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         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
 276         kvm_mmu_destroy(vcpu);
 277         vcpu_put(vcpu);
 278         kvm_arch_ops->vcpu_free(vcpu);
 279 }
 280
 281 static void kvm_free_vcpus(struct kvm *kvm)
 282 {
 283         unsigned int i;
 284
 285         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 286                 kvm_free_vcpu(&kvm->vcpus[i]);
 287 }
 288
 289 static int kvm_dev_release(struct inode *inode, struct file *filp)
 290 {
 291         struct kvm *kvm = filp->private_data;
 292
 293         kvm_free_vcpus(kvm);
 294         kvm_free_physmem(kvm);
 295         kfree(kvm);
 296         return 0;
 297 }
 298
 299 static void inject_gp(struct kvm_vcpu *vcpu)
 300 {
 301         kvm_arch_ops->inject_gp(vcpu, 0);
 302 }
 303
 304 /*
 305  * Load the pae pdptrs.  Return true is they are all valid.
 306  */
 307 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 308 {
 309         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 310         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 311         int i;
 312         u64 pdpte;
 313         u64 *pdpt;
 314         int ret;
 315         struct kvm_memory_slot *memslot;
 316
 317         spin_lock(&vcpu->kvm->lock);
 318         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 319         /* FIXME: !memslot - emulate? 0xff? */
 320         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 321
 322         ret = 1;
 323         for (i = 0; i < 4; ++i) {
 324                 pdpte = pdpt[offset + i];
 325                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 326                         ret = 0;
 327                         goto out;
 328                 }
 329         }
 330
 331         for (i = 0; i < 4; ++i)
 332                 vcpu->pdptrs[i] = pdpt[offset + i];
 333
 334 out:
 335         kunmap_atomic(pdpt, KM_USER0);
 336         spin_unlock(&vcpu->kvm->lock);
 337
 338         return ret;
 339 }
 340
 341 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 342 {
 343         if (cr0 & CR0_RESEVED_BITS) {
 344                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 345                        cr0, vcpu->cr0);
 346                 inject_gp(vcpu);
 347                 return;
 348         }
 349
 350         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 351                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 352                 inject_gp(vcpu);
 353                 return;
 354         }
 355
 356         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 357                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 358                        "and a clear PE flag\n");
 359                 inject_gp(vcpu);
 360                 return;
 361         }
 362
 363         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 364 #ifdef CONFIG_X86_64
 365                 if ((vcpu->shadow_efer & EFER_LME)) {
 366                         int cs_db, cs_l;
 367
 368                         if (!is_pae(vcpu)) {
 369                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 370                                        "in long mode while PAE is disabled\n");
 371                                 inject_gp(vcpu);
 372                                 return;
 373                         }
 374                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 375                         if (cs_l) {
 376                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 377                                        "in long mode while CS.L == 1\n");
 378                                 inject_gp(vcpu);
 379                                 return;
 380
 381                         }
 382                 } else
 383 #endif
 384                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 385                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 386                                "reserved bits\n");
 387                         inject_gp(vcpu);
 388                         return;
 389                 }
 390
 391         }
 392
 393         kvm_arch_ops->set_cr0(vcpu, cr0);
 394         vcpu->cr0 = cr0;
 395
 396         spin_lock(&vcpu->kvm->lock);
 397         kvm_mmu_reset_context(vcpu);
 398         spin_unlock(&vcpu->kvm->lock);
 399         return;
 400 }
 401 EXPORT_SYMBOL_GPL(set_cr0);
 402
 403 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 404 {
 405         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 406         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 407 }
 408 EXPORT_SYMBOL_GPL(lmsw);
 409
 410 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 411 {
 412         if (cr4 & CR4_RESEVED_BITS) {
 413                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 414                 inject_gp(vcpu);
 415                 return;
 416         }
 417
 418         if (is_long_mode(vcpu)) {
 419                 if (!(cr4 & CR4_PAE_MASK)) {
 420                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 421                                "in long mode\n");
 422                         inject_gp(vcpu);
 423                         return;
 424                 }
 425         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 426                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 427                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 428                 inject_gp(vcpu);
 429         }
 430
 431         if (cr4 & CR4_VMXE_MASK) {
 432                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 433                 inject_gp(vcpu);
 434                 return;
 435         }
 436         kvm_arch_ops->set_cr4(vcpu, cr4);
 437         spin_lock(&vcpu->kvm->lock);
 438         kvm_mmu_reset_context(vcpu);
 439         spin_unlock(&vcpu->kvm->lock);
 440 }
 441 EXPORT_SYMBOL_GPL(set_cr4);
 442
 443 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 444 {
 445         if (is_long_mode(vcpu)) {
 446                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
 447                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 448                         inject_gp(vcpu);
 449                         return;
 450                 }
 451         } else {
 452                 if (cr3 & CR3_RESEVED_BITS) {
 453                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 454                         inject_gp(vcpu);
 455                         return;
 456                 }
 457                 if (is_paging(vcpu) && is_pae(vcpu) &&
 458                     !load_pdptrs(vcpu, cr3)) {
 459                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 460                                "reserved bits\n");
 461                         inject_gp(vcpu);
 462                         return;
 463                 }
 464         }
 465
 466         vcpu->cr3 = cr3;
 467         spin_lock(&vcpu->kvm->lock);
 468         /*
 469          * Does the new cr3 value map to physical memory? (Note, we
 470          * catch an invalid cr3 even in real-mode, because it would
 471          * cause trouble later on when we turn on paging anyway.)
 472          *
 473          * A real CPU would silently accept an invalid cr3 and would
 474          * attempt to use it - with largely undefined (and often hard
 475          * to debug) behavior on the guest side.
 476          */
 477         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 478                 inject_gp(vcpu);
 479         else
 480                 vcpu->mmu.new_cr3(vcpu);
 481         spin_unlock(&vcpu->kvm->lock);
 482 }
 483 EXPORT_SYMBOL_GPL(set_cr3);
 484
 485 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 486 {
 487         if ( cr8 & CR8_RESEVED_BITS) {
 488                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 489                 inject_gp(vcpu);
 490                 return;
 491         }
 492         vcpu->cr8 = cr8;
 493 }
 494 EXPORT_SYMBOL_GPL(set_cr8);
 495
 496 void fx_init(struct kvm_vcpu *vcpu)
 497 {
 498         struct __attribute__ ((__packed__)) fx_image_s {
 499                 u16 control; //fcw
 500                 u16 status; //fsw
 501                 u16 tag; // ftw
 502                 u16 opcode; //fop
 503                 u64 ip; // fpu ip
 504                 u64 operand;// fpu dp
 505                 u32 mxcsr;
 506                 u32 mxcsr_mask;
 507
 508         } *fx_image;
 509
 510         fx_save(vcpu->host_fx_image);
 511         fpu_init();
 512         fx_save(vcpu->guest_fx_image);
 513         fx_restore(vcpu->host_fx_image);
 514
 515         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 516         fx_image->mxcsr = 0x1f80;
 517         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 518                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 519 }
 520 EXPORT_SYMBOL_GPL(fx_init);
 521
 522 /*
 523  * Creates some virtual cpus.  Good luck creating more than one.
 524  */
 525 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 526 {
 527         int r;
 528         struct kvm_vcpu *vcpu;
 529
 530         r = -EINVAL;
 531         if (!valid_vcpu(n))
 532                 goto out;
 533
 534         vcpu = &kvm->vcpus[n];
 535
 536         mutex_lock(&vcpu->mutex);
 537
 538         if (vcpu->vmcs) {
 539                 mutex_unlock(&vcpu->mutex);
 540                 return -EEXIST;
 541         }
 542
 543         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 544                                            FX_IMAGE_ALIGN);
 545         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 546
 547         vcpu->cpu = -1;  /* First load will set up TR */
 548         r = kvm_arch_ops->vcpu_create(vcpu);
 549         if (r < 0)
 550                 goto out_free_vcpus;
 551
 552         r = kvm_mmu_create(vcpu);
 553         if (r < 0)
 554                 goto out_free_vcpus;
 555
 556         kvm_arch_ops->vcpu_load(vcpu);
 557         r = kvm_mmu_setup(vcpu);
 558         if (r >= 0)
 559                 r = kvm_arch_ops->vcpu_setup(vcpu);
 560         vcpu_put(vcpu);
 561
 562         if (r < 0)
 563                 goto out_free_vcpus;
 564
 565         return 0;
 566
 567 out_free_vcpus:
 568         kvm_free_vcpu(vcpu);
 569         mutex_unlock(&vcpu->mutex);
 570 out:
 571         return r;
 572 }
 573
 574 /*
 575  * Allocate some memory and give it an address in the guest physical address
 576  * space.
 577  *
 578  * Discontiguous memory is allowed, mostly for framebuffers.
 579  */
 580 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 581                                            struct kvm_memory_region *mem)
 582 {
 583         int r;
 584         gfn_t base_gfn;
 585         unsigned long npages;
 586         unsigned long i;
 587         struct kvm_memory_slot *memslot;
 588         struct kvm_memory_slot old, new;
 589         int memory_config_version;
 590
 591         r = -EINVAL;
 592         /* General sanity checks */
 593         if (mem->memory_size & (PAGE_SIZE - 1))
 594                 goto out;
 595         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 596                 goto out;
 597         if (mem->slot >= KVM_MEMORY_SLOTS)
 598                 goto out;
 599         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 600                 goto out;
 601
 602         memslot = &kvm->memslots[mem->slot];
 603         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 604         npages = mem->memory_size >> PAGE_SHIFT;
 605
 606         if (!npages)
 607                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 608
 609 raced:
 610         spin_lock(&kvm->lock);
 611
 612         memory_config_version = kvm->memory_config_version;
 613         new = old = *memslot;
 614
 615         new.base_gfn = base_gfn;
 616         new.npages = npages;
 617         new.flags = mem->flags;
 618
 619         /* Disallow changing a memory slot's size. */
 620         r = -EINVAL;
 621         if (npages && old.npages && npages != old.npages)
 622                 goto out_unlock;
 623
 624         /* Check for overlaps */
 625         r = -EEXIST;
 626         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 627                 struct kvm_memory_slot *s = &kvm->memslots[i];
 628
 629                 if (s == memslot)
 630                         continue;
 631                 if (!((base_gfn + npages <= s->base_gfn) ||
 632                       (base_gfn >= s->base_gfn + s->npages)))
 633                         goto out_unlock;
 634         }
 635         /*
 636          * Do memory allocations outside lock.  memory_config_version will
 637          * detect any races.
 638          */
 639         spin_unlock(&kvm->lock);
 640
 641         /* Deallocate if slot is being removed */
 642         if (!npages)
 643                 new.phys_mem = 0;
 644
 645         /* Free page dirty bitmap if unneeded */
 646         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 647                 new.dirty_bitmap = 0;
 648
 649         r = -ENOMEM;
 650
 651         /* Allocate if a slot is being created */
 652         if (npages && !new.phys_mem) {
 653                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 654
 655                 if (!new.phys_mem)
 656                         goto out_free;
 657
 658                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 659                 for (i = 0; i < npages; ++i) {
 660                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 661                                                      | __GFP_ZERO);
 662                         if (!new.phys_mem[i])
 663                                 goto out_free;
 664                         new.phys_mem[i]->private = 0;
 665                 }
 666         }
 667
 668         /* Allocate page dirty bitmap if needed */
 669         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 670                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 671
 672                 new.dirty_bitmap = vmalloc(dirty_bytes);
 673                 if (!new.dirty_bitmap)
 674                         goto out_free;
 675                 memset(new.dirty_bitmap, 0, dirty_bytes);
 676         }
 677
 678         spin_lock(&kvm->lock);
 679
 680         if (memory_config_version != kvm->memory_config_version) {
 681                 spin_unlock(&kvm->lock);
 682                 kvm_free_physmem_slot(&new, &old);
 683                 goto raced;
 684         }
 685
 686         r = -EAGAIN;
 687         if (kvm->busy)
 688                 goto out_unlock;
 689
 690         if (mem->slot >= kvm->nmemslots)
 691                 kvm->nmemslots = mem->slot + 1;
 692
 693         *memslot = new;
 694         ++kvm->memory_config_version;
 695
 696         spin_unlock(&kvm->lock);
 697
 698         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 699                 struct kvm_vcpu *vcpu;
 700
 701                 vcpu = vcpu_load(kvm, i);
 702                 if (!vcpu)
 703                         continue;
 704                 kvm_mmu_reset_context(vcpu);
 705                 vcpu_put(vcpu);
 706         }
 707
 708         kvm_free_physmem_slot(&old, &new);
 709         return 0;
 710
 711 out_unlock:
 712         spin_unlock(&kvm->lock);
 713 out_free:
 714         kvm_free_physmem_slot(&new, &old);
 715 out:
 716         return r;
 717 }
 718
 719 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 720 {
 721         spin_lock(&vcpu->kvm->lock);
 722         kvm_mmu_slot_remove_write_access(vcpu, slot);
 723         spin_unlock(&vcpu->kvm->lock);
 724 }
 725
 726 /*
 727  * Get (and clear) the dirty memory log for a memory slot.
 728  */
 729 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 730                                        struct kvm_dirty_log *log)
 731 {
 732         struct kvm_memory_slot *memslot;
 733         int r, i;
 734         int n;
 735         int cleared;
 736         unsigned long any = 0;
 737
 738         spin_lock(&kvm->lock);
 739
 740         /*
 741          * Prevent changes to guest memory configuration even while the lock
 742          * is not taken.
 743          */
 744         ++kvm->busy;
 745         spin_unlock(&kvm->lock);
 746         r = -EINVAL;
 747         if (log->slot >= KVM_MEMORY_SLOTS)
 748                 goto out;
 749
 750         memslot = &kvm->memslots[log->slot];
 751         r = -ENOENT;
 752         if (!memslot->dirty_bitmap)
 753                 goto out;
 754
 755         n = ALIGN(memslot->npages, 8) / 8;
 756
 757         for (i = 0; !any && i < n; ++i)
 758                 any = memslot->dirty_bitmap[i];
 759
 760         r = -EFAULT;
 761         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 762                 goto out;
 763
 764
 765         if (any) {
 766                 cleared = 0;
 767                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 768                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 769
 770                         if (!vcpu)
 771                                 continue;
 772                         if (!cleared) {
 773                                 do_remove_write_access(vcpu, log->slot);
 774                                 memset(memslot->dirty_bitmap, 0, n);
 775                                 cleared = 1;
 776                         }
 777                         kvm_arch_ops->tlb_flush(vcpu);
 778                         vcpu_put(vcpu);
 779                 }
 780         }
 781
 782         r = 0;
 783
 784 out:
 785         spin_lock(&kvm->lock);
 786         --kvm->busy;
 787         spin_unlock(&kvm->lock);
 788         return r;
 789 }
 790
 791 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 792 {
 793         int i;
 794
 795         for (i = 0; i < kvm->nmemslots; ++i) {
 796                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 797
 798                 if (gfn >= memslot->base_gfn
 799                     && gfn < memslot->base_gfn + memslot->npages)
 800                         return memslot;
 801         }
 802         return 0;
 803 }
 804 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 805
 806 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 807 {
 808         int i;
 809         struct kvm_memory_slot *memslot = 0;
 810         unsigned long rel_gfn;
 811
 812         for (i = 0; i < kvm->nmemslots; ++i) {
 813                 memslot = &kvm->memslots[i];
 814
 815                 if (gfn >= memslot->base_gfn
 816                     && gfn < memslot->base_gfn + memslot->npages) {
 817
 818                         if (!memslot || !memslot->dirty_bitmap)
 819                                 return;
 820
 821                         rel_gfn = gfn - memslot->base_gfn;
 822
 823                         /* avoid RMW */
 824                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 825                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 826                         return;
 827                 }
 828         }
 829 }
 830
 831 static int emulator_read_std(unsigned long addr,
 832                              unsigned long *val,
 833                              unsigned int bytes,
 834                              struct x86_emulate_ctxt *ctxt)
 835 {
 836         struct kvm_vcpu *vcpu = ctxt->vcpu;
 837         void *data = val;
 838
 839         while (bytes) {
 840                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 841                 unsigned offset = addr & (PAGE_SIZE-1);
 842                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 843                 unsigned long pfn;
 844                 struct kvm_memory_slot *memslot;
 845                 void *page;
 846
 847                 if (gpa == UNMAPPED_GVA)
 848                         return X86EMUL_PROPAGATE_FAULT;
 849                 pfn = gpa >> PAGE_SHIFT;
 850                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 851                 if (!memslot)
 852                         return X86EMUL_UNHANDLEABLE;
 853                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 854
 855                 memcpy(data, page + offset, tocopy);
 856
 857                 kunmap_atomic(page, KM_USER0);
 858
 859                 bytes -= tocopy;
 860                 data += tocopy;
 861                 addr += tocopy;
 862         }
 863
 864         return X86EMUL_CONTINUE;
 865 }
 866
 867 static int emulator_write_std(unsigned long addr,
 868                               unsigned long val,
 869                               unsigned int bytes,
 870                               struct x86_emulate_ctxt *ctxt)
 871 {
 872         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 873                addr, bytes);
 874         return X86EMUL_UNHANDLEABLE;
 875 }
 876
 877 static int emulator_read_emulated(unsigned long addr,
 878                                   unsigned long *val,
 879                                   unsigned int bytes,
 880                                   struct x86_emulate_ctxt *ctxt)
 881 {
 882         struct kvm_vcpu *vcpu = ctxt->vcpu;
 883
 884         if (vcpu->mmio_read_completed) {
 885                 memcpy(val, vcpu->mmio_data, bytes);
 886                 vcpu->mmio_read_completed = 0;
 887                 return X86EMUL_CONTINUE;
 888         } else if (emulator_read_std(addr, val, bytes, ctxt)
 889                    == X86EMUL_CONTINUE)
 890                 return X86EMUL_CONTINUE;
 891         else {
 892                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 893                 if (gpa == UNMAPPED_GVA)
 894                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
 895                 vcpu->mmio_needed = 1;
 896                 vcpu->mmio_phys_addr = gpa;
 897                 vcpu->mmio_size = bytes;
 898                 vcpu->mmio_is_write = 0;
 899
 900                 return X86EMUL_UNHANDLEABLE;
 901         }
 902 }
 903
 904 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 905                                unsigned long val, int bytes)
 906 {
 907         struct kvm_memory_slot *m;
 908         struct page *page;
 909         void *virt;
 910
 911         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 912                 return 0;
 913         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 914         if (!m)
 915                 return 0;
 916         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 917         kvm_mmu_pre_write(vcpu, gpa, bytes);
 918         virt = kmap_atomic(page, KM_USER0);
 919         memcpy(virt + offset_in_page(gpa), &val, bytes);
 920         kunmap_atomic(virt, KM_USER0);
 921         kvm_mmu_post_write(vcpu, gpa, bytes);
 922         return 1;
 923 }
 924
 925 static int emulator_write_emulated(unsigned long addr,
 926                                    unsigned long val,
 927                                    unsigned int bytes,
 928                                    struct x86_emulate_ctxt *ctxt)
 929 {
 930         struct kvm_vcpu *vcpu = ctxt->vcpu;
 931         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 932
 933         if (gpa == UNMAPPED_GVA)
 934                 return X86EMUL_PROPAGATE_FAULT;
 935
 936         if (emulator_write_phys(vcpu, gpa, val, bytes))
 937                 return X86EMUL_CONTINUE;
 938
 939         vcpu->mmio_needed = 1;
 940         vcpu->mmio_phys_addr = gpa;
 941         vcpu->mmio_size = bytes;
 942         vcpu->mmio_is_write = 1;
 943         memcpy(vcpu->mmio_data, &val, bytes);
 944
 945         return X86EMUL_CONTINUE;
 946 }
 947
 948 static int emulator_cmpxchg_emulated(unsigned long addr,
 949                                      unsigned long old,
 950                                      unsigned long new,
 951                                      unsigned int bytes,
 952                                      struct x86_emulate_ctxt *ctxt)
 953 {
 954         static int reported;
 955
 956         if (!reported) {
 957                 reported = 1;
 958                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 959         }
 960         return emulator_write_emulated(addr, new, bytes, ctxt);
 961 }
 962
 963 #ifdef CONFIG_X86_32
 964
 965 static int emulator_cmpxchg8b_emulated(unsigned long addr,
 966                                        unsigned long old_lo,
 967                                        unsigned long old_hi,
 968                                        unsigned long new_lo,
 969                                        unsigned long new_hi,
 970                                        struct x86_emulate_ctxt *ctxt)
 971 {
 972         static int reported;
 973         int r;
 974
 975         if (!reported) {
 976                 reported = 1;
 977                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
 978         }
 979         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
 980         if (r != X86EMUL_CONTINUE)
 981                 return r;
 982         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
 983 }
 984
 985 #endif
 986
 987 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 988 {
 989         return kvm_arch_ops->get_segment_base(vcpu, seg);
 990 }
 991
 992 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 993 {
 994         return X86EMUL_CONTINUE;
 995 }
 996
 997 int emulate_clts(struct kvm_vcpu *vcpu)
 998 {
 999         unsigned long cr0;
1000
1001         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1002         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1003         kvm_arch_ops->set_cr0(vcpu, cr0);
1004         return X86EMUL_CONTINUE;
1005 }
1006
1007 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1008 {
1009         struct kvm_vcpu *vcpu = ctxt->vcpu;
1010
1011         switch (dr) {
1012         case 0 ... 3:
1013                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1014                 return X86EMUL_CONTINUE;
1015         default:
1016                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1017                        __FUNCTION__, dr);
1018                 return X86EMUL_UNHANDLEABLE;
1019         }
1020 }
1021
1022 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1023 {
1024         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1025         int exception;
1026
1027         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1028         if (exception) {
1029                 /* FIXME: better handling */
1030                 return X86EMUL_UNHANDLEABLE;
1031         }
1032         return X86EMUL_CONTINUE;
1033 }
1034
1035 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1036 {
1037         static int reported;
1038         u8 opcodes[4];
1039         unsigned long rip = ctxt->vcpu->rip;
1040         unsigned long rip_linear;
1041
1042         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1043
1044         if (reported)
1045                 return;
1046
1047         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1048
1049         printk(KERN_ERR "emulation failed but !mmio_needed?"
1050                " rip %lx %02x %02x %02x %02x\n",
1051                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1052         reported = 1;
1053 }
1054
1055 struct x86_emulate_ops emulate_ops = {
1056         .read_std            = emulator_read_std,
1057         .write_std           = emulator_write_std,
1058         .read_emulated       = emulator_read_emulated,
1059         .write_emulated      = emulator_write_emulated,
1060         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1061 #ifdef CONFIG_X86_32
1062         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1063 #endif
1064 };
1065
1066 int emulate_instruction(struct kvm_vcpu *vcpu,
1067                         struct kvm_run *run,
1068                         unsigned long cr2,
1069                         u16 error_code)
1070 {
1071         struct x86_emulate_ctxt emulate_ctxt;
1072         int r;
1073         int cs_db, cs_l;
1074
1075         kvm_arch_ops->cache_regs(vcpu);
1076
1077         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1078
1079         emulate_ctxt.vcpu = vcpu;
1080         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1081         emulate_ctxt.cr2 = cr2;
1082         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1083                 ? X86EMUL_MODE_REAL : cs_l
1084                 ? X86EMUL_MODE_PROT64 : cs_db
1085                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1086
1087         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1088                 emulate_ctxt.cs_base = 0;
1089                 emulate_ctxt.ds_base = 0;
1090                 emulate_ctxt.es_base = 0;
1091                 emulate_ctxt.ss_base = 0;
1092         } else {
1093                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1094                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1095                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1096                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1097         }
1098
1099         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1100         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1101
1102         vcpu->mmio_is_write = 0;
1103         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1104
1105         if ((r || vcpu->mmio_is_write) && run) {
1106                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1107                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1108                 run->mmio.len = vcpu->mmio_size;
1109                 run->mmio.is_write = vcpu->mmio_is_write;
1110         }
1111
1112         if (r) {
1113                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1114                         return EMULATE_DONE;
1115                 if (!vcpu->mmio_needed) {
1116                         report_emulation_failure(&emulate_ctxt);
1117                         return EMULATE_FAIL;
1118                 }
1119                 return EMULATE_DO_MMIO;
1120         }
1121
1122         kvm_arch_ops->decache_regs(vcpu);
1123         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1124
1125         if (vcpu->mmio_is_write)
1126                 return EMULATE_DO_MMIO;
1127
1128         return EMULATE_DONE;
1129 }
1130 EXPORT_SYMBOL_GPL(emulate_instruction);
1131
1132 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1133 {
1134         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1135 }
1136
1137 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1138 {
1139         struct descriptor_table dt = { limit, base };
1140
1141         kvm_arch_ops->set_gdt(vcpu, &dt);
1142 }
1143
1144 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1145 {
1146         struct descriptor_table dt = { limit, base };
1147
1148         kvm_arch_ops->set_idt(vcpu, &dt);
1149 }
1150
1151 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1152                    unsigned long *rflags)
1153 {
1154         lmsw(vcpu, msw);
1155         *rflags = kvm_arch_ops->get_rflags(vcpu);
1156 }
1157
1158 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1159 {
1160         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1161         switch (cr) {
1162         case 0:
1163                 return vcpu->cr0;
1164         case 2:
1165                 return vcpu->cr2;
1166         case 3:
1167                 return vcpu->cr3;
1168         case 4:
1169                 return vcpu->cr4;
1170         default:
1171                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1172                 return 0;
1173         }
1174 }
1175
1176 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1177                      unsigned long *rflags)
1178 {
1179         switch (cr) {
1180         case 0:
1181                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1182                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1183                 break;
1184         case 2:
1185                 vcpu->cr2 = val;
1186                 break;
1187         case 3:
1188                 set_cr3(vcpu, val);
1189                 break;
1190         case 4:
1191                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1192                 break;
1193         default:
1194                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1195         }
1196 }
1197
1198 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1199 {
1200         u64 data;
1201
1202         switch (msr) {
1203         case 0xc0010010: /* SYSCFG */
1204         case 0xc0010015: /* HWCR */
1205         case MSR_IA32_PLATFORM_ID:
1206         case MSR_IA32_P5_MC_ADDR:
1207         case MSR_IA32_P5_MC_TYPE:
1208         case MSR_IA32_MC0_CTL:
1209         case MSR_IA32_MCG_STATUS:
1210         case MSR_IA32_MCG_CAP:
1211         case MSR_IA32_MC0_MISC:
1212         case MSR_IA32_MC0_MISC+4:
1213         case MSR_IA32_MC0_MISC+8:
1214         case MSR_IA32_MC0_MISC+12:
1215         case MSR_IA32_MC0_MISC+16:
1216         case MSR_IA32_UCODE_REV:
1217         case MSR_IA32_PERF_STATUS:
1218                 /* MTRR registers */
1219         case 0xfe:
1220         case 0x200 ... 0x2ff:
1221                 data = 0;
1222                 break;
1223         case 0xcd: /* fsb frequency */
1224                 data = 3;
1225                 break;
1226         case MSR_IA32_APICBASE:
1227                 data = vcpu->apic_base;
1228                 break;
1229         case MSR_IA32_MISC_ENABLE:
1230                 data = vcpu->ia32_misc_enable_msr;
1231                 break;
1232 #ifdef CONFIG_X86_64
1233         case MSR_EFER:
1234                 data = vcpu->shadow_efer;
1235                 break;
1236 #endif
1237         default:
1238                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1239                 return 1;
1240         }
1241         *pdata = data;
1242         return 0;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1245
1246 /*
1247  * Reads an msr value (of 'msr_index') into 'pdata'.
1248  * Returns 0 on success, non-0 otherwise.
1249  * Assumes vcpu_load() was already called.
1250  */
1251 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1252 {
1253         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1254 }
1255
1256 #ifdef CONFIG_X86_64
1257
1258 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1259 {
1260         if (efer & EFER_RESERVED_BITS) {
1261                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1262                        efer);
1263                 inject_gp(vcpu);
1264                 return;
1265         }
1266
1267         if (is_paging(vcpu)
1268             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1269                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1270                 inject_gp(vcpu);
1271                 return;
1272         }
1273
1274         kvm_arch_ops->set_efer(vcpu, efer);
1275
1276         efer &= ~EFER_LMA;
1277         efer |= vcpu->shadow_efer & EFER_LMA;
1278
1279         vcpu->shadow_efer = efer;
1280 }
1281
1282 #endif
1283
1284 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1285 {
1286         switch (msr) {
1287 #ifdef CONFIG_X86_64
1288         case MSR_EFER:
1289                 set_efer(vcpu, data);
1290                 break;
1291 #endif
1292         case MSR_IA32_MC0_STATUS:
1293                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1294                        __FUNCTION__, data);
1295                 break;
1296         case MSR_IA32_UCODE_REV:
1297         case MSR_IA32_UCODE_WRITE:
1298         case 0x200 ... 0x2ff: /* MTRRs */
1299                 break;
1300         case MSR_IA32_APICBASE:
1301                 vcpu->apic_base = data;
1302                 break;
1303         case MSR_IA32_MISC_ENABLE:
1304                 vcpu->ia32_misc_enable_msr = data;
1305                 break;
1306         default:
1307                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1308                 return 1;
1309         }
1310         return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1313
1314 /*
1315  * Writes msr value into into the appropriate "register".
1316  * Returns 0 on success, non-0 otherwise.
1317  * Assumes vcpu_load() was already called.
1318  */
1319 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1320 {
1321         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1322 }
1323
1324 void kvm_resched(struct kvm_vcpu *vcpu)
1325 {
1326         vcpu_put(vcpu);
1327         cond_resched();
1328         /* Cannot fail -  no vcpu unplug yet. */
1329         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_resched);
1332
1333 void load_msrs(struct vmx_msr_entry *e, int n)
1334 {
1335         int i;
1336
1337         for (i = 0; i < n; ++i)
1338                 wrmsrl(e[i].index, e[i].data);
1339 }
1340 EXPORT_SYMBOL_GPL(load_msrs);
1341
1342 void save_msrs(struct vmx_msr_entry *e, int n)
1343 {
1344         int i;
1345
1346         for (i = 0; i < n; ++i)
1347                 rdmsrl(e[i].index, e[i].data);
1348 }
1349 EXPORT_SYMBOL_GPL(save_msrs);
1350
1351 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1352 {
1353         struct kvm_vcpu *vcpu;
1354         int r;
1355
1356         if (!valid_vcpu(kvm_run->vcpu))
1357                 return -EINVAL;
1358
1359         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1360         if (!vcpu)
1361                 return -ENOENT;
1362
1363         if (kvm_run->emulated) {
1364                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1365                 kvm_run->emulated = 0;
1366         }
1367
1368         if (kvm_run->mmio_completed) {
1369                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1370                 vcpu->mmio_read_completed = 1;
1371         }
1372
1373         vcpu->mmio_needed = 0;
1374
1375         r = kvm_arch_ops->run(vcpu, kvm_run);
1376
1377         vcpu_put(vcpu);
1378         return r;
1379 }
1380
1381 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1382 {
1383         struct kvm_vcpu *vcpu;
1384
1385         if (!valid_vcpu(regs->vcpu))
1386                 return -EINVAL;
1387
1388         vcpu = vcpu_load(kvm, regs->vcpu);
1389         if (!vcpu)
1390                 return -ENOENT;
1391
1392         kvm_arch_ops->cache_regs(vcpu);
1393
1394         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1395         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1396         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1397         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1398         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1399         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1400         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1401         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1402 #ifdef CONFIG_X86_64
1403         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1404         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1405         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1406         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1407         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1408         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1409         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1410         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1411 #endif
1412
1413         regs->rip = vcpu->rip;
1414         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1415
1416         /*
1417          * Don't leak debug flags in case they were set for guest debugging
1418          */
1419         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1420                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1421
1422         vcpu_put(vcpu);
1423
1424         return 0;
1425 }
1426
1427 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1428 {
1429         struct kvm_vcpu *vcpu;
1430
1431         if (!valid_vcpu(regs->vcpu))
1432                 return -EINVAL;
1433
1434         vcpu = vcpu_load(kvm, regs->vcpu);
1435         if (!vcpu)
1436                 return -ENOENT;
1437
1438         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1439         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1440         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1441         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1442         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1443         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1444         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1445         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1446 #ifdef CONFIG_X86_64
1447         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1448         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1449         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1450         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1451         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1452         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1453         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1454         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1455 #endif
1456
1457         vcpu->rip = regs->rip;
1458         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1459
1460         kvm_arch_ops->decache_regs(vcpu);
1461
1462         vcpu_put(vcpu);
1463
1464         return 0;
1465 }
1466
1467 static void get_segment(struct kvm_vcpu *vcpu,
1468                         struct kvm_segment *var, int seg)
1469 {
1470         return kvm_arch_ops->get_segment(vcpu, var, seg);
1471 }
1472
1473 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1474 {
1475         struct kvm_vcpu *vcpu;
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         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1485         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1486         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1487         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1488         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1489         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1490
1491         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1492         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1493
1494         kvm_arch_ops->get_idt(vcpu, &dt);
1495         sregs->idt.limit = dt.limit;
1496         sregs->idt.base = dt.base;
1497         kvm_arch_ops->get_gdt(vcpu, &dt);
1498         sregs->gdt.limit = dt.limit;
1499         sregs->gdt.base = dt.base;
1500
1501         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1502         sregs->cr0 = vcpu->cr0;
1503         sregs->cr2 = vcpu->cr2;
1504         sregs->cr3 = vcpu->cr3;
1505         sregs->cr4 = vcpu->cr4;
1506         sregs->cr8 = vcpu->cr8;
1507         sregs->efer = vcpu->shadow_efer;
1508         sregs->apic_base = vcpu->apic_base;
1509
1510         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1511                sizeof sregs->interrupt_bitmap);
1512
1513         vcpu_put(vcpu);
1514
1515         return 0;
1516 }
1517
1518 static void set_segment(struct kvm_vcpu *vcpu,
1519                         struct kvm_segment *var, int seg)
1520 {
1521         return kvm_arch_ops->set_segment(vcpu, var, seg);
1522 }
1523
1524 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1525 {
1526         struct kvm_vcpu *vcpu;
1527         int mmu_reset_needed = 0;
1528         int i;
1529         struct descriptor_table dt;
1530
1531         if (!valid_vcpu(sregs->vcpu))
1532                 return -EINVAL;
1533         vcpu = vcpu_load(kvm, sregs->vcpu);
1534         if (!vcpu)
1535                 return -ENOENT;
1536
1537         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1538         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1539         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1540         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1541         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1542         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1543
1544         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1545         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1546
1547         dt.limit = sregs->idt.limit;
1548         dt.base = sregs->idt.base;
1549         kvm_arch_ops->set_idt(vcpu, &dt);
1550         dt.limit = sregs->gdt.limit;
1551         dt.base = sregs->gdt.base;
1552         kvm_arch_ops->set_gdt(vcpu, &dt);
1553
1554         vcpu->cr2 = sregs->cr2;
1555         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1556         vcpu->cr3 = sregs->cr3;
1557
1558         vcpu->cr8 = sregs->cr8;
1559
1560         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1561 #ifdef CONFIG_X86_64
1562         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1563 #endif
1564         vcpu->apic_base = sregs->apic_base;
1565
1566         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1567
1568         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1569         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1570
1571         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1572         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1573         if (!is_long_mode(vcpu) && is_pae(vcpu))
1574                 load_pdptrs(vcpu, vcpu->cr3);
1575
1576         if (mmu_reset_needed)
1577                 kvm_mmu_reset_context(vcpu);
1578
1579         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1580                sizeof vcpu->irq_pending);
1581         vcpu->irq_summary = 0;
1582         for (i = 0; i < NR_IRQ_WORDS; ++i)
1583                 if (vcpu->irq_pending[i])
1584                         __set_bit(i, &vcpu->irq_summary);
1585
1586         vcpu_put(vcpu);
1587
1588         return 0;
1589 }
1590
1591 /*
1592  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1593  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1594  *
1595  * This list is modified at module load time to reflect the
1596  * capabilities of the host cpu.
1597  */
1598 static u32 msrs_to_save[] = {
1599         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1600         MSR_K6_STAR,
1601 #ifdef CONFIG_X86_64
1602         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1603 #endif
1604         MSR_IA32_TIME_STAMP_COUNTER,
1605 };
1606
1607 static unsigned num_msrs_to_save;
1608
1609 static u32 emulated_msrs[] = {
1610         MSR_IA32_MISC_ENABLE,
1611 };
1612
1613 static __init void kvm_init_msr_list(void)
1614 {
1615         u32 dummy[2];
1616         unsigned i, j;
1617
1618         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1619                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1620                         continue;
1621                 if (j < i)
1622                         msrs_to_save[j] = msrs_to_save[i];
1623                 j++;
1624         }
1625         num_msrs_to_save = j;
1626 }
1627
1628 /*
1629  * Adapt set_msr() to msr_io()'s calling convention
1630  */
1631 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1632 {
1633         return set_msr(vcpu, index, *data);
1634 }
1635
1636 /*
1637  * Read or write a bunch of msrs. All parameters are kernel addresses.
1638  *
1639  * @return number of msrs set successfully.
1640  */
1641 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1642                     struct kvm_msr_entry *entries,
1643                     int (*do_msr)(struct kvm_vcpu *vcpu,
1644                                   unsigned index, u64 *data))
1645 {
1646         struct kvm_vcpu *vcpu;
1647         int i;
1648
1649         if (!valid_vcpu(msrs->vcpu))
1650                 return -EINVAL;
1651
1652         vcpu = vcpu_load(kvm, msrs->vcpu);
1653         if (!vcpu)
1654                 return -ENOENT;
1655
1656         for (i = 0; i < msrs->nmsrs; ++i)
1657                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1658                         break;
1659
1660         vcpu_put(vcpu);
1661
1662         return i;
1663 }
1664
1665 /*
1666  * Read or write a bunch of msrs. Parameters are user addresses.
1667  *
1668  * @return number of msrs set successfully.
1669  */
1670 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1671                   int (*do_msr)(struct kvm_vcpu *vcpu,
1672                                 unsigned index, u64 *data),
1673                   int writeback)
1674 {
1675         struct kvm_msrs msrs;
1676         struct kvm_msr_entry *entries;
1677         int r, n;
1678         unsigned size;
1679
1680         r = -EFAULT;
1681         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1682                 goto out;
1683
1684         r = -E2BIG;
1685         if (msrs.nmsrs >= MAX_IO_MSRS)
1686                 goto out;
1687
1688         r = -ENOMEM;
1689         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1690         entries = vmalloc(size);
1691         if (!entries)
1692                 goto out;
1693
1694         r = -EFAULT;
1695         if (copy_from_user(entries, user_msrs->entries, size))
1696                 goto out_free;
1697
1698         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1699         if (r < 0)
1700                 goto out_free;
1701
1702         r = -EFAULT;
1703         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1704                 goto out_free;
1705
1706         r = n;
1707
1708 out_free:
1709         vfree(entries);
1710 out:
1711         return r;
1712 }
1713
1714 /*
1715  * Translate a guest virtual address to a guest physical address.
1716  */
1717 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1718 {
1719         unsigned long vaddr = tr->linear_address;
1720         struct kvm_vcpu *vcpu;
1721         gpa_t gpa;
1722
1723         vcpu = vcpu_load(kvm, tr->vcpu);
1724         if (!vcpu)
1725                 return -ENOENT;
1726         spin_lock(&kvm->lock);
1727         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1728         tr->physical_address = gpa;
1729         tr->valid = gpa != UNMAPPED_GVA;
1730         tr->writeable = 1;
1731         tr->usermode = 0;
1732         spin_unlock(&kvm->lock);
1733         vcpu_put(vcpu);
1734
1735         return 0;
1736 }
1737
1738 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1739 {
1740         struct kvm_vcpu *vcpu;
1741
1742         if (!valid_vcpu(irq->vcpu))
1743                 return -EINVAL;
1744         if (irq->irq < 0 || irq->irq >= 256)
1745                 return -EINVAL;
1746         vcpu = vcpu_load(kvm, irq->vcpu);
1747         if (!vcpu)
1748                 return -ENOENT;
1749
1750         set_bit(irq->irq, vcpu->irq_pending);
1751         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1752
1753         vcpu_put(vcpu);
1754
1755         return 0;
1756 }
1757
1758 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1759                                      struct kvm_debug_guest *dbg)
1760 {
1761         struct kvm_vcpu *vcpu;
1762         int r;
1763
1764         if (!valid_vcpu(dbg->vcpu))
1765                 return -EINVAL;
1766         vcpu = vcpu_load(kvm, dbg->vcpu);
1767         if (!vcpu)
1768                 return -ENOENT;
1769
1770         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1771
1772         vcpu_put(vcpu);
1773
1774         return r;
1775 }
1776
1777 static long kvm_dev_ioctl(struct file *filp,
1778                           unsigned int ioctl, unsigned long arg)
1779 {
1780         struct kvm *kvm = filp->private_data;
1781         int r = -EINVAL;
1782
1783         switch (ioctl) {
1784         case KVM_GET_API_VERSION:
1785                 r = KVM_API_VERSION;
1786                 break;
1787         case KVM_CREATE_VCPU: {
1788                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1789                 if (r)
1790                         goto out;
1791                 break;
1792         }
1793         case KVM_RUN: {
1794                 struct kvm_run kvm_run;
1795
1796                 r = -EFAULT;
1797                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1798                         goto out;
1799                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1800                 if (r < 0 &&  r != -EINTR)
1801                         goto out;
1802                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1803                         r = -EFAULT;
1804                         goto out;
1805                 }
1806                 break;
1807         }
1808         case KVM_GET_REGS: {
1809                 struct kvm_regs kvm_regs;
1810
1811                 r = -EFAULT;
1812                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1813                         goto out;
1814                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1815                 if (r)
1816                         goto out;
1817                 r = -EFAULT;
1818                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_SET_REGS: {
1824                 struct kvm_regs kvm_regs;
1825
1826                 r = -EFAULT;
1827                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1828                         goto out;
1829                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_GET_SREGS: {
1836                 struct kvm_sregs kvm_sregs;
1837
1838                 r = -EFAULT;
1839                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1840                         goto out;
1841                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1842                 if (r)
1843                         goto out;
1844                 r = -EFAULT;
1845                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_SET_SREGS: {
1851                 struct kvm_sregs kvm_sregs;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1855                         goto out;
1856                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1857                 if (r)
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_TRANSLATE: {
1863                 struct kvm_translation tr;
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1867                         goto out;
1868                 r = kvm_dev_ioctl_translate(kvm, &tr);
1869                 if (r)
1870                         goto out;
1871                 r = -EFAULT;
1872                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_INTERRUPT: {
1878                 struct kvm_interrupt irq;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1882                         goto out;
1883                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1884                 if (r)
1885                         goto out;
1886                 r = 0;
1887                 break;
1888         }
1889         case KVM_DEBUG_GUEST: {
1890                 struct kvm_debug_guest dbg;
1891
1892                 r = -EFAULT;
1893                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1894                         goto out;
1895                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1896                 if (r)
1897                         goto out;
1898                 r = 0;
1899                 break;
1900         }
1901         case KVM_SET_MEMORY_REGION: {
1902                 struct kvm_memory_region kvm_mem;
1903
1904                 r = -EFAULT;
1905                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1906                         goto out;
1907                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1908                 if (r)
1909                         goto out;
1910                 break;
1911         }
1912         case KVM_GET_DIRTY_LOG: {
1913                 struct kvm_dirty_log log;
1914
1915                 r = -EFAULT;
1916                 if (copy_from_user(&log, (void *)arg, sizeof log))
1917                         goto out;
1918                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1919                 if (r)
1920                         goto out;
1921                 break;
1922         }
1923         case KVM_GET_MSRS:
1924                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1925                 break;
1926         case KVM_SET_MSRS:
1927                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1928                 break;
1929         case KVM_GET_MSR_INDEX_LIST: {
1930                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1931                 struct kvm_msr_list msr_list;
1932                 unsigned n;
1933
1934                 r = -EFAULT;
1935                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1936                         goto out;
1937                 n = msr_list.nmsrs;
1938                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1939                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1940                         goto out;
1941                 r = -E2BIG;
1942                 if (n < num_msrs_to_save)
1943                         goto out;
1944                 r = -EFAULT;
1945                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1946                                  num_msrs_to_save * sizeof(u32)))
1947                         goto out;
1948                 if (copy_to_user(user_msr_list->indices
1949                                  + num_msrs_to_save * sizeof(u32),
1950                                  &emulated_msrs,
1951                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1952                         goto out;
1953                 r = 0;
1954                 break;
1955         }
1956         default:
1957                 ;
1958         }
1959 out:
1960         return r;
1961 }
1962
1963 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1964                                    unsigned long address,
1965                                    int *type)
1966 {
1967         struct kvm *kvm = vma->vm_file->private_data;
1968         unsigned long pgoff;
1969         struct kvm_memory_slot *slot;
1970         struct page *page;
1971
1972         *type = VM_FAULT_MINOR;
1973         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1974         slot = gfn_to_memslot(kvm, pgoff);
1975         if (!slot)
1976                 return NOPAGE_SIGBUS;
1977         page = gfn_to_page(slot, pgoff);
1978         if (!page)
1979                 return NOPAGE_SIGBUS;
1980         get_page(page);
1981         return page;
1982 }
1983
1984 static struct vm_operations_struct kvm_dev_vm_ops = {
1985         .nopage = kvm_dev_nopage,
1986 };
1987
1988 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1989 {
1990         vma->vm_ops = &kvm_dev_vm_ops;
1991         return 0;
1992 }
1993
1994 static struct file_operations kvm_chardev_ops = {
1995         .open           = kvm_dev_open,
1996         .release        = kvm_dev_release,
1997         .unlocked_ioctl = kvm_dev_ioctl,
1998         .compat_ioctl   = kvm_dev_ioctl,
1999         .mmap           = kvm_dev_mmap,
2000 };
2001
2002 static struct miscdevice kvm_dev = {
2003         MISC_DYNAMIC_MINOR,
2004         "kvm",
2005         &kvm_chardev_ops,
2006 };
2007
2008 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2009                        void *v)
2010 {
2011         if (val == SYS_RESTART) {
2012                 /*
2013                  * Some (well, at least mine) BIOSes hang on reboot if
2014                  * in vmx root mode.
2015                  */
2016                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2017                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2018         }
2019         return NOTIFY_OK;
2020 }
2021
2022 static struct notifier_block kvm_reboot_notifier = {
2023         .notifier_call = kvm_reboot,
2024         .priority = 0,
2025 };
2026
2027 static __init void kvm_init_debug(void)
2028 {
2029         struct kvm_stats_debugfs_item *p;
2030
2031         debugfs_dir = debugfs_create_dir("kvm", 0);
2032         for (p = debugfs_entries; p->name; ++p)
2033                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2034                                                p->data);
2035 }
2036
2037 static void kvm_exit_debug(void)
2038 {
2039         struct kvm_stats_debugfs_item *p;
2040
2041         for (p = debugfs_entries; p->name; ++p)
2042                 debugfs_remove(p->dentry);
2043         debugfs_remove(debugfs_dir);
2044 }
2045
2046 hpa_t bad_page_address;
2047
2048 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2049 {
2050         int r;
2051
2052         if (kvm_arch_ops) {
2053                 printk(KERN_ERR "kvm: already loaded the other module\n");
2054                 return -EEXIST;
2055         }
2056
2057         if (!ops->cpu_has_kvm_support()) {
2058                 printk(KERN_ERR "kvm: no hardware support\n");
2059                 return -EOPNOTSUPP;
2060         }
2061         if (ops->disabled_by_bios()) {
2062                 printk(KERN_ERR "kvm: disabled by bios\n");
2063                 return -EOPNOTSUPP;
2064         }
2065
2066         kvm_arch_ops = ops;
2067
2068         r = kvm_arch_ops->hardware_setup();
2069         if (r < 0)
2070             return r;
2071
2072         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2073         register_reboot_notifier(&kvm_reboot_notifier);
2074
2075         kvm_chardev_ops.owner = module;
2076
2077         r = misc_register(&kvm_dev);
2078         if (r) {
2079                 printk (KERN_ERR "kvm: misc device register failed\n");
2080                 goto out_free;
2081         }
2082
2083         return r;
2084
2085 out_free:
2086         unregister_reboot_notifier(&kvm_reboot_notifier);
2087         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2088         kvm_arch_ops->hardware_unsetup();
2089         return r;
2090 }
2091
2092 void kvm_exit_arch(void)
2093 {
2094         misc_deregister(&kvm_dev);
2095
2096         unregister_reboot_notifier(&kvm_reboot_notifier);
2097         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2098         kvm_arch_ops->hardware_unsetup();
2099         kvm_arch_ops = NULL;
2100 }
2101
2102 static __init int kvm_init(void)
2103 {
2104         static struct page *bad_page;
2105         int r = 0;
2106
2107         kvm_init_debug();
2108
2109         kvm_init_msr_list();
2110
2111         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2112                 r = -ENOMEM;
2113                 goto out;
2114         }
2115
2116         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2117         memset(__va(bad_page_address), 0, PAGE_SIZE);
2118
2119         return r;
2120
2121 out:
2122         kvm_exit_debug();
2123         return r;
2124 }
2125
2126 static __exit void kvm_exit(void)
2127 {
2128         kvm_exit_debug();
2129         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2130 }
2131
2132 module_init(kvm_init)
2133 module_exit(kvm_exit)
2134
2135 EXPORT_SYMBOL_GPL(kvm_init_arch);
2136 EXPORT_SYMBOL_GPL(kvm_exit_arch);