SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/kvm/kvm_main.c

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