Pileus Git - ~andy/linux/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <asm/processor.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <asm/msr.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/reboot.h>
  32 #include <asm/io.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <asm/desc.h>
  37
  38 #include "x86_emulate.h"
  39 #include "segment_descriptor.h"
  40
  41 MODULE_AUTHOR("Qumranet");
  42 MODULE_LICENSE("GPL");
  43
  44 static DEFINE_SPINLOCK(kvm_lock);
  45 static LIST_HEAD(vm_list);
  46
  47 struct kvm_arch_ops *kvm_arch_ops;
  48 struct kvm_stat kvm_stat;
  49 EXPORT_SYMBOL_GPL(kvm_stat);
  50
  51 static struct kvm_stats_debugfs_item {
  52         const char *name;
  53         u32 *data;
  54         struct dentry *dentry;
  55 } debugfs_entries[] = {
  56         { "pf_fixed", &kvm_stat.pf_fixed },
  57         { "pf_guest", &kvm_stat.pf_guest },
  58         { "tlb_flush", &kvm_stat.tlb_flush },
  59         { "invlpg", &kvm_stat.invlpg },
  60         { "exits", &kvm_stat.exits },
  61         { "io_exits", &kvm_stat.io_exits },
  62         { "mmio_exits", &kvm_stat.mmio_exits },
  63         { "signal_exits", &kvm_stat.signal_exits },
  64         { "irq_window", &kvm_stat.irq_window_exits },
  65         { "halt_exits", &kvm_stat.halt_exits },
  66         { "request_irq", &kvm_stat.request_irq_exits },
  67         { "irq_exits", &kvm_stat.irq_exits },
  68         { NULL, NULL }
  69 };
  70
  71 static struct dentry *debugfs_dir;
  72
  73 #define MAX_IO_MSRS 256
  74
  75 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  76 #define LMSW_GUEST_MASK 0x0eULL
  77 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  78 #define CR8_RESEVED_BITS (~0x0fULL)
  79 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  80
  81 #ifdef CONFIG_X86_64
  82 // LDT or TSS descriptor in the GDT. 16 bytes.
  83 struct segment_descriptor_64 {
  84         struct segment_descriptor s;
  85         u32 base_higher;
  86         u32 pad_zero;
  87 };
  88
  89 #endif
  90
  91 unsigned long segment_base(u16 selector)
  92 {
  93         struct descriptor_table gdt;
  94         struct segment_descriptor *d;
  95         unsigned long table_base;
  96         typedef unsigned long ul;
  97         unsigned long v;
  98
  99         if (selector == 0)
 100                 return 0;
 101
 102         asm ("sgdt %0" : "=m"(gdt));
 103         table_base = gdt.base;
 104
 105         if (selector & 4) {           /* from ldt */
 106                 u16 ldt_selector;
 107
 108                 asm ("sldt %0" : "=g"(ldt_selector));
 109                 table_base = segment_base(ldt_selector);
 110         }
 111         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 112         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 113 #ifdef CONFIG_X86_64
 114         if (d->system == 0
 115             && (d->type == 2 || d->type == 9 || d->type == 11))
 116                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 117 #endif
 118         return v;
 119 }
 120 EXPORT_SYMBOL_GPL(segment_base);
 121
 122 static inline int valid_vcpu(int n)
 123 {
 124         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 125 }
 126
 127 int kvm_read_guest(struct kvm_vcpu *vcpu,
 128                              gva_t addr,
 129                              unsigned long size,
 130                              void *dest)
 131 {
 132         unsigned char *host_buf = dest;
 133         unsigned long req_size = size;
 134
 135         while (size) {
 136                 hpa_t paddr;
 137                 unsigned now;
 138                 unsigned offset;
 139                 hva_t guest_buf;
 140
 141                 paddr = gva_to_hpa(vcpu, addr);
 142
 143                 if (is_error_hpa(paddr))
 144                         break;
 145
 146                 guest_buf = (hva_t)kmap_atomic(
 147                                         pfn_to_page(paddr >> PAGE_SHIFT),
 148                                         KM_USER0);
 149                 offset = addr & ~PAGE_MASK;
 150                 guest_buf |= offset;
 151                 now = min(size, PAGE_SIZE - offset);
 152                 memcpy(host_buf, (void*)guest_buf, now);
 153                 host_buf += now;
 154                 addr += now;
 155                 size -= now;
 156                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 157         }
 158         return req_size - size;
 159 }
 160 EXPORT_SYMBOL_GPL(kvm_read_guest);
 161
 162 int kvm_write_guest(struct kvm_vcpu *vcpu,
 163                              gva_t addr,
 164                              unsigned long size,
 165                              void *data)
 166 {
 167         unsigned char *host_buf = data;
 168         unsigned long req_size = size;
 169
 170         while (size) {
 171                 hpa_t paddr;
 172                 unsigned now;
 173                 unsigned offset;
 174                 hva_t guest_buf;
 175
 176                 paddr = gva_to_hpa(vcpu, addr);
 177
 178                 if (is_error_hpa(paddr))
 179                         break;
 180
 181                 guest_buf = (hva_t)kmap_atomic(
 182                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 183                 offset = addr & ~PAGE_MASK;
 184                 guest_buf |= offset;
 185                 now = min(size, PAGE_SIZE - offset);
 186                 memcpy((void*)guest_buf, host_buf, now);
 187                 host_buf += now;
 188                 addr += now;
 189                 size -= now;
 190                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 191         }
 192         return req_size - size;
 193 }
 194 EXPORT_SYMBOL_GPL(kvm_write_guest);
 195
 196 static int vcpu_slot(struct kvm_vcpu *vcpu)
 197 {
 198         return vcpu - vcpu->kvm->vcpus;
 199 }
 200
 201 /*
 202  * Switches to specified vcpu, until a matching vcpu_put()
 203  */
 204 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
 205 {
 206         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
 207
 208         mutex_lock(&vcpu->mutex);
 209         if (unlikely(!vcpu->vmcs)) {
 210                 mutex_unlock(&vcpu->mutex);
 211                 return NULL;
 212         }
 213         return kvm_arch_ops->vcpu_load(vcpu);
 214 }
 215
 216 static void vcpu_put(struct kvm_vcpu *vcpu)
 217 {
 218         kvm_arch_ops->vcpu_put(vcpu);
 219         mutex_unlock(&vcpu->mutex);
 220 }
 221
 222 static int kvm_dev_open(struct inode *inode, struct file *filp)
 223 {
 224         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 225         int i;
 226
 227         if (!kvm)
 228                 return -ENOMEM;
 229
 230         spin_lock_init(&kvm->lock);
 231         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 232         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 233                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 234
 235                 mutex_init(&vcpu->mutex);
 236                 vcpu->cpu = -1;
 237                 vcpu->kvm = kvm;
 238                 vcpu->mmu.root_hpa = INVALID_PAGE;
 239                 INIT_LIST_HEAD(&vcpu->free_pages);
 240                 spin_lock(&kvm_lock);
 241                 list_add(&kvm->vm_list, &vm_list);
 242                 spin_unlock(&kvm_lock);
 243         }
 244         filp->private_data = kvm;
 245         return 0;
 246 }
 247
 248 /*
 249  * Free any memory in @free but not in @dont.
 250  */
 251 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 252                                   struct kvm_memory_slot *dont)
 253 {
 254         int i;
 255
 256         if (!dont || free->phys_mem != dont->phys_mem)
 257                 if (free->phys_mem) {
 258                         for (i = 0; i < free->npages; ++i)
 259                                 if (free->phys_mem[i])
 260                                         __free_page(free->phys_mem[i]);
 261                         vfree(free->phys_mem);
 262                 }
 263
 264         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 265                 vfree(free->dirty_bitmap);
 266
 267         free->phys_mem = NULL;
 268         free->npages = 0;
 269         free->dirty_bitmap = NULL;
 270 }
 271
 272 static void kvm_free_physmem(struct kvm *kvm)
 273 {
 274         int i;
 275
 276         for (i = 0; i < kvm->nmemslots; ++i)
 277                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 278 }
 279
 280 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 281 {
 282         if (!vcpu_load(vcpu->kvm, vcpu_slot(vcpu)))
 283                 return;
 284
 285         kvm_mmu_destroy(vcpu);
 286         vcpu_put(vcpu);
 287         kvm_arch_ops->vcpu_free(vcpu);
 288 }
 289
 290 static void kvm_free_vcpus(struct kvm *kvm)
 291 {
 292         unsigned int i;
 293
 294         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 295                 kvm_free_vcpu(&kvm->vcpus[i]);
 296 }
 297
 298 static int kvm_dev_release(struct inode *inode, struct file *filp)
 299 {
 300         struct kvm *kvm = filp->private_data;
 301
 302         spin_lock(&kvm_lock);
 303         list_del(&kvm->vm_list);
 304         spin_unlock(&kvm_lock);
 305         kvm_free_vcpus(kvm);
 306         kvm_free_physmem(kvm);
 307         kfree(kvm);
 308         return 0;
 309 }
 310
 311 static void inject_gp(struct kvm_vcpu *vcpu)
 312 {
 313         kvm_arch_ops->inject_gp(vcpu, 0);
 314 }
 315
 316 /*
 317  * Load the pae pdptrs.  Return true is they are all valid.
 318  */
 319 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 320 {
 321         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 322         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 323         int i;
 324         u64 pdpte;
 325         u64 *pdpt;
 326         int ret;
 327         struct kvm_memory_slot *memslot;
 328
 329         spin_lock(&vcpu->kvm->lock);
 330         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 331         /* FIXME: !memslot - emulate? 0xff? */
 332         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 333
 334         ret = 1;
 335         for (i = 0; i < 4; ++i) {
 336                 pdpte = pdpt[offset + i];
 337                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 338                         ret = 0;
 339                         goto out;
 340                 }
 341         }
 342
 343         for (i = 0; i < 4; ++i)
 344                 vcpu->pdptrs[i] = pdpt[offset + i];
 345
 346 out:
 347         kunmap_atomic(pdpt, KM_USER0);
 348         spin_unlock(&vcpu->kvm->lock);
 349
 350         return ret;
 351 }
 352
 353 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 354 {
 355         if (cr0 & CR0_RESEVED_BITS) {
 356                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 357                        cr0, vcpu->cr0);
 358                 inject_gp(vcpu);
 359                 return;
 360         }
 361
 362         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 363                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 364                 inject_gp(vcpu);
 365                 return;
 366         }
 367
 368         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 369                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 370                        "and a clear PE flag\n");
 371                 inject_gp(vcpu);
 372                 return;
 373         }
 374
 375         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 376 #ifdef CONFIG_X86_64
 377                 if ((vcpu->shadow_efer & EFER_LME)) {
 378                         int cs_db, cs_l;
 379
 380                         if (!is_pae(vcpu)) {
 381                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 382                                        "in long mode while PAE is disabled\n");
 383                                 inject_gp(vcpu);
 384                                 return;
 385                         }
 386                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 387                         if (cs_l) {
 388                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 389                                        "in long mode while CS.L == 1\n");
 390                                 inject_gp(vcpu);
 391                                 return;
 392
 393                         }
 394                 } else
 395 #endif
 396                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 397                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 398                                "reserved bits\n");
 399                         inject_gp(vcpu);
 400                         return;
 401                 }
 402
 403         }
 404
 405         kvm_arch_ops->set_cr0(vcpu, cr0);
 406         vcpu->cr0 = cr0;
 407
 408         spin_lock(&vcpu->kvm->lock);
 409         kvm_mmu_reset_context(vcpu);
 410         spin_unlock(&vcpu->kvm->lock);
 411         return;
 412 }
 413 EXPORT_SYMBOL_GPL(set_cr0);
 414
 415 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 416 {
 417         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 418         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 419 }
 420 EXPORT_SYMBOL_GPL(lmsw);
 421
 422 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 423 {
 424         if (cr4 & CR4_RESEVED_BITS) {
 425                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 426                 inject_gp(vcpu);
 427                 return;
 428         }
 429
 430         if (is_long_mode(vcpu)) {
 431                 if (!(cr4 & CR4_PAE_MASK)) {
 432                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 433                                "in long mode\n");
 434                         inject_gp(vcpu);
 435                         return;
 436                 }
 437         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 438                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 439                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 440                 inject_gp(vcpu);
 441         }
 442
 443         if (cr4 & CR4_VMXE_MASK) {
 444                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 445                 inject_gp(vcpu);
 446                 return;
 447         }
 448         kvm_arch_ops->set_cr4(vcpu, cr4);
 449         spin_lock(&vcpu->kvm->lock);
 450         kvm_mmu_reset_context(vcpu);
 451         spin_unlock(&vcpu->kvm->lock);
 452 }
 453 EXPORT_SYMBOL_GPL(set_cr4);
 454
 455 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 456 {
 457         if (is_long_mode(vcpu)) {
 458                 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
 459                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 460                         inject_gp(vcpu);
 461                         return;
 462                 }
 463         } else {
 464                 if (cr3 & CR3_RESEVED_BITS) {
 465                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 466                         inject_gp(vcpu);
 467                         return;
 468                 }
 469                 if (is_paging(vcpu) && is_pae(vcpu) &&
 470                     !load_pdptrs(vcpu, cr3)) {
 471                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 472                                "reserved bits\n");
 473                         inject_gp(vcpu);
 474                         return;
 475                 }
 476         }
 477
 478         vcpu->cr3 = cr3;
 479         spin_lock(&vcpu->kvm->lock);
 480         /*
 481          * Does the new cr3 value map to physical memory? (Note, we
 482          * catch an invalid cr3 even in real-mode, because it would
 483          * cause trouble later on when we turn on paging anyway.)
 484          *
 485          * A real CPU would silently accept an invalid cr3 and would
 486          * attempt to use it - with largely undefined (and often hard
 487          * to debug) behavior on the guest side.
 488          */
 489         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 490                 inject_gp(vcpu);
 491         else
 492                 vcpu->mmu.new_cr3(vcpu);
 493         spin_unlock(&vcpu->kvm->lock);
 494 }
 495 EXPORT_SYMBOL_GPL(set_cr3);
 496
 497 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 498 {
 499         if ( cr8 & CR8_RESEVED_BITS) {
 500                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 501                 inject_gp(vcpu);
 502                 return;
 503         }
 504         vcpu->cr8 = cr8;
 505 }
 506 EXPORT_SYMBOL_GPL(set_cr8);
 507
 508 void fx_init(struct kvm_vcpu *vcpu)
 509 {
 510         struct __attribute__ ((__packed__)) fx_image_s {
 511                 u16 control; //fcw
 512                 u16 status; //fsw
 513                 u16 tag; // ftw
 514                 u16 opcode; //fop
 515                 u64 ip; // fpu ip
 516                 u64 operand;// fpu dp
 517                 u32 mxcsr;
 518                 u32 mxcsr_mask;
 519
 520         } *fx_image;
 521
 522         fx_save(vcpu->host_fx_image);
 523         fpu_init();
 524         fx_save(vcpu->guest_fx_image);
 525         fx_restore(vcpu->host_fx_image);
 526
 527         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 528         fx_image->mxcsr = 0x1f80;
 529         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 530                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 531 }
 532 EXPORT_SYMBOL_GPL(fx_init);
 533
 534 /*
 535  * Creates some virtual cpus.  Good luck creating more than one.
 536  */
 537 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 538 {
 539         int r;
 540         struct kvm_vcpu *vcpu;
 541
 542         r = -EINVAL;
 543         if (!valid_vcpu(n))
 544                 goto out;
 545
 546         vcpu = &kvm->vcpus[n];
 547
 548         mutex_lock(&vcpu->mutex);
 549
 550         if (vcpu->vmcs) {
 551                 mutex_unlock(&vcpu->mutex);
 552                 return -EEXIST;
 553         }
 554
 555         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 556                                            FX_IMAGE_ALIGN);
 557         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 558
 559         r = kvm_arch_ops->vcpu_create(vcpu);
 560         if (r < 0)
 561                 goto out_free_vcpus;
 562
 563         r = kvm_mmu_create(vcpu);
 564         if (r < 0)
 565                 goto out_free_vcpus;
 566
 567         kvm_arch_ops->vcpu_load(vcpu);
 568         r = kvm_mmu_setup(vcpu);
 569         if (r >= 0)
 570                 r = kvm_arch_ops->vcpu_setup(vcpu);
 571         vcpu_put(vcpu);
 572
 573         if (r < 0)
 574                 goto out_free_vcpus;
 575
 576         return 0;
 577
 578 out_free_vcpus:
 579         kvm_free_vcpu(vcpu);
 580         mutex_unlock(&vcpu->mutex);
 581 out:
 582         return r;
 583 }
 584
 585 /*
 586  * Allocate some memory and give it an address in the guest physical address
 587  * space.
 588  *
 589  * Discontiguous memory is allowed, mostly for framebuffers.
 590  */
 591 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 592                                            struct kvm_memory_region *mem)
 593 {
 594         int r;
 595         gfn_t base_gfn;
 596         unsigned long npages;
 597         unsigned long i;
 598         struct kvm_memory_slot *memslot;
 599         struct kvm_memory_slot old, new;
 600         int memory_config_version;
 601
 602         r = -EINVAL;
 603         /* General sanity checks */
 604         if (mem->memory_size & (PAGE_SIZE - 1))
 605                 goto out;
 606         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 607                 goto out;
 608         if (mem->slot >= KVM_MEMORY_SLOTS)
 609                 goto out;
 610         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 611                 goto out;
 612
 613         memslot = &kvm->memslots[mem->slot];
 614         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 615         npages = mem->memory_size >> PAGE_SHIFT;
 616
 617         if (!npages)
 618                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 619
 620 raced:
 621         spin_lock(&kvm->lock);
 622
 623         memory_config_version = kvm->memory_config_version;
 624         new = old = *memslot;
 625
 626         new.base_gfn = base_gfn;
 627         new.npages = npages;
 628         new.flags = mem->flags;
 629
 630         /* Disallow changing a memory slot's size. */
 631         r = -EINVAL;
 632         if (npages && old.npages && npages != old.npages)
 633                 goto out_unlock;
 634
 635         /* Check for overlaps */
 636         r = -EEXIST;
 637         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 638                 struct kvm_memory_slot *s = &kvm->memslots[i];
 639
 640                 if (s == memslot)
 641                         continue;
 642                 if (!((base_gfn + npages <= s->base_gfn) ||
 643                       (base_gfn >= s->base_gfn + s->npages)))
 644                         goto out_unlock;
 645         }
 646         /*
 647          * Do memory allocations outside lock.  memory_config_version will
 648          * detect any races.
 649          */
 650         spin_unlock(&kvm->lock);
 651
 652         /* Deallocate if slot is being removed */
 653         if (!npages)
 654                 new.phys_mem = NULL;
 655
 656         /* Free page dirty bitmap if unneeded */
 657         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 658                 new.dirty_bitmap = NULL;
 659
 660         r = -ENOMEM;
 661
 662         /* Allocate if a slot is being created */
 663         if (npages && !new.phys_mem) {
 664                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 665
 666                 if (!new.phys_mem)
 667                         goto out_free;
 668
 669                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 670                 for (i = 0; i < npages; ++i) {
 671                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 672                                                      | __GFP_ZERO);
 673                         if (!new.phys_mem[i])
 674                                 goto out_free;
 675                         new.phys_mem[i]->private = 0;
 676                 }
 677         }
 678
 679         /* Allocate page dirty bitmap if needed */
 680         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 681                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 682
 683                 new.dirty_bitmap = vmalloc(dirty_bytes);
 684                 if (!new.dirty_bitmap)
 685                         goto out_free;
 686                 memset(new.dirty_bitmap, 0, dirty_bytes);
 687         }
 688
 689         spin_lock(&kvm->lock);
 690
 691         if (memory_config_version != kvm->memory_config_version) {
 692                 spin_unlock(&kvm->lock);
 693                 kvm_free_physmem_slot(&new, &old);
 694                 goto raced;
 695         }
 696
 697         r = -EAGAIN;
 698         if (kvm->busy)
 699                 goto out_unlock;
 700
 701         if (mem->slot >= kvm->nmemslots)
 702                 kvm->nmemslots = mem->slot + 1;
 703
 704         *memslot = new;
 705         ++kvm->memory_config_version;
 706
 707         spin_unlock(&kvm->lock);
 708
 709         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 710                 struct kvm_vcpu *vcpu;
 711
 712                 vcpu = vcpu_load(kvm, i);
 713                 if (!vcpu)
 714                         continue;
 715                 kvm_mmu_reset_context(vcpu);
 716                 vcpu_put(vcpu);
 717         }
 718
 719         kvm_free_physmem_slot(&old, &new);
 720         return 0;
 721
 722 out_unlock:
 723         spin_unlock(&kvm->lock);
 724 out_free:
 725         kvm_free_physmem_slot(&new, &old);
 726 out:
 727         return r;
 728 }
 729
 730 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 731 {
 732         spin_lock(&vcpu->kvm->lock);
 733         kvm_mmu_slot_remove_write_access(vcpu, slot);
 734         spin_unlock(&vcpu->kvm->lock);
 735 }
 736
 737 /*
 738  * Get (and clear) the dirty memory log for a memory slot.
 739  */
 740 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 741                                        struct kvm_dirty_log *log)
 742 {
 743         struct kvm_memory_slot *memslot;
 744         int r, i;
 745         int n;
 746         int cleared;
 747         unsigned long any = 0;
 748
 749         spin_lock(&kvm->lock);
 750
 751         /*
 752          * Prevent changes to guest memory configuration even while the lock
 753          * is not taken.
 754          */
 755         ++kvm->busy;
 756         spin_unlock(&kvm->lock);
 757         r = -EINVAL;
 758         if (log->slot >= KVM_MEMORY_SLOTS)
 759                 goto out;
 760
 761         memslot = &kvm->memslots[log->slot];
 762         r = -ENOENT;
 763         if (!memslot->dirty_bitmap)
 764                 goto out;
 765
 766         n = ALIGN(memslot->npages, 8) / 8;
 767
 768         for (i = 0; !any && i < n; ++i)
 769                 any = memslot->dirty_bitmap[i];
 770
 771         r = -EFAULT;
 772         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 773                 goto out;
 774
 775
 776         if (any) {
 777                 cleared = 0;
 778                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 779                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 780
 781                         if (!vcpu)
 782                                 continue;
 783                         if (!cleared) {
 784                                 do_remove_write_access(vcpu, log->slot);
 785                                 memset(memslot->dirty_bitmap, 0, n);
 786                                 cleared = 1;
 787                         }
 788                         kvm_arch_ops->tlb_flush(vcpu);
 789                         vcpu_put(vcpu);
 790                 }
 791         }
 792
 793         r = 0;
 794
 795 out:
 796         spin_lock(&kvm->lock);
 797         --kvm->busy;
 798         spin_unlock(&kvm->lock);
 799         return r;
 800 }
 801
 802 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 803 {
 804         int i;
 805
 806         for (i = 0; i < kvm->nmemslots; ++i) {
 807                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 808
 809                 if (gfn >= memslot->base_gfn
 810                     && gfn < memslot->base_gfn + memslot->npages)
 811                         return memslot;
 812         }
 813         return NULL;
 814 }
 815 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 816
 817 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 818 {
 819         int i;
 820         struct kvm_memory_slot *memslot = NULL;
 821         unsigned long rel_gfn;
 822
 823         for (i = 0; i < kvm->nmemslots; ++i) {
 824                 memslot = &kvm->memslots[i];
 825
 826                 if (gfn >= memslot->base_gfn
 827                     && gfn < memslot->base_gfn + memslot->npages) {
 828
 829                         if (!memslot || !memslot->dirty_bitmap)
 830                                 return;
 831
 832                         rel_gfn = gfn - memslot->base_gfn;
 833
 834                         /* avoid RMW */
 835                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 836                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 837                         return;
 838                 }
 839         }
 840 }
 841
 842 static int emulator_read_std(unsigned long addr,
 843                              unsigned long *val,
 844                              unsigned int bytes,
 845                              struct x86_emulate_ctxt *ctxt)
 846 {
 847         struct kvm_vcpu *vcpu = ctxt->vcpu;
 848         void *data = val;
 849
 850         while (bytes) {
 851                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 852                 unsigned offset = addr & (PAGE_SIZE-1);
 853                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 854                 unsigned long pfn;
 855                 struct kvm_memory_slot *memslot;
 856                 void *page;
 857
 858                 if (gpa == UNMAPPED_GVA)
 859                         return X86EMUL_PROPAGATE_FAULT;
 860                 pfn = gpa >> PAGE_SHIFT;
 861                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 862                 if (!memslot)
 863                         return X86EMUL_UNHANDLEABLE;
 864                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 865
 866                 memcpy(data, page + offset, tocopy);
 867
 868                 kunmap_atomic(page, KM_USER0);
 869
 870                 bytes -= tocopy;
 871                 data += tocopy;
 872                 addr += tocopy;
 873         }
 874
 875         return X86EMUL_CONTINUE;
 876 }
 877
 878 static int emulator_write_std(unsigned long addr,
 879                               unsigned long val,
 880                               unsigned int bytes,
 881                               struct x86_emulate_ctxt *ctxt)
 882 {
 883         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 884                addr, bytes);
 885         return X86EMUL_UNHANDLEABLE;
 886 }
 887
 888 static int emulator_read_emulated(unsigned long addr,
 889                                   unsigned long *val,
 890                                   unsigned int bytes,
 891                                   struct x86_emulate_ctxt *ctxt)
 892 {
 893         struct kvm_vcpu *vcpu = ctxt->vcpu;
 894
 895         if (vcpu->mmio_read_completed) {
 896                 memcpy(val, vcpu->mmio_data, bytes);
 897                 vcpu->mmio_read_completed = 0;
 898                 return X86EMUL_CONTINUE;
 899         } else if (emulator_read_std(addr, val, bytes, ctxt)
 900                    == X86EMUL_CONTINUE)
 901                 return X86EMUL_CONTINUE;
 902         else {
 903                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 904                 if (gpa == UNMAPPED_GVA)
 905                         return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
 906                 vcpu->mmio_needed = 1;
 907                 vcpu->mmio_phys_addr = gpa;
 908                 vcpu->mmio_size = bytes;
 909                 vcpu->mmio_is_write = 0;
 910
 911                 return X86EMUL_UNHANDLEABLE;
 912         }
 913 }
 914
 915 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 916                                unsigned long val, int bytes)
 917 {
 918         struct kvm_memory_slot *m;
 919         struct page *page;
 920         void *virt;
 921
 922         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 923                 return 0;
 924         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 925         if (!m)
 926                 return 0;
 927         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 928         kvm_mmu_pre_write(vcpu, gpa, bytes);
 929         virt = kmap_atomic(page, KM_USER0);
 930         memcpy(virt + offset_in_page(gpa), &val, bytes);
 931         kunmap_atomic(virt, KM_USER0);
 932         kvm_mmu_post_write(vcpu, gpa, bytes);
 933         return 1;
 934 }
 935
 936 static int emulator_write_emulated(unsigned long addr,
 937                                    unsigned long val,
 938                                    unsigned int bytes,
 939                                    struct x86_emulate_ctxt *ctxt)
 940 {
 941         struct kvm_vcpu *vcpu = ctxt->vcpu;
 942         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 943
 944         if (gpa == UNMAPPED_GVA)
 945                 return X86EMUL_PROPAGATE_FAULT;
 946
 947         if (emulator_write_phys(vcpu, gpa, val, bytes))
 948                 return X86EMUL_CONTINUE;
 949
 950         vcpu->mmio_needed = 1;
 951         vcpu->mmio_phys_addr = gpa;
 952         vcpu->mmio_size = bytes;
 953         vcpu->mmio_is_write = 1;
 954         memcpy(vcpu->mmio_data, &val, bytes);
 955
 956         return X86EMUL_CONTINUE;
 957 }
 958
 959 static int emulator_cmpxchg_emulated(unsigned long addr,
 960                                      unsigned long old,
 961                                      unsigned long new,
 962                                      unsigned int bytes,
 963                                      struct x86_emulate_ctxt *ctxt)
 964 {
 965         static int reported;
 966
 967         if (!reported) {
 968                 reported = 1;
 969                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 970         }
 971         return emulator_write_emulated(addr, new, bytes, ctxt);
 972 }
 973
 974 #ifdef CONFIG_X86_32
 975
 976 static int emulator_cmpxchg8b_emulated(unsigned long addr,
 977                                        unsigned long old_lo,
 978                                        unsigned long old_hi,
 979                                        unsigned long new_lo,
 980                                        unsigned long new_hi,
 981                                        struct x86_emulate_ctxt *ctxt)
 982 {
 983         static int reported;
 984         int r;
 985
 986         if (!reported) {
 987                 reported = 1;
 988                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
 989         }
 990         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
 991         if (r != X86EMUL_CONTINUE)
 992                 return r;
 993         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
 994 }
 995
 996 #endif
 997
 998 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 999 {
1000         return kvm_arch_ops->get_segment_base(vcpu, seg);
1001 }
1002
1003 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1004 {
1005         return X86EMUL_CONTINUE;
1006 }
1007
1008 int emulate_clts(struct kvm_vcpu *vcpu)
1009 {
1010         unsigned long cr0;
1011
1012         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1013         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1014         kvm_arch_ops->set_cr0(vcpu, cr0);
1015         return X86EMUL_CONTINUE;
1016 }
1017
1018 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1019 {
1020         struct kvm_vcpu *vcpu = ctxt->vcpu;
1021
1022         switch (dr) {
1023         case 0 ... 3:
1024                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1025                 return X86EMUL_CONTINUE;
1026         default:
1027                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1028                        __FUNCTION__, dr);
1029                 return X86EMUL_UNHANDLEABLE;
1030         }
1031 }
1032
1033 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1034 {
1035         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1036         int exception;
1037
1038         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1039         if (exception) {
1040                 /* FIXME: better handling */
1041                 return X86EMUL_UNHANDLEABLE;
1042         }
1043         return X86EMUL_CONTINUE;
1044 }
1045
1046 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1047 {
1048         static int reported;
1049         u8 opcodes[4];
1050         unsigned long rip = ctxt->vcpu->rip;
1051         unsigned long rip_linear;
1052
1053         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1054
1055         if (reported)
1056                 return;
1057
1058         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1059
1060         printk(KERN_ERR "emulation failed but !mmio_needed?"
1061                " rip %lx %02x %02x %02x %02x\n",
1062                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1063         reported = 1;
1064 }
1065
1066 struct x86_emulate_ops emulate_ops = {
1067         .read_std            = emulator_read_std,
1068         .write_std           = emulator_write_std,
1069         .read_emulated       = emulator_read_emulated,
1070         .write_emulated      = emulator_write_emulated,
1071         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1072 #ifdef CONFIG_X86_32
1073         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1074 #endif
1075 };
1076
1077 int emulate_instruction(struct kvm_vcpu *vcpu,
1078                         struct kvm_run *run,
1079                         unsigned long cr2,
1080                         u16 error_code)
1081 {
1082         struct x86_emulate_ctxt emulate_ctxt;
1083         int r;
1084         int cs_db, cs_l;
1085
1086         kvm_arch_ops->cache_regs(vcpu);
1087
1088         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1089
1090         emulate_ctxt.vcpu = vcpu;
1091         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1092         emulate_ctxt.cr2 = cr2;
1093         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1094                 ? X86EMUL_MODE_REAL : cs_l
1095                 ? X86EMUL_MODE_PROT64 : cs_db
1096                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1097
1098         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1099                 emulate_ctxt.cs_base = 0;
1100                 emulate_ctxt.ds_base = 0;
1101                 emulate_ctxt.es_base = 0;
1102                 emulate_ctxt.ss_base = 0;
1103         } else {
1104                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1105                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1106                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1107                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1108         }
1109
1110         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1111         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1112
1113         vcpu->mmio_is_write = 0;
1114         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1115
1116         if ((r || vcpu->mmio_is_write) && run) {
1117                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1118                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1119                 run->mmio.len = vcpu->mmio_size;
1120                 run->mmio.is_write = vcpu->mmio_is_write;
1121         }
1122
1123         if (r) {
1124                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1125                         return EMULATE_DONE;
1126                 if (!vcpu->mmio_needed) {
1127                         report_emulation_failure(&emulate_ctxt);
1128                         return EMULATE_FAIL;
1129                 }
1130                 return EMULATE_DO_MMIO;
1131         }
1132
1133         kvm_arch_ops->decache_regs(vcpu);
1134         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1135
1136         if (vcpu->mmio_is_write)
1137                 return EMULATE_DO_MMIO;
1138
1139         return EMULATE_DONE;
1140 }
1141 EXPORT_SYMBOL_GPL(emulate_instruction);
1142
1143 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1144 {
1145         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1146 }
1147
1148 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1149 {
1150         struct descriptor_table dt = { limit, base };
1151
1152         kvm_arch_ops->set_gdt(vcpu, &dt);
1153 }
1154
1155 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1156 {
1157         struct descriptor_table dt = { limit, base };
1158
1159         kvm_arch_ops->set_idt(vcpu, &dt);
1160 }
1161
1162 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1163                    unsigned long *rflags)
1164 {
1165         lmsw(vcpu, msw);
1166         *rflags = kvm_arch_ops->get_rflags(vcpu);
1167 }
1168
1169 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1170 {
1171         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1172         switch (cr) {
1173         case 0:
1174                 return vcpu->cr0;
1175         case 2:
1176                 return vcpu->cr2;
1177         case 3:
1178                 return vcpu->cr3;
1179         case 4:
1180                 return vcpu->cr4;
1181         default:
1182                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1183                 return 0;
1184         }
1185 }
1186
1187 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1188                      unsigned long *rflags)
1189 {
1190         switch (cr) {
1191         case 0:
1192                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1193                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1194                 break;
1195         case 2:
1196                 vcpu->cr2 = val;
1197                 break;
1198         case 3:
1199                 set_cr3(vcpu, val);
1200                 break;
1201         case 4:
1202                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1203                 break;
1204         default:
1205                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1206         }
1207 }
1208
1209 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1210 {
1211         u64 data;
1212
1213         switch (msr) {
1214         case 0xc0010010: /* SYSCFG */
1215         case 0xc0010015: /* HWCR */
1216         case MSR_IA32_PLATFORM_ID:
1217         case MSR_IA32_P5_MC_ADDR:
1218         case MSR_IA32_P5_MC_TYPE:
1219         case MSR_IA32_MC0_CTL:
1220         case MSR_IA32_MCG_STATUS:
1221         case MSR_IA32_MCG_CAP:
1222         case MSR_IA32_MC0_MISC:
1223         case MSR_IA32_MC0_MISC+4:
1224         case MSR_IA32_MC0_MISC+8:
1225         case MSR_IA32_MC0_MISC+12:
1226         case MSR_IA32_MC0_MISC+16:
1227         case MSR_IA32_UCODE_REV:
1228         case MSR_IA32_PERF_STATUS:
1229                 /* MTRR registers */
1230         case 0xfe:
1231         case 0x200 ... 0x2ff:
1232                 data = 0;
1233                 break;
1234         case 0xcd: /* fsb frequency */
1235                 data = 3;
1236                 break;
1237         case MSR_IA32_APICBASE:
1238                 data = vcpu->apic_base;
1239                 break;
1240         case MSR_IA32_MISC_ENABLE:
1241                 data = vcpu->ia32_misc_enable_msr;
1242                 break;
1243 #ifdef CONFIG_X86_64
1244         case MSR_EFER:
1245                 data = vcpu->shadow_efer;
1246                 break;
1247 #endif
1248         default:
1249                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1250                 return 1;
1251         }
1252         *pdata = data;
1253         return 0;
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1256
1257 /*
1258  * Reads an msr value (of 'msr_index') into 'pdata'.
1259  * Returns 0 on success, non-0 otherwise.
1260  * Assumes vcpu_load() was already called.
1261  */
1262 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1263 {
1264         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1265 }
1266
1267 #ifdef CONFIG_X86_64
1268
1269 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1270 {
1271         if (efer & EFER_RESERVED_BITS) {
1272                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1273                        efer);
1274                 inject_gp(vcpu);
1275                 return;
1276         }
1277
1278         if (is_paging(vcpu)
1279             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1280                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1281                 inject_gp(vcpu);
1282                 return;
1283         }
1284
1285         kvm_arch_ops->set_efer(vcpu, efer);
1286
1287         efer &= ~EFER_LMA;
1288         efer |= vcpu->shadow_efer & EFER_LMA;
1289
1290         vcpu->shadow_efer = efer;
1291 }
1292
1293 #endif
1294
1295 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1296 {
1297         switch (msr) {
1298 #ifdef CONFIG_X86_64
1299         case MSR_EFER:
1300                 set_efer(vcpu, data);
1301                 break;
1302 #endif
1303         case MSR_IA32_MC0_STATUS:
1304                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1305                        __FUNCTION__, data);
1306                 break;
1307         case MSR_IA32_UCODE_REV:
1308         case MSR_IA32_UCODE_WRITE:
1309         case 0x200 ... 0x2ff: /* MTRRs */
1310                 break;
1311         case MSR_IA32_APICBASE:
1312                 vcpu->apic_base = data;
1313                 break;
1314         case MSR_IA32_MISC_ENABLE:
1315                 vcpu->ia32_misc_enable_msr = data;
1316                 break;
1317         default:
1318                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1319                 return 1;
1320         }
1321         return 0;
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1324
1325 /*
1326  * Writes msr value into into the appropriate "register".
1327  * Returns 0 on success, non-0 otherwise.
1328  * Assumes vcpu_load() was already called.
1329  */
1330 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1331 {
1332         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1333 }
1334
1335 void kvm_resched(struct kvm_vcpu *vcpu)
1336 {
1337         vcpu_put(vcpu);
1338         cond_resched();
1339         /* Cannot fail -  no vcpu unplug yet. */
1340         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1341 }
1342 EXPORT_SYMBOL_GPL(kvm_resched);
1343
1344 void load_msrs(struct vmx_msr_entry *e, int n)
1345 {
1346         int i;
1347
1348         for (i = 0; i < n; ++i)
1349                 wrmsrl(e[i].index, e[i].data);
1350 }
1351 EXPORT_SYMBOL_GPL(load_msrs);
1352
1353 void save_msrs(struct vmx_msr_entry *e, int n)
1354 {
1355         int i;
1356
1357         for (i = 0; i < n; ++i)
1358                 rdmsrl(e[i].index, e[i].data);
1359 }
1360 EXPORT_SYMBOL_GPL(save_msrs);
1361
1362 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1363 {
1364         struct kvm_vcpu *vcpu;
1365         int r;
1366
1367         if (!valid_vcpu(kvm_run->vcpu))
1368                 return -EINVAL;
1369
1370         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1371         if (!vcpu)
1372                 return -ENOENT;
1373
1374         /* re-sync apic's tpr */
1375         vcpu->cr8 = kvm_run->cr8;
1376
1377         if (kvm_run->emulated) {
1378                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1379                 kvm_run->emulated = 0;
1380         }
1381
1382         if (kvm_run->mmio_completed) {
1383                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1384                 vcpu->mmio_read_completed = 1;
1385         }
1386
1387         vcpu->mmio_needed = 0;
1388
1389         r = kvm_arch_ops->run(vcpu, kvm_run);
1390
1391         vcpu_put(vcpu);
1392         return r;
1393 }
1394
1395 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1396 {
1397         struct kvm_vcpu *vcpu;
1398
1399         if (!valid_vcpu(regs->vcpu))
1400                 return -EINVAL;
1401
1402         vcpu = vcpu_load(kvm, regs->vcpu);
1403         if (!vcpu)
1404                 return -ENOENT;
1405
1406         kvm_arch_ops->cache_regs(vcpu);
1407
1408         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1409         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1410         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1411         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1412         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1413         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1414         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1415         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1416 #ifdef CONFIG_X86_64
1417         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1418         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1419         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1420         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1421         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1422         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1423         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1424         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1425 #endif
1426
1427         regs->rip = vcpu->rip;
1428         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1429
1430         /*
1431          * Don't leak debug flags in case they were set for guest debugging
1432          */
1433         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1434                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1435
1436         vcpu_put(vcpu);
1437
1438         return 0;
1439 }
1440
1441 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1442 {
1443         struct kvm_vcpu *vcpu;
1444
1445         if (!valid_vcpu(regs->vcpu))
1446                 return -EINVAL;
1447
1448         vcpu = vcpu_load(kvm, regs->vcpu);
1449         if (!vcpu)
1450                 return -ENOENT;
1451
1452         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1453         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1454         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1455         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1456         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1457         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1458         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1459         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1460 #ifdef CONFIG_X86_64
1461         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1462         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1463         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1464         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1465         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1466         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1467         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1468         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1469 #endif
1470
1471         vcpu->rip = regs->rip;
1472         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1473
1474         kvm_arch_ops->decache_regs(vcpu);
1475
1476         vcpu_put(vcpu);
1477
1478         return 0;
1479 }
1480
1481 static void get_segment(struct kvm_vcpu *vcpu,
1482                         struct kvm_segment *var, int seg)
1483 {
1484         return kvm_arch_ops->get_segment(vcpu, var, seg);
1485 }
1486
1487 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1488 {
1489         struct kvm_vcpu *vcpu;
1490         struct descriptor_table dt;
1491
1492         if (!valid_vcpu(sregs->vcpu))
1493                 return -EINVAL;
1494         vcpu = vcpu_load(kvm, sregs->vcpu);
1495         if (!vcpu)
1496                 return -ENOENT;
1497
1498         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1499         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1500         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1501         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1502         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1503         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1504
1505         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1506         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1507
1508         kvm_arch_ops->get_idt(vcpu, &dt);
1509         sregs->idt.limit = dt.limit;
1510         sregs->idt.base = dt.base;
1511         kvm_arch_ops->get_gdt(vcpu, &dt);
1512         sregs->gdt.limit = dt.limit;
1513         sregs->gdt.base = dt.base;
1514
1515         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1516         sregs->cr0 = vcpu->cr0;
1517         sregs->cr2 = vcpu->cr2;
1518         sregs->cr3 = vcpu->cr3;
1519         sregs->cr4 = vcpu->cr4;
1520         sregs->cr8 = vcpu->cr8;
1521         sregs->efer = vcpu->shadow_efer;
1522         sregs->apic_base = vcpu->apic_base;
1523
1524         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1525                sizeof sregs->interrupt_bitmap);
1526
1527         vcpu_put(vcpu);
1528
1529         return 0;
1530 }
1531
1532 static void set_segment(struct kvm_vcpu *vcpu,
1533                         struct kvm_segment *var, int seg)
1534 {
1535         return kvm_arch_ops->set_segment(vcpu, var, seg);
1536 }
1537
1538 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1539 {
1540         struct kvm_vcpu *vcpu;
1541         int mmu_reset_needed = 0;
1542         int i;
1543         struct descriptor_table dt;
1544
1545         if (!valid_vcpu(sregs->vcpu))
1546                 return -EINVAL;
1547         vcpu = vcpu_load(kvm, sregs->vcpu);
1548         if (!vcpu)
1549                 return -ENOENT;
1550
1551         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1552         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1553         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1554         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1555         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1556         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1557
1558         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1559         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1560
1561         dt.limit = sregs->idt.limit;
1562         dt.base = sregs->idt.base;
1563         kvm_arch_ops->set_idt(vcpu, &dt);
1564         dt.limit = sregs->gdt.limit;
1565         dt.base = sregs->gdt.base;
1566         kvm_arch_ops->set_gdt(vcpu, &dt);
1567
1568         vcpu->cr2 = sregs->cr2;
1569         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1570         vcpu->cr3 = sregs->cr3;
1571
1572         vcpu->cr8 = sregs->cr8;
1573
1574         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1575 #ifdef CONFIG_X86_64
1576         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1577 #endif
1578         vcpu->apic_base = sregs->apic_base;
1579
1580         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1581
1582         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1583         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1584
1585         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1586         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1587         if (!is_long_mode(vcpu) && is_pae(vcpu))
1588                 load_pdptrs(vcpu, vcpu->cr3);
1589
1590         if (mmu_reset_needed)
1591                 kvm_mmu_reset_context(vcpu);
1592
1593         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1594                sizeof vcpu->irq_pending);
1595         vcpu->irq_summary = 0;
1596         for (i = 0; i < NR_IRQ_WORDS; ++i)
1597                 if (vcpu->irq_pending[i])
1598                         __set_bit(i, &vcpu->irq_summary);
1599
1600         vcpu_put(vcpu);
1601
1602         return 0;
1603 }
1604
1605 /*
1606  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1607  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1608  *
1609  * This list is modified at module load time to reflect the
1610  * capabilities of the host cpu.
1611  */
1612 static u32 msrs_to_save[] = {
1613         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1614         MSR_K6_STAR,
1615 #ifdef CONFIG_X86_64
1616         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1617 #endif
1618         MSR_IA32_TIME_STAMP_COUNTER,
1619 };
1620
1621 static unsigned num_msrs_to_save;
1622
1623 static u32 emulated_msrs[] = {
1624         MSR_IA32_MISC_ENABLE,
1625 };
1626
1627 static __init void kvm_init_msr_list(void)
1628 {
1629         u32 dummy[2];
1630         unsigned i, j;
1631
1632         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1633                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1634                         continue;
1635                 if (j < i)
1636                         msrs_to_save[j] = msrs_to_save[i];
1637                 j++;
1638         }
1639         num_msrs_to_save = j;
1640 }
1641
1642 /*
1643  * Adapt set_msr() to msr_io()'s calling convention
1644  */
1645 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1646 {
1647         return set_msr(vcpu, index, *data);
1648 }
1649
1650 /*
1651  * Read or write a bunch of msrs. All parameters are kernel addresses.
1652  *
1653  * @return number of msrs set successfully.
1654  */
1655 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1656                     struct kvm_msr_entry *entries,
1657                     int (*do_msr)(struct kvm_vcpu *vcpu,
1658                                   unsigned index, u64 *data))
1659 {
1660         struct kvm_vcpu *vcpu;
1661         int i;
1662
1663         if (!valid_vcpu(msrs->vcpu))
1664                 return -EINVAL;
1665
1666         vcpu = vcpu_load(kvm, msrs->vcpu);
1667         if (!vcpu)
1668                 return -ENOENT;
1669
1670         for (i = 0; i < msrs->nmsrs; ++i)
1671                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1672                         break;
1673
1674         vcpu_put(vcpu);
1675
1676         return i;
1677 }
1678
1679 /*
1680  * Read or write a bunch of msrs. Parameters are user addresses.
1681  *
1682  * @return number of msrs set successfully.
1683  */
1684 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1685                   int (*do_msr)(struct kvm_vcpu *vcpu,
1686                                 unsigned index, u64 *data),
1687                   int writeback)
1688 {
1689         struct kvm_msrs msrs;
1690         struct kvm_msr_entry *entries;
1691         int r, n;
1692         unsigned size;
1693
1694         r = -EFAULT;
1695         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1696                 goto out;
1697
1698         r = -E2BIG;
1699         if (msrs.nmsrs >= MAX_IO_MSRS)
1700                 goto out;
1701
1702         r = -ENOMEM;
1703         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1704         entries = vmalloc(size);
1705         if (!entries)
1706                 goto out;
1707
1708         r = -EFAULT;
1709         if (copy_from_user(entries, user_msrs->entries, size))
1710                 goto out_free;
1711
1712         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1713         if (r < 0)
1714                 goto out_free;
1715
1716         r = -EFAULT;
1717         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1718                 goto out_free;
1719
1720         r = n;
1721
1722 out_free:
1723         vfree(entries);
1724 out:
1725         return r;
1726 }
1727
1728 /*
1729  * Translate a guest virtual address to a guest physical address.
1730  */
1731 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1732 {
1733         unsigned long vaddr = tr->linear_address;
1734         struct kvm_vcpu *vcpu;
1735         gpa_t gpa;
1736
1737         vcpu = vcpu_load(kvm, tr->vcpu);
1738         if (!vcpu)
1739                 return -ENOENT;
1740         spin_lock(&kvm->lock);
1741         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1742         tr->physical_address = gpa;
1743         tr->valid = gpa != UNMAPPED_GVA;
1744         tr->writeable = 1;
1745         tr->usermode = 0;
1746         spin_unlock(&kvm->lock);
1747         vcpu_put(vcpu);
1748
1749         return 0;
1750 }
1751
1752 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1753 {
1754         struct kvm_vcpu *vcpu;
1755
1756         if (!valid_vcpu(irq->vcpu))
1757                 return -EINVAL;
1758         if (irq->irq < 0 || irq->irq >= 256)
1759                 return -EINVAL;
1760         vcpu = vcpu_load(kvm, irq->vcpu);
1761         if (!vcpu)
1762                 return -ENOENT;
1763
1764         set_bit(irq->irq, vcpu->irq_pending);
1765         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1766
1767         vcpu_put(vcpu);
1768
1769         return 0;
1770 }
1771
1772 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1773                                      struct kvm_debug_guest *dbg)
1774 {
1775         struct kvm_vcpu *vcpu;
1776         int r;
1777
1778         if (!valid_vcpu(dbg->vcpu))
1779                 return -EINVAL;
1780         vcpu = vcpu_load(kvm, dbg->vcpu);
1781         if (!vcpu)
1782                 return -ENOENT;
1783
1784         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1785
1786         vcpu_put(vcpu);
1787
1788         return r;
1789 }
1790
1791 static long kvm_dev_ioctl(struct file *filp,
1792                           unsigned int ioctl, unsigned long arg)
1793 {
1794         struct kvm *kvm = filp->private_data;
1795         void __user *argp = (void __user *)arg;
1796         int r = -EINVAL;
1797
1798         switch (ioctl) {
1799         case KVM_GET_API_VERSION:
1800                 r = KVM_API_VERSION;
1801                 break;
1802         case KVM_CREATE_VCPU: {
1803                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1804                 if (r)
1805                         goto out;
1806                 break;
1807         }
1808         case KVM_RUN: {
1809                 struct kvm_run kvm_run;
1810
1811                 r = -EFAULT;
1812                 if (copy_from_user(&kvm_run, argp, sizeof kvm_run))
1813                         goto out;
1814                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1815                 if (r < 0 &&  r != -EINTR)
1816                         goto out;
1817                 if (copy_to_user(argp, &kvm_run, sizeof kvm_run)) {
1818                         r = -EFAULT;
1819                         goto out;
1820                 }
1821                 break;
1822         }
1823         case KVM_GET_REGS: {
1824                 struct kvm_regs kvm_regs;
1825
1826                 r = -EFAULT;
1827                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1828                         goto out;
1829                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1830                 if (r)
1831                         goto out;
1832                 r = -EFAULT;
1833                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_SET_REGS: {
1839                 struct kvm_regs kvm_regs;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1843                         goto out;
1844                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1845                 if (r)
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_GET_SREGS: {
1851                 struct kvm_sregs kvm_sregs;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1855                         goto out;
1856                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1857                 if (r)
1858                         goto out;
1859                 r = -EFAULT;
1860                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_SET_SREGS: {
1866                 struct kvm_sregs kvm_sregs;
1867
1868                 r = -EFAULT;
1869                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1870                         goto out;
1871                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1872                 if (r)
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_TRANSLATE: {
1878                 struct kvm_translation tr;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&tr, argp, sizeof tr))
1882                         goto out;
1883                 r = kvm_dev_ioctl_translate(kvm, &tr);
1884                 if (r)
1885                         goto out;
1886                 r = -EFAULT;
1887                 if (copy_to_user(argp, &tr, sizeof tr))
1888                         goto out;
1889                 r = 0;
1890                 break;
1891         }
1892         case KVM_INTERRUPT: {
1893                 struct kvm_interrupt irq;
1894
1895                 r = -EFAULT;
1896                 if (copy_from_user(&irq, argp, sizeof irq))
1897                         goto out;
1898                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1899                 if (r)
1900                         goto out;
1901                 r = 0;
1902                 break;
1903         }
1904         case KVM_DEBUG_GUEST: {
1905                 struct kvm_debug_guest dbg;
1906
1907                 r = -EFAULT;
1908                 if (copy_from_user(&dbg, argp, sizeof dbg))
1909                         goto out;
1910                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1911                 if (r)
1912                         goto out;
1913                 r = 0;
1914                 break;
1915         }
1916         case KVM_SET_MEMORY_REGION: {
1917                 struct kvm_memory_region kvm_mem;
1918
1919                 r = -EFAULT;
1920                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1921                         goto out;
1922                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1923                 if (r)
1924                         goto out;
1925                 break;
1926         }
1927         case KVM_GET_DIRTY_LOG: {
1928                 struct kvm_dirty_log log;
1929
1930                 r = -EFAULT;
1931                 if (copy_from_user(&log, argp, sizeof log))
1932                         goto out;
1933                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1934                 if (r)
1935                         goto out;
1936                 break;
1937         }
1938         case KVM_GET_MSRS:
1939                 r = msr_io(kvm, argp, get_msr, 1);
1940                 break;
1941         case KVM_SET_MSRS:
1942                 r = msr_io(kvm, argp, do_set_msr, 0);
1943                 break;
1944         case KVM_GET_MSR_INDEX_LIST: {
1945                 struct kvm_msr_list __user *user_msr_list = argp;
1946                 struct kvm_msr_list msr_list;
1947                 unsigned n;
1948
1949                 r = -EFAULT;
1950                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1951                         goto out;
1952                 n = msr_list.nmsrs;
1953                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1954                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1955                         goto out;
1956                 r = -E2BIG;
1957                 if (n < num_msrs_to_save)
1958                         goto out;
1959                 r = -EFAULT;
1960                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1961                                  num_msrs_to_save * sizeof(u32)))
1962                         goto out;
1963                 if (copy_to_user(user_msr_list->indices
1964                                  + num_msrs_to_save * sizeof(u32),
1965                                  &emulated_msrs,
1966                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1967                         goto out;
1968                 r = 0;
1969                 break;
1970         }
1971         default:
1972                 ;
1973         }
1974 out:
1975         return r;
1976 }
1977
1978 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1979                                    unsigned long address,
1980                                    int *type)
1981 {
1982         struct kvm *kvm = vma->vm_file->private_data;
1983         unsigned long pgoff;
1984         struct kvm_memory_slot *slot;
1985         struct page *page;
1986
1987         *type = VM_FAULT_MINOR;
1988         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1989         slot = gfn_to_memslot(kvm, pgoff);
1990         if (!slot)
1991                 return NOPAGE_SIGBUS;
1992         page = gfn_to_page(slot, pgoff);
1993         if (!page)
1994                 return NOPAGE_SIGBUS;
1995         get_page(page);
1996         return page;
1997 }
1998
1999 static struct vm_operations_struct kvm_dev_vm_ops = {
2000         .nopage = kvm_dev_nopage,
2001 };
2002
2003 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
2004 {
2005         vma->vm_ops = &kvm_dev_vm_ops;
2006         return 0;
2007 }
2008
2009 static struct file_operations kvm_chardev_ops = {
2010         .open           = kvm_dev_open,
2011         .release        = kvm_dev_release,
2012         .unlocked_ioctl = kvm_dev_ioctl,
2013         .compat_ioctl   = kvm_dev_ioctl,
2014         .mmap           = kvm_dev_mmap,
2015 };
2016
2017 static struct miscdevice kvm_dev = {
2018         MISC_DYNAMIC_MINOR,
2019         "kvm",
2020         &kvm_chardev_ops,
2021 };
2022
2023 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2024                        void *v)
2025 {
2026         if (val == SYS_RESTART) {
2027                 /*
2028                  * Some (well, at least mine) BIOSes hang on reboot if
2029                  * in vmx root mode.
2030                  */
2031                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2032                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2033         }
2034         return NOTIFY_OK;
2035 }
2036
2037 static struct notifier_block kvm_reboot_notifier = {
2038         .notifier_call = kvm_reboot,
2039         .priority = 0,
2040 };
2041
2042 static __init void kvm_init_debug(void)
2043 {
2044         struct kvm_stats_debugfs_item *p;
2045
2046         debugfs_dir = debugfs_create_dir("kvm", NULL);
2047         for (p = debugfs_entries; p->name; ++p)
2048                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2049                                                p->data);
2050 }
2051
2052 static void kvm_exit_debug(void)
2053 {
2054         struct kvm_stats_debugfs_item *p;
2055
2056         for (p = debugfs_entries; p->name; ++p)
2057                 debugfs_remove(p->dentry);
2058         debugfs_remove(debugfs_dir);
2059 }
2060
2061 hpa_t bad_page_address;
2062
2063 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2064 {
2065         int r;
2066
2067         if (kvm_arch_ops) {
2068                 printk(KERN_ERR "kvm: already loaded the other module\n");
2069                 return -EEXIST;
2070         }
2071
2072         if (!ops->cpu_has_kvm_support()) {
2073                 printk(KERN_ERR "kvm: no hardware support\n");
2074                 return -EOPNOTSUPP;
2075         }
2076         if (ops->disabled_by_bios()) {
2077                 printk(KERN_ERR "kvm: disabled by bios\n");
2078                 return -EOPNOTSUPP;
2079         }
2080
2081         kvm_arch_ops = ops;
2082
2083         r = kvm_arch_ops->hardware_setup();
2084         if (r < 0)
2085             return r;
2086
2087         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2088         register_reboot_notifier(&kvm_reboot_notifier);
2089
2090         kvm_chardev_ops.owner = module;
2091
2092         r = misc_register(&kvm_dev);
2093         if (r) {
2094                 printk (KERN_ERR "kvm: misc device register failed\n");
2095                 goto out_free;
2096         }
2097
2098         return r;
2099
2100 out_free:
2101         unregister_reboot_notifier(&kvm_reboot_notifier);
2102         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2103         kvm_arch_ops->hardware_unsetup();
2104         return r;
2105 }
2106
2107 void kvm_exit_arch(void)
2108 {
2109         misc_deregister(&kvm_dev);
2110
2111         unregister_reboot_notifier(&kvm_reboot_notifier);
2112         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2113         kvm_arch_ops->hardware_unsetup();
2114         kvm_arch_ops = NULL;
2115 }
2116
2117 static __init int kvm_init(void)
2118 {
2119         static struct page *bad_page;
2120         int r = 0;
2121
2122         kvm_init_debug();
2123
2124         kvm_init_msr_list();
2125
2126         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2127                 r = -ENOMEM;
2128                 goto out;
2129         }
2130
2131         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2132         memset(__va(bad_page_address), 0, PAGE_SIZE);
2133
2134         return r;
2135
2136 out:
2137         kvm_exit_debug();
2138         return r;
2139 }
2140
2141 static __exit void kvm_exit(void)
2142 {
2143         kvm_exit_debug();
2144         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2145 }
2146
2147 module_init(kvm_init)
2148 module_exit(kvm_exit)
2149
2150 EXPORT_SYMBOL_GPL(kvm_init_arch);
2151 EXPORT_SYMBOL_GPL(kvm_exit_arch);