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->kvm = kvm;
 234                 vcpu->mmu.root_hpa = INVALID_PAGE;
 235                 INIT_LIST_HEAD(&vcpu->free_pages);
 236         }
 237         filp->private_data = kvm;
 238         return 0;
 239 }
 240
 241 /*
 242  * Free any memory in @free but not in @dont.
 243  */
 244 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 245                                   struct kvm_memory_slot *dont)
 246 {
 247         int i;
 248
 249         if (!dont || free->phys_mem != dont->phys_mem)
 250                 if (free->phys_mem) {
 251                         for (i = 0; i < free->npages; ++i)
 252                                 if (free->phys_mem[i])
 253                                         __free_page(free->phys_mem[i]);
 254                         vfree(free->phys_mem);
 255                 }
 256
 257         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 258                 vfree(free->dirty_bitmap);
 259
 260         free->phys_mem = 0;
 261         free->npages = 0;
 262         free->dirty_bitmap = 0;
 263 }
 264
 265 static void kvm_free_physmem(struct kvm *kvm)
 266 {
 267         int i;
 268
 269         for (i = 0; i < kvm->nmemslots; ++i)
 270                 kvm_free_physmem_slot(&kvm->memslots[i], 0);
 271 }
 272
 273 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 274 {
 275         kvm_mmu_destroy(vcpu);
 276         kvm_arch_ops->vcpu_free(vcpu);
 277 }
 278
 279 static void kvm_free_vcpus(struct kvm *kvm)
 280 {
 281         unsigned int i;
 282
 283         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 284                 kvm_free_vcpu(&kvm->vcpus[i]);
 285 }
 286
 287 static int kvm_dev_release(struct inode *inode, struct file *filp)
 288 {
 289         struct kvm *kvm = filp->private_data;
 290
 291         kvm_free_vcpus(kvm);
 292         kvm_free_physmem(kvm);
 293         kfree(kvm);
 294         return 0;
 295 }
 296
 297 static void inject_gp(struct kvm_vcpu *vcpu)
 298 {
 299         kvm_arch_ops->inject_gp(vcpu, 0);
 300 }
 301
 302 /*
 303  * Load the pae pdptrs.  Return true is they are all valid.
 304  */
 305 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 306 {
 307         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 308         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 309         int i;
 310         u64 pdpte;
 311         u64 *pdpt;
 312         int ret;
 313         struct kvm_memory_slot *memslot;
 314
 315         spin_lock(&vcpu->kvm->lock);
 316         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 317         /* FIXME: !memslot - emulate? 0xff? */
 318         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 319
 320         ret = 1;
 321         for (i = 0; i < 4; ++i) {
 322                 pdpte = pdpt[offset + i];
 323                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 324                         ret = 0;
 325                         goto out;
 326                 }
 327         }
 328
 329         for (i = 0; i < 4; ++i)
 330                 vcpu->pdptrs[i] = pdpt[offset + i];
 331
 332 out:
 333         kunmap_atomic(pdpt, KM_USER0);
 334         spin_unlock(&vcpu->kvm->lock);
 335
 336         return ret;
 337 }
 338
 339 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 340 {
 341         if (cr0 & CR0_RESEVED_BITS) {
 342                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 343                        cr0, vcpu->cr0);
 344                 inject_gp(vcpu);
 345                 return;
 346         }
 347
 348         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 349                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 350                 inject_gp(vcpu);
 351                 return;
 352         }
 353
 354         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 355                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 356                        "and a clear PE flag\n");
 357                 inject_gp(vcpu);
 358                 return;
 359         }
 360
 361         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 362 #ifdef CONFIG_X86_64
 363                 if ((vcpu->shadow_efer & EFER_LME)) {
 364                         int cs_db, cs_l;
 365
 366                         if (!is_pae(vcpu)) {
 367                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 368                                        "in long mode while PAE is disabled\n");
 369                                 inject_gp(vcpu);
 370                                 return;
 371                         }
 372                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 373                         if (cs_l) {
 374                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 375                                        "in long mode while CS.L == 1\n");
 376                                 inject_gp(vcpu);
 377                                 return;
 378
 379                         }
 380                 } else
 381 #endif
 382                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 383                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 384                                "reserved bits\n");
 385                         inject_gp(vcpu);
 386                         return;
 387                 }
 388
 389         }
 390
 391         kvm_arch_ops->set_cr0(vcpu, cr0);
 392         vcpu->cr0 = cr0;
 393
 394         spin_lock(&vcpu->kvm->lock);
 395         kvm_mmu_reset_context(vcpu);
 396         spin_unlock(&vcpu->kvm->lock);
 397         return;
 398 }
 399 EXPORT_SYMBOL_GPL(set_cr0);
 400
 401 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 402 {
 403         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 404         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 405 }
 406 EXPORT_SYMBOL_GPL(lmsw);
 407
 408 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 409 {
 410         if (cr4 & CR4_RESEVED_BITS) {
 411                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 412                 inject_gp(vcpu);
 413                 return;
 414         }
 415
 416         if (is_long_mode(vcpu)) {
 417                 if (!(cr4 & CR4_PAE_MASK)) {
 418                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 419                                "in long mode\n");
 420                         inject_gp(vcpu);
 421                         return;
 422                 }
 423         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 424                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 425                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 426                 inject_gp(vcpu);
 427         }
 428
 429         if (cr4 & CR4_VMXE_MASK) {
 430                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 431                 inject_gp(vcpu);
 432                 return;
 433         }
 434         kvm_arch_ops->set_cr4(vcpu, cr4);
 435         spin_lock(&vcpu->kvm->lock);
 436         kvm_mmu_reset_context(vcpu);
 437         spin_unlock(&vcpu->kvm->lock);
 438 }
 439 EXPORT_SYMBOL_GPL(set_cr4);
 440
 441 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 442 {
 443         if (is_long_mode(vcpu)) {
 444                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
 445                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 446                         inject_gp(vcpu);
 447                         return;
 448                 }
 449         } else {
 450                 if (cr3 & CR3_RESEVED_BITS) {
 451                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 452                         inject_gp(vcpu);
 453                         return;
 454                 }
 455                 if (is_paging(vcpu) && is_pae(vcpu) &&
 456                     !load_pdptrs(vcpu, cr3)) {
 457                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 458                                "reserved bits\n");
 459                         inject_gp(vcpu);
 460                         return;
 461                 }
 462         }
 463
 464         vcpu->cr3 = cr3;
 465         spin_lock(&vcpu->kvm->lock);
 466         /*
 467          * Does the new cr3 value map to physical memory? (Note, we
 468          * catch an invalid cr3 even in real-mode, because it would
 469          * cause trouble later on when we turn on paging anyway.)
 470          *
 471          * A real CPU would silently accept an invalid cr3 and would
 472          * attempt to use it - with largely undefined (and often hard
 473          * to debug) behavior on the guest side.
 474          */
 475         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 476                 inject_gp(vcpu);
 477         else
 478                 vcpu->mmu.new_cr3(vcpu);
 479         spin_unlock(&vcpu->kvm->lock);
 480 }
 481 EXPORT_SYMBOL_GPL(set_cr3);
 482
 483 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 484 {
 485         if ( cr8 & CR8_RESEVED_BITS) {
 486                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 487                 inject_gp(vcpu);
 488                 return;
 489         }
 490         vcpu->cr8 = cr8;
 491 }
 492 EXPORT_SYMBOL_GPL(set_cr8);
 493
 494 void fx_init(struct kvm_vcpu *vcpu)
 495 {
 496         struct __attribute__ ((__packed__)) fx_image_s {
 497                 u16 control; //fcw
 498                 u16 status; //fsw
 499                 u16 tag; // ftw
 500                 u16 opcode; //fop
 501                 u64 ip; // fpu ip
 502                 u64 operand;// fpu dp
 503                 u32 mxcsr;
 504                 u32 mxcsr_mask;
 505
 506         } *fx_image;
 507
 508         fx_save(vcpu->host_fx_image);
 509         fpu_init();
 510         fx_save(vcpu->guest_fx_image);
 511         fx_restore(vcpu->host_fx_image);
 512
 513         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 514         fx_image->mxcsr = 0x1f80;
 515         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 516                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 517 }
 518 EXPORT_SYMBOL_GPL(fx_init);
 519
 520 /*
 521  * Creates some virtual cpus.  Good luck creating more than one.
 522  */
 523 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 524 {
 525         int r;
 526         struct kvm_vcpu *vcpu;
 527
 528         r = -EINVAL;
 529         if (!valid_vcpu(n))
 530                 goto out;
 531
 532         vcpu = &kvm->vcpus[n];
 533
 534         mutex_lock(&vcpu->mutex);
 535
 536         if (vcpu->vmcs) {
 537                 mutex_unlock(&vcpu->mutex);
 538                 return -EEXIST;
 539         }
 540
 541         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 542                                            FX_IMAGE_ALIGN);
 543         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 544
 545         vcpu->cpu = -1;  /* First load will set up TR */
 546         r = kvm_arch_ops->vcpu_create(vcpu);
 547         if (r < 0)
 548                 goto out_free_vcpus;
 549
 550         r = kvm_mmu_create(vcpu);
 551         if (r < 0)
 552                 goto out_free_vcpus;
 553
 554         kvm_arch_ops->vcpu_load(vcpu);
 555         r = kvm_mmu_setup(vcpu);
 556         if (r >= 0)
 557                 r = kvm_arch_ops->vcpu_setup(vcpu);
 558         vcpu_put(vcpu);
 559
 560         if (r < 0)
 561                 goto out_free_vcpus;
 562
 563         return 0;
 564
 565 out_free_vcpus:
 566         kvm_free_vcpu(vcpu);
 567         mutex_unlock(&vcpu->mutex);
 568 out:
 569         return r;
 570 }
 571
 572 /*
 573  * Allocate some memory and give it an address in the guest physical address
 574  * space.
 575  *
 576  * Discontiguous memory is allowed, mostly for framebuffers.
 577  */
 578 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 579                                            struct kvm_memory_region *mem)
 580 {
 581         int r;
 582         gfn_t base_gfn;
 583         unsigned long npages;
 584         unsigned long i;
 585         struct kvm_memory_slot *memslot;
 586         struct kvm_memory_slot old, new;
 587         int memory_config_version;
 588
 589         r = -EINVAL;
 590         /* General sanity checks */
 591         if (mem->memory_size & (PAGE_SIZE - 1))
 592                 goto out;
 593         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 594                 goto out;
 595         if (mem->slot >= KVM_MEMORY_SLOTS)
 596                 goto out;
 597         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 598                 goto out;
 599
 600         memslot = &kvm->memslots[mem->slot];
 601         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 602         npages = mem->memory_size >> PAGE_SHIFT;
 603
 604         if (!npages)
 605                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 606
 607 raced:
 608         spin_lock(&kvm->lock);
 609
 610         memory_config_version = kvm->memory_config_version;
 611         new = old = *memslot;
 612
 613         new.base_gfn = base_gfn;
 614         new.npages = npages;
 615         new.flags = mem->flags;
 616
 617         /* Disallow changing a memory slot's size. */
 618         r = -EINVAL;
 619         if (npages && old.npages && npages != old.npages)
 620                 goto out_unlock;
 621
 622         /* Check for overlaps */
 623         r = -EEXIST;
 624         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 625                 struct kvm_memory_slot *s = &kvm->memslots[i];
 626
 627                 if (s == memslot)
 628                         continue;
 629                 if (!((base_gfn + npages <= s->base_gfn) ||
 630                       (base_gfn >= s->base_gfn + s->npages)))
 631                         goto out_unlock;
 632         }
 633         /*
 634          * Do memory allocations outside lock.  memory_config_version will
 635          * detect any races.
 636          */
 637         spin_unlock(&kvm->lock);
 638
 639         /* Deallocate if slot is being removed */
 640         if (!npages)
 641                 new.phys_mem = 0;
 642
 643         /* Free page dirty bitmap if unneeded */
 644         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 645                 new.dirty_bitmap = 0;
 646
 647         r = -ENOMEM;
 648
 649         /* Allocate if a slot is being created */
 650         if (npages && !new.phys_mem) {
 651                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 652
 653                 if (!new.phys_mem)
 654                         goto out_free;
 655
 656                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 657                 for (i = 0; i < npages; ++i) {
 658                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 659                                                      | __GFP_ZERO);
 660                         if (!new.phys_mem[i])
 661                                 goto out_free;
 662                         new.phys_mem[i]->private = 0;
 663                 }
 664         }
 665
 666         /* Allocate page dirty bitmap if needed */
 667         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 668                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 669
 670                 new.dirty_bitmap = vmalloc(dirty_bytes);
 671                 if (!new.dirty_bitmap)
 672                         goto out_free;
 673                 memset(new.dirty_bitmap, 0, dirty_bytes);
 674         }
 675
 676         spin_lock(&kvm->lock);
 677
 678         if (memory_config_version != kvm->memory_config_version) {
 679                 spin_unlock(&kvm->lock);
 680                 kvm_free_physmem_slot(&new, &old);
 681                 goto raced;
 682         }
 683
 684         r = -EAGAIN;
 685         if (kvm->busy)
 686                 goto out_unlock;
 687
 688         if (mem->slot >= kvm->nmemslots)
 689                 kvm->nmemslots = mem->slot + 1;
 690
 691         *memslot = new;
 692         ++kvm->memory_config_version;
 693
 694         spin_unlock(&kvm->lock);
 695
 696         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 697                 struct kvm_vcpu *vcpu;
 698
 699                 vcpu = vcpu_load(kvm, i);
 700                 if (!vcpu)
 701                         continue;
 702                 kvm_mmu_reset_context(vcpu);
 703                 vcpu_put(vcpu);
 704         }
 705
 706         kvm_free_physmem_slot(&old, &new);
 707         return 0;
 708
 709 out_unlock:
 710         spin_unlock(&kvm->lock);
 711 out_free:
 712         kvm_free_physmem_slot(&new, &old);
 713 out:
 714         return r;
 715 }
 716
 717 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 718 {
 719         spin_lock(&vcpu->kvm->lock);
 720         kvm_mmu_slot_remove_write_access(vcpu, slot);
 721         spin_unlock(&vcpu->kvm->lock);
 722 }
 723
 724 /*
 725  * Get (and clear) the dirty memory log for a memory slot.
 726  */
 727 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 728                                        struct kvm_dirty_log *log)
 729 {
 730         struct kvm_memory_slot *memslot;
 731         int r, i;
 732         int n;
 733         int cleared;
 734         unsigned long any = 0;
 735
 736         spin_lock(&kvm->lock);
 737
 738         /*
 739          * Prevent changes to guest memory configuration even while the lock
 740          * is not taken.
 741          */
 742         ++kvm->busy;
 743         spin_unlock(&kvm->lock);
 744         r = -EINVAL;
 745         if (log->slot >= KVM_MEMORY_SLOTS)
 746                 goto out;
 747
 748         memslot = &kvm->memslots[log->slot];
 749         r = -ENOENT;
 750         if (!memslot->dirty_bitmap)
 751                 goto out;
 752
 753         n = ALIGN(memslot->npages, 8) / 8;
 754
 755         for (i = 0; !any && i < n; ++i)
 756                 any = memslot->dirty_bitmap[i];
 757
 758         r = -EFAULT;
 759         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 760                 goto out;
 761
 762
 763         if (any) {
 764                 cleared = 0;
 765                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 766                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 767
 768                         if (!vcpu)
 769                                 continue;
 770                         if (!cleared) {
 771                                 do_remove_write_access(vcpu, log->slot);
 772                                 memset(memslot->dirty_bitmap, 0, n);
 773                                 cleared = 1;
 774                         }
 775                         kvm_arch_ops->tlb_flush(vcpu);
 776                         vcpu_put(vcpu);
 777                 }
 778         }
 779
 780         r = 0;
 781
 782 out:
 783         spin_lock(&kvm->lock);
 784         --kvm->busy;
 785         spin_unlock(&kvm->lock);
 786         return r;
 787 }
 788
 789 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 790 {
 791         int i;
 792
 793         for (i = 0; i < kvm->nmemslots; ++i) {
 794                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 795
 796                 if (gfn >= memslot->base_gfn
 797                     && gfn < memslot->base_gfn + memslot->npages)
 798                         return memslot;
 799         }
 800         return 0;
 801 }
 802 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 803
 804 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 805 {
 806         int i;
 807         struct kvm_memory_slot *memslot = 0;
 808         unsigned long rel_gfn;
 809
 810         for (i = 0; i < kvm->nmemslots; ++i) {
 811                 memslot = &kvm->memslots[i];
 812
 813                 if (gfn >= memslot->base_gfn
 814                     && gfn < memslot->base_gfn + memslot->npages) {
 815
 816                         if (!memslot || !memslot->dirty_bitmap)
 817                                 return;
 818
 819                         rel_gfn = gfn - memslot->base_gfn;
 820
 821                         /* avoid RMW */
 822                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 823                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 824                         return;
 825                 }
 826         }
 827 }
 828
 829 static int emulator_read_std(unsigned long addr,
 830                              unsigned long *val,
 831                              unsigned int bytes,
 832                              struct x86_emulate_ctxt *ctxt)
 833 {
 834         struct kvm_vcpu *vcpu = ctxt->vcpu;
 835         void *data = val;
 836
 837         while (bytes) {
 838                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 839                 unsigned offset = addr & (PAGE_SIZE-1);
 840                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 841                 unsigned long pfn;
 842                 struct kvm_memory_slot *memslot;
 843                 void *page;
 844
 845                 if (gpa == UNMAPPED_GVA)
 846                         return X86EMUL_PROPAGATE_FAULT;
 847                 pfn = gpa >> PAGE_SHIFT;
 848                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 849                 if (!memslot)
 850                         return X86EMUL_UNHANDLEABLE;
 851                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 852
 853                 memcpy(data, page + offset, tocopy);
 854
 855                 kunmap_atomic(page, KM_USER0);
 856
 857                 bytes -= tocopy;
 858                 data += tocopy;
 859                 addr += tocopy;
 860         }
 861
 862         return X86EMUL_CONTINUE;
 863 }
 864
 865 static int emulator_write_std(unsigned long addr,
 866                               unsigned long val,
 867                               unsigned int bytes,
 868                               struct x86_emulate_ctxt *ctxt)
 869 {
 870         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 871                addr, bytes);
 872         return X86EMUL_UNHANDLEABLE;
 873 }
 874
 875 static int emulator_read_emulated(unsigned long addr,
 876                                   unsigned long *val,
 877                                   unsigned int bytes,
 878                                   struct x86_emulate_ctxt *ctxt)
 879 {
 880         struct kvm_vcpu *vcpu = ctxt->vcpu;
 881
 882         if (vcpu->mmio_read_completed) {
 883                 memcpy(val, vcpu->mmio_data, bytes);
 884                 vcpu->mmio_read_completed = 0;
 885                 return X86EMUL_CONTINUE;
 886         } else if (emulator_read_std(addr, val, bytes, ctxt)
 887                    == X86EMUL_CONTINUE)
 888                 return X86EMUL_CONTINUE;
 889         else {
 890                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 891                 if (gpa == UNMAPPED_GVA)
 892                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
 893                 vcpu->mmio_needed = 1;
 894                 vcpu->mmio_phys_addr = gpa;
 895                 vcpu->mmio_size = bytes;
 896                 vcpu->mmio_is_write = 0;
 897
 898                 return X86EMUL_UNHANDLEABLE;
 899         }
 900 }
 901
 902 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 903                                unsigned long val, int bytes)
 904 {
 905         struct kvm_memory_slot *m;
 906         struct page *page;
 907         void *virt;
 908
 909         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 910                 return 0;
 911         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 912         if (!m)
 913                 return 0;
 914         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 915         kvm_mmu_pre_write(vcpu, gpa, bytes);
 916         virt = kmap_atomic(page, KM_USER0);
 917         memcpy(virt + offset_in_page(gpa), &val, bytes);
 918         kunmap_atomic(virt, KM_USER0);
 919         kvm_mmu_post_write(vcpu, gpa, bytes);
 920         return 1;
 921 }
 922
 923 static int emulator_write_emulated(unsigned long addr,
 924                                    unsigned long val,
 925                                    unsigned int bytes,
 926                                    struct x86_emulate_ctxt *ctxt)
 927 {
 928         struct kvm_vcpu *vcpu = ctxt->vcpu;
 929         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 930
 931         if (gpa == UNMAPPED_GVA)
 932                 return X86EMUL_PROPAGATE_FAULT;
 933
 934         if (emulator_write_phys(vcpu, gpa, val, bytes))
 935                 return X86EMUL_CONTINUE;
 936
 937         vcpu->mmio_needed = 1;
 938         vcpu->mmio_phys_addr = gpa;
 939         vcpu->mmio_size = bytes;
 940         vcpu->mmio_is_write = 1;
 941         memcpy(vcpu->mmio_data, &val, bytes);
 942
 943         return X86EMUL_CONTINUE;
 944 }
 945
 946 static int emulator_cmpxchg_emulated(unsigned long addr,
 947                                      unsigned long old,
 948                                      unsigned long new,
 949                                      unsigned int bytes,
 950                                      struct x86_emulate_ctxt *ctxt)
 951 {
 952         static int reported;
 953
 954         if (!reported) {
 955                 reported = 1;
 956                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 957         }
 958         return emulator_write_emulated(addr, new, bytes, ctxt);
 959 }
 960
 961 #ifdef CONFIG_X86_32
 962
 963 static int emulator_cmpxchg8b_emulated(unsigned long addr,
 964                                        unsigned long old_lo,
 965                                        unsigned long old_hi,
 966                                        unsigned long new_lo,
 967                                        unsigned long new_hi,
 968                                        struct x86_emulate_ctxt *ctxt)
 969 {
 970         static int reported;
 971         int r;
 972
 973         if (!reported) {
 974                 reported = 1;
 975                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
 976         }
 977         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
 978         if (r != X86EMUL_CONTINUE)
 979                 return r;
 980         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
 981 }
 982
 983 #endif
 984
 985 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 986 {
 987         return kvm_arch_ops->get_segment_base(vcpu, seg);
 988 }
 989
 990 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 991 {
 992         return X86EMUL_CONTINUE;
 993 }
 994
 995 int emulate_clts(struct kvm_vcpu *vcpu)
 996 {
 997         unsigned long cr0;
 998
 999         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1000         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1001         kvm_arch_ops->set_cr0(vcpu, cr0);
1002         return X86EMUL_CONTINUE;
1003 }
1004
1005 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1006 {
1007         struct kvm_vcpu *vcpu = ctxt->vcpu;
1008
1009         switch (dr) {
1010         case 0 ... 3:
1011                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1012                 return X86EMUL_CONTINUE;
1013         default:
1014                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1015                        __FUNCTION__, dr);
1016                 return X86EMUL_UNHANDLEABLE;
1017         }
1018 }
1019
1020 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1021 {
1022         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1023         int exception;
1024
1025         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1026         if (exception) {
1027                 /* FIXME: better handling */
1028                 return X86EMUL_UNHANDLEABLE;
1029         }
1030         return X86EMUL_CONTINUE;
1031 }
1032
1033 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1034 {
1035         static int reported;
1036         u8 opcodes[4];
1037         unsigned long rip = ctxt->vcpu->rip;
1038         unsigned long rip_linear;
1039
1040         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1041
1042         if (reported)
1043                 return;
1044
1045         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1046
1047         printk(KERN_ERR "emulation failed but !mmio_needed?"
1048                " rip %lx %02x %02x %02x %02x\n",
1049                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1050         reported = 1;
1051 }
1052
1053 struct x86_emulate_ops emulate_ops = {
1054         .read_std            = emulator_read_std,
1055         .write_std           = emulator_write_std,
1056         .read_emulated       = emulator_read_emulated,
1057         .write_emulated      = emulator_write_emulated,
1058         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1059 #ifdef CONFIG_X86_32
1060         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1061 #endif
1062 };
1063
1064 int emulate_instruction(struct kvm_vcpu *vcpu,
1065                         struct kvm_run *run,
1066                         unsigned long cr2,
1067                         u16 error_code)
1068 {
1069         struct x86_emulate_ctxt emulate_ctxt;
1070         int r;
1071         int cs_db, cs_l;
1072
1073         kvm_arch_ops->cache_regs(vcpu);
1074
1075         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1076
1077         emulate_ctxt.vcpu = vcpu;
1078         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1079         emulate_ctxt.cr2 = cr2;
1080         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1081                 ? X86EMUL_MODE_REAL : cs_l
1082                 ? X86EMUL_MODE_PROT64 : cs_db
1083                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1084
1085         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1086                 emulate_ctxt.cs_base = 0;
1087                 emulate_ctxt.ds_base = 0;
1088                 emulate_ctxt.es_base = 0;
1089                 emulate_ctxt.ss_base = 0;
1090         } else {
1091                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1092                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1093                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1094                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1095         }
1096
1097         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1098         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1099
1100         vcpu->mmio_is_write = 0;
1101         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1102
1103         if ((r || vcpu->mmio_is_write) && run) {
1104                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1105                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1106                 run->mmio.len = vcpu->mmio_size;
1107                 run->mmio.is_write = vcpu->mmio_is_write;
1108         }
1109
1110         if (r) {
1111                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1112                         return EMULATE_DONE;
1113                 if (!vcpu->mmio_needed) {
1114                         report_emulation_failure(&emulate_ctxt);
1115                         return EMULATE_FAIL;
1116                 }
1117                 return EMULATE_DO_MMIO;
1118         }
1119
1120         kvm_arch_ops->decache_regs(vcpu);
1121         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1122
1123         if (vcpu->mmio_is_write)
1124                 return EMULATE_DO_MMIO;
1125
1126         return EMULATE_DONE;
1127 }
1128 EXPORT_SYMBOL_GPL(emulate_instruction);
1129
1130 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1131 {
1132         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1133 }
1134
1135 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1136 {
1137         struct descriptor_table dt = { limit, base };
1138
1139         kvm_arch_ops->set_gdt(vcpu, &dt);
1140 }
1141
1142 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1143 {
1144         struct descriptor_table dt = { limit, base };
1145
1146         kvm_arch_ops->set_idt(vcpu, &dt);
1147 }
1148
1149 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1150                    unsigned long *rflags)
1151 {
1152         lmsw(vcpu, msw);
1153         *rflags = kvm_arch_ops->get_rflags(vcpu);
1154 }
1155
1156 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1157 {
1158         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1159         switch (cr) {
1160         case 0:
1161                 return vcpu->cr0;
1162         case 2:
1163                 return vcpu->cr2;
1164         case 3:
1165                 return vcpu->cr3;
1166         case 4:
1167                 return vcpu->cr4;
1168         default:
1169                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1170                 return 0;
1171         }
1172 }
1173
1174 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1175                      unsigned long *rflags)
1176 {
1177         switch (cr) {
1178         case 0:
1179                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1180                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1181                 break;
1182         case 2:
1183                 vcpu->cr2 = val;
1184                 break;
1185         case 3:
1186                 set_cr3(vcpu, val);
1187                 break;
1188         case 4:
1189                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1190                 break;
1191         default:
1192                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1193         }
1194 }
1195
1196 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1197 {
1198         u64 data;
1199
1200         switch (msr) {
1201         case 0xc0010010: /* SYSCFG */
1202         case 0xc0010015: /* HWCR */
1203         case MSR_IA32_PLATFORM_ID:
1204         case MSR_IA32_P5_MC_ADDR:
1205         case MSR_IA32_P5_MC_TYPE:
1206         case MSR_IA32_MC0_CTL:
1207         case MSR_IA32_MCG_STATUS:
1208         case MSR_IA32_MCG_CAP:
1209         case MSR_IA32_MC0_MISC:
1210         case MSR_IA32_MC0_MISC+4:
1211         case MSR_IA32_MC0_MISC+8:
1212         case MSR_IA32_MC0_MISC+12:
1213         case MSR_IA32_MC0_MISC+16:
1214         case MSR_IA32_UCODE_REV:
1215         case MSR_IA32_PERF_STATUS:
1216                 /* MTRR registers */
1217         case 0xfe:
1218         case 0x200 ... 0x2ff:
1219                 data = 0;
1220                 break;
1221         case 0xcd: /* fsb frequency */
1222                 data = 3;
1223                 break;
1224         case MSR_IA32_APICBASE:
1225                 data = vcpu->apic_base;
1226                 break;
1227 #ifdef CONFIG_X86_64
1228         case MSR_EFER:
1229                 data = vcpu->shadow_efer;
1230                 break;
1231 #endif
1232         default:
1233                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1234                 return 1;
1235         }
1236         *pdata = data;
1237         return 0;
1238 }
1239 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1240
1241 /*
1242  * Reads an msr value (of 'msr_index') into 'pdata'.
1243  * Returns 0 on success, non-0 otherwise.
1244  * Assumes vcpu_load() was already called.
1245  */
1246 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1247 {
1248         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1249 }
1250
1251 #ifdef CONFIG_X86_64
1252
1253 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1254 {
1255         if (efer & EFER_RESERVED_BITS) {
1256                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1257                        efer);
1258                 inject_gp(vcpu);
1259                 return;
1260         }
1261
1262         if (is_paging(vcpu)
1263             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1264                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1265                 inject_gp(vcpu);
1266                 return;
1267         }
1268
1269         kvm_arch_ops->set_efer(vcpu, efer);
1270
1271         efer &= ~EFER_LMA;
1272         efer |= vcpu->shadow_efer & EFER_LMA;
1273
1274         vcpu->shadow_efer = efer;
1275 }
1276
1277 #endif
1278
1279 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1280 {
1281         switch (msr) {
1282 #ifdef CONFIG_X86_64
1283         case MSR_EFER:
1284                 set_efer(vcpu, data);
1285                 break;
1286 #endif
1287         case MSR_IA32_MC0_STATUS:
1288                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1289                        __FUNCTION__, data);
1290                 break;
1291         case MSR_IA32_UCODE_REV:
1292         case MSR_IA32_UCODE_WRITE:
1293         case 0x200 ... 0x2ff: /* MTRRs */
1294                 break;
1295         case MSR_IA32_APICBASE:
1296                 vcpu->apic_base = data;
1297                 break;
1298         default:
1299                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1300                 return 1;
1301         }
1302         return 0;
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1305
1306 /*
1307  * Writes msr value into into the appropriate "register".
1308  * Returns 0 on success, non-0 otherwise.
1309  * Assumes vcpu_load() was already called.
1310  */
1311 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1312 {
1313         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1314 }
1315
1316 void kvm_resched(struct kvm_vcpu *vcpu)
1317 {
1318         vcpu_put(vcpu);
1319         cond_resched();
1320         /* Cannot fail -  no vcpu unplug yet. */
1321         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_resched);
1324
1325 void load_msrs(struct vmx_msr_entry *e, int n)
1326 {
1327         int i;
1328
1329         for (i = 0; i < n; ++i)
1330                 wrmsrl(e[i].index, e[i].data);
1331 }
1332 EXPORT_SYMBOL_GPL(load_msrs);
1333
1334 void save_msrs(struct vmx_msr_entry *e, int n)
1335 {
1336         int i;
1337
1338         for (i = 0; i < n; ++i)
1339                 rdmsrl(e[i].index, e[i].data);
1340 }
1341 EXPORT_SYMBOL_GPL(save_msrs);
1342
1343 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1344 {
1345         struct kvm_vcpu *vcpu;
1346         int r;
1347
1348         if (!valid_vcpu(kvm_run->vcpu))
1349                 return -EINVAL;
1350
1351         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1352         if (!vcpu)
1353                 return -ENOENT;
1354
1355         if (kvm_run->emulated) {
1356                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1357                 kvm_run->emulated = 0;
1358         }
1359
1360         if (kvm_run->mmio_completed) {
1361                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1362                 vcpu->mmio_read_completed = 1;
1363         }
1364
1365         vcpu->mmio_needed = 0;
1366
1367         r = kvm_arch_ops->run(vcpu, kvm_run);
1368
1369         vcpu_put(vcpu);
1370         return r;
1371 }
1372
1373 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1374 {
1375         struct kvm_vcpu *vcpu;
1376
1377         if (!valid_vcpu(regs->vcpu))
1378                 return -EINVAL;
1379
1380         vcpu = vcpu_load(kvm, regs->vcpu);
1381         if (!vcpu)
1382                 return -ENOENT;
1383
1384         kvm_arch_ops->cache_regs(vcpu);
1385
1386         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1387         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1388         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1389         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1390         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1391         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1392         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1393         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1394 #ifdef CONFIG_X86_64
1395         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1396         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1397         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1398         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1399         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1400         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1401         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1402         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1403 #endif
1404
1405         regs->rip = vcpu->rip;
1406         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1407
1408         /*
1409          * Don't leak debug flags in case they were set for guest debugging
1410          */
1411         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1412                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1413
1414         vcpu_put(vcpu);
1415
1416         return 0;
1417 }
1418
1419 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1420 {
1421         struct kvm_vcpu *vcpu;
1422
1423         if (!valid_vcpu(regs->vcpu))
1424                 return -EINVAL;
1425
1426         vcpu = vcpu_load(kvm, regs->vcpu);
1427         if (!vcpu)
1428                 return -ENOENT;
1429
1430         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1431         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1432         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1433         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1434         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1435         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1436         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1437         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1438 #ifdef CONFIG_X86_64
1439         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1440         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1441         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1442         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1443         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1444         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1445         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1446         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1447 #endif
1448
1449         vcpu->rip = regs->rip;
1450         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1451
1452         kvm_arch_ops->decache_regs(vcpu);
1453
1454         vcpu_put(vcpu);
1455
1456         return 0;
1457 }
1458
1459 static void get_segment(struct kvm_vcpu *vcpu,
1460                         struct kvm_segment *var, int seg)
1461 {
1462         return kvm_arch_ops->get_segment(vcpu, var, seg);
1463 }
1464
1465 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1466 {
1467         struct kvm_vcpu *vcpu;
1468         struct descriptor_table dt;
1469
1470         if (!valid_vcpu(sregs->vcpu))
1471                 return -EINVAL;
1472         vcpu = vcpu_load(kvm, sregs->vcpu);
1473         if (!vcpu)
1474                 return -ENOENT;
1475
1476         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1477         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1478         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1479         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1480         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1481         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1482
1483         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1484         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1485
1486         kvm_arch_ops->get_idt(vcpu, &dt);
1487         sregs->idt.limit = dt.limit;
1488         sregs->idt.base = dt.base;
1489         kvm_arch_ops->get_gdt(vcpu, &dt);
1490         sregs->gdt.limit = dt.limit;
1491         sregs->gdt.base = dt.base;
1492
1493         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1494         sregs->cr0 = vcpu->cr0;
1495         sregs->cr2 = vcpu->cr2;
1496         sregs->cr3 = vcpu->cr3;
1497         sregs->cr4 = vcpu->cr4;
1498         sregs->cr8 = vcpu->cr8;
1499         sregs->efer = vcpu->shadow_efer;
1500         sregs->apic_base = vcpu->apic_base;
1501
1502         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1503                sizeof sregs->interrupt_bitmap);
1504
1505         vcpu_put(vcpu);
1506
1507         return 0;
1508 }
1509
1510 static void set_segment(struct kvm_vcpu *vcpu,
1511                         struct kvm_segment *var, int seg)
1512 {
1513         return kvm_arch_ops->set_segment(vcpu, var, seg);
1514 }
1515
1516 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1517 {
1518         struct kvm_vcpu *vcpu;
1519         int mmu_reset_needed = 0;
1520         int i;
1521         struct descriptor_table dt;
1522
1523         if (!valid_vcpu(sregs->vcpu))
1524                 return -EINVAL;
1525         vcpu = vcpu_load(kvm, sregs->vcpu);
1526         if (!vcpu)
1527                 return -ENOENT;
1528
1529         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1530         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1531         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1532         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1533         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1534         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1535
1536         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1537         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1538
1539         dt.limit = sregs->idt.limit;
1540         dt.base = sregs->idt.base;
1541         kvm_arch_ops->set_idt(vcpu, &dt);
1542         dt.limit = sregs->gdt.limit;
1543         dt.base = sregs->gdt.base;
1544         kvm_arch_ops->set_gdt(vcpu, &dt);
1545
1546         vcpu->cr2 = sregs->cr2;
1547         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1548         vcpu->cr3 = sregs->cr3;
1549
1550         vcpu->cr8 = sregs->cr8;
1551
1552         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1553 #ifdef CONFIG_X86_64
1554         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1555 #endif
1556         vcpu->apic_base = sregs->apic_base;
1557
1558         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1559
1560         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1561         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1562
1563         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1564         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1565         if (!is_long_mode(vcpu) && is_pae(vcpu))
1566                 load_pdptrs(vcpu, vcpu->cr3);
1567
1568         if (mmu_reset_needed)
1569                 kvm_mmu_reset_context(vcpu);
1570
1571         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1572                sizeof vcpu->irq_pending);
1573         vcpu->irq_summary = 0;
1574         for (i = 0; i < NR_IRQ_WORDS; ++i)
1575                 if (vcpu->irq_pending[i])
1576                         __set_bit(i, &vcpu->irq_summary);
1577
1578         vcpu_put(vcpu);
1579
1580         return 0;
1581 }
1582
1583 /*
1584  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1585  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1586  *
1587  * This list is modified at module load time to reflect the
1588  * capabilities of the host cpu.
1589  */
1590 static u32 msrs_to_save[] = {
1591         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1592         MSR_K6_STAR,
1593 #ifdef CONFIG_X86_64
1594         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1595 #endif
1596         MSR_IA32_TIME_STAMP_COUNTER,
1597 };
1598
1599 static unsigned num_msrs_to_save;
1600
1601 static __init void kvm_init_msr_list(void)
1602 {
1603         u32 dummy[2];
1604         unsigned i, j;
1605
1606         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1607                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1608                         continue;
1609                 if (j < i)
1610                         msrs_to_save[j] = msrs_to_save[i];
1611                 j++;
1612         }
1613         num_msrs_to_save = j;
1614 }
1615
1616 /*
1617  * Adapt set_msr() to msr_io()'s calling convention
1618  */
1619 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1620 {
1621         return set_msr(vcpu, index, *data);
1622 }
1623
1624 /*
1625  * Read or write a bunch of msrs. All parameters are kernel addresses.
1626  *
1627  * @return number of msrs set successfully.
1628  */
1629 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1630                     struct kvm_msr_entry *entries,
1631                     int (*do_msr)(struct kvm_vcpu *vcpu,
1632                                   unsigned index, u64 *data))
1633 {
1634         struct kvm_vcpu *vcpu;
1635         int i;
1636
1637         if (!valid_vcpu(msrs->vcpu))
1638                 return -EINVAL;
1639
1640         vcpu = vcpu_load(kvm, msrs->vcpu);
1641         if (!vcpu)
1642                 return -ENOENT;
1643
1644         for (i = 0; i < msrs->nmsrs; ++i)
1645                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1646                         break;
1647
1648         vcpu_put(vcpu);
1649
1650         return i;
1651 }
1652
1653 /*
1654  * Read or write a bunch of msrs. Parameters are user addresses.
1655  *
1656  * @return number of msrs set successfully.
1657  */
1658 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1659                   int (*do_msr)(struct kvm_vcpu *vcpu,
1660                                 unsigned index, u64 *data),
1661                   int writeback)
1662 {
1663         struct kvm_msrs msrs;
1664         struct kvm_msr_entry *entries;
1665         int r, n;
1666         unsigned size;
1667
1668         r = -EFAULT;
1669         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1670                 goto out;
1671
1672         r = -E2BIG;
1673         if (msrs.nmsrs >= MAX_IO_MSRS)
1674                 goto out;
1675
1676         r = -ENOMEM;
1677         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1678         entries = vmalloc(size);
1679         if (!entries)
1680                 goto out;
1681
1682         r = -EFAULT;
1683         if (copy_from_user(entries, user_msrs->entries, size))
1684                 goto out_free;
1685
1686         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1687         if (r < 0)
1688                 goto out_free;
1689
1690         r = -EFAULT;
1691         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1692                 goto out_free;
1693
1694         r = n;
1695
1696 out_free:
1697         vfree(entries);
1698 out:
1699         return r;
1700 }
1701
1702 /*
1703  * Translate a guest virtual address to a guest physical address.
1704  */
1705 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1706 {
1707         unsigned long vaddr = tr->linear_address;
1708         struct kvm_vcpu *vcpu;
1709         gpa_t gpa;
1710
1711         vcpu = vcpu_load(kvm, tr->vcpu);
1712         if (!vcpu)
1713                 return -ENOENT;
1714         spin_lock(&kvm->lock);
1715         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1716         tr->physical_address = gpa;
1717         tr->valid = gpa != UNMAPPED_GVA;
1718         tr->writeable = 1;
1719         tr->usermode = 0;
1720         spin_unlock(&kvm->lock);
1721         vcpu_put(vcpu);
1722
1723         return 0;
1724 }
1725
1726 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1727 {
1728         struct kvm_vcpu *vcpu;
1729
1730         if (!valid_vcpu(irq->vcpu))
1731                 return -EINVAL;
1732         if (irq->irq < 0 || irq->irq >= 256)
1733                 return -EINVAL;
1734         vcpu = vcpu_load(kvm, irq->vcpu);
1735         if (!vcpu)
1736                 return -ENOENT;
1737
1738         set_bit(irq->irq, vcpu->irq_pending);
1739         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1740
1741         vcpu_put(vcpu);
1742
1743         return 0;
1744 }
1745
1746 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1747                                      struct kvm_debug_guest *dbg)
1748 {
1749         struct kvm_vcpu *vcpu;
1750         int r;
1751
1752         if (!valid_vcpu(dbg->vcpu))
1753                 return -EINVAL;
1754         vcpu = vcpu_load(kvm, dbg->vcpu);
1755         if (!vcpu)
1756                 return -ENOENT;
1757
1758         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1759
1760         vcpu_put(vcpu);
1761
1762         return r;
1763 }
1764
1765 static long kvm_dev_ioctl(struct file *filp,
1766                           unsigned int ioctl, unsigned long arg)
1767 {
1768         struct kvm *kvm = filp->private_data;
1769         int r = -EINVAL;
1770
1771         switch (ioctl) {
1772         case KVM_GET_API_VERSION:
1773                 r = KVM_API_VERSION;
1774                 break;
1775         case KVM_CREATE_VCPU: {
1776                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1777                 if (r)
1778                         goto out;
1779                 break;
1780         }
1781         case KVM_RUN: {
1782                 struct kvm_run kvm_run;
1783
1784                 r = -EFAULT;
1785                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1786                         goto out;
1787                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1788                 if (r < 0 &&  r != -EINTR)
1789                         goto out;
1790                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1791                         r = -EFAULT;
1792                         goto out;
1793                 }
1794                 break;
1795         }
1796         case KVM_GET_REGS: {
1797                 struct kvm_regs kvm_regs;
1798
1799                 r = -EFAULT;
1800                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1801                         goto out;
1802                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1803                 if (r)
1804                         goto out;
1805                 r = -EFAULT;
1806                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1807                         goto out;
1808                 r = 0;
1809                 break;
1810         }
1811         case KVM_SET_REGS: {
1812                 struct kvm_regs kvm_regs;
1813
1814                 r = -EFAULT;
1815                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1816                         goto out;
1817                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1818                 if (r)
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_GET_SREGS: {
1824                 struct kvm_sregs kvm_sregs;
1825
1826                 r = -EFAULT;
1827                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1828                         goto out;
1829                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1830                 if (r)
1831                         goto out;
1832                 r = -EFAULT;
1833                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_SET_SREGS: {
1839                 struct kvm_sregs kvm_sregs;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1843                         goto out;
1844                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1845                 if (r)
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_TRANSLATE: {
1851                 struct kvm_translation tr;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1855                         goto out;
1856                 r = kvm_dev_ioctl_translate(kvm, &tr);
1857                 if (r)
1858                         goto out;
1859                 r = -EFAULT;
1860                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_INTERRUPT: {
1866                 struct kvm_interrupt irq;
1867
1868                 r = -EFAULT;
1869                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1870                         goto out;
1871                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1872                 if (r)
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_DEBUG_GUEST: {
1878                 struct kvm_debug_guest dbg;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1882                         goto out;
1883                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1884                 if (r)
1885                         goto out;
1886                 r = 0;
1887                 break;
1888         }
1889         case KVM_SET_MEMORY_REGION: {
1890                 struct kvm_memory_region kvm_mem;
1891
1892                 r = -EFAULT;
1893                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1894                         goto out;
1895                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1896                 if (r)
1897                         goto out;
1898                 break;
1899         }
1900         case KVM_GET_DIRTY_LOG: {
1901                 struct kvm_dirty_log log;
1902
1903                 r = -EFAULT;
1904                 if (copy_from_user(&log, (void *)arg, sizeof log))
1905                         goto out;
1906                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1907                 if (r)
1908                         goto out;
1909                 break;
1910         }
1911         case KVM_GET_MSRS:
1912                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1913                 break;
1914         case KVM_SET_MSRS:
1915                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1916                 break;
1917         case KVM_GET_MSR_INDEX_LIST: {
1918                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1919                 struct kvm_msr_list msr_list;
1920                 unsigned n;
1921
1922                 r = -EFAULT;
1923                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1924                         goto out;
1925                 n = msr_list.nmsrs;
1926                 msr_list.nmsrs = num_msrs_to_save;
1927                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1928                         goto out;
1929                 r = -E2BIG;
1930                 if (n < num_msrs_to_save)
1931                         goto out;
1932                 r = -EFAULT;
1933                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1934                                  num_msrs_to_save * sizeof(u32)))
1935                         goto out;
1936                 r = 0;
1937                 break;
1938         }
1939         default:
1940                 ;
1941         }
1942 out:
1943         return r;
1944 }
1945
1946 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1947                                    unsigned long address,
1948                                    int *type)
1949 {
1950         struct kvm *kvm = vma->vm_file->private_data;
1951         unsigned long pgoff;
1952         struct kvm_memory_slot *slot;
1953         struct page *page;
1954
1955         *type = VM_FAULT_MINOR;
1956         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1957         slot = gfn_to_memslot(kvm, pgoff);
1958         if (!slot)
1959                 return NOPAGE_SIGBUS;
1960         page = gfn_to_page(slot, pgoff);
1961         if (!page)
1962                 return NOPAGE_SIGBUS;
1963         get_page(page);
1964         return page;
1965 }
1966
1967 static struct vm_operations_struct kvm_dev_vm_ops = {
1968         .nopage = kvm_dev_nopage,
1969 };
1970
1971 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1972 {
1973         vma->vm_ops = &kvm_dev_vm_ops;
1974         return 0;
1975 }
1976
1977 static struct file_operations kvm_chardev_ops = {
1978         .open           = kvm_dev_open,
1979         .release        = kvm_dev_release,
1980         .unlocked_ioctl = kvm_dev_ioctl,
1981         .compat_ioctl   = kvm_dev_ioctl,
1982         .mmap           = kvm_dev_mmap,
1983 };
1984
1985 static struct miscdevice kvm_dev = {
1986         MISC_DYNAMIC_MINOR,
1987         "kvm",
1988         &kvm_chardev_ops,
1989 };
1990
1991 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1992                        void *v)
1993 {
1994         if (val == SYS_RESTART) {
1995                 /*
1996                  * Some (well, at least mine) BIOSes hang on reboot if
1997                  * in vmx root mode.
1998                  */
1999                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2000                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2001         }
2002         return NOTIFY_OK;
2003 }
2004
2005 static struct notifier_block kvm_reboot_notifier = {
2006         .notifier_call = kvm_reboot,
2007         .priority = 0,
2008 };
2009
2010 static __init void kvm_init_debug(void)
2011 {
2012         struct kvm_stats_debugfs_item *p;
2013
2014         debugfs_dir = debugfs_create_dir("kvm", 0);
2015         for (p = debugfs_entries; p->name; ++p)
2016                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2017                                                p->data);
2018 }
2019
2020 static void kvm_exit_debug(void)
2021 {
2022         struct kvm_stats_debugfs_item *p;
2023
2024         for (p = debugfs_entries; p->name; ++p)
2025                 debugfs_remove(p->dentry);
2026         debugfs_remove(debugfs_dir);
2027 }
2028
2029 hpa_t bad_page_address;
2030
2031 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2032 {
2033         int r;
2034
2035         if (kvm_arch_ops) {
2036                 printk(KERN_ERR "kvm: already loaded the other module\n");
2037                 return -EEXIST;
2038         }
2039
2040         if (!ops->cpu_has_kvm_support()) {
2041                 printk(KERN_ERR "kvm: no hardware support\n");
2042                 return -EOPNOTSUPP;
2043         }
2044         if (ops->disabled_by_bios()) {
2045                 printk(KERN_ERR "kvm: disabled by bios\n");
2046                 return -EOPNOTSUPP;
2047         }
2048
2049         kvm_arch_ops = ops;
2050
2051         r = kvm_arch_ops->hardware_setup();
2052         if (r < 0)
2053             return r;
2054
2055         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2056         register_reboot_notifier(&kvm_reboot_notifier);
2057
2058         kvm_chardev_ops.owner = module;
2059
2060         r = misc_register(&kvm_dev);
2061         if (r) {
2062                 printk (KERN_ERR "kvm: misc device register failed\n");
2063                 goto out_free;
2064         }
2065
2066         return r;
2067
2068 out_free:
2069         unregister_reboot_notifier(&kvm_reboot_notifier);
2070         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2071         kvm_arch_ops->hardware_unsetup();
2072         return r;
2073 }
2074
2075 void kvm_exit_arch(void)
2076 {
2077         misc_deregister(&kvm_dev);
2078
2079         unregister_reboot_notifier(&kvm_reboot_notifier);
2080         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2081         kvm_arch_ops->hardware_unsetup();
2082         kvm_arch_ops = NULL;
2083 }
2084
2085 static __init int kvm_init(void)
2086 {
2087         static struct page *bad_page;
2088         int r = 0;
2089
2090         kvm_init_debug();
2091
2092         kvm_init_msr_list();
2093
2094         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2095                 r = -ENOMEM;
2096                 goto out;
2097         }
2098
2099         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2100         memset(__va(bad_page_address), 0, PAGE_SIZE);
2101
2102         return r;
2103
2104 out:
2105         kvm_exit_debug();
2106         return r;
2107 }
2108
2109 static __exit void kvm_exit(void)
2110 {
2111         kvm_exit_debug();
2112         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2113 }
2114
2115 module_init(kvm_init)
2116 module_exit(kvm_exit)
2117
2118 EXPORT_SYMBOL_GPL(kvm_init_arch);
2119 EXPORT_SYMBOL_GPL(kvm_exit_arch);