {
 164         unsigned char *host_buf = data;
 165         unsigned long req_size = size;
 166
 167         while (size) {
 168                 hpa_t paddr;
 169                 unsigned now;
 170                 unsigned offset;
 171                 hva_t guest_buf;
 172
 173                 paddr = gva_to_hpa(vcpu, addr);
 174
 175                 if (is_error_hpa(paddr))
 176                         break;
 177
 178                 guest_buf = (hva_t)kmap_atomic(
 179                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 180                 offset = addr & ~PAGE_MASK;
 181                 guest_buf |= offset;
 182                 now = min(size, PAGE_SIZE - offset);
 183                 memcpy((void*)guest_buf, host_buf, now);
 184                 host_buf += now;
 185                 addr += now;
 186                 size -= now;
 187                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 188         }
 189         return req_size - size;
 190 }
 191 EXPORT_SYMBOL_GPL(kvm_write_guest);
 192
 193 static int vcpu_slot(struct kvm_vcpu *vcpu)
 194 {
 195         return vcpu - vcpu->kvm->vcpus;
 196 }
 197
 198 /*
 199  * Switches to specified vcpu, until a matching vcpu_put()
 200  */
 201 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
 202 {
 203         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
 204
 205         mutex_lock(&vcpu->mutex);
 206         if (unlikely(!vcpu->vmcs)) {
 207                 mutex_unlock(&vcpu->mutex);
 208                 return 0;
 209         }
 210         return kvm_arch_ops->vcpu_load(vcpu);
 211 }
 212
 213 static void vcpu_put(struct kvm_vcpu *vcpu)
 214 {
 215         kvm_arch_ops->vcpu_put(vcpu);
 216         mutex_unlock(&vcpu->mutex);
 217 }
 218
 219 static int kvm_dev_open(struct inode *inode, struct file *filp)
 220 {
 221         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 222         int i;
 223
 224         if (!kvm)
 225                 return -ENOMEM;
 226
 227         spin_lock_init(&kvm->lock);
 228         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 229         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 230                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 231
 232                 mutex_init(&vcpu->mutex);
 233                 vcpu->mmu.root_hpa = INVALID_PAGE;
 234                 INIT_LIST_HEAD(&vcpu->free_pages);
 235         }
 236         filp->private_data = kvm;
 237         return 0;
 238 }
 239
 240 /*
 241  * Free any memory in @free but not in @dont.
 242  */
 243 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 244                                   struct kvm_memory_slot *dont)
 245 {
 246         int i;
 247
 248         if (!dont || free->phys_mem != dont->phys_mem)
 249                 if (free->phys_mem) {
 250                         for (i = 0; i < free->npages; ++i)
 251                                 if (free->phys_mem[i])
 252                                         __free_page(free->phys_mem[i]);
 253                         vfree(free->phys_mem);
 254                 }
 255
 256         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 257                 vfree(free->dirty_bitmap);
 258
 259         free->phys_mem = 0;
 260         free->npages = 0;
 261         free->dirty_bitmap = 0;
 262 }
 263
 264 static void kvm_free_physmem(struct kvm *kvm)
 265 {
 266         int i;
 267
 268         for (i = 0; i < kvm->nmemslots; ++i)
 269                 kvm_free_physmem_slot(&kvm->memslots[i], 0);
 270 }
 271
 272 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 273 {
 274         kvm_arch_ops->vcpu_free(vcpu);
 275         kvm_mmu_destroy(vcpu);
 276 }
 277
 278 static void kvm_free_vcpus(struct kvm *kvm)
 279 {
 280         unsigned int i;
 281
 282         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 283                 kvm_free_vcpu(&kvm->vcpus[i]);
 284 }
 285
 286 static int kvm_dev_release(struct inode *inode, struct file *filp)
 287 {
 288         struct kvm *kvm = filp->private_data;
 289
 290         kvm_free_vcpus(kvm);
 291         kvm_free_physmem(kvm);
 292         kfree(kvm);
 293         return 0;
 294 }
 295
 296 static void inject_gp(struct kvm_vcpu *vcpu)
 297 {
 298         kvm_arch_ops->inject_gp(vcpu, 0);
 299 }
 300
 301 /*
 302  * Load the pae pdptrs.  Return true is they are all valid.
 303  */
 304 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 305 {
 306         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 307         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 308         int i;
 309         u64 pdpte;
 310         u64 *pdpt;
 311         int ret;
 312         struct kvm_memory_slot *memslot;
 313
 314         spin_lock(&vcpu->kvm->lock);
 315         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 316         /* FIXME: !memslot - emulate? 0xff? */
 317         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 318
 319         ret = 1;
 320         for (i = 0; i < 4; ++i) {
 321                 pdpte = pdpt[offset + i];
 322                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 323                         ret = 0;
 324                         goto out;
 325                 }
 326         }
 327
 328         for (i = 0; i < 4; ++i)
 329                 vcpu->pdptrs[i] = pdpt[offset + i];
 330
 331 out:
 332         kunmap_atomic(pdpt, KM_USER0);
 333         spin_unlock(&vcpu->kvm->lock);
 334
 335         return ret;
 336 }
 337
 338 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 339 {
 340         if (cr0 & CR0_RESEVED_BITS) {
 341                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 342                        cr0, vcpu->cr0);
 343                 inject_gp(vcpu);
 344                 return;
 345         }
 346
 347         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 348                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 349                 inject_gp(vcpu);
 350                 return;
 351         }
 352
 353         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 354                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 355                        "and a clear PE flag\n");
 356                 inject_gp(vcpu);
 357                 return;
 358         }
 359
 360         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 361 #ifdef CONFIG_X86_64
 362                 if ((vcpu->shadow_efer & EFER_LME)) {
 363                         int cs_db, cs_l;
 364
 365                         if (!is_pae(vcpu)) {
 366                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 367                                        "in long mode while PAE is disabled\n");
 368                                 inject_gp(vcpu);
 369                                 return;
 370                         }
 371                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 372                         if (cs_l) {
 373                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 374                                        "in long mode while CS.L == 1\n");
 375                                 inject_gp(vcpu);
 376                                 return;
 377
 378                         }
 379                 } else
 380 #endif
 381                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 382                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 383                                "reserved bits\n");
 384                         inject_gp(vcpu);
 385                         return;
 386                 }
 387
 388         }
 389
 390         kvm_arch_ops->set_cr0(vcpu, cr0);
 391         vcpu->cr0 = cr0;
 392
 393         spin_lock(&vcpu->kvm->lock);
 394         kvm_mmu_reset_context(vcpu);
 395         spin_unlock(&vcpu->kvm->lock);
 396         return;
 397 }
 398 EXPORT_SYMBOL_GPL(set_cr0);
 399
 400 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 401 {
 402         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 403         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 404 }
 405 EXPORT_SYMBOL_GPL(lmsw);
 406
 407 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 408 {
 409         if (cr4 & CR4_RESEVED_BITS) {
 410                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 411                 inject_gp(vcpu);
 412                 return;
 413         }
 414
 415         if (is_long_mode(vcpu)) {
 416                 if (!(cr4 & CR4_PAE_MASK)) {
 417                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 418                                "in long mode\n");
 419                         inject_gp(vcpu);
 420                         return;
 421                 }
 422         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 423                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 424                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 425                 inject_gp(vcpu);
 426         }
 427
 428         if (cr4 & CR4_VMXE_MASK) {
 429                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 430                 inject_gp(vcpu);
 431                 return;
 432         }
 433         kvm_arch_ops->set_cr4(vcpu, cr4);
 434         spin_lock(&vcpu->kvm->lock);
 435         kvm_mmu_reset_context(vcpu);
 436         spin_unlock(&vcpu->kvm->lock);
 437 }
 438 EXPORT_SYMBOL_GPL(set_cr4);
 439
 440 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 441 {
 442         if (is_long_mode(vcpu)) {
 443                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
 444                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 445                         inject_gp(vcpu);
 446                         return;
 447                 }
 448         } else {
 449                 if (cr3 & CR3_RESEVED_BITS) {
 450                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 451                         inject_gp(vcpu);
 452                         return;
 453                 }
 454                 if (is_paging(vcpu) && is_pae(vcpu) &&
 455                     !load_pdptrs(vcpu, cr3)) {
 456                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 457                                "reserved bits\n");
 458                         inject_gp(vcpu);
 459                         return;
 460                 }
 461         }
 462
 463         vcpu->cr3 = cr3;
 464         spin_lock(&vcpu->kvm->lock);
 465         vcpu->mmu.new_cr3(vcpu);
 466         spin_unlock(&vcpu->kvm->lock);
 467 }
 468 EXPORT_SYMBOL_GPL(set_cr3);
 469
 470 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 471 {
 472         if ( cr8 & CR8_RESEVED_BITS) {
 473                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 474                 inject_gp(vcpu);
 475                 return;
 476         }
 477         vcpu->cr8 = cr8;
 478 }
 479 EXPORT_SYMBOL_GPL(set_cr8);
 480
 481 void fx_init(struct kvm_vcpu *vcpu)
 482 {
 483         struct __attribute__ ((__packed__)) fx_image_s {
 484                 u16 control; //fcw
 485                 u16 status; //fsw
 486                 u16 tag; // ftw
 487                 u16 opcode; //fop
 488                 u64 ip; // fpu ip
 489                 u64 operand;// fpu dp
 490                 u32 mxcsr;
 491                 u32 mxcsr_mask;
 492
 493         } *fx_image;
 494
 495         fx_save(vcpu->host_fx_image);
 496         fpu_init();
 497         fx_save(vcpu->guest_fx_image);
 498         fx_restore(vcpu->host_fx_image);
 499
 500         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 501         fx_image->mxcsr = 0x1f80;
 502         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 503                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 504 }
 505 EXPORT_SYMBOL_GPL(fx_init);
 506
 507 /*
 508  * Creates some virtual cpus.  Good luck creating more than one.
 509  */
 510 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 511 {
 512         int r;
 513         struct kvm_vcpu *vcpu;
 514
 515         r = -EINVAL;
 516         if (!valid_vcpu(n))
 517                 goto out;
 518
 519         vcpu = &kvm->vcpus[n];
 520
 521         mutex_lock(&vcpu->mutex);
 522
 523         if (vcpu->vmcs) {
 524                 mutex_unlock(&vcpu->mutex);
 525                 return -EEXIST;
 526         }
 527
 528         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 529                                            FX_IMAGE_ALIGN);
 530         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 531
 532         vcpu->cpu = -1;  /* First load will set up TR */
 533         vcpu->kvm = kvm;
 534         r = kvm_arch_ops->vcpu_create(vcpu);
 535         if (r < 0)
 536                 goto out_free_vcpus;
 537
 538         r = kvm_mmu_create(vcpu);
 539         if (r < 0)
 540                 goto out_free_vcpus;
 541
 542         kvm_arch_ops->vcpu_load(vcpu);
 543         r = kvm_mmu_setup(vcpu);
 544         if (r >= 0)
 545                 r = kvm_arch_ops->vcpu_setup(vcpu);
 546         vcpu_put(vcpu);
 547
 548         if (r < 0)
 549                 goto out_free_vcpus;
 550
 551         return 0;
 552
 553 out_free_vcpus:
 554         kvm_free_vcpu(vcpu);
 555         mutex_unlock(&vcpu->mutex);
 556 out:
 557         return r;
 558 }
 559
 560 /*
 561  * Allocate some memory and give it an address in the guest physical address
 562  * space.
 563  *
 564  * Discontiguous memory is allowed, mostly for framebuffers.
 565  */
 566 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 567                                            struct kvm_memory_region *mem)
 568 {
 569         int r;
 570         gfn_t base_gfn;
 571         unsigned long npages;
 572         unsigned long i;
 573         struct kvm_memory_slot *memslot;
 574         struct kvm_memory_slot old, new;
 575         int memory_config_version;
 576
 577         r = -EINVAL;
 578         /* General sanity checks */
 579         if (mem->memory_size & (PAGE_SIZE - 1))
 580                 goto out;
 581         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 582                 goto out;
 583         if (mem->slot >= KVM_MEMORY_SLOTS)
 584                 goto out;
 585         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 586                 goto out;
 587
 588         memslot = &kvm->memslots[mem->slot];
 589         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 590         npages = mem->memory_size >> PAGE_SHIFT;
 591
 592         if (!npages)
 593                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 594
 595 raced:
 596         spin_lock(&kvm->lock);
 597
 598         memory_config_version = kvm->memory_config_version;
 599         new = old = *memslot;
 600
 601         new.base_gfn = base_gfn;
 602         new.npages = npages;
 603         new.flags = mem->flags;
 604
 605         /* Disallow changing a memory slot's size. */
 606         r = -EINVAL;
 607         if (npages && old.npages && npages != old.npages)
 608                 goto out_unlock;
 609
 610         /* Check for overlaps */
 611         r = -EEXIST;
 612         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 613                 struct kvm_memory_slot *s = &kvm->memslots[i];
 614
 615                 if (s == memslot)
 616                         continue;
 617                 if (!((base_gfn + npages <= s->base_gfn) ||
 618                       (base_gfn >= s->base_gfn + s->npages)))
 619                         goto out_unlock;
 620         }
 621         /*
 622          * Do memory allocations outside lock.  memory_config_version will
 623          * detect any races.
 624          */
 625         spin_unlock(&kvm->lock);
 626
 627         /* Deallocate if slot is being removed */
 628         if (!npages)
 629                 new.phys_mem = 0;
 630
 631         /* Free page dirty bitmap if unneeded */
 632         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 633                 new.dirty_bitmap = 0;
 634
 635         r = -ENOMEM;
 636
 637         /* Allocate if a slot is being created */
 638         if (npages && !new.phys_mem) {
 639                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 640
 641                 if (!new.phys_mem)
 642                         goto out_free;
 643
 644                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 645                 for (i = 0; i < npages; ++i) {
 646                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 647                                                      | __GFP_ZERO);
 648                         if (!new.phys_mem[i])
 649                                 goto out_free;
 650                         new.phys_mem[i]->private = 0;
 651                 }
 652         }
 653
 654         /* Allocate page dirty bitmap if needed */
 655         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 656                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 657
 658                 new.dirty_bitmap = vmalloc(dirty_bytes);
 659                 if (!new.dirty_bitmap)
 660                         goto out_free;
 661                 memset(new.dirty_bitmap, 0, dirty_bytes);
 662         }
 663
 664         spin_lock(&kvm->lock);
 665
 666         if (memory_config_version != kvm->memory_config_version) {
 667                 spin_unlock(&kvm->lock);
 668                 kvm_free_physmem_slot(&new, &old);
 669                 goto raced;
 670         }
 671
 672         r = -EAGAIN;
 673         if (kvm->busy)
 674                 goto out_unlock;
 675
 676         if (mem->slot >= kvm->nmemslots)
 677                 kvm->nmemslots = mem->slot + 1;
 678
 679         *memslot = new;
 680         ++kvm->memory_config_version;
 681
 682         spin_unlock(&kvm->lock);
 683
 684         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 685                 struct kvm_vcpu *vcpu;
 686
 687                 vcpu = vcpu_load(kvm, i);
 688                 if (!vcpu)
 689                         continue;
 690                 kvm_mmu_reset_context(vcpu);
 691                 vcpu_put(vcpu);
 692         }
 693
 694         kvm_free_physmem_slot(&old, &new);
 695         return 0;
 696
 697 out_unlock:
 698         spin_unlock(&kvm->lock);
 699 out_free:
 700         kvm_free_physmem_slot(&new, &old);
 701 out:
 702         return r;
 703 }
 704
 705 /*
 706  * Get (and clear) the dirty memory log for a memory slot.
 707  */
 708 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 709                                        struct kvm_dirty_log *log)
 710 {
 711         struct kvm_memory_slot *memslot;
 712         int r, i;
 713         int n;
 714         unsigned long any = 0;
 715
 716         spin_lock(&kvm->lock);
 717
 718         /*
 719          * Prevent changes to guest memory configuration even while the lock
 720          * is not taken.
 721          */
 722         ++kvm->busy;
 723         spin_unlock(&kvm->lock);
 724         r = -EINVAL;
 725         if (log->slot >= KVM_MEMORY_SLOTS)
 726                 goto out;
 727
 728         memslot = &kvm->memslots[log->slot];
 729         r = -ENOENT;
 730         if (!memslot->dirty_bitmap)
 731                 goto out;
 732
 733         n = ALIGN(memslot->npages, 8) / 8;
 734
 735         for (i = 0; !any && i < n; ++i)
 736                 any = memslot->dirty_bitmap[i];
 737
 738         r = -EFAULT;
 739         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 740                 goto out;
 741
 742
 743         if (any) {
 744                 spin_lock(&kvm->lock);
 745                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 746                 spin_unlock(&kvm->lock);
 747                 memset(memslot->dirty_bitmap, 0, n);
 748                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 749                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 750
 751                         if (!vcpu)
 752                                 continue;
 753                         kvm_arch_ops->tlb_flush(vcpu);
 754                         vcpu_put(vcpu);
 755                 }
 756         }
 757
 758         r = 0;
 759
 760 out:
 761         spin_lock(&kvm->lock);
 762         --kvm->busy;
 763         spin_unlock(&kvm->lock);
 764         return r;
 765 }
 766
 767 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 768 {
 769         int i;
 770
 771         for (i = 0; i < kvm->nmemslots; ++i) {
 772                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 773
 774                 if (gfn >= memslot->base_gfn
 775                     && gfn < memslot->base_gfn + memslot->npages)
 776                         return memslot;
 777         }
 778         return 0;
 779 }
 780 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 781
 782 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 783 {
 784         int i;
 785         struct kvm_memory_slot *memslot = 0;
 786         unsigned long rel_gfn;
 787
 788         for (i = 0; i < kvm->nmemslots; ++i) {
 789                 memslot = &kvm->memslots[i];
 790
 791                 if (gfn >= memslot->base_gfn
 792                     && gfn < memslot->base_gfn + memslot->npages) {
 793
 794                         if (!memslot || !memslot->dirty_bitmap)
 795                                 return;
 796
 797                         rel_gfn = gfn - memslot->base_gfn;
 798
 799                         /* avoid RMW */
 800                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 801                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 802                         return;
 803                 }
 804         }
 805 }
 806
 807 static int emulator_read_std(unsigned long addr,
 808                              unsigned long *val,
 809                              unsigned int bytes,
 810                              struct x86_emulate_ctxt *ctxt)
 811 {
 812         struct kvm_vcpu *vcpu = ctxt->vcpu;
 813         void *data = val;
 814
 815         while (bytes) {
 816                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 817                 unsigned offset = addr & (PAGE_SIZE-1);
 818                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 819                 unsigned long pfn;
 820                 struct kvm_memory_slot *memslot;
 821                 void *page;
 822
 823                 if (gpa == UNMAPPED_GVA)
 824                         return X86EMUL_PROPAGATE_FAULT;
 825                 pfn = gpa >> PAGE_SHIFT;
 826                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 827                 if (!memslot)
 828                         return X86EMUL_UNHANDLEABLE;
 829                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 830
 831                 memcpy(data, page + offset, tocopy);
 832
 833                 kunmap_atomic(page, KM_USER0);
 834
 835                 bytes -= tocopy;
 836                 data += tocopy;
 837                 addr += tocopy;
 838         }
 839
 840         return X86EMUL_CONTINUE;
 841 }
 842
 843 static int emulator_write_std(unsigned long addr,
 844                               unsigned long val,
 845                               unsigned int bytes,
 846                               struct x86_emulate_ctxt *ctxt)
 847 {
 848         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 849                addr, bytes);
 850         return X86EMUL_UNHANDLEABLE;
 851 }
 852
 853 static int emulator_read_emulated(unsigned long addr,
 854                                   unsigned long *val,
 855                                   unsigned int bytes,
 856                                   struct x86_emulate_ctxt *ctxt)
 857 {
 858         struct kvm_vcpu *vcpu = ctxt->vcpu;
 859
 860         if (vcpu->mmio_read_completed) {
 861                 memcpy(val, vcpu->mmio_data, bytes);
 862                 vcpu->mmio_read_completed = 0;
 863                 return X86EMUL_CONTINUE;
 864         } else if (emulator_read_std(addr, val, bytes, ctxt)
 865                    == X86EMUL_CONTINUE)
 866                 return X86EMUL_CONTINUE;
 867         else {
 868                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 869                 if (gpa == UNMAPPED_GVA)
 870                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
 871                 vcpu->mmio_needed = 1;
 872                 vcpu->mmio_phys_addr = gpa;
 873                 vcpu->mmio_size = bytes;
 874                 vcpu->mmio_is_write = 0;
 875
 876                 return X86EMUL_UNHANDLEABLE;
 877         }
 878 }
 879
 880 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 881                                unsigned long val, int bytes)
 882 {
 883         struct kvm_memory_slot *m;
 884         struct page *page;
 885         void *virt;
 886
 887         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 888                 return 0;
 889         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 890         if (!m)
 891                 return 0;
 892         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 893         kvm_mmu_pre_write(vcpu, gpa, bytes);
 894         virt = kmap_atomic(page, KM_USER0);
 895         memcpy(virt + offset_in_page(gpa), &val, bytes);
 896         kunmap_atomic(virt, KM_USER0);
 897         kvm_mmu_post_write(vcpu, gpa, bytes);
 898         return 1;
 899 }
 900
 901 static int emulator_write_emulated(unsigned long addr,
 902                                    unsigned long val,
 903                                    unsigned int bytes,
 904                                    struct x86_emulate_ctxt *ctxt)
 905 {
 906         struct kvm_vcpu *vcpu = ctxt->vcpu;
 907         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 908
 909         if (gpa == UNMAPPED_GVA)
 910                 return X86EMUL_PROPAGATE_FAULT;
 911
 912         if (emulator_write_phys(vcpu, gpa, val, bytes))
 913                 return X86EMUL_CONTINUE;
 914
 915         vcpu->mmio_needed = 1;
 916         vcpu->mmio_phys_addr = gpa;
 917         vcpu->mmio_size = bytes;
 918         vcpu->mmio_is_write = 1;
 919         memcpy(vcpu->mmio_data, &val, bytes);
 920
 921         return X86EMUL_CONTINUE;
 922 }
 923
 924 static int emulator_cmpxchg_emulated(unsigned long addr,
 925                                      unsigned long old,
 926                                      unsigned long new,
 927                                      unsigned int bytes,
 928                                      struct x86_emulate_ctxt *ctxt)
 929 {
 930         static int reported;
 931
 932         if (!reported) {
 933                 reported = 1;
 934                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 935         }
 936         return emulator_write_emulated(addr, new, bytes, ctxt);
 937 }
 938
 939 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 940 {
 941         return kvm_arch_ops->get_segment_base(vcpu, seg);
 942 }
 943
 944 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 945 {
 946         return X86EMUL_CONTINUE;
 947 }
 948
 949 int emulate_clts(struct kvm_vcpu *vcpu)
 950 {
 951         unsigned long cr0;
 952
 953         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 954         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
 955         kvm_arch_ops->set_cr0(vcpu, cr0);
 956         return X86EMUL_CONTINUE;
 957 }
 958
 959 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
 960 {
 961         struct kvm_vcpu *vcpu = ctxt->vcpu;
 962
 963         switch (dr) {
 964         case 0 ... 3:
 965                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
 966                 return X86EMUL_CONTINUE;
 967         default:
 968                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
 969                        __FUNCTION__, dr);
 970                 return X86EMUL_UNHANDLEABLE;
 971         }
 972 }
 973
 974 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
 975 {
 976         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
 977         int exception;
 978
 979         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
 980         if (exception) {
 981                 /* FIXME: better handling */
 982                 return X86EMUL_UNHANDLEABLE;
 983         }
 984         return X86EMUL_CONTINUE;
 985 }
 986
 987 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
 988 {
 989         static int reported;
 990         u8 opcodes[4];
 991         unsigned long rip = ctxt->vcpu->rip;
 992         unsigned long rip_linear;
 993
 994         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
 995
 996         if (reported)
 997                 return;
 998
 999         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1000
1001         printk(KERN_ERR "emulation failed but !mmio_needed?"
1002                " rip %lx %02x %02x %02x %02x\n",
1003                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1004         reported = 1;
1005 }
1006
1007 struct x86_emulate_ops emulate_ops = {
1008         .read_std            = emulator_read_std,
1009         .write_std           = emulator_write_std,
1010         .read_emulated       = emulator_read_emulated,
1011         .write_emulated      = emulator_write_emulated,
1012         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1013 };
1014
1015 int emulate_instruction(struct kvm_vcpu *vcpu,
1016                         struct kvm_run *run,
1017                         unsigned long cr2,
1018                         u16 error_code)
1019 {
1020         struct x86_emulate_ctxt emulate_ctxt;
1021         int r;
1022         int cs_db, cs_l;
1023
1024         kvm_arch_ops->cache_regs(vcpu);
1025
1026         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1027
1028         emulate_ctxt.vcpu = vcpu;
1029         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1030         emulate_ctxt.cr2 = cr2;
1031         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1032                 ? X86EMUL_MODE_REAL : cs_l
1033                 ? X86EMUL_MODE_PROT64 : cs_db
1034                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1035
1036         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1037                 emulate_ctxt.cs_base = 0;
1038                 emulate_ctxt.ds_base = 0;
1039                 emulate_ctxt.es_base = 0;
1040                 emulate_ctxt.ss_base = 0;
1041         } else {
1042                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1043                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1044                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1045                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1046         }
1047
1048         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1049         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1050
1051         vcpu->mmio_is_write = 0;
1052         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1053
1054         if ((r || vcpu->mmio_is_write) && run) {
1055                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1056                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1057                 run->mmio.len = vcpu->mmio_size;
1058                 run->mmio.is_write = vcpu->mmio_is_write;
1059         }
1060
1061         if (r) {
1062                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1063                         return EMULATE_DONE;
1064                 if (!vcpu->mmio_needed) {
1065                         report_emulation_failure(&emulate_ctxt);
1066                         return EMULATE_FAIL;
1067                 }
1068                 return EMULATE_DO_MMIO;
1069         }
1070
1071         kvm_arch_ops->decache_regs(vcpu);
1072         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1073
1074         if (vcpu->mmio_is_write)
1075                 return EMULATE_DO_MMIO;
1076
1077         return EMULATE_DONE;
1078 }
1079 EXPORT_SYMBOL_GPL(emulate_instruction);
1080
1081 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1082 {
1083         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1084 }
1085
1086 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1087 {
1088         struct descriptor_table dt = { limit, base };
1089
1090         kvm_arch_ops->set_gdt(vcpu, &dt);
1091 }
1092
1093 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1094 {
1095         struct descriptor_table dt = { limit, base };
1096
1097         kvm_arch_ops->set_idt(vcpu, &dt);
1098 }
1099
1100 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1101                    unsigned long *rflags)
1102 {
1103         lmsw(vcpu, msw);
1104         *rflags = kvm_arch_ops->get_rflags(vcpu);
1105 }
1106
1107 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1108 {
1109         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1110         switch (cr) {
1111         case 0:
1112                 return vcpu->cr0;
1113         case 2:
1114                 return vcpu->cr2;
1115         case 3:
1116                 return vcpu->cr3;
1117         case 4:
1118                 return vcpu->cr4;
1119         default:
1120                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1121                 return 0;
1122         }
1123 }
1124
1125 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1126                      unsigned long *rflags)
1127 {
1128         switch (cr) {
1129         case 0:
1130                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1131                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1132                 break;
1133         case 2:
1134                 vcpu->cr2 = val;
1135                 break;
1136         case 3:
1137                 set_cr3(vcpu, val);
1138                 break;
1139         case 4:
1140                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1141                 break;
1142         default:
1143                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1144         }
1145 }
1146
1147 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1148 {
1149         u64 data;
1150
1151         switch (msr) {
1152         case 0xc0010010: /* SYSCFG */
1153         case 0xc0010015: /* HWCR */
1154         case MSR_IA32_PLATFORM_ID:
1155         case MSR_IA32_P5_MC_ADDR:
1156         case MSR_IA32_P5_MC_TYPE:
1157         case MSR_IA32_MC0_CTL:
1158         case MSR_IA32_MCG_STATUS:
1159         case MSR_IA32_MCG_CAP:
1160         case MSR_IA32_MC0_MISC:
1161         case MSR_IA32_MC0_MISC+4:
1162         case MSR_IA32_MC0_MISC+8:
1163         case MSR_IA32_MC0_MISC+12:
1164         case MSR_IA32_MC0_MISC+16:
1165         case MSR_IA32_UCODE_REV:
1166         case MSR_IA32_PERF_STATUS:
1167                 /* MTRR registers */
1168         case 0xfe:
1169         case 0x200 ... 0x2ff:
1170                 data = 0;
1171                 break;
1172         case 0xcd: /* fsb frequency */
1173                 data = 3;
1174                 break;
1175         case MSR_IA32_APICBASE:
1176                 data = vcpu->apic_base;
1177                 break;
1178 #ifdef CONFIG_X86_64
1179         case MSR_EFER:
1180                 data = vcpu->shadow_efer;
1181                 break;
1182 #endif
1183         default:
1184                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1185                 return 1;
1186         }
1187         *pdata = data;
1188         return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1191
1192 /*
1193  * Reads an msr value (of 'msr_index') into 'pdata'.
1194  * Returns 0 on success, non-0 otherwise.
1195  * Assumes vcpu_load() was already called.
1196  */
1197 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1198 {
1199         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1200 }
1201
1202 #ifdef CONFIG_X86_64
1203
1204 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1205 {
1206         if (efer & EFER_RESERVED_BITS) {
1207                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1208                        efer);
1209                 inject_gp(vcpu);
1210                 return;
1211         }
1212
1213         if (is_paging(vcpu)
1214             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1215                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1216                 inject_gp(vcpu);
1217                 return;
1218         }
1219
1220         kvm_arch_ops->set_efer(vcpu, efer);
1221
1222         efer &= ~EFER_LMA;
1223         efer |= vcpu->shadow_efer & EFER_LMA;
1224
1225         vcpu->shadow_efer = efer;
1226 }
1227
1228 #endif
1229
1230 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1231 {
1232         switch (msr) {
1233 #ifdef CONFIG_X86_64
1234         case MSR_EFER:
1235                 set_efer(vcpu, data);
1236                 break;
1237 #endif
1238         case MSR_IA32_MC0_STATUS:
1239                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1240                        __FUNCTION__, data);
1241                 break;
1242         case MSR_IA32_UCODE_REV:
1243         case MSR_IA32_UCODE_WRITE:
1244         case 0x200 ... 0x2ff: /* MTRRs */
1245                 break;
1246         case MSR_IA32_APICBASE:
1247                 vcpu->apic_base = data;
1248                 break;
1249         default:
1250                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1251                 return 1;
1252         }
1253         return 0;
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1256
1257 /*
1258  * Writes msr value into into the appropriate "register".
1259  * Returns 0 on success, non-0 otherwise.
1260  * Assumes vcpu_load() was already called.
1261  */
1262 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1263 {
1264         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1265 }
1266
1267 void kvm_resched(struct kvm_vcpu *vcpu)
1268 {
1269         vcpu_put(vcpu);
1270         cond_resched();
1271         /* Cannot fail -  no vcpu unplug yet. */
1272         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_resched);
1275
1276 void load_msrs(struct vmx_msr_entry *e, int n)
1277 {
1278         int i;
1279
1280         for (i = 0; i < n; ++i)
1281                 wrmsrl(e[i].index, e[i].data);
1282 }
1283 EXPORT_SYMBOL_GPL(load_msrs);
1284
1285 void save_msrs(struct vmx_msr_entry *e, int n)
1286 {
1287         int i;
1288
1289         for (i = 0; i < n; ++i)
1290                 rdmsrl(e[i].index, e[i].data);
1291 }
1292 EXPORT_SYMBOL_GPL(save_msrs);
1293
1294 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1295 {
1296         struct kvm_vcpu *vcpu;
1297         int r;
1298
1299         if (!valid_vcpu(kvm_run->vcpu))
1300                 return -EINVAL;
1301
1302         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1303         if (!vcpu)
1304                 return -ENOENT;
1305
1306         if (kvm_run->emulated) {
1307                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1308                 kvm_run->emulated = 0;
1309         }
1310
1311         if (kvm_run->mmio_completed) {
1312                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1313                 vcpu->mmio_read_completed = 1;
1314         }
1315
1316         vcpu->mmio_needed = 0;
1317
1318         r = kvm_arch_ops->run(vcpu, kvm_run);
1319
1320         vcpu_put(vcpu);
1321         return r;
1322 }
1323
1324 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1325 {
1326         struct kvm_vcpu *vcpu;
1327
1328         if (!valid_vcpu(regs->vcpu))
1329                 return -EINVAL;
1330
1331         vcpu = vcpu_load(kvm, regs->vcpu);
1332         if (!vcpu)
1333                 return -ENOENT;
1334
1335         kvm_arch_ops->cache_regs(vcpu);
1336
1337         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1338         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1339         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1340         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1341         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1342         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1343         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1344         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1345 #ifdef CONFIG_X86_64
1346         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1347         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1348         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1349         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1350         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1351         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1352         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1353         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1354 #endif
1355
1356         regs->rip = vcpu->rip;
1357         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1358
1359         /*
1360          * Don't leak debug flags in case they were set for guest debugging
1361          */
1362         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1363                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1364
1365         vcpu_put(vcpu);
1366
1367         return 0;
1368 }
1369
1370 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1371 {
1372         struct kvm_vcpu *vcpu;
1373
1374         if (!valid_vcpu(regs->vcpu))
1375                 return -EINVAL;
1376
1377         vcpu = vcpu_load(kvm, regs->vcpu);
1378         if (!vcpu)
1379                 return -ENOENT;
1380
1381         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1382         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1383         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1384         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1385         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1386         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1387         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1388         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1389 #ifdef CONFIG_X86_64
1390         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1391         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1392         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1393         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1394         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1395         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1396         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1397         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1398 #endif
1399
1400         vcpu->rip = regs->rip;
1401         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1402
1403         kvm_arch_ops->decache_regs(vcpu);
1404
1405         vcpu_put(vcpu);
1406
1407         return 0;
1408 }
1409
1410 static void get_segment(struct kvm_vcpu *vcpu,
1411                         struct kvm_segment *var, int seg)
1412 {
1413         return kvm_arch_ops->get_segment(vcpu, var, seg);
1414 }
1415
1416 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1417 {
1418         struct kvm_vcpu *vcpu;
1419         struct descriptor_table dt;
1420
1421         if (!valid_vcpu(sregs->vcpu))
1422                 return -EINVAL;
1423         vcpu = vcpu_load(kvm, sregs->vcpu);
1424         if (!vcpu)
1425                 return -ENOENT;
1426
1427         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1428         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1429         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1430         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1431         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1432         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1433
1434         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1435         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1436
1437         kvm_arch_ops->get_idt(vcpu, &dt);
1438         sregs->idt.limit = dt.limit;
1439         sregs->idt.base = dt.base;
1440         kvm_arch_ops->get_gdt(vcpu, &dt);
1441         sregs->gdt.limit = dt.limit;
1442         sregs->gdt.base = dt.base;
1443
1444         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1445         sregs->cr0 = vcpu->cr0;
1446         sregs->cr2 = vcpu->cr2;
1447         sregs->cr3 = vcpu->cr3;
1448         sregs->cr4 = vcpu->cr4;
1449         sregs->cr8 = vcpu->cr8;
1450         sregs->efer = vcpu->shadow_efer;
1451         sregs->apic_base = vcpu->apic_base;
1452
1453         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1454                sizeof sregs->interrupt_bitmap);
1455
1456         vcpu_put(vcpu);
1457
1458         return 0;
1459 }
1460
1461 static void set_segment(struct kvm_vcpu *vcpu,
1462                         struct kvm_segment *var, int seg)
1463 {
1464         return kvm_arch_ops->set_segment(vcpu, var, seg);
1465 }
1466
1467 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1468 {
1469         struct kvm_vcpu *vcpu;
1470         int mmu_reset_needed = 0;
1471         int i;
1472         struct descriptor_table dt;
1473
1474         if (!valid_vcpu(sregs->vcpu))
1475                 return -EINVAL;
1476         vcpu = vcpu_load(kvm, sregs->vcpu);
1477         if (!vcpu)
1478                 return -ENOENT;
1479
1480         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1481         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1482         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1483         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1484         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1485         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1486
1487         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1488         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1489
1490         dt.limit = sregs->idt.limit;
1491         dt.base = sregs->idt.base;
1492         kvm_arch_ops->set_idt(vcpu, &dt);
1493         dt.limit = sregs->gdt.limit;
1494         dt.base = sregs->gdt.base;
1495         kvm_arch_ops->set_gdt(vcpu, &dt);
1496
1497         vcpu->cr2 = sregs->cr2;
1498         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1499         vcpu->cr3 = sregs->cr3;
1500
1501         vcpu->cr8 = sregs->cr8;
1502
1503         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1504 #ifdef CONFIG_X86_64
1505         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1506 #endif
1507         vcpu->apic_base = sregs->apic_base;
1508
1509         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1510
1511         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1512         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1513
1514         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1515         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1516         if (!is_long_mode(vcpu) && is_pae(vcpu))
1517                 load_pdptrs(vcpu, vcpu->cr3);
1518
1519         if (mmu_reset_needed)
1520                 kvm_mmu_reset_context(vcpu);
1521
1522         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1523                sizeof vcpu->irq_pending);
1524         vcpu->irq_summary = 0;
1525         for (i = 0; i < NR_IRQ_WORDS; ++i)
1526                 if (vcpu->irq_pending[i])
1527                         __set_bit(i, &vcpu->irq_summary);
1528
1529         vcpu_put(vcpu);
1530
1531         return 0;
1532 }
1533
1534 /*
1535  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1536  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1537  *
1538  * This list is modified at module load time to reflect the
1539  * capabilities of the host cpu.
1540  */
1541 static u32 msrs_to_save[] = {
1542         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1543         MSR_K6_STAR,
1544 #ifdef CONFIG_X86_64
1545         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1546 #endif
1547         MSR_IA32_TIME_STAMP_COUNTER,
1548 };
1549
1550 static unsigned num_msrs_to_save;
1551
1552 static __init void kvm_init_msr_list(void)
1553 {
1554         u32 dummy[2];
1555         unsigned i, j;
1556
1557         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1558                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1559                         continue;
1560                 if (j < i)
1561                         msrs_to_save[j] = msrs_to_save[i];
1562                 j++;
1563         }
1564         num_msrs_to_save = j;
1565 }
1566
1567 /*
1568  * Adapt set_msr() to msr_io()'s calling convention
1569  */
1570 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1571 {
1572         return set_msr(vcpu, index, *data);
1573 }
1574
1575 /*
1576  * Read or write a bunch of msrs. All parameters are kernel addresses.
1577  *
1578  * @return number of msrs set successfully.
1579  */
1580 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1581                     struct kvm_msr_entry *entries,
1582                     int (*do_msr)(struct kvm_vcpu *vcpu,
1583                                   unsigned index, u64 *data))
1584 {
1585         struct kvm_vcpu *vcpu;
1586         int i;
1587
1588         if (!valid_vcpu(msrs->vcpu))
1589                 return -EINVAL;
1590
1591         vcpu = vcpu_load(kvm, msrs->vcpu);
1592         if (!vcpu)
1593                 return -ENOENT;
1594
1595         for (i = 0; i < msrs->nmsrs; ++i)
1596                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1597                         break;
1598
1599         vcpu_put(vcpu);
1600
1601         return i;
1602 }
1603
1604 /*
1605  * Read or write a bunch of msrs. Parameters are user addresses.
1606  *
1607  * @return number of msrs set successfully.
1608  */
1609 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1610                   int (*do_msr)(struct kvm_vcpu *vcpu,
1611                                 unsigned index, u64 *data),
1612                   int writeback)
1613 {
1614         struct kvm_msrs msrs;
1615         struct kvm_msr_entry *entries;
1616         int r, n;
1617         unsigned size;
1618
1619         r = -EFAULT;
1620         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1621                 goto out;
1622
1623         r = -E2BIG;
1624         if (msrs.nmsrs >= MAX_IO_MSRS)
1625                 goto out;
1626
1627         r = -ENOMEM;
1628         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1629         entries = vmalloc(size);
1630         if (!entries)
1631                 goto out;
1632
1633         r = -EFAULT;
1634         if (copy_from_user(entries, user_msrs->entries, size))
1635                 goto out_free;
1636
1637         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1638         if (r < 0)
1639                 goto out_free;
1640
1641         r = -EFAULT;
1642         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1643                 goto out_free;
1644
1645         r = n;
1646
1647 out_free:
1648         vfree(entries);
1649 out:
1650         return r;
1651 }
1652
1653 /*
1654  * Translate a guest virtual address to a guest physical address.
1655  */
1656 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1657 {
1658         unsigned long vaddr = tr->linear_address;
1659         struct kvm_vcpu *vcpu;
1660         gpa_t gpa;
1661
1662         vcpu = vcpu_load(kvm, tr->vcpu);
1663         if (!vcpu)
1664                 return -ENOENT;
1665         spin_lock(&kvm->lock);
1666         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1667         tr->physical_address = gpa;
1668         tr->valid = gpa != UNMAPPED_GVA;
1669         tr->writeable = 1;
1670         tr->usermode = 0;
1671         spin_unlock(&kvm->lock);
1672         vcpu_put(vcpu);
1673
1674         return 0;
1675 }
1676
1677 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1678 {
1679         struct kvm_vcpu *vcpu;
1680
1681         if (!valid_vcpu(irq->vcpu))
1682                 return -EINVAL;
1683         if (irq->irq < 0 || irq->irq >= 256)
1684                 return -EINVAL;
1685         vcpu = vcpu_load(kvm, irq->vcpu);
1686         if (!vcpu)
1687                 return -ENOENT;
1688
1689         set_bit(irq->irq, vcpu->irq_pending);
1690         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1691
1692         vcpu_put(vcpu);
1693
1694         return 0;
1695 }
1696
1697 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1698                                      struct kvm_debug_guest *dbg)
1699 {
1700         struct kvm_vcpu *vcpu;
1701         int r;
1702
1703         if (!valid_vcpu(dbg->vcpu))
1704                 return -EINVAL;
1705         vcpu = vcpu_load(kvm, dbg->vcpu);
1706         if (!vcpu)
1707                 return -ENOENT;
1708
1709         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1710
1711         vcpu_put(vcpu);
1712
1713         return r;
1714 }
1715
1716 static long kvm_dev_ioctl(struct file *filp,
1717                           unsigned int ioctl, unsigned long arg)
1718 {
1719         struct kvm *kvm = filp->private_data;
1720         int r = -EINVAL;
1721
1722         switch (ioctl) {
1723         case KVM_GET_API_VERSION:
1724                 r = KVM_API_VERSION;
1725                 break;
1726         case KVM_CREATE_VCPU: {
1727                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1728                 if (r)
1729                         goto out;
1730                 break;
1731         }
1732         case KVM_RUN: {
1733                 struct kvm_run kvm_run;
1734
1735                 r = -EFAULT;
1736                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1737                         goto out;
1738                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1739                 if (r < 0 &&  r != -EINTR)
1740                         goto out;
1741                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1742                         r = -EFAULT;
1743                         goto out;
1744                 }
1745                 break;
1746         }
1747         case KVM_GET_REGS: {
1748                 struct kvm_regs kvm_regs;
1749
1750                 r = -EFAULT;
1751                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1752                         goto out;
1753                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1754                 if (r)
1755                         goto out;
1756                 r = -EFAULT;
1757                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1758                         goto out;
1759                 r = 0;
1760                 break;
1761         }
1762         case KVM_SET_REGS: {
1763                 struct kvm_regs kvm_regs;
1764
1765                 r = -EFAULT;
1766                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1767                         goto out;
1768                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1769                 if (r)
1770                         goto out;
1771                 r = 0;
1772                 break;
1773         }
1774         case KVM_GET_SREGS: {
1775                 struct kvm_sregs kvm_sregs;
1776
1777                 r = -EFAULT;
1778                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1779                         goto out;
1780                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1781                 if (r)
1782                         goto out;
1783                 r = -EFAULT;
1784                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1785                         goto out;
1786                 r = 0;
1787                 break;
1788         }
1789         case KVM_SET_SREGS: {
1790                 struct kvm_sregs kvm_sregs;
1791
1792                 r = -EFAULT;
1793                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1794                         goto out;
1795                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1796                 if (r)
1797                         goto out;
1798                 r = 0;
1799                 break;
1800         }
1801         case KVM_TRANSLATE: {
1802                 struct kvm_translation tr;
1803
1804                 r = -EFAULT;
1805                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1806                         goto out;
1807                 r = kvm_dev_ioctl_translate(kvm, &tr);
1808                 if (r)
1809                         goto out;
1810                 r = -EFAULT;
1811                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1812                         goto out;
1813                 r = 0;
1814                 break;
1815         }
1816         case KVM_INTERRUPT: {
1817                 struct kvm_interrupt irq;
1818
1819                 r = -EFAULT;
1820                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1821                         goto out;
1822                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1823                 if (r)
1824                         goto out;
1825                 r = 0;
1826                 break;
1827         }
1828         case KVM_DEBUG_GUEST: {
1829                 struct kvm_debug_guest dbg;
1830
1831                 r = -EFAULT;
1832                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1833                         goto out;
1834                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1835                 if (r)
1836                         goto out;
1837                 r = 0;
1838                 break;
1839         }
1840         case KVM_SET_MEMORY_REGION: {
1841                 struct kvm_memory_region kvm_mem;
1842
1843                 r = -EFAULT;
1844                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1845                         goto out;
1846                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1847                 if (r)
1848                         goto out;
1849                 break;
1850         }
1851         case KVM_GET_DIRTY_LOG: {
1852                 struct kvm_dirty_log log;
1853
1854                 r = -EFAULT;
1855                 if (copy_from_user(&log, (void *)arg, sizeof log))
1856                         goto out;
1857                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1858                 if (r)
1859                         goto out;
1860                 break;
1861         }
1862         case KVM_GET_MSRS:
1863                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1864                 break;
1865         case KVM_SET_MSRS:
1866                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1867                 break;
1868         case KVM_GET_MSR_INDEX_LIST: {
1869                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1870                 struct kvm_msr_list msr_list;
1871                 unsigned n;
1872
1873                 r = -EFAULT;
1874                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1875                         goto out;
1876                 n = msr_list.nmsrs;
1877                 msr_list.nmsrs = num_msrs_to_save;
1878                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1879                         goto out;
1880                 r = -E2BIG;
1881                 if (n < num_msrs_to_save)
1882                         goto out;
1883                 r = -EFAULT;
1884                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1885                                  num_msrs_to_save * sizeof(u32)))
1886                         goto out;
1887                 r = 0;
1888         }
1889         default:
1890                 ;
1891         }
1892 out:
1893         return r;
1894 }
1895
1896 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1897                                    unsigned long address,
1898                                    int *type)
1899 {
1900         struct kvm *kvm = vma->vm_file->private_data;
1901         unsigned long pgoff;
1902         struct kvm_memory_slot *slot;
1903         struct page *page;
1904
1905         *type = VM_FAULT_MINOR;
1906         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1907         slot = gfn_to_memslot(kvm, pgoff);
1908         if (!slot)
1909                 return NOPAGE_SIGBUS;
1910         page = gfn_to_page(slot, pgoff);
1911         if (!page)
1912                 return NOPAGE_SIGBUS;
1913         get_page(page);
1914         return page;
1915 }
1916
1917 static struct vm_operations_struct kvm_dev_vm_ops = {
1918         .nopage = kvm_dev_nopage,
1919 };
1920
1921 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1922 {
1923         vma->vm_ops = &kvm_dev_vm_ops;
1924         return 0;
1925 }
1926
1927 static struct file_operations kvm_chardev_ops = {
1928         .open           = kvm_dev_open,
1929         .release        = kvm_dev_release,
1930         .unlocked_ioctl = kvm_dev_ioctl,
1931         .compat_ioctl   = kvm_dev_ioctl,
1932         .mmap           = kvm_dev_mmap,
1933 };
1934
1935 static struct miscdevice kvm_dev = {
1936         MISC_DYNAMIC_MINOR,
1937         "kvm",
1938         &kvm_chardev_ops,
1939 };
1940
1941 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1942                        void *v)
1943 {
1944         if (val == SYS_RESTART) {
1945                 /*
1946                  * Some (well, at least mine) BIOSes hang on reboot if
1947                  * in vmx root mode.
1948                  */
1949                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1950                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1951         }
1952         return NOTIFY_OK;
1953 }
1954
1955 static struct notifier_block kvm_reboot_notifier = {
1956         .notifier_call = kvm_reboot,
1957         .priority = 0,
1958 };
1959
1960 static __init void kvm_init_debug(void)
1961 {
1962         struct kvm_stats_debugfs_item *p;
1963
1964         debugfs_dir = debugfs_create_dir("kvm", 0);
1965         for (p = debugfs_entries; p->name; ++p)
1966                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1967                                                p->data);
1968 }
1969
1970 static void kvm_exit_debug(void)
1971 {
1972         struct kvm_stats_debugfs_item *p;
1973
1974         for (p = debugfs_entries; p->name; ++p)
1975                 debugfs_remove(p->dentry);
1976         debugfs_remove(debugfs_dir);
1977 }
1978
1979 hpa_t bad_page_address;
1980
1981 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1982 {
1983         int r;
1984
1985         if (kvm_arch_ops) {
1986                 printk(KERN_ERR "kvm: already loaded the other module\n");
1987                 return -EEXIST;
1988         }
1989
1990         if (!ops->cpu_has_kvm_support()) {
1991                 printk(KERN_ERR "kvm: no hardware support\n");
1992                 return -EOPNOTSUPP;
1993         }
1994         if (ops->disabled_by_bios()) {
1995                 printk(KERN_ERR "kvm: disabled by bios\n");
1996                 return -EOPNOTSUPP;
1997         }
1998
1999         kvm_arch_ops = ops;
2000
2001         r = kvm_arch_ops->hardware_setup();
2002         if (r < 0)
2003             return r;
2004
2005         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2006         register_reboot_notifier(&kvm_reboot_notifier);
2007
2008         kvm_chardev_ops.owner = module;
2009
2010         r = misc_register(&kvm_dev);
2011         if (r) {
2012                 printk (KERN_ERR "kvm: misc device register failed\n");
2013                 goto out_free;
2014         }
2015
2016         return r;
2017
2018 out_free:
2019         unregister_reboot_notifier(&kvm_reboot_notifier);
2020         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2021         kvm_arch_ops->hardware_unsetup();
2022         return r;
2023 }
2024
2025 void kvm_exit_arch(void)
2026 {
2027         misc_deregister(&kvm_dev);
2028
2029         unregister_reboot_notifier(&kvm_reboot_notifier);
2030         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2031         kvm_arch_ops->hardware_unsetup();
2032         kvm_arch_ops = NULL;
2033 }
2034
2035 static __init int kvm_init(void)
2036 {
2037         static struct page *bad_page;
2038         int r = 0;
2039
2040         kvm_init_debug();
2041
2042         kvm_init_msr_list();
2043
2044         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2045                 r = -ENOMEM;
2046                 goto out;
2047         }
2048
2049         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2050         memset(__va(bad_page_address), 0, PAGE_SIZE);
2051
2052         return r;
2053
2054 out:
2055         kvm_exit_debug();
2056         return r;
2057 }
2058
2059 static __exit void kvm_exit(void)
2060 {
2061         kvm_exit_debug();
2062         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2063 }
2064
2065 module_init(kvm_init)
2066 module_exit(kvm_exit)
2067
2068 EXPORT_SYMBOL_GPL(kvm_init_arch);
2069 EXPORT_SYMBOL_GPL(kvm_exit_arch);