err.no Git - 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 #include <linux/sysdev.h>
  38 #include <linux/cpu.h>
  39
  40 #include "x86_emulate.h"
  41 #include "segment_descriptor.h"
  42
  43 MODULE_AUTHOR("Qumranet");
  44 MODULE_LICENSE("GPL");
  45
  46 static DEFINE_SPINLOCK(kvm_lock);
  47 static LIST_HEAD(vm_list);
  48
  49 struct kvm_arch_ops *kvm_arch_ops;
  50 struct kvm_stat kvm_stat;
  51 EXPORT_SYMBOL_GPL(kvm_stat);
  52
  53 static struct kvm_stats_debugfs_item {
  54         const char *name;
  55         u32 *data;
  56         struct dentry *dentry;
  57 } debugfs_entries[] = {
  58         { "pf_fixed", &kvm_stat.pf_fixed },
  59         { "pf_guest", &kvm_stat.pf_guest },
  60         { "tlb_flush", &kvm_stat.tlb_flush },
  61         { "invlpg", &kvm_stat.invlpg },
  62         { "exits", &kvm_stat.exits },
  63         { "io_exits", &kvm_stat.io_exits },
  64         { "mmio_exits", &kvm_stat.mmio_exits },
  65         { "signal_exits", &kvm_stat.signal_exits },
  66         { "irq_window", &kvm_stat.irq_window_exits },
  67         { "halt_exits", &kvm_stat.halt_exits },
  68         { "request_irq", &kvm_stat.request_irq_exits },
  69         { "irq_exits", &kvm_stat.irq_exits },
  70         { NULL, NULL }
  71 };
  72
  73 static struct dentry *debugfs_dir;
  74
  75 #define MAX_IO_MSRS 256
  76
  77 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  78 #define LMSW_GUEST_MASK 0x0eULL
  79 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  80 #define CR8_RESEVED_BITS (~0x0fULL)
  81 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  82
  83 #ifdef CONFIG_X86_64
  84 // LDT or TSS descriptor in the GDT. 16 bytes.
  85 struct segment_descriptor_64 {
  86         struct segment_descriptor s;
  87         u32 base_higher;
  88         u32 pad_zero;
  89 };
  90
  91 #endif
  92
  93 unsigned long segment_base(u16 selector)
  94 {
  95         struct descriptor_table gdt;
  96         struct segment_descriptor *d;
  97         unsigned long table_base;
  98         typedef unsigned long ul;
  99         unsigned long v;
 100
 101         if (selector == 0)
 102                 return 0;
 103
 104         asm ("sgdt %0" : "=m"(gdt));
 105         table_base = gdt.base;
 106
 107         if (selector & 4) {           /* from ldt */
 108                 u16 ldt_selector;
 109
 110                 asm ("sldt %0" : "=g"(ldt_selector));
 111                 table_base = segment_base(ldt_selector);
 112         }
 113         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 114         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 115 #ifdef CONFIG_X86_64
 116         if (d->system == 0
 117             && (d->type == 2 || d->type == 9 || d->type == 11))
 118                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 119 #endif
 120         return v;
 121 }
 122 EXPORT_SYMBOL_GPL(segment_base);
 123
 124 static inline int valid_vcpu(int n)
 125 {
 126         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 127 }
 128
 129 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, 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, gva_t addr, unsigned long size,
 163                     void *data)
 164 {
 165         unsigned char *host_buf = data;
 166         unsigned long req_size = size;
 167
 168         while (size) {
 169                 hpa_t paddr;
 170                 unsigned now;
 171                 unsigned offset;
 172                 hva_t guest_buf;
 173
 174                 paddr = gva_to_hpa(vcpu, addr);
 175
 176                 if (is_error_hpa(paddr))
 177                         break;
 178
 179                 guest_buf = (hva_t)kmap_atomic(
 180                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 181                 offset = addr & ~PAGE_MASK;
 182                 guest_buf |= offset;
 183                 now = min(size, PAGE_SIZE - offset);
 184                 memcpy((void*)guest_buf, host_buf, now);
 185                 host_buf += now;
 186                 addr += now;
 187                 size -= now;
 188                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 189         }
 190         return req_size - size;
 191 }
 192 EXPORT_SYMBOL_GPL(kvm_write_guest);
 193
 194 static int vcpu_slot(struct kvm_vcpu *vcpu)
 195 {
 196         return vcpu - vcpu->kvm->vcpus;
 197 }
 198
 199 /*
 200  * Switches to specified vcpu, until a matching vcpu_put()
 201  */
 202 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
 203 {
 204         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
 205
 206         mutex_lock(&vcpu->mutex);
 207         if (unlikely(!vcpu->vmcs)) {
 208                 mutex_unlock(&vcpu->mutex);
 209                 return NULL;
 210         }
 211         return kvm_arch_ops->vcpu_load(vcpu);
 212 }
 213
 214 static void vcpu_put(struct kvm_vcpu *vcpu)
 215 {
 216         kvm_arch_ops->vcpu_put(vcpu);
 217         mutex_unlock(&vcpu->mutex);
 218 }
 219
 220 static int kvm_dev_open(struct inode *inode, struct file *filp)
 221 {
 222         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 223         int i;
 224
 225         if (!kvm)
 226                 return -ENOMEM;
 227
 228         spin_lock_init(&kvm->lock);
 229         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 230         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 231                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 232
 233                 mutex_init(&vcpu->mutex);
 234                 vcpu->cpu = -1;
 235                 vcpu->kvm = kvm;
 236                 vcpu->mmu.root_hpa = INVALID_PAGE;
 237                 INIT_LIST_HEAD(&vcpu->free_pages);
 238                 spin_lock(&kvm_lock);
 239                 list_add(&kvm->vm_list, &vm_list);
 240                 spin_unlock(&kvm_lock);
 241         }
 242         filp->private_data = kvm;
 243         return 0;
 244 }
 245
 246 /*
 247  * Free any memory in @free but not in @dont.
 248  */
 249 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 250                                   struct kvm_memory_slot *dont)
 251 {
 252         int i;
 253
 254         if (!dont || free->phys_mem != dont->phys_mem)
 255                 if (free->phys_mem) {
 256                         for (i = 0; i < free->npages; ++i)
 257                                 if (free->phys_mem[i])
 258                                         __free_page(free->phys_mem[i]);
 259                         vfree(free->phys_mem);
 260                 }
 261
 262         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 263                 vfree(free->dirty_bitmap);
 264
 265         free->phys_mem = NULL;
 266         free->npages = 0;
 267         free->dirty_bitmap = NULL;
 268 }
 269
 270 static void kvm_free_physmem(struct kvm *kvm)
 271 {
 272         int i;
 273
 274         for (i = 0; i < kvm->nmemslots; ++i)
 275                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 276 }
 277
 278 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 279 {
 280         if (!vcpu_load(vcpu->kvm, vcpu_slot(vcpu)))
 281                 return;
 282
 283         kvm_mmu_destroy(vcpu);
 284         vcpu_put(vcpu);
 285         kvm_arch_ops->vcpu_free(vcpu);
 286 }
 287
 288 static void kvm_free_vcpus(struct kvm *kvm)
 289 {
 290         unsigned int i;
 291
 292         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 293                 kvm_free_vcpu(&kvm->vcpus[i]);
 294 }
 295
 296 static int kvm_dev_release(struct inode *inode, struct file *filp)
 297 {
 298         struct kvm *kvm = filp->private_data;
 299
 300         spin_lock(&kvm_lock);
 301         list_del(&kvm->vm_list);
 302         spin_unlock(&kvm_lock);
 303         kvm_free_vcpus(kvm);
 304         kvm_free_physmem(kvm);
 305         kfree(kvm);
 306         return 0;
 307 }
 308
 309 static void inject_gp(struct kvm_vcpu *vcpu)
 310 {
 311         kvm_arch_ops->inject_gp(vcpu, 0);
 312 }
 313
 314 /*
 315  * Load the pae pdptrs.  Return true is they are all valid.
 316  */
 317 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 318 {
 319         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 320         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 321         int i;
 322         u64 pdpte;
 323         u64 *pdpt;
 324         int ret;
 325         struct kvm_memory_slot *memslot;
 326
 327         spin_lock(&vcpu->kvm->lock);
 328         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 329         /* FIXME: !memslot - emulate? 0xff? */
 330         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 331
 332         ret = 1;
 333         for (i = 0; i < 4; ++i) {
 334                 pdpte = pdpt[offset + i];
 335                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 336                         ret = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         for (i = 0; i < 4; ++i)
 342                 vcpu->pdptrs[i] = pdpt[offset + i];
 343
 344 out:
 345         kunmap_atomic(pdpt, KM_USER0);
 346         spin_unlock(&vcpu->kvm->lock);
 347
 348         return ret;
 349 }
 350
 351 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 352 {
 353         if (cr0 & CR0_RESEVED_BITS) {
 354                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 355                        cr0, vcpu->cr0);
 356                 inject_gp(vcpu);
 357                 return;
 358         }
 359
 360         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 361                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 362                 inject_gp(vcpu);
 363                 return;
 364         }
 365
 366         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 367                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 368                        "and a clear PE flag\n");
 369                 inject_gp(vcpu);
 370                 return;
 371         }
 372
 373         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 374 #ifdef CONFIG_X86_64
 375                 if ((vcpu->shadow_efer & EFER_LME)) {
 376                         int cs_db, cs_l;
 377
 378                         if (!is_pae(vcpu)) {
 379                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 380                                        "in long mode while PAE is disabled\n");
 381                                 inject_gp(vcpu);
 382                                 return;
 383                         }
 384                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 385                         if (cs_l) {
 386                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 387                                        "in long mode while CS.L == 1\n");
 388                                 inject_gp(vcpu);
 389                                 return;
 390
 391                         }
 392                 } else
 393 #endif
 394                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 395                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 396                                "reserved bits\n");
 397                         inject_gp(vcpu);
 398                         return;
 399                 }
 400
 401         }
 402
 403         kvm_arch_ops->set_cr0(vcpu, cr0);
 404         vcpu->cr0 = cr0;
 405
 406         spin_lock(&vcpu->kvm->lock);
 407         kvm_mmu_reset_context(vcpu);
 408         spin_unlock(&vcpu->kvm->lock);
 409         return;
 410 }
 411 EXPORT_SYMBOL_GPL(set_cr0);
 412
 413 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 414 {
 415         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 416         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 417 }
 418 EXPORT_SYMBOL_GPL(lmsw);
 419
 420 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 421 {
 422         if (cr4 & CR4_RESEVED_BITS) {
 423                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 424                 inject_gp(vcpu);
 425                 return;
 426         }
 427
 428         if (is_long_mode(vcpu)) {
 429                 if (!(cr4 & CR4_PAE_MASK)) {
 430                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 431                                "in long mode\n");
 432                         inject_gp(vcpu);
 433                         return;
 434                 }
 435         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 436                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 437                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 438                 inject_gp(vcpu);
 439         }
 440
 441         if (cr4 & CR4_VMXE_MASK) {
 442                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 443                 inject_gp(vcpu);
 444                 return;
 445         }
 446         kvm_arch_ops->set_cr4(vcpu, cr4);
 447         spin_lock(&vcpu->kvm->lock);
 448         kvm_mmu_reset_context(vcpu);
 449         spin_unlock(&vcpu->kvm->lock);
 450 }
 451 EXPORT_SYMBOL_GPL(set_cr4);
 452
 453 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 454 {
 455         if (is_long_mode(vcpu)) {
 456                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 457                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 458                         inject_gp(vcpu);
 459                         return;
 460                 }
 461         } else {
 462                 if (cr3 & CR3_RESEVED_BITS) {
 463                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 464                         inject_gp(vcpu);
 465                         return;
 466                 }
 467                 if (is_paging(vcpu) && is_pae(vcpu) &&
 468                     !load_pdptrs(vcpu, cr3)) {
 469                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 470                                "reserved bits\n");
 471                         inject_gp(vcpu);
 472                         return;
 473                 }
 474         }
 475
 476         vcpu->cr3 = cr3;
 477         spin_lock(&vcpu->kvm->lock);
 478         /*
 479          * Does the new cr3 value map to physical memory? (Note, we
 480          * catch an invalid cr3 even in real-mode, because it would
 481          * cause trouble later on when we turn on paging anyway.)
 482          *
 483          * A real CPU would silently accept an invalid cr3 and would
 484          * attempt to use it - with largely undefined (and often hard
 485          * to debug) behavior on the guest side.
 486          */
 487         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 488                 inject_gp(vcpu);
 489         else
 490                 vcpu->mmu.new_cr3(vcpu);
 491         spin_unlock(&vcpu->kvm->lock);
 492 }
 493 EXPORT_SYMBOL_GPL(set_cr3);
 494
 495 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 496 {
 497         if ( cr8 & CR8_RESEVED_BITS) {
 498                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 499                 inject_gp(vcpu);
 500                 return;
 501         }
 502         vcpu->cr8 = cr8;
 503 }
 504 EXPORT_SYMBOL_GPL(set_cr8);
 505
 506 void fx_init(struct kvm_vcpu *vcpu)
 507 {
 508         struct __attribute__ ((__packed__)) fx_image_s {
 509                 u16 control; //fcw
 510                 u16 status; //fsw
 511                 u16 tag; // ftw
 512                 u16 opcode; //fop
 513                 u64 ip; // fpu ip
 514                 u64 operand;// fpu dp
 515                 u32 mxcsr;
 516                 u32 mxcsr_mask;
 517
 518         } *fx_image;
 519
 520         fx_save(vcpu->host_fx_image);
 521         fpu_init();
 522         fx_save(vcpu->guest_fx_image);
 523         fx_restore(vcpu->host_fx_image);
 524
 525         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 526         fx_image->mxcsr = 0x1f80;
 527         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 528                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 529 }
 530 EXPORT_SYMBOL_GPL(fx_init);
 531
 532 /*
 533  * Creates some virtual cpus.  Good luck creating more than one.
 534  */
 535 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 536 {
 537         int r;
 538         struct kvm_vcpu *vcpu;
 539
 540         r = -EINVAL;
 541         if (!valid_vcpu(n))
 542                 goto out;
 543
 544         vcpu = &kvm->vcpus[n];
 545
 546         mutex_lock(&vcpu->mutex);
 547
 548         if (vcpu->vmcs) {
 549                 mutex_unlock(&vcpu->mutex);
 550                 return -EEXIST;
 551         }
 552
 553         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 554                                            FX_IMAGE_ALIGN);
 555         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 556
 557         r = kvm_arch_ops->vcpu_create(vcpu);
 558         if (r < 0)
 559                 goto out_free_vcpus;
 560
 561         r = kvm_mmu_create(vcpu);
 562         if (r < 0)
 563                 goto out_free_vcpus;
 564
 565         kvm_arch_ops->vcpu_load(vcpu);
 566         r = kvm_mmu_setup(vcpu);
 567         if (r >= 0)
 568                 r = kvm_arch_ops->vcpu_setup(vcpu);
 569         vcpu_put(vcpu);
 570
 571         if (r < 0)
 572                 goto out_free_vcpus;
 573
 574         return 0;
 575
 576 out_free_vcpus:
 577         kvm_free_vcpu(vcpu);
 578         mutex_unlock(&vcpu->mutex);
 579 out:
 580         return r;
 581 }
 582
 583 /*
 584  * Allocate some memory and give it an address in the guest physical address
 585  * space.
 586  *
 587  * Discontiguous memory is allowed, mostly for framebuffers.
 588  */
 589 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 590                                            struct kvm_memory_region *mem)
 591 {
 592         int r;
 593         gfn_t base_gfn;
 594         unsigned long npages;
 595         unsigned long i;
 596         struct kvm_memory_slot *memslot;
 597         struct kvm_memory_slot old, new;
 598         int memory_config_version;
 599
 600         r = -EINVAL;
 601         /* General sanity checks */
 602         if (mem->memory_size & (PAGE_SIZE - 1))
 603                 goto out;
 604         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 605                 goto out;
 606         if (mem->slot >= KVM_MEMORY_SLOTS)
 607                 goto out;
 608         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 609                 goto out;
 610
 611         memslot = &kvm->memslots[mem->slot];
 612         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 613         npages = mem->memory_size >> PAGE_SHIFT;
 614
 615         if (!npages)
 616                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 617
 618 raced:
 619         spin_lock(&kvm->lock);
 620
 621         memory_config_version = kvm->memory_config_version;
 622         new = old = *memslot;
 623
 624         new.base_gfn = base_gfn;
 625         new.npages = npages;
 626         new.flags = mem->flags;
 627
 628         /* Disallow changing a memory slot's size. */
 629         r = -EINVAL;
 630         if (npages && old.npages && npages != old.npages)
 631                 goto out_unlock;
 632
 633         /* Check for overlaps */
 634         r = -EEXIST;
 635         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 636                 struct kvm_memory_slot *s = &kvm->memslots[i];
 637
 638                 if (s == memslot)
 639                         continue;
 640                 if (!((base_gfn + npages <= s->base_gfn) ||
 641                       (base_gfn >= s->base_gfn + s->npages)))
 642                         goto out_unlock;
 643         }
 644         /*
 645          * Do memory allocations outside lock.  memory_config_version will
 646          * detect any races.
 647          */
 648         spin_unlock(&kvm->lock);
 649
 650         /* Deallocate if slot is being removed */
 651         if (!npages)
 652                 new.phys_mem = NULL;
 653
 654         /* Free page dirty bitmap if unneeded */
 655         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 656                 new.dirty_bitmap = NULL;
 657
 658         r = -ENOMEM;
 659
 660         /* Allocate if a slot is being created */
 661         if (npages && !new.phys_mem) {
 662                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 663
 664                 if (!new.phys_mem)
 665                         goto out_free;
 666
 667                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 668                 for (i = 0; i < npages; ++i) {
 669                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 670                                                      | __GFP_ZERO);
 671                         if (!new.phys_mem[i])
 672                                 goto out_free;
 673                         set_page_private(new.phys_mem[i],0);
 674                 }
 675         }
 676
 677         /* Allocate page dirty bitmap if needed */
 678         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 679                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 680
 681                 new.dirty_bitmap = vmalloc(dirty_bytes);
 682                 if (!new.dirty_bitmap)
 683                         goto out_free;
 684                 memset(new.dirty_bitmap, 0, dirty_bytes);
 685         }
 686
 687         spin_lock(&kvm->lock);
 688
 689         if (memory_config_version != kvm->memory_config_version) {
 690                 spin_unlock(&kvm->lock);
 691                 kvm_free_physmem_slot(&new, &old);
 692                 goto raced;
 693         }
 694
 695         r = -EAGAIN;
 696         if (kvm->busy)
 697                 goto out_unlock;
 698
 699         if (mem->slot >= kvm->nmemslots)
 700                 kvm->nmemslots = mem->slot + 1;
 701
 702         *memslot = new;
 703         ++kvm->memory_config_version;
 704
 705         spin_unlock(&kvm->lock);
 706
 707         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 708                 struct kvm_vcpu *vcpu;
 709
 710                 vcpu = vcpu_load(kvm, i);
 711                 if (!vcpu)
 712                         continue;
 713                 kvm_mmu_reset_context(vcpu);
 714                 vcpu_put(vcpu);
 715         }
 716
 717         kvm_free_physmem_slot(&old, &new);
 718         return 0;
 719
 720 out_unlock:
 721         spin_unlock(&kvm->lock);
 722 out_free:
 723         kvm_free_physmem_slot(&new, &old);
 724 out:
 725         return r;
 726 }
 727
 728 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 729 {
 730         spin_lock(&vcpu->kvm->lock);
 731         kvm_mmu_slot_remove_write_access(vcpu, slot);
 732         spin_unlock(&vcpu->kvm->lock);
 733 }
 734
 735 /*
 736  * Get (and clear) the dirty memory log for a memory slot.
 737  */
 738 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 739                                        struct kvm_dirty_log *log)
 740 {
 741         struct kvm_memory_slot *memslot;
 742         int r, i;
 743         int n;
 744         int cleared;
 745         unsigned long any = 0;
 746
 747         spin_lock(&kvm->lock);
 748
 749         /*
 750          * Prevent changes to guest memory configuration even while the lock
 751          * is not taken.
 752          */
 753         ++kvm->busy;
 754         spin_unlock(&kvm->lock);
 755         r = -EINVAL;
 756         if (log->slot >= KVM_MEMORY_SLOTS)
 757                 goto out;
 758
 759         memslot = &kvm->memslots[log->slot];
 760         r = -ENOENT;
 761         if (!memslot->dirty_bitmap)
 762                 goto out;
 763
 764         n = ALIGN(memslot->npages, 8) / 8;
 765
 766         for (i = 0; !any && i < n; ++i)
 767                 any = memslot->dirty_bitmap[i];
 768
 769         r = -EFAULT;
 770         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 771                 goto out;
 772
 773         if (any) {
 774                 cleared = 0;
 775                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 776                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 777
 778                         if (!vcpu)
 779                                 continue;
 780                         if (!cleared) {
 781                                 do_remove_write_access(vcpu, log->slot);
 782                                 memset(memslot->dirty_bitmap, 0, n);
 783                                 cleared = 1;
 784                         }
 785                         kvm_arch_ops->tlb_flush(vcpu);
 786                         vcpu_put(vcpu);
 787                 }
 788         }
 789
 790         r = 0;
 791
 792 out:
 793         spin_lock(&kvm->lock);
 794         --kvm->busy;
 795         spin_unlock(&kvm->lock);
 796         return r;
 797 }
 798
 799 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 800 {
 801         int i;
 802
 803         for (i = 0; i < kvm->nmemslots; ++i) {
 804                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 805
 806                 if (gfn >= memslot->base_gfn
 807                     && gfn < memslot->base_gfn + memslot->npages)
 808                         return memslot;
 809         }
 810         return NULL;
 811 }
 812 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 813
 814 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 815 {
 816         int i;
 817         struct kvm_memory_slot *memslot = NULL;
 818         unsigned long rel_gfn;
 819
 820         for (i = 0; i < kvm->nmemslots; ++i) {
 821                 memslot = &kvm->memslots[i];
 822
 823                 if (gfn >= memslot->base_gfn
 824                     && gfn < memslot->base_gfn + memslot->npages) {
 825
 826                         if (!memslot || !memslot->dirty_bitmap)
 827                                 return;
 828
 829                         rel_gfn = gfn - memslot->base_gfn;
 830
 831                         /* avoid RMW */
 832                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 833                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 834                         return;
 835                 }
 836         }
 837 }
 838
 839 static int emulator_read_std(unsigned long addr,
 840                              unsigned long *val,
 841                              unsigned int bytes,
 842                              struct x86_emulate_ctxt *ctxt)
 843 {
 844         struct kvm_vcpu *vcpu = ctxt->vcpu;
 845         void *data = val;
 846
 847         while (bytes) {
 848                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 849                 unsigned offset = addr & (PAGE_SIZE-1);
 850                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 851                 unsigned long pfn;
 852                 struct kvm_memory_slot *memslot;
 853                 void *page;
 854
 855                 if (gpa == UNMAPPED_GVA)
 856                         return X86EMUL_PROPAGATE_FAULT;
 857                 pfn = gpa >> PAGE_SHIFT;
 858                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 859                 if (!memslot)
 860                         return X86EMUL_UNHANDLEABLE;
 861                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 862
 863                 memcpy(data, page + offset, tocopy);
 864
 865                 kunmap_atomic(page, KM_USER0);
 866
 867                 bytes -= tocopy;
 868                 data += tocopy;
 869                 addr += tocopy;
 870         }
 871
 872         return X86EMUL_CONTINUE;
 873 }
 874
 875 static int emulator_write_std(unsigned long addr,
 876                               unsigned long val,
 877                               unsigned int bytes,
 878                               struct x86_emulate_ctxt *ctxt)
 879 {
 880         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 881                addr, bytes);
 882         return X86EMUL_UNHANDLEABLE;
 883 }
 884
 885 static int emulator_read_emulated(unsigned long addr,
 886                                   unsigned long *val,
 887                                   unsigned int bytes,
 888                                   struct x86_emulate_ctxt *ctxt)
 889 {
 890         struct kvm_vcpu *vcpu = ctxt->vcpu;
 891
 892         if (vcpu->mmio_read_completed) {
 893                 memcpy(val, vcpu->mmio_data, bytes);
 894                 vcpu->mmio_read_completed = 0;
 895                 return X86EMUL_CONTINUE;
 896         } else if (emulator_read_std(addr, val, bytes, ctxt)
 897                    == X86EMUL_CONTINUE)
 898                 return X86EMUL_CONTINUE;
 899         else {
 900                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 901
 902                 if (gpa == UNMAPPED_GVA)
 903                         return X86EMUL_PROPAGATE_FAULT;
 904                 vcpu->mmio_needed = 1;
 905                 vcpu->mmio_phys_addr = gpa;
 906                 vcpu->mmio_size = bytes;
 907                 vcpu->mmio_is_write = 0;
 908
 909                 return X86EMUL_UNHANDLEABLE;
 910         }
 911 }
 912
 913 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 914                                unsigned long val, int bytes)
 915 {
 916         struct kvm_memory_slot *m;
 917         struct page *page;
 918         void *virt;
 919
 920         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 921                 return 0;
 922         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 923         if (!m)
 924                 return 0;
 925         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 926         kvm_mmu_pre_write(vcpu, gpa, bytes);
 927         virt = kmap_atomic(page, KM_USER0);
 928         memcpy(virt + offset_in_page(gpa), &val, bytes);
 929         kunmap_atomic(virt, KM_USER0);
 930         kvm_mmu_post_write(vcpu, gpa, bytes);
 931         return 1;
 932 }
 933
 934 static int emulator_write_emulated(unsigned long addr,
 935                                    unsigned long val,
 936                                    unsigned int bytes,
 937                                    struct x86_emulate_ctxt *ctxt)
 938 {
 939         struct kvm_vcpu *vcpu = ctxt->vcpu;
 940         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 941
 942         if (gpa == UNMAPPED_GVA)
 943                 return X86EMUL_PROPAGATE_FAULT;
 944
 945         if (emulator_write_phys(vcpu, gpa, val, bytes))
 946                 return X86EMUL_CONTINUE;
 947
 948         vcpu->mmio_needed = 1;
 949         vcpu->mmio_phys_addr = gpa;
 950         vcpu->mmio_size = bytes;
 951         vcpu->mmio_is_write = 1;
 952         memcpy(vcpu->mmio_data, &val, bytes);
 953
 954         return X86EMUL_CONTINUE;
 955 }
 956
 957 static int emulator_cmpxchg_emulated(unsigned long addr,
 958                                      unsigned long old,
 959                                      unsigned long new,
 960                                      unsigned int bytes,
 961                                      struct x86_emulate_ctxt *ctxt)
 962 {
 963         static int reported;
 964
 965         if (!reported) {
 966                 reported = 1;
 967                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 968         }
 969         return emulator_write_emulated(addr, new, bytes, ctxt);
 970 }
 971
 972 #ifdef CONFIG_X86_32
 973
 974 static int emulator_cmpxchg8b_emulated(unsigned long addr,
 975                                        unsigned long old_lo,
 976                                        unsigned long old_hi,
 977                                        unsigned long new_lo,
 978                                        unsigned long new_hi,
 979                                        struct x86_emulate_ctxt *ctxt)
 980 {
 981         static int reported;
 982         int r;
 983
 984         if (!reported) {
 985                 reported = 1;
 986                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
 987         }
 988         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
 989         if (r != X86EMUL_CONTINUE)
 990                 return r;
 991         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
 992 }
 993
 994 #endif
 995
 996 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 997 {
 998         return kvm_arch_ops->get_segment_base(vcpu, seg);
 999 }
1000
1001 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1002 {
1003         return X86EMUL_CONTINUE;
1004 }
1005
1006 int emulate_clts(struct kvm_vcpu *vcpu)
1007 {
1008         unsigned long cr0;
1009
1010         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1011         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1012         kvm_arch_ops->set_cr0(vcpu, cr0);
1013         return X86EMUL_CONTINUE;
1014 }
1015
1016 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1017 {
1018         struct kvm_vcpu *vcpu = ctxt->vcpu;
1019
1020         switch (dr) {
1021         case 0 ... 3:
1022                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1023                 return X86EMUL_CONTINUE;
1024         default:
1025                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1026                        __FUNCTION__, dr);
1027                 return X86EMUL_UNHANDLEABLE;
1028         }
1029 }
1030
1031 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1032 {
1033         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1034         int exception;
1035
1036         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1037         if (exception) {
1038                 /* FIXME: better handling */
1039                 return X86EMUL_UNHANDLEABLE;
1040         }
1041         return X86EMUL_CONTINUE;
1042 }
1043
1044 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1045 {
1046         static int reported;
1047         u8 opcodes[4];
1048         unsigned long rip = ctxt->vcpu->rip;
1049         unsigned long rip_linear;
1050
1051         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1052
1053         if (reported)
1054                 return;
1055
1056         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1057
1058         printk(KERN_ERR "emulation failed but !mmio_needed?"
1059                " rip %lx %02x %02x %02x %02x\n",
1060                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1061         reported = 1;
1062 }
1063
1064 struct x86_emulate_ops emulate_ops = {
1065         .read_std            = emulator_read_std,
1066         .write_std           = emulator_write_std,
1067         .read_emulated       = emulator_read_emulated,
1068         .write_emulated      = emulator_write_emulated,
1069         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1070 #ifdef CONFIG_X86_32
1071         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1072 #endif
1073 };
1074
1075 int emulate_instruction(struct kvm_vcpu *vcpu,
1076                         struct kvm_run *run,
1077                         unsigned long cr2,
1078                         u16 error_code)
1079 {
1080         struct x86_emulate_ctxt emulate_ctxt;
1081         int r;
1082         int cs_db, cs_l;
1083
1084         kvm_arch_ops->cache_regs(vcpu);
1085
1086         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1087
1088         emulate_ctxt.vcpu = vcpu;
1089         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1090         emulate_ctxt.cr2 = cr2;
1091         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1092                 ? X86EMUL_MODE_REAL : cs_l
1093                 ? X86EMUL_MODE_PROT64 : cs_db
1094                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1095
1096         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1097                 emulate_ctxt.cs_base = 0;
1098                 emulate_ctxt.ds_base = 0;
1099                 emulate_ctxt.es_base = 0;
1100                 emulate_ctxt.ss_base = 0;
1101         } else {
1102                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1103                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1104                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1105                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1106         }
1107
1108         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1109         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1110
1111         vcpu->mmio_is_write = 0;
1112         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1113
1114         if ((r || vcpu->mmio_is_write) && run) {
1115                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1116                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1117                 run->mmio.len = vcpu->mmio_size;
1118                 run->mmio.is_write = vcpu->mmio_is_write;
1119         }
1120
1121         if (r) {
1122                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1123                         return EMULATE_DONE;
1124                 if (!vcpu->mmio_needed) {
1125                         report_emulation_failure(&emulate_ctxt);
1126                         return EMULATE_FAIL;
1127                 }
1128                 return EMULATE_DO_MMIO;
1129         }
1130
1131         kvm_arch_ops->decache_regs(vcpu);
1132         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1133
1134         if (vcpu->mmio_is_write)
1135                 return EMULATE_DO_MMIO;
1136
1137         return EMULATE_DONE;
1138 }
1139 EXPORT_SYMBOL_GPL(emulate_instruction);
1140
1141 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1142 {
1143         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1144 }
1145
1146 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1147 {
1148         struct descriptor_table dt = { limit, base };
1149
1150         kvm_arch_ops->set_gdt(vcpu, &dt);
1151 }
1152
1153 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1154 {
1155         struct descriptor_table dt = { limit, base };
1156
1157         kvm_arch_ops->set_idt(vcpu, &dt);
1158 }
1159
1160 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1161                    unsigned long *rflags)
1162 {
1163         lmsw(vcpu, msw);
1164         *rflags = kvm_arch_ops->get_rflags(vcpu);
1165 }
1166
1167 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1168 {
1169         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1170         switch (cr) {
1171         case 0:
1172                 return vcpu->cr0;
1173         case 2:
1174                 return vcpu->cr2;
1175         case 3:
1176                 return vcpu->cr3;
1177         case 4:
1178                 return vcpu->cr4;
1179         default:
1180                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1181                 return 0;
1182         }
1183 }
1184
1185 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1186                      unsigned long *rflags)
1187 {
1188         switch (cr) {
1189         case 0:
1190                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1191                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1192                 break;
1193         case 2:
1194                 vcpu->cr2 = val;
1195                 break;
1196         case 3:
1197                 set_cr3(vcpu, val);
1198                 break;
1199         case 4:
1200                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1201                 break;
1202         default:
1203                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1204         }
1205 }
1206
1207 /*
1208  * Register the para guest with the host:
1209  */
1210 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1211 {
1212         struct kvm_vcpu_para_state *para_state;
1213         hpa_t para_state_hpa, hypercall_hpa;
1214         struct page *para_state_page;
1215         unsigned char *hypercall;
1216         gpa_t hypercall_gpa;
1217
1218         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1219         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1220
1221         /*
1222          * Needs to be page aligned:
1223          */
1224         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1225                 goto err_gp;
1226
1227         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1228         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1229         if (is_error_hpa(para_state_hpa))
1230                 goto err_gp;
1231
1232         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1233         para_state = kmap_atomic(para_state_page, KM_USER0);
1234
1235         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1236         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1237
1238         para_state->host_version = KVM_PARA_API_VERSION;
1239         /*
1240          * We cannot support guests that try to register themselves
1241          * with a newer API version than the host supports:
1242          */
1243         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1244                 para_state->ret = -KVM_EINVAL;
1245                 goto err_kunmap_skip;
1246         }
1247
1248         hypercall_gpa = para_state->hypercall_gpa;
1249         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1250         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1251         if (is_error_hpa(hypercall_hpa)) {
1252                 para_state->ret = -KVM_EINVAL;
1253                 goto err_kunmap_skip;
1254         }
1255
1256         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1257         vcpu->para_state_page = para_state_page;
1258         vcpu->para_state_gpa = para_state_gpa;
1259         vcpu->hypercall_gpa = hypercall_gpa;
1260
1261         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1262                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1263         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1264         kunmap_atomic(hypercall, KM_USER1);
1265
1266         para_state->ret = 0;
1267 err_kunmap_skip:
1268         kunmap_atomic(para_state, KM_USER0);
1269         return 0;
1270 err_gp:
1271         return 1;
1272 }
1273
1274 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1275 {
1276         u64 data;
1277
1278         switch (msr) {
1279         case 0xc0010010: /* SYSCFG */
1280         case 0xc0010015: /* HWCR */
1281         case MSR_IA32_PLATFORM_ID:
1282         case MSR_IA32_P5_MC_ADDR:
1283         case MSR_IA32_P5_MC_TYPE:
1284         case MSR_IA32_MC0_CTL:
1285         case MSR_IA32_MCG_STATUS:
1286         case MSR_IA32_MCG_CAP:
1287         case MSR_IA32_MC0_MISC:
1288         case MSR_IA32_MC0_MISC+4:
1289         case MSR_IA32_MC0_MISC+8:
1290         case MSR_IA32_MC0_MISC+12:
1291         case MSR_IA32_MC0_MISC+16:
1292         case MSR_IA32_UCODE_REV:
1293         case MSR_IA32_PERF_STATUS:
1294                 /* MTRR registers */
1295         case 0xfe:
1296         case 0x200 ... 0x2ff:
1297                 data = 0;
1298                 break;
1299         case 0xcd: /* fsb frequency */
1300                 data = 3;
1301                 break;
1302         case MSR_IA32_APICBASE:
1303                 data = vcpu->apic_base;
1304                 break;
1305         case MSR_IA32_MISC_ENABLE:
1306                 data = vcpu->ia32_misc_enable_msr;
1307                 break;
1308 #ifdef CONFIG_X86_64
1309         case MSR_EFER:
1310                 data = vcpu->shadow_efer;
1311                 break;
1312 #endif
1313         default:
1314                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1315                 return 1;
1316         }
1317         *pdata = data;
1318         return 0;
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1321
1322 /*
1323  * Reads an msr value (of 'msr_index') into 'pdata'.
1324  * Returns 0 on success, non-0 otherwise.
1325  * Assumes vcpu_load() was already called.
1326  */
1327 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1328 {
1329         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1330 }
1331
1332 #ifdef CONFIG_X86_64
1333
1334 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1335 {
1336         if (efer & EFER_RESERVED_BITS) {
1337                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1338                        efer);
1339                 inject_gp(vcpu);
1340                 return;
1341         }
1342
1343         if (is_paging(vcpu)
1344             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1345                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1346                 inject_gp(vcpu);
1347                 return;
1348         }
1349
1350         kvm_arch_ops->set_efer(vcpu, efer);
1351
1352         efer &= ~EFER_LMA;
1353         efer |= vcpu->shadow_efer & EFER_LMA;
1354
1355         vcpu->shadow_efer = efer;
1356 }
1357
1358 #endif
1359
1360 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1361 {
1362         switch (msr) {
1363 #ifdef CONFIG_X86_64
1364         case MSR_EFER:
1365                 set_efer(vcpu, data);
1366                 break;
1367 #endif
1368         case MSR_IA32_MC0_STATUS:
1369                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1370                        __FUNCTION__, data);
1371                 break;
1372         case MSR_IA32_UCODE_REV:
1373         case MSR_IA32_UCODE_WRITE:
1374         case 0x200 ... 0x2ff: /* MTRRs */
1375                 break;
1376         case MSR_IA32_APICBASE:
1377                 vcpu->apic_base = data;
1378                 break;
1379         case MSR_IA32_MISC_ENABLE:
1380                 vcpu->ia32_misc_enable_msr = data;
1381                 break;
1382         /*
1383          * This is the 'probe whether the host is KVM' logic:
1384          */
1385         case MSR_KVM_API_MAGIC:
1386                 return vcpu_register_para(vcpu, data);
1387
1388         default:
1389                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1390                 return 1;
1391         }
1392         return 0;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1395
1396 /*
1397  * Writes msr value into into the appropriate "register".
1398  * Returns 0 on success, non-0 otherwise.
1399  * Assumes vcpu_load() was already called.
1400  */
1401 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1402 {
1403         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1404 }
1405
1406 void kvm_resched(struct kvm_vcpu *vcpu)
1407 {
1408         vcpu_put(vcpu);
1409         cond_resched();
1410         /* Cannot fail -  no vcpu unplug yet. */
1411         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_resched);
1414
1415 void load_msrs(struct vmx_msr_entry *e, int n)
1416 {
1417         int i;
1418
1419         for (i = 0; i < n; ++i)
1420                 wrmsrl(e[i].index, e[i].data);
1421 }
1422 EXPORT_SYMBOL_GPL(load_msrs);
1423
1424 void save_msrs(struct vmx_msr_entry *e, int n)
1425 {
1426         int i;
1427
1428         for (i = 0; i < n; ++i)
1429                 rdmsrl(e[i].index, e[i].data);
1430 }
1431 EXPORT_SYMBOL_GPL(save_msrs);
1432
1433 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1434 {
1435         struct kvm_vcpu *vcpu;
1436         int r;
1437
1438         if (!valid_vcpu(kvm_run->vcpu))
1439                 return -EINVAL;
1440
1441         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1442         if (!vcpu)
1443                 return -ENOENT;
1444
1445         /* re-sync apic's tpr */
1446         vcpu->cr8 = kvm_run->cr8;
1447
1448         if (kvm_run->emulated) {
1449                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1450                 kvm_run->emulated = 0;
1451         }
1452
1453         if (kvm_run->mmio_completed) {
1454                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1455                 vcpu->mmio_read_completed = 1;
1456         }
1457
1458         vcpu->mmio_needed = 0;
1459
1460         r = kvm_arch_ops->run(vcpu, kvm_run);
1461
1462         vcpu_put(vcpu);
1463         return r;
1464 }
1465
1466 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1467 {
1468         struct kvm_vcpu *vcpu;
1469
1470         if (!valid_vcpu(regs->vcpu))
1471                 return -EINVAL;
1472
1473         vcpu = vcpu_load(kvm, regs->vcpu);
1474         if (!vcpu)
1475                 return -ENOENT;
1476
1477         kvm_arch_ops->cache_regs(vcpu);
1478
1479         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1480         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1481         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1482         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1483         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1484         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1485         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1486         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1487 #ifdef CONFIG_X86_64
1488         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1489         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1490         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1491         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1492         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1493         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1494         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1495         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1496 #endif
1497
1498         regs->rip = vcpu->rip;
1499         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1500
1501         /*
1502          * Don't leak debug flags in case they were set for guest debugging
1503          */
1504         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1505                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1506
1507         vcpu_put(vcpu);
1508
1509         return 0;
1510 }
1511
1512 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1513 {
1514         struct kvm_vcpu *vcpu;
1515
1516         if (!valid_vcpu(regs->vcpu))
1517                 return -EINVAL;
1518
1519         vcpu = vcpu_load(kvm, regs->vcpu);
1520         if (!vcpu)
1521                 return -ENOENT;
1522
1523         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1524         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1525         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1526         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1527         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1528         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1529         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1530         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1531 #ifdef CONFIG_X86_64
1532         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1533         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1534         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1535         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1536         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1537         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1538         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1539         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1540 #endif
1541
1542         vcpu->rip = regs->rip;
1543         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1544
1545         kvm_arch_ops->decache_regs(vcpu);
1546
1547         vcpu_put(vcpu);
1548
1549         return 0;
1550 }
1551
1552 static void get_segment(struct kvm_vcpu *vcpu,
1553                         struct kvm_segment *var, int seg)
1554 {
1555         return kvm_arch_ops->get_segment(vcpu, var, seg);
1556 }
1557
1558 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1559 {
1560         struct kvm_vcpu *vcpu;
1561         struct descriptor_table dt;
1562
1563         if (!valid_vcpu(sregs->vcpu))
1564                 return -EINVAL;
1565         vcpu = vcpu_load(kvm, sregs->vcpu);
1566         if (!vcpu)
1567                 return -ENOENT;
1568
1569         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1570         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1571         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1572         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1573         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1574         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1575
1576         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1577         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1578
1579         kvm_arch_ops->get_idt(vcpu, &dt);
1580         sregs->idt.limit = dt.limit;
1581         sregs->idt.base = dt.base;
1582         kvm_arch_ops->get_gdt(vcpu, &dt);
1583         sregs->gdt.limit = dt.limit;
1584         sregs->gdt.base = dt.base;
1585
1586         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1587         sregs->cr0 = vcpu->cr0;
1588         sregs->cr2 = vcpu->cr2;
1589         sregs->cr3 = vcpu->cr3;
1590         sregs->cr4 = vcpu->cr4;
1591         sregs->cr8 = vcpu->cr8;
1592         sregs->efer = vcpu->shadow_efer;
1593         sregs->apic_base = vcpu->apic_base;
1594
1595         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1596                sizeof sregs->interrupt_bitmap);
1597
1598         vcpu_put(vcpu);
1599
1600         return 0;
1601 }
1602
1603 static void set_segment(struct kvm_vcpu *vcpu,
1604                         struct kvm_segment *var, int seg)
1605 {
1606         return kvm_arch_ops->set_segment(vcpu, var, seg);
1607 }
1608
1609 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1610 {
1611         struct kvm_vcpu *vcpu;
1612         int mmu_reset_needed = 0;
1613         int i;
1614         struct descriptor_table dt;
1615
1616         if (!valid_vcpu(sregs->vcpu))
1617                 return -EINVAL;
1618         vcpu = vcpu_load(kvm, sregs->vcpu);
1619         if (!vcpu)
1620                 return -ENOENT;
1621
1622         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1623         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1624         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1625         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1626         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1627         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1628
1629         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1630         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1631
1632         dt.limit = sregs->idt.limit;
1633         dt.base = sregs->idt.base;
1634         kvm_arch_ops->set_idt(vcpu, &dt);
1635         dt.limit = sregs->gdt.limit;
1636         dt.base = sregs->gdt.base;
1637         kvm_arch_ops->set_gdt(vcpu, &dt);
1638
1639         vcpu->cr2 = sregs->cr2;
1640         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1641         vcpu->cr3 = sregs->cr3;
1642
1643         vcpu->cr8 = sregs->cr8;
1644
1645         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1646 #ifdef CONFIG_X86_64
1647         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1648 #endif
1649         vcpu->apic_base = sregs->apic_base;
1650
1651         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1652
1653         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1654         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1655
1656         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1657         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1658         if (!is_long_mode(vcpu) && is_pae(vcpu))
1659                 load_pdptrs(vcpu, vcpu->cr3);
1660
1661         if (mmu_reset_needed)
1662                 kvm_mmu_reset_context(vcpu);
1663
1664         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1665                sizeof vcpu->irq_pending);
1666         vcpu->irq_summary = 0;
1667         for (i = 0; i < NR_IRQ_WORDS; ++i)
1668                 if (vcpu->irq_pending[i])
1669                         __set_bit(i, &vcpu->irq_summary);
1670
1671         vcpu_put(vcpu);
1672
1673         return 0;
1674 }
1675
1676 /*
1677  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1678  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1679  *
1680  * This list is modified at module load time to reflect the
1681  * capabilities of the host cpu.
1682  */
1683 static u32 msrs_to_save[] = {
1684         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1685         MSR_K6_STAR,
1686 #ifdef CONFIG_X86_64
1687         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1688 #endif
1689         MSR_IA32_TIME_STAMP_COUNTER,
1690 };
1691
1692 static unsigned num_msrs_to_save;
1693
1694 static u32 emulated_msrs[] = {
1695         MSR_IA32_MISC_ENABLE,
1696 };
1697
1698 static __init void kvm_init_msr_list(void)
1699 {
1700         u32 dummy[2];
1701         unsigned i, j;
1702
1703         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1704                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1705                         continue;
1706                 if (j < i)
1707                         msrs_to_save[j] = msrs_to_save[i];
1708                 j++;
1709         }
1710         num_msrs_to_save = j;
1711 }
1712
1713 /*
1714  * Adapt set_msr() to msr_io()'s calling convention
1715  */
1716 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1717 {
1718         return set_msr(vcpu, index, *data);
1719 }
1720
1721 /*
1722  * Read or write a bunch of msrs. All parameters are kernel addresses.
1723  *
1724  * @return number of msrs set successfully.
1725  */
1726 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1727                     struct kvm_msr_entry *entries,
1728                     int (*do_msr)(struct kvm_vcpu *vcpu,
1729                                   unsigned index, u64 *data))
1730 {
1731         struct kvm_vcpu *vcpu;
1732         int i;
1733
1734         if (!valid_vcpu(msrs->vcpu))
1735                 return -EINVAL;
1736
1737         vcpu = vcpu_load(kvm, msrs->vcpu);
1738         if (!vcpu)
1739                 return -ENOENT;
1740
1741         for (i = 0; i < msrs->nmsrs; ++i)
1742                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1743                         break;
1744
1745         vcpu_put(vcpu);
1746
1747         return i;
1748 }
1749
1750 /*
1751  * Read or write a bunch of msrs. Parameters are user addresses.
1752  *
1753  * @return number of msrs set successfully.
1754  */
1755 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1756                   int (*do_msr)(struct kvm_vcpu *vcpu,
1757                                 unsigned index, u64 *data),
1758                   int writeback)
1759 {
1760         struct kvm_msrs msrs;
1761         struct kvm_msr_entry *entries;
1762         int r, n;
1763         unsigned size;
1764
1765         r = -EFAULT;
1766         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1767                 goto out;
1768
1769         r = -E2BIG;
1770         if (msrs.nmsrs >= MAX_IO_MSRS)
1771                 goto out;
1772
1773         r = -ENOMEM;
1774         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1775         entries = vmalloc(size);
1776         if (!entries)
1777                 goto out;
1778
1779         r = -EFAULT;
1780         if (copy_from_user(entries, user_msrs->entries, size))
1781                 goto out_free;
1782
1783         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1784         if (r < 0)
1785                 goto out_free;
1786
1787         r = -EFAULT;
1788         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1789                 goto out_free;
1790
1791         r = n;
1792
1793 out_free:
1794         vfree(entries);
1795 out:
1796         return r;
1797 }
1798
1799 /*
1800  * Translate a guest virtual address to a guest physical address.
1801  */
1802 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1803 {
1804         unsigned long vaddr = tr->linear_address;
1805         struct kvm_vcpu *vcpu;
1806         gpa_t gpa;
1807
1808         vcpu = vcpu_load(kvm, tr->vcpu);
1809         if (!vcpu)
1810                 return -ENOENT;
1811         spin_lock(&kvm->lock);
1812         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1813         tr->physical_address = gpa;
1814         tr->valid = gpa != UNMAPPED_GVA;
1815         tr->writeable = 1;
1816         tr->usermode = 0;
1817         spin_unlock(&kvm->lock);
1818         vcpu_put(vcpu);
1819
1820         return 0;
1821 }
1822
1823 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1824 {
1825         struct kvm_vcpu *vcpu;
1826
1827         if (!valid_vcpu(irq->vcpu))
1828                 return -EINVAL;
1829         if (irq->irq < 0 || irq->irq >= 256)
1830                 return -EINVAL;
1831         vcpu = vcpu_load(kvm, irq->vcpu);
1832         if (!vcpu)
1833                 return -ENOENT;
1834
1835         set_bit(irq->irq, vcpu->irq_pending);
1836         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1837
1838         vcpu_put(vcpu);
1839
1840         return 0;
1841 }
1842
1843 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1844                                      struct kvm_debug_guest *dbg)
1845 {
1846         struct kvm_vcpu *vcpu;
1847         int r;
1848
1849         if (!valid_vcpu(dbg->vcpu))
1850                 return -EINVAL;
1851         vcpu = vcpu_load(kvm, dbg->vcpu);
1852         if (!vcpu)
1853                 return -ENOENT;
1854
1855         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1856
1857         vcpu_put(vcpu);
1858
1859         return r;
1860 }
1861
1862 static long kvm_dev_ioctl(struct file *filp,
1863                           unsigned int ioctl, unsigned long arg)
1864 {
1865         struct kvm *kvm = filp->private_data;
1866         void __user *argp = (void __user *)arg;
1867         int r = -EINVAL;
1868
1869         switch (ioctl) {
1870         case KVM_GET_API_VERSION:
1871                 r = KVM_API_VERSION;
1872                 break;
1873         case KVM_CREATE_VCPU:
1874                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1875                 if (r)
1876                         goto out;
1877                 break;
1878         case KVM_RUN: {
1879                 struct kvm_run kvm_run;
1880
1881                 r = -EFAULT;
1882                 if (copy_from_user(&kvm_run, argp, sizeof kvm_run))
1883                         goto out;
1884                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1885                 if (r < 0 &&  r != -EINTR)
1886                         goto out;
1887                 if (copy_to_user(argp, &kvm_run, sizeof kvm_run)) {
1888                         r = -EFAULT;
1889                         goto out;
1890                 }
1891                 break;
1892         }
1893         case KVM_GET_REGS: {
1894                 struct kvm_regs kvm_regs;
1895
1896                 r = -EFAULT;
1897                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1898                         goto out;
1899                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1900                 if (r)
1901                         goto out;
1902                 r = -EFAULT;
1903                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1904                         goto out;
1905                 r = 0;
1906                 break;
1907         }
1908         case KVM_SET_REGS: {
1909                 struct kvm_regs kvm_regs;
1910
1911                 r = -EFAULT;
1912                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1913                         goto out;
1914                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1915                 if (r)
1916                         goto out;
1917                 r = 0;
1918                 break;
1919         }
1920         case KVM_GET_SREGS: {
1921                 struct kvm_sregs kvm_sregs;
1922
1923                 r = -EFAULT;
1924                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1925                         goto out;
1926                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1927                 if (r)
1928                         goto out;
1929                 r = -EFAULT;
1930                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1931                         goto out;
1932                 r = 0;
1933                 break;
1934         }
1935         case KVM_SET_SREGS: {
1936                 struct kvm_sregs kvm_sregs;
1937
1938                 r = -EFAULT;
1939                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1940                         goto out;
1941                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1942                 if (r)
1943                         goto out;
1944                 r = 0;
1945                 break;
1946         }
1947         case KVM_TRANSLATE: {
1948                 struct kvm_translation tr;
1949
1950                 r = -EFAULT;
1951                 if (copy_from_user(&tr, argp, sizeof tr))
1952                         goto out;
1953                 r = kvm_dev_ioctl_translate(kvm, &tr);
1954                 if (r)
1955                         goto out;
1956                 r = -EFAULT;
1957                 if (copy_to_user(argp, &tr, sizeof tr))
1958                         goto out;
1959                 r = 0;
1960                 break;
1961         }
1962         case KVM_INTERRUPT: {
1963                 struct kvm_interrupt irq;
1964
1965                 r = -EFAULT;
1966                 if (copy_from_user(&irq, argp, sizeof irq))
1967                         goto out;
1968                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1969                 if (r)
1970                         goto out;
1971                 r = 0;
1972                 break;
1973         }
1974         case KVM_DEBUG_GUEST: {
1975                 struct kvm_debug_guest dbg;
1976
1977                 r = -EFAULT;
1978                 if (copy_from_user(&dbg, argp, sizeof dbg))
1979                         goto out;
1980                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1981                 if (r)
1982                         goto out;
1983                 r = 0;
1984                 break;
1985         }
1986         case KVM_SET_MEMORY_REGION: {
1987                 struct kvm_memory_region kvm_mem;
1988
1989                 r = -EFAULT;
1990                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1991                         goto out;
1992                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1993                 if (r)
1994                         goto out;
1995                 break;
1996         }
1997         case KVM_GET_DIRTY_LOG: {
1998                 struct kvm_dirty_log log;
1999
2000                 r = -EFAULT;
2001                 if (copy_from_user(&log, argp, sizeof log))
2002                         goto out;
2003                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
2004                 if (r)
2005                         goto out;
2006                 break;
2007         }
2008         case KVM_GET_MSRS:
2009                 r = msr_io(kvm, argp, get_msr, 1);
2010                 break;
2011         case KVM_SET_MSRS:
2012                 r = msr_io(kvm, argp, do_set_msr, 0);
2013                 break;
2014         case KVM_GET_MSR_INDEX_LIST: {
2015                 struct kvm_msr_list __user *user_msr_list = argp;
2016                 struct kvm_msr_list msr_list;
2017                 unsigned n;
2018
2019                 r = -EFAULT;
2020                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2021                         goto out;
2022                 n = msr_list.nmsrs;
2023                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2024                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2025                         goto out;
2026                 r = -E2BIG;
2027                 if (n < num_msrs_to_save)
2028                         goto out;
2029                 r = -EFAULT;
2030                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2031                                  num_msrs_to_save * sizeof(u32)))
2032                         goto out;
2033                 if (copy_to_user(user_msr_list->indices
2034                                  + num_msrs_to_save * sizeof(u32),
2035                                  &emulated_msrs,
2036                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2037                         goto out;
2038                 r = 0;
2039                 break;
2040         }
2041         default:
2042                 ;
2043         }
2044 out:
2045         return r;
2046 }
2047
2048 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
2049                                    unsigned long address,
2050                                    int *type)
2051 {
2052         struct kvm *kvm = vma->vm_file->private_data;
2053         unsigned long pgoff;
2054         struct kvm_memory_slot *slot;
2055         struct page *page;
2056
2057         *type = VM_FAULT_MINOR;
2058         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2059         slot = gfn_to_memslot(kvm, pgoff);
2060         if (!slot)
2061                 return NOPAGE_SIGBUS;
2062         page = gfn_to_page(slot, pgoff);
2063         if (!page)
2064                 return NOPAGE_SIGBUS;
2065         get_page(page);
2066         return page;
2067 }
2068
2069 static struct vm_operations_struct kvm_dev_vm_ops = {
2070         .nopage = kvm_dev_nopage,
2071 };
2072
2073 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
2074 {
2075         vma->vm_ops = &kvm_dev_vm_ops;
2076         return 0;
2077 }
2078
2079 static struct file_operations kvm_chardev_ops = {
2080         .open           = kvm_dev_open,
2081         .release        = kvm_dev_release,
2082         .unlocked_ioctl = kvm_dev_ioctl,
2083         .compat_ioctl   = kvm_dev_ioctl,
2084         .mmap           = kvm_dev_mmap,
2085 };
2086
2087 static struct miscdevice kvm_dev = {
2088         MISC_DYNAMIC_MINOR,
2089         "kvm",
2090         &kvm_chardev_ops,
2091 };
2092
2093 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2094                        void *v)
2095 {
2096         if (val == SYS_RESTART) {
2097                 /*
2098                  * Some (well, at least mine) BIOSes hang on reboot if
2099                  * in vmx root mode.
2100                  */
2101                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2102                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2103         }
2104         return NOTIFY_OK;
2105 }
2106
2107 static struct notifier_block kvm_reboot_notifier = {
2108         .notifier_call = kvm_reboot,
2109         .priority = 0,
2110 };
2111
2112 /*
2113  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2114  * cached on it.
2115  */
2116 static void decache_vcpus_on_cpu(int cpu)
2117 {
2118         struct kvm *vm;
2119         struct kvm_vcpu *vcpu;
2120         int i;
2121
2122         spin_lock(&kvm_lock);
2123         list_for_each_entry(vm, &vm_list, vm_list)
2124                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2125                         vcpu = &vm->vcpus[i];
2126                         /*
2127                          * If the vcpu is locked, then it is running on some
2128                          * other cpu and therefore it is not cached on the
2129                          * cpu in question.
2130                          *
2131                          * If it's not locked, check the last cpu it executed
2132                          * on.
2133                          */
2134                         if (mutex_trylock(&vcpu->mutex)) {
2135                                 if (vcpu->cpu == cpu) {
2136                                         kvm_arch_ops->vcpu_decache(vcpu);
2137                                         vcpu->cpu = -1;
2138                                 }
2139                                 mutex_unlock(&vcpu->mutex);
2140                         }
2141                 }
2142         spin_unlock(&kvm_lock);
2143 }
2144
2145 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2146                            void *v)
2147 {
2148         int cpu = (long)v;
2149
2150         switch (val) {
2151         case CPU_DOWN_PREPARE:
2152         case CPU_UP_CANCELED:
2153                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2154                        cpu);
2155                 decache_vcpus_on_cpu(cpu);
2156                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2157                                          NULL, 0, 1);
2158                 break;
2159         case CPU_ONLINE:
2160                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2161                        cpu);
2162                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2163                                          NULL, 0, 1);
2164                 break;
2165         }
2166         return NOTIFY_OK;
2167 }
2168
2169 static struct notifier_block kvm_cpu_notifier = {
2170         .notifier_call = kvm_cpu_hotplug,
2171         .priority = 20, /* must be > scheduler priority */
2172 };
2173
2174 static __init void kvm_init_debug(void)
2175 {
2176         struct kvm_stats_debugfs_item *p;
2177
2178         debugfs_dir = debugfs_create_dir("kvm", NULL);
2179         for (p = debugfs_entries; p->name; ++p)
2180                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2181                                                p->data);
2182 }
2183
2184 static void kvm_exit_debug(void)
2185 {
2186         struct kvm_stats_debugfs_item *p;
2187
2188         for (p = debugfs_entries; p->name; ++p)
2189                 debugfs_remove(p->dentry);
2190         debugfs_remove(debugfs_dir);
2191 }
2192
2193 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2194 {
2195         decache_vcpus_on_cpu(raw_smp_processor_id());
2196         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2197         return 0;
2198 }
2199
2200 static int kvm_resume(struct sys_device *dev)
2201 {
2202         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2203         return 0;
2204 }
2205
2206 static struct sysdev_class kvm_sysdev_class = {
2207         set_kset_name("kvm"),
2208         .suspend = kvm_suspend,
2209         .resume = kvm_resume,
2210 };
2211
2212 static struct sys_device kvm_sysdev = {
2213         .id = 0,
2214         .cls = &kvm_sysdev_class,
2215 };
2216
2217 hpa_t bad_page_address;
2218
2219 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2220 {
2221         int r;
2222
2223         if (kvm_arch_ops) {
2224                 printk(KERN_ERR "kvm: already loaded the other module\n");
2225                 return -EEXIST;
2226         }
2227
2228         if (!ops->cpu_has_kvm_support()) {
2229                 printk(KERN_ERR "kvm: no hardware support\n");
2230                 return -EOPNOTSUPP;
2231         }
2232         if (ops->disabled_by_bios()) {
2233                 printk(KERN_ERR "kvm: disabled by bios\n");
2234                 return -EOPNOTSUPP;
2235         }
2236
2237         kvm_arch_ops = ops;
2238
2239         r = kvm_arch_ops->hardware_setup();
2240         if (r < 0)
2241             return r;
2242
2243         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2244         r = register_cpu_notifier(&kvm_cpu_notifier);
2245         if (r)
2246                 goto out_free_1;
2247         register_reboot_notifier(&kvm_reboot_notifier);
2248
2249         r = sysdev_class_register(&kvm_sysdev_class);
2250         if (r)
2251                 goto out_free_2;
2252
2253         r = sysdev_register(&kvm_sysdev);
2254         if (r)
2255                 goto out_free_3;
2256
2257         kvm_chardev_ops.owner = module;
2258
2259         r = misc_register(&kvm_dev);
2260         if (r) {
2261                 printk (KERN_ERR "kvm: misc device register failed\n");
2262                 goto out_free;
2263         }
2264
2265         return r;
2266
2267 out_free:
2268         sysdev_unregister(&kvm_sysdev);
2269 out_free_3:
2270         sysdev_class_unregister(&kvm_sysdev_class);
2271 out_free_2:
2272         unregister_reboot_notifier(&kvm_reboot_notifier);
2273         unregister_cpu_notifier(&kvm_cpu_notifier);
2274 out_free_1:
2275         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2276         kvm_arch_ops->hardware_unsetup();
2277         return r;
2278 }
2279
2280 void kvm_exit_arch(void)
2281 {
2282         misc_deregister(&kvm_dev);
2283         sysdev_unregister(&kvm_sysdev);
2284         sysdev_class_unregister(&kvm_sysdev_class);
2285         unregister_reboot_notifier(&kvm_reboot_notifier);
2286         unregister_cpu_notifier(&kvm_cpu_notifier);
2287         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2288         kvm_arch_ops->hardware_unsetup();
2289         kvm_arch_ops = NULL;
2290 }
2291
2292 static __init int kvm_init(void)
2293 {
2294         static struct page *bad_page;
2295         int r = 0;
2296
2297         kvm_init_debug();
2298
2299         kvm_init_msr_list();
2300
2301         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2302                 r = -ENOMEM;
2303                 goto out;
2304         }
2305
2306         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2307         memset(__va(bad_page_address), 0, PAGE_SIZE);
2308
2309         return r;
2310
2311 out:
2312         kvm_exit_debug();
2313         return r;
2314 }
2315
2316 static __exit void kvm_exit(void)
2317 {
2318         kvm_exit_debug();
2319         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2320 }
2321
2322 module_init(kvm_init)
2323 module_exit(kvm_exit)
2324
2325 EXPORT_SYMBOL_GPL(kvm_init_arch);
2326 EXPORT_SYMBOL_GPL(kvm_exit_arch);