drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <asm/processor.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <asm/msr.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/reboot.h>
  32 #include <asm/io.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <asm/desc.h>
  37
  38 #include "x86_emulate.h"
  39 #include "segment_descriptor.h"
  40
  41 MODULE_AUTHOR("Qumranet");
  42 MODULE_LICENSE("GPL");
  43
  44 struct kvm_arch_ops *kvm_arch_ops;
  45 struct kvm_stat kvm_stat;
  46 EXPORT_SYMBOL_GPL(kvm_stat);
  47
  48 static struct kvm_stats_debugfs_item {
  49         const char *name;
  50         u32 *data;
  51         struct dentry *dentry;
  52 } debugfs_entries[] = {
  53         { "pf_fixed", &kvm_stat.pf_fixed },
  54         { "pf_guest", &kvm_stat.pf_guest },
  55         { "tlb_flush", &kvm_stat.tlb_flush },
  56         { "invlpg", &kvm_stat.invlpg },
  57         { "exits", &kvm_stat.exits },
  58         { "io_exits", &kvm_stat.io_exits },
  59         { "mmio_exits", &kvm_stat.mmio_exits },
  60         { "signal_exits", &kvm_stat.signal_exits },
  61         { "irq_window", &kvm_stat.irq_window_exits },
  62         { "halt_exits", &kvm_stat.halt_exits },
  63         { "request_irq", &kvm_stat.request_irq_exits },
  64         { "irq_exits", &kvm_stat.irq_exits },
  65         { 0, 0 }
  66 };
  67
  68 static struct dentry *debugfs_dir;
  69
  70 #define MAX_IO_MSRS 256
  71
  72 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  73 #define LMSW_GUEST_MASK 0x0eULL
  74 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  75 #define CR8_RESEVED_BITS (~0x0fULL)
  76 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  77
  78 #ifdef CONFIG_X86_64
  79 // LDT or TSS descriptor in the GDT. 16 bytes.
  80 struct segment_descriptor_64 {
  81         struct segment_descriptor s;
  82         u32 base_higher;
  83         u32 pad_zero;
  84 };
  85
  86 #endif
  87
  88 unsigned long segment_base(u16 selector)
  89 {
  90         struct descriptor_table gdt;
  91         struct segment_descriptor *d;
  92         unsigned long table_base;
  93         typedef unsigned long ul;
  94         unsigned long v;
  95
  96         if (selector == 0)
  97                 return 0;
  98
  99         asm ("sgdt %0" : "=m"(gdt));
 100         table_base = gdt.base;
 101
 102         if (selector & 4) {           /* from ldt */
 103                 u16 ldt_selector;
 104
 105                 asm ("sldt %0" : "=g"(ldt_selector));
 106                 table_base = segment_base(ldt_selector);
 107         }
 108         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 109         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 110 #ifdef CONFIG_X86_64
 111         if (d->system == 0
 112             && (d->type == 2 || d->type == 9 || d->type == 11))
 113                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 114 #endif
 115         return v;
 116 }
 117 EXPORT_SYMBOL_GPL(segment_base);
 118
 119 static inline int valid_vcpu(int n)
 120 {
 121         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 122 }
 123
 124 int kvm_read_guest(struct kvm_vcpu *vcpu,
 125                              gva_t addr,
 126                              unsigned long size,
 127                              void *dest)
 128 {
 129         unsigned char *host_buf = dest;
 130         unsigned long req_size = size;
 131
 132         while (size) {
 133                 hpa_t paddr;
 134                 unsigned now;
 135                 unsigned offset;
 136                 hva_t guest_buf;
 137
 138                 paddr = gva_to_hpa(vcpu, addr);
 139
 140                 if (is_error_hpa(paddr))
 141                         break;
 142
 143                 guest_buf = (hva_t)kmap_atomic(
 144                                         pfn_to_page(paddr >> PAGE_SHIFT),
 145                                         KM_USER0);
 146                 offset = addr & ~PAGE_MASK;
 147                 guest_buf |= offset;
 148                 now = min(size, PAGE_SIZE - offset);
 149                 memcpy(host_buf, (void*)guest_buf, now);
 150                 host_buf += now;
 151                 addr += now;
 152                 size -= now;
 153                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 154         }
 155         return req_size - size;
 156 }
 157 EXPORT_SYMBOL_GPL(kvm_read_guest);
 158
 159 int kvm_write_guest(struct kvm_vcpu *vcpu,
 160                              gva_t addr,
 161                              unsigned long size,
 162                              void *data)
 163 {
 164         unsigned char *host_buf = data;
 165         unsigned long req_size = size;
 166
 167         while (size) {
 168                 hpa_t paddr;
 169                 unsigned now;
 170                 unsigned offset;
 171                 hva_t guest_buf;
 172
 173                 paddr = gva_to_hpa(vcpu, addr);
 174
 175                 if (is_error_hpa(paddr))
 176                         break;
 177
 178                 guest_buf = (hva_t)kmap_atomic(
 179                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 180                 offset = addr & ~PAGE_MASK;
 181                 guest_buf |= offset;
 182                 now = min(size, PAGE_SIZE - offset);
 183                 memcpy((void*)guest_buf, host_buf, now);
 184                 host_buf += now;
 185                 addr += now;
 186                 size -= now;
 187                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 188         }
 189         return req_size - size;
 190 }
 191 EXPORT_SYMBOL_GPL(kvm_write_guest);
 192
 193 static int vcpu_slot(struct kvm_vcpu *vcpu)
 194 {
 195         return vcpu - vcpu->kvm->vcpus;
 196 }
 197
 198 /*
 199  * Switches to specified vcpu, until a matching vcpu_put()
 200  */
 201 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
 202 {
 203         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
 204
 205         mutex_lock(&vcpu->mutex);
 206         if (unlikely(!vcpu->vmcs)) {
 207                 mutex_unlock(&vcpu->mutex);
 208                 return 0;
 209         }
 210         return kvm_arch_ops->vcpu_load(vcpu);
 211 }
 212
 213 static void vcpu_put(struct kvm_vcpu *vcpu)
 214 {
 215         kvm_arch_ops->vcpu_put(vcpu);
 216         mutex_unlock(&vcpu->mutex);
 217 }
 218
 219 static int kvm_dev_open(struct inode *inode, struct file *filp)
 220 {
 221         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 222         int i;
 223
 224         if (!kvm)
 225                 return -ENOMEM;
 226
 227         spin_lock_init(&kvm->lock);
 228         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 229         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 230                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 231
 232                 mutex_init(&vcpu->mutex);
 233                 vcpu->mmu.root_hpa = INVALID_PAGE;
 234                 INIT_LIST_HEAD(&vcpu->free_pages);
 235         }
 236         filp->private_data = kvm;
 237         return 0;
 238 }
 239
 240 /*
 241  * Free any memory in @free but not in @dont.
 242  */
 243 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 244                                   struct kvm_memory_slot *dont)
 245 {
 246         int i;
 247
 248         if (!dont || free->phys_mem != dont->phys_mem)
 249                 if (free->phys_mem) {
 250                         for (i = 0; i < free->npages; ++i)
 251                                 if (free->phys_mem[i])
 252                                         __free_page(free->phys_mem[i]);
 253                         vfree(free->phys_mem);
 254                 }
 255
 256         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 257                 vfree(free->dirty_bitmap);
 258
 259         free->phys_mem = 0;
 260         free->npages = 0;
 261         free->dirty_bitmap = 0;
 262 }
 263
 264 static void kvm_free_physmem(struct kvm *kvm)
 265 {
 266         int i;
 267
 268         for (i = 0; i < kvm->nmemslots; ++i)
 269                 kvm_free_physmem_slot(&kvm->memslots[i], 0);
 270 }
 271
 272 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 273 {
 274         kvm_arch_ops->vcpu_free(vcpu);
 275         kvm_mmu_destroy(vcpu);
 276 }
 277
 278 static void kvm_free_vcpus(struct kvm *kvm)
 279 {
 280         unsigned int i;
 281
 282         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 283                 kvm_free_vcpu(&kvm->vcpus[i]);
 284 }
 285
 286 static int kvm_dev_release(struct inode *inode, struct file *filp)
 287 {
 288         struct kvm *kvm = filp->private_data;
 289
 290         kvm_free_vcpus(kvm);
 291         kvm_free_physmem(kvm);
 292         kfree(kvm);
 293         return 0;
 294 }
 295
 296 static void inject_gp(struct kvm_vcpu *vcpu)
 297 {
 298         kvm_arch_ops->inject_gp(vcpu, 0);
 299 }
 300
 301 static int pdptrs_have_reserved_bits_set(struct kvm_vcpu *vcpu,
 302                                          unsigned long cr3)
 303 {
 304         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 305         unsigned offset = (cr3 & (PAGE_SIZE-1)) >> 5;
 306         int i;
 307         u64 pdpte;
 308         u64 *pdpt;
 309         struct kvm_memory_slot *memslot;
 310
 311         spin_lock(&vcpu->kvm->lock);
 312         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 313         /* FIXME: !memslot - emulate? 0xff? */
 314         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 315
 316         for (i = 0; i < 4; ++i) {
 317                 pdpte = pdpt[offset + i];
 318                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull))
 319                         break;
 320         }
 321
 322         kunmap_atomic(pdpt, KM_USER0);
 323         spin_unlock(&vcpu->kvm->lock);
 324
 325         return i != 4;
 326 }
 327
 328 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 329 {
 330         if (cr0 & CR0_RESEVED_BITS) {
 331                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 332                        cr0, vcpu->cr0);
 333                 inject_gp(vcpu);
 334                 return;
 335         }
 336
 337         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 338                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 339                 inject_gp(vcpu);
 340                 return;
 341         }
 342
 343         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 344                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 345                        "and a clear PE flag\n");
 346                 inject_gp(vcpu);
 347                 return;
 348         }
 349
 350         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 351 #ifdef CONFIG_X86_64
 352                 if ((vcpu->shadow_efer & EFER_LME)) {
 353                         int cs_db, cs_l;
 354
 355                         if (!is_pae(vcpu)) {
 356                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 357                                        "in long mode while PAE is disabled\n");
 358                                 inject_gp(vcpu);
 359                                 return;
 360                         }
 361                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 362                         if (cs_l) {
 363                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 364                                        "in long mode while CS.L == 1\n");
 365                                 inject_gp(vcpu);
 366                                 return;
 367
 368                         }
 369                 } else
 370 #endif
 371                 if (is_pae(vcpu) &&
 372                             pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
 373                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 374                                "reserved bits\n");
 375                         inject_gp(vcpu);
 376                         return;
 377                 }
 378
 379         }
 380
 381         kvm_arch_ops->set_cr0(vcpu, cr0);
 382         vcpu->cr0 = cr0;
 383
 384         spin_lock(&vcpu->kvm->lock);
 385         kvm_mmu_reset_context(vcpu);
 386         spin_unlock(&vcpu->kvm->lock);
 387         return;
 388 }
 389 EXPORT_SYMBOL_GPL(set_cr0);
 390
 391 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 392 {
 393         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 394         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 395 }
 396 EXPORT_SYMBOL_GPL(lmsw);
 397
 398 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 399 {
 400         if (cr4 & CR4_RESEVED_BITS) {
 401                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 402                 inject_gp(vcpu);
 403                 return;
 404         }
 405
 406         if (is_long_mode(vcpu)) {
 407                 if (!(cr4 & CR4_PAE_MASK)) {
 408                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 409                                "in long mode\n");
 410                         inject_gp(vcpu);
 411                         return;
 412                 }
 413         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 414                    && pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
 415                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 416                 inject_gp(vcpu);
 417         }
 418
 419         if (cr4 & CR4_VMXE_MASK) {
 420                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 421                 inject_gp(vcpu);
 422                 return;
 423         }
 424         kvm_arch_ops->set_cr4(vcpu, cr4);
 425         spin_lock(&vcpu->kvm->lock);
 426         kvm_mmu_reset_context(vcpu);
 427         spin_unlock(&vcpu->kvm->lock);
 428 }
 429 EXPORT_SYMBOL_GPL(set_cr4);
 430
 431 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 432 {
 433         if (is_long_mode(vcpu)) {
 434                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
 435                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 436                         inject_gp(vcpu);
 437                         return;
 438                 }
 439         } else {
 440                 if (cr3 & CR3_RESEVED_BITS) {
 441                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 442                         inject_gp(vcpu);
 443                         return;
 444                 }
 445                 if (is_paging(vcpu) && is_pae(vcpu) &&
 446                     pdptrs_have_reserved_bits_set(vcpu, cr3)) {
 447                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 448                                "reserved bits\n");
 449                         inject_gp(vcpu);
 450                         return;
 451                 }
 452         }
 453
 454         vcpu->cr3 = cr3;
 455         spin_lock(&vcpu->kvm->lock);
 456         vcpu->mmu.new_cr3(vcpu);
 457         spin_unlock(&vcpu->kvm->lock);
 458 }
 459 EXPORT_SYMBOL_GPL(set_cr3);
 460
 461 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 462 {
 463         if ( cr8 & CR8_RESEVED_BITS) {
 464                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 465                 inject_gp(vcpu);
 466                 return;
 467         }
 468         vcpu->cr8 = cr8;
 469 }
 470 EXPORT_SYMBOL_GPL(set_cr8);
 471
 472 void fx_init(struct kvm_vcpu *vcpu)
 473 {
 474         struct __attribute__ ((__packed__)) fx_image_s {
 475                 u16 control; //fcw
 476                 u16 status; //fsw
 477                 u16 tag; // ftw
 478                 u16 opcode; //fop
 479                 u64 ip; // fpu ip
 480                 u64 operand;// fpu dp
 481                 u32 mxcsr;
 482                 u32 mxcsr_mask;
 483
 484         } *fx_image;
 485
 486         fx_save(vcpu->host_fx_image);
 487         fpu_init();
 488         fx_save(vcpu->guest_fx_image);
 489         fx_restore(vcpu->host_fx_image);
 490
 491         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 492         fx_image->mxcsr = 0x1f80;
 493         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 494                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 495 }
 496 EXPORT_SYMBOL_GPL(fx_init);
 497
 498 /*
 499  * Creates some virtual cpus.  Good luck creating more than one.
 500  */
 501 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 502 {
 503         int r;
 504         struct kvm_vcpu *vcpu;
 505
 506         r = -EINVAL;
 507         if (!valid_vcpu(n))
 508                 goto out;
 509
 510         vcpu = &kvm->vcpus[n];
 511
 512         mutex_lock(&vcpu->mutex);
 513
 514         if (vcpu->vmcs) {
 515                 mutex_unlock(&vcpu->mutex);
 516                 return -EEXIST;
 517         }
 518
 519         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 520                                            FX_IMAGE_ALIGN);
 521         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 522
 523         vcpu->cpu = -1;  /* First load will set up TR */
 524         vcpu->kvm = kvm;
 525         r = kvm_arch_ops->vcpu_create(vcpu);
 526         if (r < 0)
 527                 goto out_free_vcpus;
 528
 529         r = kvm_mmu_create(vcpu);
 530         if (r < 0)
 531                 goto out_free_vcpus;
 532
 533         kvm_arch_ops->vcpu_load(vcpu);
 534         r = kvm_mmu_setup(vcpu);
 535         if (r >= 0)
 536                 r = kvm_arch_ops->vcpu_setup(vcpu);
 537         vcpu_put(vcpu);
 538
 539         if (r < 0)
 540                 goto out_free_vcpus;
 541
 542         return 0;
 543
 544 out_free_vcpus:
 545         kvm_free_vcpu(vcpu);
 546         mutex_unlock(&vcpu->mutex);
 547 out:
 548         return r;
 549 }
 550
 551 /*
 552  * Allocate some memory and give it an address in the guest physical address
 553  * space.
 554  *
 555  * Discontiguous memory is allowed, mostly for framebuffers.
 556  */
 557 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 558                                            struct kvm_memory_region *mem)
 559 {
 560         int r;
 561         gfn_t base_gfn;
 562         unsigned long npages;
 563         unsigned long i;
 564         struct kvm_memory_slot *memslot;
 565         struct kvm_memory_slot old, new;
 566         int memory_config_version;
 567
 568         r = -EINVAL;
 569         /* General sanity checks */
 570         if (mem->memory_size & (PAGE_SIZE - 1))
 571                 goto out;
 572         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 573                 goto out;
 574         if (mem->slot >= KVM_MEMORY_SLOTS)
 575                 goto out;
 576         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 577                 goto out;
 578
 579         memslot = &kvm->memslots[mem->slot];
 580         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 581         npages = mem->memory_size >> PAGE_SHIFT;
 582
 583         if (!npages)
 584                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 585
 586 raced:
 587         spin_lock(&kvm->lock);
 588
 589         memory_config_version = kvm->memory_config_version;
 590         new = old = *memslot;
 591
 592         new.base_gfn = base_gfn;
 593         new.npages = npages;
 594         new.flags = mem->flags;
 595
 596         /* Disallow changing a memory slot's size. */
 597         r = -EINVAL;
 598         if (npages && old.npages && npages != old.npages)
 599                 goto out_unlock;
 600
 601         /* Check for overlaps */
 602         r = -EEXIST;
 603         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 604                 struct kvm_memory_slot *s = &kvm->memslots[i];
 605
 606                 if (s == memslot)
 607                         continue;
 608                 if (!((base_gfn + npages <= s->base_gfn) ||
 609                       (base_gfn >= s->base_gfn + s->npages)))
 610                         goto out_unlock;
 611         }
 612         /*
 613          * Do memory allocations outside lock.  memory_config_version will
 614          * detect any races.
 615          */
 616         spin_unlock(&kvm->lock);
 617
 618         /* Deallocate if slot is being removed */
 619         if (!npages)
 620                 new.phys_mem = 0;
 621
 622         /* Free page dirty bitmap if unneeded */
 623         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 624                 new.dirty_bitmap = 0;
 625
 626         r = -ENOMEM;
 627
 628         /* Allocate if a slot is being created */
 629         if (npages && !new.phys_mem) {
 630                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 631
 632                 if (!new.phys_mem)
 633                         goto out_free;
 634
 635                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 636                 for (i = 0; i < npages; ++i) {
 637                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 638                                                      | __GFP_ZERO);
 639                         if (!new.phys_mem[i])
 640                                 goto out_free;
 641                 }
 642         }
 643
 644         /* Allocate page dirty bitmap if needed */
 645         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 646                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 647
 648                 new.dirty_bitmap = vmalloc(dirty_bytes);
 649                 if (!new.dirty_bitmap)
 650                         goto out_free;
 651                 memset(new.dirty_bitmap, 0, dirty_bytes);
 652         }
 653
 654         spin_lock(&kvm->lock);
 655
 656         if (memory_config_version != kvm->memory_config_version) {
 657                 spin_unlock(&kvm->lock);
 658                 kvm_free_physmem_slot(&new, &old);
 659                 goto raced;
 660         }
 661
 662         r = -EAGAIN;
 663         if (kvm->busy)
 664                 goto out_unlock;
 665
 666         if (mem->slot >= kvm->nmemslots)
 667                 kvm->nmemslots = mem->slot + 1;
 668
 669         *memslot = new;
 670         ++kvm->memory_config_version;
 671
 672         spin_unlock(&kvm->lock);
 673
 674         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 675                 struct kvm_vcpu *vcpu;
 676
 677                 vcpu = vcpu_load(kvm, i);
 678                 if (!vcpu)
 679                         continue;
 680                 kvm_mmu_reset_context(vcpu);
 681                 vcpu_put(vcpu);
 682         }
 683
 684         kvm_free_physmem_slot(&old, &new);
 685         return 0;
 686
 687 out_unlock:
 688         spin_unlock(&kvm->lock);
 689 out_free:
 690         kvm_free_physmem_slot(&new, &old);
 691 out:
 692         return r;
 693 }
 694
 695 /*
 696  * Get (and clear) the dirty memory log for a memory slot.
 697  */
 698 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 699                                        struct kvm_dirty_log *log)
 700 {
 701         struct kvm_memory_slot *memslot;
 702         int r, i;
 703         int n;
 704         unsigned long any = 0;
 705
 706         spin_lock(&kvm->lock);
 707
 708         /*
 709          * Prevent changes to guest memory configuration even while the lock
 710          * is not taken.
 711          */
 712         ++kvm->busy;
 713         spin_unlock(&kvm->lock);
 714         r = -EINVAL;
 715         if (log->slot >= KVM_MEMORY_SLOTS)
 716                 goto out;
 717
 718         memslot = &kvm->memslots[log->slot];
 719         r = -ENOENT;
 720         if (!memslot->dirty_bitmap)
 721                 goto out;
 722
 723         n = ALIGN(memslot->npages, 8) / 8;
 724
 725         for (i = 0; !any && i < n; ++i)
 726                 any = memslot->dirty_bitmap[i];
 727
 728         r = -EFAULT;
 729         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 730                 goto out;
 731
 732
 733         if (any) {
 734                 spin_lock(&kvm->lock);
 735                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 736                 spin_unlock(&kvm->lock);
 737                 memset(memslot->dirty_bitmap, 0, n);
 738                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 739                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 740
 741                         if (!vcpu)
 742                                 continue;
 743                         kvm_arch_ops->tlb_flush(vcpu);
 744                         vcpu_put(vcpu);
 745                 }
 746         }
 747
 748         r = 0;
 749
 750 out:
 751         spin_lock(&kvm->lock);
 752         --kvm->busy;
 753         spin_unlock(&kvm->lock);
 754         return r;
 755 }
 756
 757 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 758 {
 759         int i;
 760
 761         for (i = 0; i < kvm->nmemslots; ++i) {
 762                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 763
 764                 if (gfn >= memslot->base_gfn
 765                     && gfn < memslot->base_gfn + memslot->npages)
 766                         return memslot;
 767         }
 768         return 0;
 769 }
 770 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 771
 772 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 773 {
 774         int i;
 775         struct kvm_memory_slot *memslot = 0;
 776         unsigned long rel_gfn;
 777
 778         for (i = 0; i < kvm->nmemslots; ++i) {
 779                 memslot = &kvm->memslots[i];
 780
 781                 if (gfn >= memslot->base_gfn
 782                     && gfn < memslot->base_gfn + memslot->npages) {
 783
 784                         if (!memslot || !memslot->dirty_bitmap)
 785                                 return;
 786
 787                         rel_gfn = gfn - memslot->base_gfn;
 788
 789                         /* avoid RMW */
 790                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 791                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 792                         return;
 793                 }
 794         }
 795 }
 796
 797 static int emulator_read_std(unsigned long addr,
 798                              unsigned long *val,
 799                              unsigned int bytes,
 800                              struct x86_emulate_ctxt *ctxt)
 801 {
 802         struct kvm_vcpu *vcpu = ctxt->vcpu;
 803         void *data = val;
 804
 805         while (bytes) {
 806                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 807                 unsigned offset = addr & (PAGE_SIZE-1);
 808                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 809                 unsigned long pfn;
 810                 struct kvm_memory_slot *memslot;
 811                 void *page;
 812
 813                 if (gpa == UNMAPPED_GVA)
 814                         return X86EMUL_PROPAGATE_FAULT;
 815                 pfn = gpa >> PAGE_SHIFT;
 816                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 817                 if (!memslot)
 818                         return X86EMUL_UNHANDLEABLE;
 819                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 820
 821                 memcpy(data, page + offset, tocopy);
 822
 823                 kunmap_atomic(page, KM_USER0);
 824
 825                 bytes -= tocopy;
 826                 data += tocopy;
 827                 addr += tocopy;
 828         }
 829
 830         return X86EMUL_CONTINUE;
 831 }
 832
 833 static int emulator_write_std(unsigned long addr,
 834                               unsigned long val,
 835                               unsigned int bytes,
 836                               struct x86_emulate_ctxt *ctxt)
 837 {
 838         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 839                addr, bytes);
 840         return X86EMUL_UNHANDLEABLE;
 841 }
 842
 843 static int emulator_read_emulated(unsigned long addr,
 844                                   unsigned long *val,
 845                                   unsigned int bytes,
 846                                   struct x86_emulate_ctxt *ctxt)
 847 {
 848         struct kvm_vcpu *vcpu = ctxt->vcpu;
 849
 850         if (vcpu->mmio_read_completed) {
 851                 memcpy(val, vcpu->mmio_data, bytes);
 852                 vcpu->mmio_read_completed = 0;
 853                 return X86EMUL_CONTINUE;
 854         } else if (emulator_read_std(addr, val, bytes, ctxt)
 855                    == X86EMUL_CONTINUE)
 856                 return X86EMUL_CONTINUE;
 857         else {
 858                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 859                 if (gpa == UNMAPPED_GVA)
 860                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
 861                 vcpu->mmio_needed = 1;
 862                 vcpu->mmio_phys_addr = gpa;
 863                 vcpu->mmio_size = bytes;
 864                 vcpu->mmio_is_write = 0;
 865
 866                 return X86EMUL_UNHANDLEABLE;
 867         }
 868 }
 869
 870 static int emulator_write_emulated(unsigned long addr,
 871                                    unsigned long val,
 872                                    unsigned int bytes,
 873                                    struct x86_emulate_ctxt *ctxt)
 874 {
 875         struct kvm_vcpu *vcpu = ctxt->vcpu;
 876         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 877
 878         if (gpa == UNMAPPED_GVA)
 879                 return X86EMUL_PROPAGATE_FAULT;
 880
 881         vcpu->mmio_needed = 1;
 882         vcpu->mmio_phys_addr = gpa;
 883         vcpu->mmio_size = bytes;
 884         vcpu->mmio_is_write = 1;
 885         memcpy(vcpu->mmio_data, &val, bytes);
 886
 887         return X86EMUL_CONTINUE;
 888 }
 889
 890 static int emulator_cmpxchg_emulated(unsigned long addr,
 891                                      unsigned long old,
 892                                      unsigned long new,
 893                                      unsigned int bytes,
 894                                      struct x86_emulate_ctxt *ctxt)
 895 {
 896         static int reported;
 897
 898         if (!reported) {
 899                 reported = 1;
 900                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 901         }
 902         return emulator_write_emulated(addr, new, bytes, ctxt);
 903 }
 904
 905 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 906 {
 907         return kvm_arch_ops->get_segment_base(vcpu, seg);
 908 }
 909
 910 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 911 {
 912         spin_lock(&vcpu->kvm->lock);
 913         vcpu->mmu.inval_page(vcpu, address);
 914         spin_unlock(&vcpu->kvm->lock);
 915         kvm_arch_ops->invlpg(vcpu, address);
 916         return X86EMUL_CONTINUE;
 917 }
 918
 919 int emulate_clts(struct kvm_vcpu *vcpu)
 920 {
 921         unsigned long cr0;
 922
 923         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 924         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
 925         kvm_arch_ops->set_cr0(vcpu, cr0);
 926         return X86EMUL_CONTINUE;
 927 }
 928
 929 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
 930 {
 931         struct kvm_vcpu *vcpu = ctxt->vcpu;
 932
 933         switch (dr) {
 934         case 0 ... 3:
 935                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
 936                 return X86EMUL_CONTINUE;
 937         default:
 938                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
 939                        __FUNCTION__, dr);
 940                 return X86EMUL_UNHANDLEABLE;
 941         }
 942 }
 943
 944 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
 945 {
 946         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
 947         int exception;
 948
 949         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
 950         if (exception) {
 951                 /* FIXME: better handling */
 952                 return X86EMUL_UNHANDLEABLE;
 953         }
 954         return X86EMUL_CONTINUE;
 955 }
 956
 957 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
 958 {
 959         static int reported;
 960         u8 opcodes[4];
 961         unsigned long rip = ctxt->vcpu->rip;
 962         unsigned long rip_linear;
 963
 964         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
 965
 966         if (reported)
 967                 return;
 968
 969         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
 970
 971         printk(KERN_ERR "emulation failed but !mmio_needed?"
 972                " rip %lx %02x %02x %02x %02x\n",
 973                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
 974         reported = 1;
 975 }
 976
 977 struct x86_emulate_ops emulate_ops = {
 978         .read_std            = emulator_read_std,
 979         .write_std           = emulator_write_std,
 980         .read_emulated       = emulator_read_emulated,
 981         .write_emulated      = emulator_write_emulated,
 982         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
 983 };
 984
 985 int emulate_instruction(struct kvm_vcpu *vcpu,
 986                         struct kvm_run *run,
 987                         unsigned long cr2,
 988                         u16 error_code)
 989 {
 990         struct x86_emulate_ctxt emulate_ctxt;
 991         int r;
 992         int cs_db, cs_l;
 993
 994         kvm_arch_ops->cache_regs(vcpu);
 995
 996         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 997
 998         emulate_ctxt.vcpu = vcpu;
 999         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1000         emulate_ctxt.cr2 = cr2;
1001         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1002                 ? X86EMUL_MODE_REAL : cs_l
1003                 ? X86EMUL_MODE_PROT64 : cs_db
1004                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1005
1006         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1007                 emulate_ctxt.cs_base = 0;
1008                 emulate_ctxt.ds_base = 0;
1009                 emulate_ctxt.es_base = 0;
1010                 emulate_ctxt.ss_base = 0;
1011         } else {
1012                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1013                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1014                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1015                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1016         }
1017
1018         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1019         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1020
1021         vcpu->mmio_is_write = 0;
1022         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1023
1024         if ((r || vcpu->mmio_is_write) && run) {
1025                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1026                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1027                 run->mmio.len = vcpu->mmio_size;
1028                 run->mmio.is_write = vcpu->mmio_is_write;
1029         }
1030
1031         if (r) {
1032                 if (!vcpu->mmio_needed) {
1033                         report_emulation_failure(&emulate_ctxt);
1034                         return EMULATE_FAIL;
1035                 }
1036                 return EMULATE_DO_MMIO;
1037         }
1038
1039         kvm_arch_ops->decache_regs(vcpu);
1040         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1041
1042         if (vcpu->mmio_is_write)
1043                 return EMULATE_DO_MMIO;
1044
1045         return EMULATE_DONE;
1046 }
1047 EXPORT_SYMBOL_GPL(emulate_instruction);
1048
1049 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1050 {
1051         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1052 }
1053
1054 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1055 {
1056         struct descriptor_table dt = { limit, base };
1057
1058         kvm_arch_ops->set_gdt(vcpu, &dt);
1059 }
1060
1061 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1062 {
1063         struct descriptor_table dt = { limit, base };
1064
1065         kvm_arch_ops->set_idt(vcpu, &dt);
1066 }
1067
1068 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1069                    unsigned long *rflags)
1070 {
1071         lmsw(vcpu, msw);
1072         *rflags = kvm_arch_ops->get_rflags(vcpu);
1073 }
1074
1075 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1076 {
1077         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1078         switch (cr) {
1079         case 0:
1080                 return vcpu->cr0;
1081         case 2:
1082                 return vcpu->cr2;
1083         case 3:
1084                 return vcpu->cr3;
1085         case 4:
1086                 return vcpu->cr4;
1087         default:
1088                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1089                 return 0;
1090         }
1091 }
1092
1093 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1094                      unsigned long *rflags)
1095 {
1096         switch (cr) {
1097         case 0:
1098                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1099                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1100                 break;
1101         case 2:
1102                 vcpu->cr2 = val;
1103                 break;
1104         case 3:
1105                 set_cr3(vcpu, val);
1106                 break;
1107         case 4:
1108                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1109                 break;
1110         default:
1111                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1112         }
1113 }
1114
1115 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1116 {
1117         u64 data;
1118
1119         switch (msr) {
1120         case 0xc0010010: /* SYSCFG */
1121         case 0xc0010015: /* HWCR */
1122         case MSR_IA32_PLATFORM_ID:
1123         case MSR_IA32_P5_MC_ADDR:
1124         case MSR_IA32_P5_MC_TYPE:
1125         case MSR_IA32_MC0_CTL:
1126         case MSR_IA32_MCG_STATUS:
1127         case MSR_IA32_MCG_CAP:
1128         case MSR_IA32_MC0_MISC:
1129         case MSR_IA32_MC0_MISC+4:
1130         case MSR_IA32_MC0_MISC+8:
1131         case MSR_IA32_MC0_MISC+12:
1132         case MSR_IA32_MC0_MISC+16:
1133         case MSR_IA32_UCODE_REV:
1134         case MSR_IA32_PERF_STATUS:
1135                 /* MTRR registers */
1136         case 0xfe:
1137         case 0x200 ... 0x2ff:
1138                 data = 0;
1139                 break;
1140         case 0xcd: /* fsb frequency */
1141                 data = 3;
1142                 break;
1143         case MSR_IA32_APICBASE:
1144                 data = vcpu->apic_base;
1145                 break;
1146 #ifdef CONFIG_X86_64
1147         case MSR_EFER:
1148                 data = vcpu->shadow_efer;
1149                 break;
1150 #endif
1151         default:
1152                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1153                 return 1;
1154         }
1155         *pdata = data;
1156         return 0;
1157 }
1158 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1159
1160 /*
1161  * Reads an msr value (of 'msr_index') into 'pdata'.
1162  * Returns 0 on success, non-0 otherwise.
1163  * Assumes vcpu_load() was already called.
1164  */
1165 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1166 {
1167         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1168 }
1169
1170 #ifdef CONFIG_X86_64
1171
1172 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1173 {
1174         if (efer & EFER_RESERVED_BITS) {
1175                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1176                        efer);
1177                 inject_gp(vcpu);
1178                 return;
1179         }
1180
1181         if (is_paging(vcpu)
1182             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1183                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1184                 inject_gp(vcpu);
1185                 return;
1186         }
1187
1188         kvm_arch_ops->set_efer(vcpu, efer);
1189
1190         efer &= ~EFER_LMA;
1191         efer |= vcpu->shadow_efer & EFER_LMA;
1192
1193         vcpu->shadow_efer = efer;
1194 }
1195
1196 #endif
1197
1198 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1199 {
1200         switch (msr) {
1201 #ifdef CONFIG_X86_64
1202         case MSR_EFER:
1203                 set_efer(vcpu, data);
1204                 break;
1205 #endif
1206         case MSR_IA32_MC0_STATUS:
1207                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1208                        __FUNCTION__, data);
1209                 break;
1210         case MSR_IA32_UCODE_REV:
1211         case MSR_IA32_UCODE_WRITE:
1212         case 0x200 ... 0x2ff: /* MTRRs */
1213                 break;
1214         case MSR_IA32_APICBASE:
1215                 vcpu->apic_base = data;
1216                 break;
1217         default:
1218                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1219                 return 1;
1220         }
1221         return 0;
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1224
1225 /*
1226  * Writes msr value into into the appropriate "register".
1227  * Returns 0 on success, non-0 otherwise.
1228  * Assumes vcpu_load() was already called.
1229  */
1230 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1231 {
1232         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1233 }
1234
1235 void kvm_resched(struct kvm_vcpu *vcpu)
1236 {
1237         vcpu_put(vcpu);
1238         cond_resched();
1239         /* Cannot fail -  no vcpu unplug yet. */
1240         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_resched);
1243
1244 void load_msrs(struct vmx_msr_entry *e, int n)
1245 {
1246         int i;
1247
1248         for (i = 0; i < n; ++i)
1249                 wrmsrl(e[i].index, e[i].data);
1250 }
1251 EXPORT_SYMBOL_GPL(load_msrs);
1252
1253 void save_msrs(struct vmx_msr_entry *e, int n)
1254 {
1255         int i;
1256
1257         for (i = 0; i < n; ++i)
1258                 rdmsrl(e[i].index, e[i].data);
1259 }
1260 EXPORT_SYMBOL_GPL(save_msrs);
1261
1262 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1263 {
1264         struct kvm_vcpu *vcpu;
1265         int r;
1266
1267         if (!valid_vcpu(kvm_run->vcpu))
1268                 return -EINVAL;
1269
1270         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1271         if (!vcpu)
1272                 return -ENOENT;
1273
1274         if (kvm_run->emulated) {
1275                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1276                 kvm_run->emulated = 0;
1277         }
1278
1279         if (kvm_run->mmio_completed) {
1280                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1281                 vcpu->mmio_read_completed = 1;
1282         }
1283
1284         vcpu->mmio_needed = 0;
1285
1286         r = kvm_arch_ops->run(vcpu, kvm_run);
1287
1288         vcpu_put(vcpu);
1289         return r;
1290 }
1291
1292 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1293 {
1294         struct kvm_vcpu *vcpu;
1295
1296         if (!valid_vcpu(regs->vcpu))
1297                 return -EINVAL;
1298
1299         vcpu = vcpu_load(kvm, regs->vcpu);
1300         if (!vcpu)
1301                 return -ENOENT;
1302
1303         kvm_arch_ops->cache_regs(vcpu);
1304
1305         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1306         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1307         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1308         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1309         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1310         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1311         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1312         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1313 #ifdef CONFIG_X86_64
1314         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1315         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1316         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1317         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1318         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1319         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1320         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1321         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1322 #endif
1323
1324         regs->rip = vcpu->rip;
1325         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1326
1327         /*
1328          * Don't leak debug flags in case they were set for guest debugging
1329          */
1330         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1331                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1332
1333         vcpu_put(vcpu);
1334
1335         return 0;
1336 }
1337
1338 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1339 {
1340         struct kvm_vcpu *vcpu;
1341
1342         if (!valid_vcpu(regs->vcpu))
1343                 return -EINVAL;
1344
1345         vcpu = vcpu_load(kvm, regs->vcpu);
1346         if (!vcpu)
1347                 return -ENOENT;
1348
1349         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1350         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1351         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1352         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1353         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1354         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1355         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1356         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1357 #ifdef CONFIG_X86_64
1358         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1359         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1360         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1361         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1362         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1363         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1364         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1365         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1366 #endif
1367
1368         vcpu->rip = regs->rip;
1369         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1370
1371         kvm_arch_ops->decache_regs(vcpu);
1372
1373         vcpu_put(vcpu);
1374
1375         return 0;
1376 }
1377
1378 static void get_segment(struct kvm_vcpu *vcpu,
1379                         struct kvm_segment *var, int seg)
1380 {
1381         return kvm_arch_ops->get_segment(vcpu, var, seg);
1382 }
1383
1384 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1385 {
1386         struct kvm_vcpu *vcpu;
1387         struct descriptor_table dt;
1388
1389         if (!valid_vcpu(sregs->vcpu))
1390                 return -EINVAL;
1391         vcpu = vcpu_load(kvm, sregs->vcpu);
1392         if (!vcpu)
1393                 return -ENOENT;
1394
1395         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1396         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1397         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1398         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1399         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1400         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1401
1402         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1403         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1404
1405         kvm_arch_ops->get_idt(vcpu, &dt);
1406         sregs->idt.limit = dt.limit;
1407         sregs->idt.base = dt.base;
1408         kvm_arch_ops->get_gdt(vcpu, &dt);
1409         sregs->gdt.limit = dt.limit;
1410         sregs->gdt.base = dt.base;
1411
1412         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1413         sregs->cr0 = vcpu->cr0;
1414         sregs->cr2 = vcpu->cr2;
1415         sregs->cr3 = vcpu->cr3;
1416         sregs->cr4 = vcpu->cr4;
1417         sregs->cr8 = vcpu->cr8;
1418         sregs->efer = vcpu->shadow_efer;
1419         sregs->apic_base = vcpu->apic_base;
1420
1421         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1422                sizeof sregs->interrupt_bitmap);
1423
1424         vcpu_put(vcpu);
1425
1426         return 0;
1427 }
1428
1429 static void set_segment(struct kvm_vcpu *vcpu,
1430                         struct kvm_segment *var, int seg)
1431 {
1432         return kvm_arch_ops->set_segment(vcpu, var, seg);
1433 }
1434
1435 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1436 {
1437         struct kvm_vcpu *vcpu;
1438         int mmu_reset_needed = 0;
1439         int i;
1440         struct descriptor_table dt;
1441
1442         if (!valid_vcpu(sregs->vcpu))
1443                 return -EINVAL;
1444         vcpu = vcpu_load(kvm, sregs->vcpu);
1445         if (!vcpu)
1446                 return -ENOENT;
1447
1448         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1449         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1450         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1451         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1452         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1453         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1454
1455         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1456         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1457
1458         dt.limit = sregs->idt.limit;
1459         dt.base = sregs->idt.base;
1460         kvm_arch_ops->set_idt(vcpu, &dt);
1461         dt.limit = sregs->gdt.limit;
1462         dt.base = sregs->gdt.base;
1463         kvm_arch_ops->set_gdt(vcpu, &dt);
1464
1465         vcpu->cr2 = sregs->cr2;
1466         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1467         vcpu->cr3 = sregs->cr3;
1468
1469         vcpu->cr8 = sregs->cr8;
1470
1471         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1472 #ifdef CONFIG_X86_64
1473         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1474 #endif
1475         vcpu->apic_base = sregs->apic_base;
1476
1477         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1478
1479         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1480         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1481
1482         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1483         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1484
1485         if (mmu_reset_needed)
1486                 kvm_mmu_reset_context(vcpu);
1487
1488         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1489                sizeof vcpu->irq_pending);
1490         vcpu->irq_summary = 0;
1491         for (i = 0; i < NR_IRQ_WORDS; ++i)
1492                 if (vcpu->irq_pending[i])
1493                         __set_bit(i, &vcpu->irq_summary);
1494
1495         vcpu_put(vcpu);
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1502  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1503  *
1504  * This list is modified at module load time to reflect the
1505  * capabilities of the host cpu.
1506  */
1507 static u32 msrs_to_save[] = {
1508         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1509         MSR_K6_STAR,
1510 #ifdef CONFIG_X86_64
1511         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1512 #endif
1513         MSR_IA32_TIME_STAMP_COUNTER,
1514 };
1515
1516 static unsigned num_msrs_to_save;
1517
1518 static __init void kvm_init_msr_list(void)
1519 {
1520         u32 dummy[2];
1521         unsigned i, j;
1522
1523         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1524                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1525                         continue;
1526                 if (j < i)
1527                         msrs_to_save[j] = msrs_to_save[i];
1528                 j++;
1529         }
1530         num_msrs_to_save = j;
1531 }
1532
1533 /*
1534  * Adapt set_msr() to msr_io()'s calling convention
1535  */
1536 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1537 {
1538         return set_msr(vcpu, index, *data);
1539 }
1540
1541 /*
1542  * Read or write a bunch of msrs. All parameters are kernel addresses.
1543  *
1544  * @return number of msrs set successfully.
1545  */
1546 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1547                     struct kvm_msr_entry *entries,
1548                     int (*do_msr)(struct kvm_vcpu *vcpu,
1549                                   unsigned index, u64 *data))
1550 {
1551         struct kvm_vcpu *vcpu;
1552         int i;
1553
1554         if (!valid_vcpu(msrs->vcpu))
1555                 return -EINVAL;
1556
1557         vcpu = vcpu_load(kvm, msrs->vcpu);
1558         if (!vcpu)
1559                 return -ENOENT;
1560
1561         for (i = 0; i < msrs->nmsrs; ++i)
1562                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1563                         break;
1564
1565         vcpu_put(vcpu);
1566
1567         return i;
1568 }
1569
1570 /*
1571  * Read or write a bunch of msrs. Parameters are user addresses.
1572  *
1573  * @return number of msrs set successfully.
1574  */
1575 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1576                   int (*do_msr)(struct kvm_vcpu *vcpu,
1577                                 unsigned index, u64 *data),
1578                   int writeback)
1579 {
1580         struct kvm_msrs msrs;
1581         struct kvm_msr_entry *entries;
1582         int r, n;
1583         unsigned size;
1584
1585         r = -EFAULT;
1586         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1587                 goto out;
1588
1589         r = -E2BIG;
1590         if (msrs.nmsrs >= MAX_IO_MSRS)
1591                 goto out;
1592
1593         r = -ENOMEM;
1594         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1595         entries = vmalloc(size);
1596         if (!entries)
1597                 goto out;
1598
1599         r = -EFAULT;
1600         if (copy_from_user(entries, user_msrs->entries, size))
1601                 goto out_free;
1602
1603         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1604         if (r < 0)
1605                 goto out_free;
1606
1607         r = -EFAULT;
1608         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1609                 goto out_free;
1610
1611         r = n;
1612
1613 out_free:
1614         vfree(entries);
1615 out:
1616         return r;
1617 }
1618
1619 /*
1620  * Translate a guest virtual address to a guest physical address.
1621  */
1622 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1623 {
1624         unsigned long vaddr = tr->linear_address;
1625         struct kvm_vcpu *vcpu;
1626         gpa_t gpa;
1627
1628         vcpu = vcpu_load(kvm, tr->vcpu);
1629         if (!vcpu)
1630                 return -ENOENT;
1631         spin_lock(&kvm->lock);
1632         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1633         tr->physical_address = gpa;
1634         tr->valid = gpa != UNMAPPED_GVA;
1635         tr->writeable = 1;
1636         tr->usermode = 0;
1637         spin_unlock(&kvm->lock);
1638         vcpu_put(vcpu);
1639
1640         return 0;
1641 }
1642
1643 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1644 {
1645         struct kvm_vcpu *vcpu;
1646
1647         if (!valid_vcpu(irq->vcpu))
1648                 return -EINVAL;
1649         if (irq->irq < 0 || irq->irq >= 256)
1650                 return -EINVAL;
1651         vcpu = vcpu_load(kvm, irq->vcpu);
1652         if (!vcpu)
1653                 return -ENOENT;
1654
1655         set_bit(irq->irq, vcpu->irq_pending);
1656         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1657
1658         vcpu_put(vcpu);
1659
1660         return 0;
1661 }
1662
1663 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1664                                      struct kvm_debug_guest *dbg)
1665 {
1666         struct kvm_vcpu *vcpu;
1667         int r;
1668
1669         if (!valid_vcpu(dbg->vcpu))
1670                 return -EINVAL;
1671         vcpu = vcpu_load(kvm, dbg->vcpu);
1672         if (!vcpu)
1673                 return -ENOENT;
1674
1675         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1676
1677         vcpu_put(vcpu);
1678
1679         return r;
1680 }
1681
1682 static long kvm_dev_ioctl(struct file *filp,
1683                           unsigned int ioctl, unsigned long arg)
1684 {
1685         struct kvm *kvm = filp->private_data;
1686         int r = -EINVAL;
1687
1688         switch (ioctl) {
1689         case KVM_GET_API_VERSION:
1690                 r = KVM_API_VERSION;
1691                 break;
1692         case KVM_CREATE_VCPU: {
1693                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1694                 if (r)
1695                         goto out;
1696                 break;
1697         }
1698         case KVM_RUN: {
1699                 struct kvm_run kvm_run;
1700
1701                 r = -EFAULT;
1702                 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1703                         goto out;
1704                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1705                 if (r < 0 &&  r != -EINTR)
1706                         goto out;
1707                 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1708                         r = -EFAULT;
1709                         goto out;
1710                 }
1711                 break;
1712         }
1713         case KVM_GET_REGS: {
1714                 struct kvm_regs kvm_regs;
1715
1716                 r = -EFAULT;
1717                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1718                         goto out;
1719                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1720                 if (r)
1721                         goto out;
1722                 r = -EFAULT;
1723                 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1724                         goto out;
1725                 r = 0;
1726                 break;
1727         }
1728         case KVM_SET_REGS: {
1729                 struct kvm_regs kvm_regs;
1730
1731                 r = -EFAULT;
1732                 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1733                         goto out;
1734                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1735                 if (r)
1736                         goto out;
1737                 r = 0;
1738                 break;
1739         }
1740         case KVM_GET_SREGS: {
1741                 struct kvm_sregs kvm_sregs;
1742
1743                 r = -EFAULT;
1744                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1745                         goto out;
1746                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1747                 if (r)
1748                         goto out;
1749                 r = -EFAULT;
1750                 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1751                         goto out;
1752                 r = 0;
1753                 break;
1754         }
1755         case KVM_SET_SREGS: {
1756                 struct kvm_sregs kvm_sregs;
1757
1758                 r = -EFAULT;
1759                 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1760                         goto out;
1761                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1762                 if (r)
1763                         goto out;
1764                 r = 0;
1765                 break;
1766         }
1767         case KVM_TRANSLATE: {
1768                 struct kvm_translation tr;
1769
1770                 r = -EFAULT;
1771                 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1772                         goto out;
1773                 r = kvm_dev_ioctl_translate(kvm, &tr);
1774                 if (r)
1775                         goto out;
1776                 r = -EFAULT;
1777                 if (copy_to_user((void *)arg, &tr, sizeof tr))
1778                         goto out;
1779                 r = 0;
1780                 break;
1781         }
1782         case KVM_INTERRUPT: {
1783                 struct kvm_interrupt irq;
1784
1785                 r = -EFAULT;
1786                 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1787                         goto out;
1788                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1789                 if (r)
1790                         goto out;
1791                 r = 0;
1792                 break;
1793         }
1794         case KVM_DEBUG_GUEST: {
1795                 struct kvm_debug_guest dbg;
1796
1797                 r = -EFAULT;
1798                 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1799                         goto out;
1800                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1801                 if (r)
1802                         goto out;
1803                 r = 0;
1804                 break;
1805         }
1806         case KVM_SET_MEMORY_REGION: {
1807                 struct kvm_memory_region kvm_mem;
1808
1809                 r = -EFAULT;
1810                 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1811                         goto out;
1812                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1813                 if (r)
1814                         goto out;
1815                 break;
1816         }
1817         case KVM_GET_DIRTY_LOG: {
1818                 struct kvm_dirty_log log;
1819
1820                 r = -EFAULT;
1821                 if (copy_from_user(&log, (void *)arg, sizeof log))
1822                         goto out;
1823                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1824                 if (r)
1825                         goto out;
1826                 break;
1827         }
1828         case KVM_GET_MSRS:
1829                 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1830                 break;
1831         case KVM_SET_MSRS:
1832                 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1833                 break;
1834         case KVM_GET_MSR_INDEX_LIST: {
1835                 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1836                 struct kvm_msr_list msr_list;
1837                 unsigned n;
1838
1839                 r = -EFAULT;
1840                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1841                         goto out;
1842                 n = msr_list.nmsrs;
1843                 msr_list.nmsrs = num_msrs_to_save;
1844                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1845                         goto out;
1846                 r = -E2BIG;
1847                 if (n < num_msrs_to_save)
1848                         goto out;
1849                 r = -EFAULT;
1850                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1851                                  num_msrs_to_save * sizeof(u32)))
1852                         goto out;
1853                 r = 0;
1854         }
1855         default:
1856                 ;
1857         }
1858 out:
1859         return r;
1860 }
1861
1862 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1863                                    unsigned long address,
1864                                    int *type)
1865 {
1866         struct kvm *kvm = vma->vm_file->private_data;
1867         unsigned long pgoff;
1868         struct kvm_memory_slot *slot;
1869         struct page *page;
1870
1871         *type = VM_FAULT_MINOR;
1872         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1873         slot = gfn_to_memslot(kvm, pgoff);
1874         if (!slot)
1875                 return NOPAGE_SIGBUS;
1876         page = gfn_to_page(slot, pgoff);
1877         if (!page)
1878                 return NOPAGE_SIGBUS;
1879         get_page(page);
1880         return page;
1881 }
1882
1883 static struct vm_operations_struct kvm_dev_vm_ops = {
1884         .nopage = kvm_dev_nopage,
1885 };
1886
1887 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1888 {
1889         vma->vm_ops = &kvm_dev_vm_ops;
1890         return 0;
1891 }
1892
1893 static struct file_operations kvm_chardev_ops = {
1894         .open           = kvm_dev_open,
1895         .release        = kvm_dev_release,
1896         .unlocked_ioctl = kvm_dev_ioctl,
1897         .compat_ioctl   = kvm_dev_ioctl,
1898         .mmap           = kvm_dev_mmap,
1899 };
1900
1901 static struct miscdevice kvm_dev = {
1902         MISC_DYNAMIC_MINOR,
1903         "kvm",
1904         &kvm_chardev_ops,
1905 };
1906
1907 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1908                        void *v)
1909 {
1910         if (val == SYS_RESTART) {
1911                 /*
1912                  * Some (well, at least mine) BIOSes hang on reboot if
1913                  * in vmx root mode.
1914                  */
1915                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1916                 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1917         }
1918         return NOTIFY_OK;
1919 }
1920
1921 static struct notifier_block kvm_reboot_notifier = {
1922         .notifier_call = kvm_reboot,
1923         .priority = 0,
1924 };
1925
1926 static __init void kvm_init_debug(void)
1927 {
1928         struct kvm_stats_debugfs_item *p;
1929
1930         debugfs_dir = debugfs_create_dir("kvm", 0);
1931         for (p = debugfs_entries; p->name; ++p)
1932                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1933                                                p->data);
1934 }
1935
1936 static void kvm_exit_debug(void)
1937 {
1938         struct kvm_stats_debugfs_item *p;
1939
1940         for (p = debugfs_entries; p->name; ++p)
1941                 debugfs_remove(p->dentry);
1942         debugfs_remove(debugfs_dir);
1943 }
1944
1945 hpa_t bad_page_address;
1946
1947 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1948 {
1949         int r;
1950
1951         if (kvm_arch_ops) {
1952                 printk(KERN_ERR "kvm: already loaded the other module\n");
1953                 return -EEXIST;
1954         }
1955
1956         if (!ops->cpu_has_kvm_support()) {
1957                 printk(KERN_ERR "kvm: no hardware support\n");
1958                 return -EOPNOTSUPP;
1959         }
1960         if (ops->disabled_by_bios()) {
1961                 printk(KERN_ERR "kvm: disabled by bios\n");
1962                 return -EOPNOTSUPP;
1963         }
1964
1965         kvm_arch_ops = ops;
1966
1967         r = kvm_arch_ops->hardware_setup();
1968         if (r < 0)
1969             return r;
1970
1971         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
1972         register_reboot_notifier(&kvm_reboot_notifier);
1973
1974         kvm_chardev_ops.owner = module;
1975
1976         r = misc_register(&kvm_dev);
1977         if (r) {
1978                 printk (KERN_ERR "kvm: misc device register failed\n");
1979                 goto out_free;
1980         }
1981
1982         return r;
1983
1984 out_free:
1985         unregister_reboot_notifier(&kvm_reboot_notifier);
1986         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1987         kvm_arch_ops->hardware_unsetup();
1988         return r;
1989 }
1990
1991 void kvm_exit_arch(void)
1992 {
1993         misc_deregister(&kvm_dev);
1994
1995         unregister_reboot_notifier(&kvm_reboot_notifier);
1996         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1997         kvm_arch_ops->hardware_unsetup();
1998         kvm_arch_ops = NULL;
1999 }
2000
2001 static __init int kvm_init(void)
2002 {
2003         static struct page *bad_page;
2004         int r = 0;
2005
2006         kvm_init_debug();
2007
2008         kvm_init_msr_list();
2009
2010         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2011                 r = -ENOMEM;
2012                 goto out;
2013         }
2014
2015         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2016         memset(__va(bad_page_address), 0, PAGE_SIZE);
2017
2018         return r;
2019
2020 out:
2021         kvm_exit_debug();
2022         return r;
2023 }
2024
2025 static __exit void kvm_exit(void)
2026 {
2027         kvm_exit_debug();
2028         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2029 }
2030
2031 module_init(kvm_init)
2032 module_exit(kvm_exit)
2033
2034 EXPORT_SYMBOL_GPL(kvm_init_arch);
2035 EXPORT_SYMBOL_GPL(kvm_exit_arch);