err.no Git - linux-2.6/blob - 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         spin_lock(&vcpu->kvm->lock);
 947         vcpu->mmu.inval_page(vcpu, address);
 948         spin_unlock(&vcpu->kvm->lock);
 949         kvm_arch_ops->invlpg(vcpu, address);
 950         return X86EMUL_CONTINUE;
 951 }
 952
 953 int emulate_clts(struct kvm_vcpu *vcpu)
 954 {
 955         unsigned long cr0;
 956
 957         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 958         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
 959         kvm_arch_ops->set_cr0(vcpu, cr0);
 960         return X86EMUL_CONTINUE;
 961 }
 962
 963 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
 964 {
 965         struct kvm_vcpu *vcpu = ctxt->vcpu;
 966
 967         switch (dr) {
 968         case 0 ... 3:
 969                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
 970                 return X86EMUL_CONTINUE;
 971         default:
 972                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
 973                        __FUNCTION__, dr);
 974                 return X86EMUL_UNHANDLEABLE;
 975         }
 976 }
 977
 978 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
 979 {
 980         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
 981         int exception;
 982
 983         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
 984         if (exception) {
 985                 /* FIXME: better handling */
 986                 return X86EMUL_UNHANDLEABLE;
 987         }
 988         return X86EMUL_CONTINUE;
 989 }
 990
 991 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
 992 {
 993         static int reported;
 994         u8 opcodes[4];
 995         unsigned long rip = ctxt->vcpu->rip;
 996         unsigned long rip_linear;
 997
 998         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
 999
1000         if (reported)
1001                 return;
1002
1003         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1004
1005         printk(KERN_ERR "emulation failed but !mmio_needed?"
1006                " rip %lx %02x %02x %02x %02x\n",
1007                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1008         reported = 1;
1009 }
1010
1011 struct x86_emulate_ops emulate_ops = {
1012         .read_std            = emulator_read_std,
1013         .write_std           = emulator_write_std,
1014         .read_emulated       = emulator_read_emulated,
1015         .write_emulated      = emulator_write_emulated,
1016         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1017 };
1018
1019 int emulate_instruction(struct kvm_vcpu *vcpu,
1020                         struct kvm_run *run,
1021                         unsigned long cr2,
1022                         u16 error_code)
1023 {
1024         struct x86_emulate_ctxt emulate_ctxt;
1025         int r;
1026         int cs_db, cs_l;
1027
1028         kvm_arch_ops->cache_regs(vcpu);
1029
1030         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1031
1032         emulate_ctxt.vcpu = vcpu;
1033         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1034         emulate_ctxt.cr2 = cr2;
1035         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1036                 ? X86EMUL_MODE_REAL : cs_l
1037                 ? X86EMUL_MODE_PROT64 : cs_db
1038                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1039
1040         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1041                 emulate_ctxt.cs_base = 0;
1042                 emulate_ctxt.ds_base = 0;
1043                 emulate_ctxt.es_base = 0;
1044                 emulate_ctxt.ss_base = 0;
1045         } else {
1046                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1047                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1048                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1049                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1050         }
1051
1052         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1053         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1054
1055         vcpu->mmio_is_write = 0;
1056         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1057
1058         if ((r || vcpu->mmio_is_write) && run) {
1059                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1060                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1061                 run->mmio.len = vcpu->mmio_size;
1062                 run->mmio.is_write = vcpu->mmio_is_write;
1063         }
1064
1065         if (r) {
1066                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1067                         return EMULATE_DONE;
1068                 if (!vcpu->mmio_needed) {
1069                         report_emulation_failure(&emulate_ctxt);
1070                         return EMULATE_FAIL;
1071                 }
1072                 return EMULATE_DO_MMIO;
1073         }
1074
1075         kvm_arch_ops->decache_regs(vcpu);
1076         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1077
1078         if (vcpu->mmio_is_write)
1079                 return EMULATE_DO_MMIO;
1080
1081         return EMULATE_DONE;
1082 }
1083 EXPORT_SYMBOL_GPL(emulate_instruction);
1084
1085 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1086 {
1087         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1088 }
1089
1090 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1091 {
1092         struct descriptor_table dt = { limit, base };
1093
1094         kvm_arch_ops->set_gdt(vcpu, &dt);
1095 }
1096
1097 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1098 {
1099         struct descriptor_table dt = { limit, base };
1100
1101         kvm_arch_ops->set_idt(vcpu, &dt);
1102 }
1103
1104 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1105                    unsigned long *rflags)
1106 {
1107         lmsw(vcpu, msw);
1108         *rflags = kvm_arch_ops->get_rflags(vcpu);
1109 }
1110
1111 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1112 {
1113         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1114         switch (cr) {
1115         case 0:
1116                 return vcpu->cr0;
1117         case 2:
1118                 return vcpu->cr2;
1119         case 3:
1120                 return vcpu->cr3;
1121         case 4:
1122                 return vcpu->cr4;
1123         default:
1124                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1125                 return 0;
1126         }
1127 }
1128
1129 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1130                      unsigned long *rflags)
1131 {
1132         switch (cr) {
1133         case 0:
1134                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1135                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1136                 break;
1137         case 2:
1138                 vcpu->cr2 = val;
1139                 break;
1140         case 3:
1141                 set_cr3(vcpu, val);
1142                 break;
1143         case 4:
1144                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1145                 break;
1146         default:
1147                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1148         }
1149 }
1150
1151 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1152 {
1153         u64 data;
1154
1155         switch (msr) {
1156         case 0xc0010010: /* SYSCFG */
1157         case 0xc0010015: /* HWCR */
1158         case MSR_IA32_PLATFORM_ID:
1159         case MSR_IA32_P5_MC_ADDR:
1160         case MSR_IA32_P5_MC_TYPE:
1161         case MSR_IA32_MC0_CTL:
1162         case MSR_IA32_MCG_STATUS:
1163         case MSR_IA32_MCG_CAP:
1164         case MSR_IA32_MC0_MISC:
1165         case MSR_IA32_MC0_MISC+4:
1166         case MSR_IA32_MC0_MISC+8:
1167         case MSR_IA32_MC0_MISC+12:
1168         case MSR_IA32_MC0_MISC+16:
1169         case MSR_IA32_UCODE_REV:
1170         case MSR_IA32_PERF_STATUS:
1171                 /* MTRR registers */
1172         case 0xfe:
1173         case 0x200 ... 0x2ff:
1174                 data = 0;
1175                 break;
1176         case 0xcd: /* fsb frequency */
1177                 data = 3;
1178                 break;
1179         case MSR_IA32_APICBASE:
1180                 data = vcpu->apic_base;
1181                 break;
1182 #ifdef CONFIG_X86_64
1183         case MSR_EFER:
1184                 data = vcpu->shadow_efer;
1185                 break;
1186 #endif
1187         default:
1188                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1189                 return 1;
1190         }
1191         *pdata = data;
1192         return 0;
1193 }
1194 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1195
1196 /*
1197  * Reads an msr value (of 'msr_index') into 'pdata'.
1198  * Returns 0 on success, non-0 otherwise.
1199  * Assumes vcpu_load() was already called.
1200  */
1201 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1202 {
1203         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1204 }
1205
1206 #ifdef CONFIG_X86_64
1207
1208 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1209 {
1210         if (efer & EFER_RESERVED_BITS) {
1211                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1212                        efer);
1213                 inject_gp(vcpu);
1214                 return;
1215         }
1216
1217         if (is_paging(vcpu)
1218             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1219                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1220                 inject_gp(vcpu);
1221                 return;
1222         }
1223
1224         kvm_arch_ops->set_efer(vcpu, efer);
1225
1226         efer &= ~EFER_LMA;
1227         efer |= vcpu->shadow_efer & EFER_LMA;
1228
1229         vcpu->shadow_efer = efer;
1230 }
1231
1232 #endif
1233
1234 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1235 {
1236         switch (msr) {
1237 #ifdef CONFIG_X86_64
1238         case MSR_EFER:
1239                 set_efer(vcpu, data);
1240                 break;
1241 #endif
1242         case MSR_IA32_MC0_STATUS:
1243                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1244                        __FUNCTION__, data);
1245                 break;
1246         case MSR_IA32_UCODE_REV:
1247         case MSR_IA32_UCODE_WRITE:
1248         case 0x200 ... 0x2ff: /* MTRRs */
1249                 break;
1250         case MSR_IA32_APICBASE:
1251                 vcpu->apic_base = data;
1252                 break;
1253         default:
1254                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1255                 return 1;
1256         }
1257         return 0;
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1260
1261 /*
1262  * Writes msr value into into the appropriate "register".
1263  * Returns 0 on success, non-0 otherwise.
1264  * Assumes vcpu_load() was already called.
1265  */
1266 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1267 {
1268         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1269 }
1270
1271 void kvm_resched(struct kvm_vcpu *vcpu)
1272 {
1273         vcpu_put(vcpu);
1274         cond_resched();
1275         /* Cannot fail -  no vcpu unplug yet. */
1276         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1277 }
1278 EXPORT_SYMBOL_GPL(kvm_resched);
1279
1280 void load_msrs(struct vmx_msr_entry *e, int n)
1281 {
1282         int i;
1283
1284         for (i = 0; i < n; ++i)
1285                 wrmsrl(e[i].index, e[i].data);
1286 }
1287 EXPORT_SYMBOL_GPL(load_msrs);
1288
1289 void save_msrs(struct vmx_msr_entry *e, int n)
1290 {
1291         int i;
1292
1293         for (i = 0; i < n; ++i)
1294                 rdmsrl(e[i].index, e[i].data);
1295 }
1296 EXPORT_SYMBOL_GPL(save_msrs);
1297
1298 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1299 {
1300         struct kvm_vcpu *vcpu;
1301         int r;
1302
1303         if (!valid_vcpu(kvm_run->vcpu))
1304                 return -EINVAL;
1305
1306         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1307         if (!vcpu)
1308                 return -ENOENT;
1309
1310         if (kvm_run->emulated) {
1311                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1312                 kvm_run->emulated = 0;
1313         }
1314
1315         if (kvm_run->mmio_completed) {
1316                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1317                 vcpu->mmio_read_completed = 1;
1318         }
1319
1320         vcpu->mmio_needed = 0;
1321
1322         r = kvm_arch_ops->run(vcpu, kvm_run);
1323
1324         vcpu_put(vcpu);
1325         return r;
1326 }
1327
1328 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1329 {
1330         struct kvm_vcpu *vcpu;
1331
1332         if (!valid_vcpu(regs->vcpu))
1333                 return -EINVAL;
1334
1335         vcpu = vcpu_load(kvm, regs->vcpu);
1336         if (!vcpu)
1337                 return -ENOENT;
1338
1339         kvm_arch_ops->cache_regs(vcpu);
1340
1341         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1342         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1343         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1344         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1345         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1346         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1347         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1348         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1349 #ifdef CONFIG_X86_64
1350         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1351         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1352         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1353         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1354         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1355         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1356         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1357         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1358 #endif
1359
1360         regs->rip = vcpu->rip;
1361         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1362
1363         /*
1364          * Don't leak debug flags in case they were set for guest debugging
1365          */
1366         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1367                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1368
1369         vcpu_put(vcpu);
1370
1371         return 0;
1372 }
1373
1374 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1375 {
1376         struct kvm_vcpu *vcpu;
1377
1378         if (!valid_vcpu(regs->vcpu))
1379                 return -EINVAL;
1380
1381         vcpu = vcpu_load(kvm, regs->vcpu);
1382         if (!vcpu)
1383                 return -ENOENT;
1384
1385         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1386         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1387         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1388         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1389         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1390         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1391         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1392         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1393 #ifdef CONFIG_X86_64
1394         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1395         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1396         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1397         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1398         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1399         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1400         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1401         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1402 #endif
1403
1404         vcpu->rip = regs->rip;
1405         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1406
1407         kvm_arch_ops->decache_regs(vcpu);
1408
1409         vcpu_put(vcpu);
1410
1411         return 0;
1412 }
1413
1414 static void get_segment(struct kvm_vcpu *vcpu,
1415                         struct kvm_segment *var, int seg)
1416 {
1417         return kvm_arch_ops->get_segment(vcpu, var, seg);
1418 }
1419
1420 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1421 {
1422         struct kvm_vcpu *vcpu;
1423         struct descriptor_table dt;
1424
1425         if (!valid_vcpu(sregs->vcpu))
1426                 return -EINVAL;
1427         vcpu = vcpu_load(kvm, sregs->vcpu);
1428         if (!vcpu)
1429                 return -ENOENT;
1430
1431         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1432         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1433         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1434         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1435         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1436         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1437
1438         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1439         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1440
1441         kvm_arch_ops->get_idt(vcpu, &dt);
1442         sregs->idt.limit = dt.limit;
1443         sregs->idt.base = dt.base;
1444         kvm_arch_ops->get_gdt(vcpu, &dt);
1445         sregs->gdt.limit = dt.limit;
1446         sregs->gdt.base = dt.base;
1447
1448         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1449         sregs->cr0 = vcpu->cr0;
1450         sregs->cr2 = vcpu->cr2;
1451         sregs->cr3 = vcpu->cr3;
1452         sregs->cr4 = vcpu->cr4;
1453         sregs->cr8 = vcpu->cr8;
1454         sregs->efer = vcpu->shadow_efer;
1455         sregs->apic_base = vcpu->apic_base;
1456
1457         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1458                sizeof sregs->interrupt_bitmap);
1459
1460         vcpu_put(vcpu);
1461
1462         return 0;
1463 }
1464
1465 static void set_segment(struct kvm_vcpu *vcpu,
1466                         struct kvm_segment *var, int seg)
1467 {
1468         return kvm_arch_ops->set_segment(vcpu, var, seg);
1469 }
1470
1471 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1472 {
1473         struct kvm_vcpu *vcpu;
1474         int mmu_reset_needed = 0;
1475         int i;
1476         struct descriptor_table dt;
1477
1478         if (!valid_vcpu(sregs->vcpu))
1479                 return -EINVAL;
1480         vcpu = vcpu_load(kvm, sregs->vcpu);
1481         if (!vcpu)
1482                 return -ENOENT;
1483
1484         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1485         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1486         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1487         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1488         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1489         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1490
1491         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1492         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1493
1494         dt.limit = sregs->idt.limit;
1495         dt.base = sregs->idt.base;
1496         kvm_arch_ops->set_idt(vcpu, &dt);
1497         dt.limit = sregs->gdt.limit;
1498         dt.base = sregs->gdt.base;
1499         kvm_arch_ops->set_gdt(vcpu, &dt);
1500
1501         vcpu->cr2 = sregs->cr2;
1502         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1503         vcpu->cr3 = sregs->cr3;
1504
1505         vcpu->cr8 = sregs->cr8;
1506
1507         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1508 #ifdef CONFIG_X86_64
1509         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1510 #endif
1511         vcpu->apic_base = sregs->apic_base;
1512
1513         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1514
1515         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1516         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1517
1518         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1519         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1520         if (!is_long_mode(vcpu) && is_pae(vcpu))
1521                 load_pdptrs(vcpu, vcpu->cr3);
1522
1523         if (mmu_reset_needed)
1524                 kvm_mmu_reset_context(vcpu);
1525
1526         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1527                sizeof vcpu->irq_pending);
1528         vcpu->irq_summary = 0;
1529         for (i = 0; i < NR_IRQ_WORDS; ++i)
1530                 if (vcpu->irq_pending[i])
1531                         __set_bit(i, &vcpu->irq_summary);
1532
1533         vcpu_put(vcpu);
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1540  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1541  *
1542  * This list is modified at module load time to reflect the
1543  * capabilities of the host cpu.
1544  */
1545 static u32 msrs_to_save[] = {
1546         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1547         MSR_K6_STAR,
1548 #ifdef CONFIG_X86_64
1549         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1550 #endif
1551         MSR_IA32_TIME_STAMP_COUNTER,
1552 };
1553
1554 static unsigned num_msrs_to_save;
1555
1556 static __init void kvm_init_msr_list(void)
1557 {
1558         u32 dummy[2];
1559         unsigned i, j;
1560
1561         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1562                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1563                         continue;
1564                 if (j < i)
1565                         msrs_to_save[j] = msrs_to_save[i];
1566                 j++;
1567         }
1568         num_msrs_to_save = j;
1569 }
1570
1571 /*
1572  * Adapt set_msr() to msr_io()'s calling convention
1573  */
1574 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1575 {
1576         return set_msr(vcpu, index, *data);
1577 }
1578
1579 /*
1580  * Read or write a bunch of msrs. All parameters are kernel addresses.
1581  *
1582  * @return number of msrs set successfully.
1583  */
1584 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1585                     struct kvm_msr_entry *entries,
1586                     int (*do_msr)(struct kvm_vcpu *vcpu,
1587                                   unsigned index, u64 *data))
1588 {
1589         struct kvm_vcpu *vcpu;
1590         int i;
1591
1592         if (!valid_vcpu(msrs->vcpu))
1593                 return -EINVAL;
1594
1595         vcpu = vcpu_load(kvm, msrs->vcpu);
1596         if (!vcpu)
1597                 return -ENOENT;
1598
1599         for (i = 0; i < msrs->nmsrs; ++i)
1600                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1601                         break;
1602
1603         vcpu_put(vcpu);
1604
1605         return i;
1606 }
1607
1608 /*
1609  * Read or write a bunch of msrs. Parameters are user addresses.
1610  *
1611  * @return number of msrs set successfully.
1612  */
1613 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1614                   int (*do_msr)(struct kvm_vcpu *vcpu,
1615                                 unsigned index, u64 *data),
1616                   int writeback)
1617 {
1618         struct kvm_msrs msrs;
1619         struct kvm_msr_entry *entries;
1620         int r, n;
1621         unsigned size;
1622
1623         r = -EFAULT;
1624         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1625                 goto out;
1626
1627         r = -E2BIG;
1628         if (msrs.nmsrs >= MAX_IO_MSRS)
1629                 goto out;
1630
1631         r = -ENOMEM;
1632         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1633         entries = vmalloc(size);
1634         if (!entries)
1635                 goto out;
1636
1637         r = -EFAULT;
1638         if (copy_from_user(entries, user_msrs->entries, size))
1639                 goto out_free;
1640
1641         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1642         if (r < 0)
1643                 goto out_free;
1644
1645         r = -EFAULT;
1646         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1647                 goto out_free;
1648
1649         r = n;
1650
1651 out_free:
1652         vfree(entries);
1653 out:
1654         return r;
1655 }
1656
1657 /*
1658  * Translate a guest virtual address to a guest physical address.
1659  */
1660 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1661 {
1662         unsigned long vaddr = tr->linear_address;
1663         struct kvm_vcpu *vcpu;
1664         gpa_t gpa;
1665
1666         vcpu = vcpu_load(kvm, tr->vcpu);
1667         if (!vcpu)
1668                 return -ENOENT;
1669         spin_lock(&kvm->lock);
1670         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1671         tr->physical_address = gpa;
1672         tr->valid = gpa != UNMAPPED_GVA;
1673         tr->writeable = 1;
1674         tr->usermode = 0;
1675         spin_unlock(&kvm->lock);
1676         vcpu_put(vcpu);
1677
1678         return 0;
1679 }
1680
1681 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1682 {
1683         struct kvm_vcpu *vcpu;
1684
1685         if (!valid_vcpu(irq->vcpu))
1686                 return -EINVAL;
1687         if (irq->irq < 0 || irq->irq >= 256)
1688                 return -EINVAL;
1689         vcpu = vcpu_load(kvm, irq->vcpu);
1690         if (!vcpu)
1691                 return -ENOENT;
1692
1693         set_bit(irq->irq, vcpu->irq_pending);
1694         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1695
1696         vcpu_put(vcpu);
1697
1698         return 0;
1699 }
1700
1701 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1702                                      struct kvm_debug_guest *dbg)
1703 {
1704         struct kvm_vcpu *vcpu;
1705         int r;
1706
1707         if (!valid_vcpu(dbg->vcpu))
1708                 return -EINVAL;
1709         vcpu = vcpu_load(kvm, dbg->vcpu);
1710         if (!vcpu)
1711                 return -ENOENT;
1712
1713         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1714
1715         vcpu_put(vcpu);
1716
1717         return r;
1718 }
1719
1720 static long kvm_dev_ioctl(struct file *filp,
1721                           unsigned int ioctl, unsigned long arg)
1722 {
1723         struct kvm *kvm = filp->private_data;
1724         int r = -EINVAL;
1725
1726         switch (ioctl) {
1727         case KVM_GET_API_VERSION:
1728                 r = KVM_API_VERSION;
1729                 break;
1730         case KVM_CREATE_VCPU: {
1731                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1732                 if (r)
1733                         goto out;
1734                 break;
1735         }
1736         case KVM_RUN: {
1737                 struct kvm_run kvm_run;
1738
1739                 r = -EFAULT;
1740                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1741                         goto out;
1742                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1743                 if (r < 0 &&  r != -EINTR)
1744                         goto out;
1745                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1746                         r = -EFAULT;
1747                         goto out;
1748                 }
1749                 break;
1750         }
1751         case KVM_GET_REGS: {
1752                 struct kvm_regs kvm_regs;
1753
1754                 r = -EFAULT;
1755                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1756                         goto out;
1757                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1758                 if (r)
1759                         goto out;
1760                 r = -EFAULT;
1761                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1762                         goto out;
1763                 r = 0;
1764                 break;
1765         }
1766         case KVM_SET_REGS: {
1767                 struct kvm_regs kvm_regs;
1768
1769                 r = -EFAULT;
1770                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1771                         goto out;
1772                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1773                 if (r)
1774                         goto out;
1775                 r = 0;
1776                 break;
1777         }
1778         case KVM_GET_SREGS: {
1779                 struct kvm_sregs kvm_sregs;
1780
1781                 r = -EFAULT;
1782                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1783                         goto out;
1784                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1785                 if (r)
1786                         goto out;
1787                 r = -EFAULT;
1788                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1789                         goto out;
1790                 r = 0;
1791                 break;
1792         }
1793         case KVM_SET_SREGS: {
1794                 struct kvm_sregs kvm_sregs;
1795
1796                 r = -EFAULT;
1797                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1798                         goto out;
1799                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1800                 if (r)
1801                         goto out;
1802                 r = 0;
1803                 break;
1804         }
1805         case KVM_TRANSLATE: {
1806                 struct kvm_translation tr;
1807
1808                 r = -EFAULT;
1809                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1810                         goto out;
1811                 r = kvm_dev_ioctl_translate(kvm, &tr);
1812                 if (r)
1813                         goto out;
1814                 r = -EFAULT;
1815                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1816                         goto out;
1817                 r = 0;
1818                 break;
1819         }
1820         case KVM_INTERRUPT: {
1821                 struct kvm_interrupt irq;
1822
1823                 r = -EFAULT;
1824                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1825                         goto out;
1826                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1827                 if (r)
1828                         goto out;
1829                 r = 0;
1830                 break;
1831         }
1832         case KVM_DEBUG_GUEST: {
1833                 struct kvm_debug_guest dbg;
1834
1835                 r = -EFAULT;
1836                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1837                         goto out;
1838                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1839                 if (r)
1840                         goto out;
1841                 r = 0;
1842                 break;
1843         }
1844         case KVM_SET_MEMORY_REGION: {
1845                 struct kvm_memory_region kvm_mem;
1846
1847                 r = -EFAULT;
1848                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1849                         goto out;
1850                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1851                 if (r)
1852                         goto out;
1853                 break;
1854         }
1855         case KVM_GET_DIRTY_LOG: {
1856                 struct kvm_dirty_log log;
1857
1858                 r = -EFAULT;
1859                 if (copy_from_user(&log, (void *)arg, sizeof log))
1860                         goto out;
1861                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1862                 if (r)
1863                         goto out;
1864                 break;
1865         }
1866         case KVM_GET_MSRS:
1867                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1868                 break;
1869         case KVM_SET_MSRS:
1870                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1871                 break;
1872         case KVM_GET_MSR_INDEX_LIST: {
1873                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1874                 struct kvm_msr_list msr_list;
1875                 unsigned n;
1876
1877                 r = -EFAULT;
1878                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1879                         goto out;
1880                 n = msr_list.nmsrs;
1881                 msr_list.nmsrs = num_msrs_to_save;
1882                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1883                         goto out;
1884                 r = -E2BIG;
1885                 if (n < num_msrs_to_save)
1886                         goto out;
1887                 r = -EFAULT;
1888                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1889                                  num_msrs_to_save * sizeof(u32)))
1890                         goto out;
1891                 r = 0;
1892         }
1893         default:
1894                 ;
1895         }
1896 out:
1897         return r;
1898 }
1899
1900 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1901                                    unsigned long address,
1902                                    int *type)
1903 {
1904         struct kvm *kvm = vma->vm_file->private_data;
1905         unsigned long pgoff;
1906         struct kvm_memory_slot *slot;
1907         struct page *page;
1908
1909         *type = VM_FAULT_MINOR;
1910         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1911         slot = gfn_to_memslot(kvm, pgoff);
1912         if (!slot)
1913                 return NOPAGE_SIGBUS;
1914         page = gfn_to_page(slot, pgoff);
1915         if (!page)
1916                 return NOPAGE_SIGBUS;
1917         get_page(page);
1918         return page;
1919 }
1920
1921 static struct vm_operations_struct kvm_dev_vm_ops = {
1922         .nopage = kvm_dev_nopage,
1923 };
1924
1925 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1926 {
1927         vma->vm_ops = &kvm_dev_vm_ops;
1928         return 0;
1929 }
1930
1931 static struct file_operations kvm_chardev_ops = {
1932         .open           = kvm_dev_open,
1933         .release        = kvm_dev_release,
1934         .unlocked_ioctl = kvm_dev_ioctl,
1935         .compat_ioctl   = kvm_dev_ioctl,
1936         .mmap           = kvm_dev_mmap,
1937 };
1938
1939 static struct miscdevice kvm_dev = {
1940         MISC_DYNAMIC_MINOR,
1941         "kvm",
1942         &kvm_chardev_ops,
1943 };
1944
1945 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1946                        void *v)
1947 {
1948         if (val == SYS_RESTART) {
1949                 /*
1950                  * Some (well, at least mine) BIOSes hang on reboot if
1951                  * in vmx root mode.
1952                  */
1953                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1954                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1955         }
1956         return NOTIFY_OK;
1957 }
1958
1959 static struct notifier_block kvm_reboot_notifier = {
1960         .notifier_call = kvm_reboot,
1961         .priority = 0,
1962 };
1963
1964 static __init void kvm_init_debug(void)
1965 {
1966         struct kvm_stats_debugfs_item *p;
1967
1968         debugfs_dir = debugfs_create_dir("kvm", 0);
1969         for (p = debugfs_entries; p->name; ++p)
1970                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1971                                                p->data);
1972 }
1973
1974 static void kvm_exit_debug(void)
1975 {
1976         struct kvm_stats_debugfs_item *p;
1977
1978         for (p = debugfs_entries; p->name; ++p)
1979                 debugfs_remove(p->dentry);
1980         debugfs_remove(debugfs_dir);
1981 }
1982
1983 hpa_t bad_page_address;
1984
1985 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1986 {
1987         int r;
1988
1989         if (kvm_arch_ops) {
1990                 printk(KERN_ERR "kvm: already loaded the other module\n");
1991                 return -EEXIST;
1992         }
1993
1994         if (!ops->cpu_has_kvm_support()) {
1995                 printk(KERN_ERR "kvm: no hardware support\n");
1996                 return -EOPNOTSUPP;
1997         }
1998         if (ops->disabled_by_bios()) {
1999                 printk(KERN_ERR "kvm: disabled by bios\n");
2000                 return -EOPNOTSUPP;
2001         }
2002
2003         kvm_arch_ops = ops;
2004
2005         r = kvm_arch_ops->hardware_setup();
2006         if (r < 0)
2007             return r;
2008
2009         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2010         register_reboot_notifier(&kvm_reboot_notifier);
2011
2012         kvm_chardev_ops.owner = module;
2013
2014         r = misc_register(&kvm_dev);
2015         if (r) {
2016                 printk (KERN_ERR "kvm: misc device register failed\n");
2017                 goto out_free;
2018         }
2019
2020         return r;
2021
2022 out_free:
2023         unregister_reboot_notifier(&kvm_reboot_notifier);
2024         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2025         kvm_arch_ops->hardware_unsetup();
2026         return r;
2027 }
2028
2029 void kvm_exit_arch(void)
2030 {
2031         misc_deregister(&kvm_dev);
2032
2033         unregister_reboot_notifier(&kvm_reboot_notifier);
2034         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2035         kvm_arch_ops->hardware_unsetup();
2036         kvm_arch_ops = NULL;
2037 }
2038
2039 static __init int kvm_init(void)
2040 {
2041         static struct page *bad_page;
2042         int r = 0;
2043
2044         kvm_init_debug();
2045
2046         kvm_init_msr_list();
2047
2048         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2049                 r = -ENOMEM;
2050                 goto out;
2051         }
2052
2053         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2054         memset(__va(bad_page_address), 0, PAGE_SIZE);
2055
2056         return r;
2057
2058 out:
2059         kvm_exit_debug();
2060         return r;
2061 }
2062
2063 static __exit void kvm_exit(void)
2064 {
2065         kvm_exit_debug();
2066         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2067 }
2068
2069 module_init(kvm_init)
2070 module_exit(kvm_exit)
2071
2072 EXPORT_SYMBOL_GPL(kvm_init_arch);
2073 EXPORT_SYMBOL_GPL(kvm_exit_arch);