git.oblomov.eu 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 <linux/magic.h>
  24 #include <asm/processor.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <asm/msr.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/reboot.h>
  33 #include <asm/io.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/highmem.h>
  36 #include <linux/file.h>
  37 #include <asm/desc.h>
  38 #include <linux/sysdev.h>
  39 #include <linux/cpu.h>
  40 #include <linux/file.h>
  41 #include <linux/fs.h>
  42 #include <linux/mount.h>
  43
  44 #include "x86_emulate.h"
  45 #include "segment_descriptor.h"
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 struct kvm_arch_ops *kvm_arch_ops;
  54 struct kvm_stat kvm_stat;
  55 EXPORT_SYMBOL_GPL(kvm_stat);
  56
  57 static struct kvm_stats_debugfs_item {
  58         const char *name;
  59         u32 *data;
  60         struct dentry *dentry;
  61 } debugfs_entries[] = {
  62         { "pf_fixed", &kvm_stat.pf_fixed },
  63         { "pf_guest", &kvm_stat.pf_guest },
  64         { "tlb_flush", &kvm_stat.tlb_flush },
  65         { "invlpg", &kvm_stat.invlpg },
  66         { "exits", &kvm_stat.exits },
  67         { "io_exits", &kvm_stat.io_exits },
  68         { "mmio_exits", &kvm_stat.mmio_exits },
  69         { "signal_exits", &kvm_stat.signal_exits },
  70         { "irq_window", &kvm_stat.irq_window_exits },
  71         { "halt_exits", &kvm_stat.halt_exits },
  72         { "request_irq", &kvm_stat.request_irq_exits },
  73         { "irq_exits", &kvm_stat.irq_exits },
  74         { NULL, NULL }
  75 };
  76
  77 static struct dentry *debugfs_dir;
  78
  79 struct vfsmount *kvmfs_mnt;
  80
  81 #define MAX_IO_MSRS 256
  82
  83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  84 #define LMSW_GUEST_MASK 0x0eULL
  85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  86 #define CR8_RESEVED_BITS (~0x0fULL)
  87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  88
  89 #ifdef CONFIG_X86_64
  90 // LDT or TSS descriptor in the GDT. 16 bytes.
  91 struct segment_descriptor_64 {
  92         struct segment_descriptor s;
  93         u32 base_higher;
  94         u32 pad_zero;
  95 };
  96
  97 #endif
  98
  99 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 100                            unsigned long arg);
 101
 102 static struct inode *kvmfs_inode(struct file_operations *fops)
 103 {
 104         int error = -ENOMEM;
 105         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 106
 107         if (!inode)
 108                 goto eexit_1;
 109
 110         inode->i_fop = fops;
 111
 112         /*
 113          * Mark the inode dirty from the very beginning,
 114          * that way it will never be moved to the dirty
 115          * list because mark_inode_dirty() will think
 116          * that it already _is_ on the dirty list.
 117          */
 118         inode->i_state = I_DIRTY;
 119         inode->i_mode = S_IRUSR | S_IWUSR;
 120         inode->i_uid = current->fsuid;
 121         inode->i_gid = current->fsgid;
 122         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 123         return inode;
 124
 125 eexit_1:
 126         return ERR_PTR(error);
 127 }
 128
 129 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 130 {
 131         struct file *file = get_empty_filp();
 132
 133         if (!file)
 134                 return ERR_PTR(-ENFILE);
 135
 136         file->f_path.mnt = mntget(kvmfs_mnt);
 137         file->f_path.dentry = d_alloc_anon(inode);
 138         if (!file->f_path.dentry)
 139                 return ERR_PTR(-ENOMEM);
 140         file->f_mapping = inode->i_mapping;
 141
 142         file->f_pos = 0;
 143         file->f_flags = O_RDWR;
 144         file->f_op = inode->i_fop;
 145         file->f_mode = FMODE_READ | FMODE_WRITE;
 146         file->f_version = 0;
 147         file->private_data = private_data;
 148         return file;
 149 }
 150
 151 unsigned long segment_base(u16 selector)
 152 {
 153         struct descriptor_table gdt;
 154         struct segment_descriptor *d;
 155         unsigned long table_base;
 156         typedef unsigned long ul;
 157         unsigned long v;
 158
 159         if (selector == 0)
 160                 return 0;
 161
 162         asm ("sgdt %0" : "=m"(gdt));
 163         table_base = gdt.base;
 164
 165         if (selector & 4) {           /* from ldt */
 166                 u16 ldt_selector;
 167
 168                 asm ("sldt %0" : "=g"(ldt_selector));
 169                 table_base = segment_base(ldt_selector);
 170         }
 171         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 172         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 173 #ifdef CONFIG_X86_64
 174         if (d->system == 0
 175             && (d->type == 2 || d->type == 9 || d->type == 11))
 176                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 177 #endif
 178         return v;
 179 }
 180 EXPORT_SYMBOL_GPL(segment_base);
 181
 182 static inline int valid_vcpu(int n)
 183 {
 184         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 185 }
 186
 187 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 188                    void *dest)
 189 {
 190         unsigned char *host_buf = dest;
 191         unsigned long req_size = size;
 192
 193         while (size) {
 194                 hpa_t paddr;
 195                 unsigned now;
 196                 unsigned offset;
 197                 hva_t guest_buf;
 198
 199                 paddr = gva_to_hpa(vcpu, addr);
 200
 201                 if (is_error_hpa(paddr))
 202                         break;
 203
 204                 guest_buf = (hva_t)kmap_atomic(
 205                                         pfn_to_page(paddr >> PAGE_SHIFT),
 206                                         KM_USER0);
 207                 offset = addr & ~PAGE_MASK;
 208                 guest_buf |= offset;
 209                 now = min(size, PAGE_SIZE - offset);
 210                 memcpy(host_buf, (void*)guest_buf, now);
 211                 host_buf += now;
 212                 addr += now;
 213                 size -= now;
 214                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 215         }
 216         return req_size - size;
 217 }
 218 EXPORT_SYMBOL_GPL(kvm_read_guest);
 219
 220 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 221                     void *data)
 222 {
 223         unsigned char *host_buf = data;
 224         unsigned long req_size = size;
 225
 226         while (size) {
 227                 hpa_t paddr;
 228                 unsigned now;
 229                 unsigned offset;
 230                 hva_t guest_buf;
 231                 gfn_t gfn;
 232
 233                 paddr = gva_to_hpa(vcpu, addr);
 234
 235                 if (is_error_hpa(paddr))
 236                         break;
 237
 238                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 239                 mark_page_dirty(vcpu->kvm, gfn);
 240                 guest_buf = (hva_t)kmap_atomic(
 241                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 242                 offset = addr & ~PAGE_MASK;
 243                 guest_buf |= offset;
 244                 now = min(size, PAGE_SIZE - offset);
 245                 memcpy((void*)guest_buf, host_buf, now);
 246                 host_buf += now;
 247                 addr += now;
 248                 size -= now;
 249                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 250         }
 251         return req_size - size;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_write_guest);
 254
 255 /*
 256  * Switches to specified vcpu, until a matching vcpu_put()
 257  */
 258 static void vcpu_load(struct kvm_vcpu *vcpu)
 259 {
 260         mutex_lock(&vcpu->mutex);
 261         kvm_arch_ops->vcpu_load(vcpu);
 262 }
 263
 264 /*
 265  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 266  * if the slot is not populated.
 267  */
 268 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 269 {
 270         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 271
 272         mutex_lock(&vcpu->mutex);
 273         if (!vcpu->vmcs) {
 274                 mutex_unlock(&vcpu->mutex);
 275                 return NULL;
 276         }
 277         kvm_arch_ops->vcpu_load(vcpu);
 278         return vcpu;
 279 }
 280
 281 static void vcpu_put(struct kvm_vcpu *vcpu)
 282 {
 283         kvm_arch_ops->vcpu_put(vcpu);
 284         mutex_unlock(&vcpu->mutex);
 285 }
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290         int i;
 291
 292         if (!kvm)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         spin_lock_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 298                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 299
 300                 mutex_init(&vcpu->mutex);
 301                 vcpu->cpu = -1;
 302                 vcpu->kvm = kvm;
 303                 vcpu->mmu.root_hpa = INVALID_PAGE;
 304                 INIT_LIST_HEAD(&vcpu->free_pages);
 305                 spin_lock(&kvm_lock);
 306                 list_add(&kvm->vm_list, &vm_list);
 307                 spin_unlock(&kvm_lock);
 308         }
 309         return kvm;
 310 }
 311
 312 static int kvm_dev_open(struct inode *inode, struct file *filp)
 313 {
 314         return 0;
 315 }
 316
 317 /*
 318  * Free any memory in @free but not in @dont.
 319  */
 320 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 321                                   struct kvm_memory_slot *dont)
 322 {
 323         int i;
 324
 325         if (!dont || free->phys_mem != dont->phys_mem)
 326                 if (free->phys_mem) {
 327                         for (i = 0; i < free->npages; ++i)
 328                                 if (free->phys_mem[i])
 329                                         __free_page(free->phys_mem[i]);
 330                         vfree(free->phys_mem);
 331                 }
 332
 333         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 334                 vfree(free->dirty_bitmap);
 335
 336         free->phys_mem = NULL;
 337         free->npages = 0;
 338         free->dirty_bitmap = NULL;
 339 }
 340
 341 static void kvm_free_physmem(struct kvm *kvm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < kvm->nmemslots; ++i)
 346                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 347 }
 348
 349 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 350 {
 351         if (!vcpu->vmcs)
 352                 return;
 353
 354         vcpu_load(vcpu);
 355         kvm_mmu_destroy(vcpu);
 356         vcpu_put(vcpu);
 357         kvm_arch_ops->vcpu_free(vcpu);
 358         free_page((unsigned long)vcpu->run);
 359         vcpu->run = NULL;
 360 }
 361
 362 static void kvm_free_vcpus(struct kvm *kvm)
 363 {
 364         unsigned int i;
 365
 366         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 367                 kvm_free_vcpu(&kvm->vcpus[i]);
 368 }
 369
 370 static int kvm_dev_release(struct inode *inode, struct file *filp)
 371 {
 372         return 0;
 373 }
 374
 375 static void kvm_destroy_vm(struct kvm *kvm)
 376 {
 377         spin_lock(&kvm_lock);
 378         list_del(&kvm->vm_list);
 379         spin_unlock(&kvm_lock);
 380         kvm_free_vcpus(kvm);
 381         kvm_free_physmem(kvm);
 382         kfree(kvm);
 383 }
 384
 385 static int kvm_vm_release(struct inode *inode, struct file *filp)
 386 {
 387         struct kvm *kvm = filp->private_data;
 388
 389         kvm_destroy_vm(kvm);
 390         return 0;
 391 }
 392
 393 static void inject_gp(struct kvm_vcpu *vcpu)
 394 {
 395         kvm_arch_ops->inject_gp(vcpu, 0);
 396 }
 397
 398 /*
 399  * Load the pae pdptrs.  Return true is they are all valid.
 400  */
 401 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 402 {
 403         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 404         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 405         int i;
 406         u64 pdpte;
 407         u64 *pdpt;
 408         int ret;
 409         struct kvm_memory_slot *memslot;
 410
 411         spin_lock(&vcpu->kvm->lock);
 412         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 413         /* FIXME: !memslot - emulate? 0xff? */
 414         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 415
 416         ret = 1;
 417         for (i = 0; i < 4; ++i) {
 418                 pdpte = pdpt[offset + i];
 419                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 420                         ret = 0;
 421                         goto out;
 422                 }
 423         }
 424
 425         for (i = 0; i < 4; ++i)
 426                 vcpu->pdptrs[i] = pdpt[offset + i];
 427
 428 out:
 429         kunmap_atomic(pdpt, KM_USER0);
 430         spin_unlock(&vcpu->kvm->lock);
 431
 432         return ret;
 433 }
 434
 435 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 436 {
 437         if (cr0 & CR0_RESEVED_BITS) {
 438                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 439                        cr0, vcpu->cr0);
 440                 inject_gp(vcpu);
 441                 return;
 442         }
 443
 444         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 445                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 446                 inject_gp(vcpu);
 447                 return;
 448         }
 449
 450         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 451                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 452                        "and a clear PE flag\n");
 453                 inject_gp(vcpu);
 454                 return;
 455         }
 456
 457         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 458 #ifdef CONFIG_X86_64
 459                 if ((vcpu->shadow_efer & EFER_LME)) {
 460                         int cs_db, cs_l;
 461
 462                         if (!is_pae(vcpu)) {
 463                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 464                                        "in long mode while PAE is disabled\n");
 465                                 inject_gp(vcpu);
 466                                 return;
 467                         }
 468                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 469                         if (cs_l) {
 470                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 471                                        "in long mode while CS.L == 1\n");
 472                                 inject_gp(vcpu);
 473                                 return;
 474
 475                         }
 476                 } else
 477 #endif
 478                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 479                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 480                                "reserved bits\n");
 481                         inject_gp(vcpu);
 482                         return;
 483                 }
 484
 485         }
 486
 487         kvm_arch_ops->set_cr0(vcpu, cr0);
 488         vcpu->cr0 = cr0;
 489
 490         spin_lock(&vcpu->kvm->lock);
 491         kvm_mmu_reset_context(vcpu);
 492         spin_unlock(&vcpu->kvm->lock);
 493         return;
 494 }
 495 EXPORT_SYMBOL_GPL(set_cr0);
 496
 497 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 498 {
 499         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 500         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 501 }
 502 EXPORT_SYMBOL_GPL(lmsw);
 503
 504 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 505 {
 506         if (cr4 & CR4_RESEVED_BITS) {
 507                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 508                 inject_gp(vcpu);
 509                 return;
 510         }
 511
 512         if (is_long_mode(vcpu)) {
 513                 if (!(cr4 & CR4_PAE_MASK)) {
 514                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 515                                "in long mode\n");
 516                         inject_gp(vcpu);
 517                         return;
 518                 }
 519         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 520                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 522                 inject_gp(vcpu);
 523         }
 524
 525         if (cr4 & CR4_VMXE_MASK) {
 526                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 527                 inject_gp(vcpu);
 528                 return;
 529         }
 530         kvm_arch_ops->set_cr4(vcpu, cr4);
 531         spin_lock(&vcpu->kvm->lock);
 532         kvm_mmu_reset_context(vcpu);
 533         spin_unlock(&vcpu->kvm->lock);
 534 }
 535 EXPORT_SYMBOL_GPL(set_cr4);
 536
 537 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 538 {
 539         if (is_long_mode(vcpu)) {
 540                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 541                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 542                         inject_gp(vcpu);
 543                         return;
 544                 }
 545         } else {
 546                 if (cr3 & CR3_RESEVED_BITS) {
 547                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 548                         inject_gp(vcpu);
 549                         return;
 550                 }
 551                 if (is_paging(vcpu) && is_pae(vcpu) &&
 552                     !load_pdptrs(vcpu, cr3)) {
 553                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 554                                "reserved bits\n");
 555                         inject_gp(vcpu);
 556                         return;
 557                 }
 558         }
 559
 560         vcpu->cr3 = cr3;
 561         spin_lock(&vcpu->kvm->lock);
 562         /*
 563          * Does the new cr3 value map to physical memory? (Note, we
 564          * catch an invalid cr3 even in real-mode, because it would
 565          * cause trouble later on when we turn on paging anyway.)
 566          *
 567          * A real CPU would silently accept an invalid cr3 and would
 568          * attempt to use it - with largely undefined (and often hard
 569          * to debug) behavior on the guest side.
 570          */
 571         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 572                 inject_gp(vcpu);
 573         else
 574                 vcpu->mmu.new_cr3(vcpu);
 575         spin_unlock(&vcpu->kvm->lock);
 576 }
 577 EXPORT_SYMBOL_GPL(set_cr3);
 578
 579 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 580 {
 581         if ( cr8 & CR8_RESEVED_BITS) {
 582                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 583                 inject_gp(vcpu);
 584                 return;
 585         }
 586         vcpu->cr8 = cr8;
 587 }
 588 EXPORT_SYMBOL_GPL(set_cr8);
 589
 590 void fx_init(struct kvm_vcpu *vcpu)
 591 {
 592         struct __attribute__ ((__packed__)) fx_image_s {
 593                 u16 control; //fcw
 594                 u16 status; //fsw
 595                 u16 tag; // ftw
 596                 u16 opcode; //fop
 597                 u64 ip; // fpu ip
 598                 u64 operand;// fpu dp
 599                 u32 mxcsr;
 600                 u32 mxcsr_mask;
 601
 602         } *fx_image;
 603
 604         fx_save(vcpu->host_fx_image);
 605         fpu_init();
 606         fx_save(vcpu->guest_fx_image);
 607         fx_restore(vcpu->host_fx_image);
 608
 609         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 610         fx_image->mxcsr = 0x1f80;
 611         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 612                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 613 }
 614 EXPORT_SYMBOL_GPL(fx_init);
 615
 616 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 617 {
 618         spin_lock(&vcpu->kvm->lock);
 619         kvm_mmu_slot_remove_write_access(vcpu, slot);
 620         spin_unlock(&vcpu->kvm->lock);
 621 }
 622
 623 /*
 624  * Allocate some memory and give it an address in the guest physical address
 625  * space.
 626  *
 627  * Discontiguous memory is allowed, mostly for framebuffers.
 628  */
 629 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 630                                           struct kvm_memory_region *mem)
 631 {
 632         int r;
 633         gfn_t base_gfn;
 634         unsigned long npages;
 635         unsigned long i;
 636         struct kvm_memory_slot *memslot;
 637         struct kvm_memory_slot old, new;
 638         int memory_config_version;
 639
 640         r = -EINVAL;
 641         /* General sanity checks */
 642         if (mem->memory_size & (PAGE_SIZE - 1))
 643                 goto out;
 644         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 645                 goto out;
 646         if (mem->slot >= KVM_MEMORY_SLOTS)
 647                 goto out;
 648         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 649                 goto out;
 650
 651         memslot = &kvm->memslots[mem->slot];
 652         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 653         npages = mem->memory_size >> PAGE_SHIFT;
 654
 655         if (!npages)
 656                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 657
 658 raced:
 659         spin_lock(&kvm->lock);
 660
 661         memory_config_version = kvm->memory_config_version;
 662         new = old = *memslot;
 663
 664         new.base_gfn = base_gfn;
 665         new.npages = npages;
 666         new.flags = mem->flags;
 667
 668         /* Disallow changing a memory slot's size. */
 669         r = -EINVAL;
 670         if (npages && old.npages && npages != old.npages)
 671                 goto out_unlock;
 672
 673         /* Check for overlaps */
 674         r = -EEXIST;
 675         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 676                 struct kvm_memory_slot *s = &kvm->memslots[i];
 677
 678                 if (s == memslot)
 679                         continue;
 680                 if (!((base_gfn + npages <= s->base_gfn) ||
 681                       (base_gfn >= s->base_gfn + s->npages)))
 682                         goto out_unlock;
 683         }
 684         /*
 685          * Do memory allocations outside lock.  memory_config_version will
 686          * detect any races.
 687          */
 688         spin_unlock(&kvm->lock);
 689
 690         /* Deallocate if slot is being removed */
 691         if (!npages)
 692                 new.phys_mem = NULL;
 693
 694         /* Free page dirty bitmap if unneeded */
 695         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 696                 new.dirty_bitmap = NULL;
 697
 698         r = -ENOMEM;
 699
 700         /* Allocate if a slot is being created */
 701         if (npages && !new.phys_mem) {
 702                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 703
 704                 if (!new.phys_mem)
 705                         goto out_free;
 706
 707                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 708                 for (i = 0; i < npages; ++i) {
 709                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 710                                                      | __GFP_ZERO);
 711                         if (!new.phys_mem[i])
 712                                 goto out_free;
 713                         set_page_private(new.phys_mem[i],0);
 714                 }
 715         }
 716
 717         /* Allocate page dirty bitmap if needed */
 718         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 719                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 720
 721                 new.dirty_bitmap = vmalloc(dirty_bytes);
 722                 if (!new.dirty_bitmap)
 723                         goto out_free;
 724                 memset(new.dirty_bitmap, 0, dirty_bytes);
 725         }
 726
 727         spin_lock(&kvm->lock);
 728
 729         if (memory_config_version != kvm->memory_config_version) {
 730                 spin_unlock(&kvm->lock);
 731                 kvm_free_physmem_slot(&new, &old);
 732                 goto raced;
 733         }
 734
 735         r = -EAGAIN;
 736         if (kvm->busy)
 737                 goto out_unlock;
 738
 739         if (mem->slot >= kvm->nmemslots)
 740                 kvm->nmemslots = mem->slot + 1;
 741
 742         *memslot = new;
 743         ++kvm->memory_config_version;
 744
 745         spin_unlock(&kvm->lock);
 746
 747         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 748                 struct kvm_vcpu *vcpu;
 749
 750                 vcpu = vcpu_load_slot(kvm, i);
 751                 if (!vcpu)
 752                         continue;
 753                 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
 754                         do_remove_write_access(vcpu, mem->slot);
 755                 kvm_mmu_reset_context(vcpu);
 756                 vcpu_put(vcpu);
 757         }
 758
 759         kvm_free_physmem_slot(&old, &new);
 760         return 0;
 761
 762 out_unlock:
 763         spin_unlock(&kvm->lock);
 764 out_free:
 765         kvm_free_physmem_slot(&new, &old);
 766 out:
 767         return r;
 768 }
 769
 770 /*
 771  * Get (and clear) the dirty memory log for a memory slot.
 772  */
 773 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 774                                       struct kvm_dirty_log *log)
 775 {
 776         struct kvm_memory_slot *memslot;
 777         int r, i;
 778         int n;
 779         int cleared;
 780         unsigned long any = 0;
 781
 782         spin_lock(&kvm->lock);
 783
 784         /*
 785          * Prevent changes to guest memory configuration even while the lock
 786          * is not taken.
 787          */
 788         ++kvm->busy;
 789         spin_unlock(&kvm->lock);
 790         r = -EINVAL;
 791         if (log->slot >= KVM_MEMORY_SLOTS)
 792                 goto out;
 793
 794         memslot = &kvm->memslots[log->slot];
 795         r = -ENOENT;
 796         if (!memslot->dirty_bitmap)
 797                 goto out;
 798
 799         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 800
 801         for (i = 0; !any && i < n/sizeof(long); ++i)
 802                 any = memslot->dirty_bitmap[i];
 803
 804         r = -EFAULT;
 805         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 806                 goto out;
 807
 808         if (any) {
 809                 cleared = 0;
 810                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 811                         struct kvm_vcpu *vcpu;
 812
 813                         vcpu = vcpu_load_slot(kvm, i);
 814                         if (!vcpu)
 815                                 continue;
 816                         if (!cleared) {
 817                                 do_remove_write_access(vcpu, log->slot);
 818                                 memset(memslot->dirty_bitmap, 0, n);
 819                                 cleared = 1;
 820                         }
 821                         kvm_arch_ops->tlb_flush(vcpu);
 822                         vcpu_put(vcpu);
 823                 }
 824         }
 825
 826         r = 0;
 827
 828 out:
 829         spin_lock(&kvm->lock);
 830         --kvm->busy;
 831         spin_unlock(&kvm->lock);
 832         return r;
 833 }
 834
 835 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 836 {
 837         int i;
 838
 839         for (i = 0; i < kvm->nmemslots; ++i) {
 840                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 841
 842                 if (gfn >= memslot->base_gfn
 843                     && gfn < memslot->base_gfn + memslot->npages)
 844                         return memslot;
 845         }
 846         return NULL;
 847 }
 848 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 849
 850 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 851 {
 852         int i;
 853         struct kvm_memory_slot *memslot = NULL;
 854         unsigned long rel_gfn;
 855
 856         for (i = 0; i < kvm->nmemslots; ++i) {
 857                 memslot = &kvm->memslots[i];
 858
 859                 if (gfn >= memslot->base_gfn
 860                     && gfn < memslot->base_gfn + memslot->npages) {
 861
 862                         if (!memslot || !memslot->dirty_bitmap)
 863                                 return;
 864
 865                         rel_gfn = gfn - memslot->base_gfn;
 866
 867                         /* avoid RMW */
 868                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 869                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 870                         return;
 871                 }
 872         }
 873 }
 874
 875 static int emulator_read_std(unsigned long addr,
 876                              unsigned long *val,
 877                              unsigned int bytes,
 878                              struct x86_emulate_ctxt *ctxt)
 879 {
 880         struct kvm_vcpu *vcpu = ctxt->vcpu;
 881         void *data = val;
 882
 883         while (bytes) {
 884                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 885                 unsigned offset = addr & (PAGE_SIZE-1);
 886                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 887                 unsigned long pfn;
 888                 struct kvm_memory_slot *memslot;
 889                 void *page;
 890
 891                 if (gpa == UNMAPPED_GVA)
 892                         return X86EMUL_PROPAGATE_FAULT;
 893                 pfn = gpa >> PAGE_SHIFT;
 894                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 895                 if (!memslot)
 896                         return X86EMUL_UNHANDLEABLE;
 897                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 898
 899                 memcpy(data, page + offset, tocopy);
 900
 901                 kunmap_atomic(page, KM_USER0);
 902
 903                 bytes -= tocopy;
 904                 data += tocopy;
 905                 addr += tocopy;
 906         }
 907
 908         return X86EMUL_CONTINUE;
 909 }
 910
 911 static int emulator_write_std(unsigned long addr,
 912                               unsigned long val,
 913                               unsigned int bytes,
 914                               struct x86_emulate_ctxt *ctxt)
 915 {
 916         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 917                addr, bytes);
 918         return X86EMUL_UNHANDLEABLE;
 919 }
 920
 921 static int emulator_read_emulated(unsigned long addr,
 922                                   unsigned long *val,
 923                                   unsigned int bytes,
 924                                   struct x86_emulate_ctxt *ctxt)
 925 {
 926         struct kvm_vcpu *vcpu = ctxt->vcpu;
 927
 928         if (vcpu->mmio_read_completed) {
 929                 memcpy(val, vcpu->mmio_data, bytes);
 930                 vcpu->mmio_read_completed = 0;
 931                 return X86EMUL_CONTINUE;
 932         } else if (emulator_read_std(addr, val, bytes, ctxt)
 933                    == X86EMUL_CONTINUE)
 934                 return X86EMUL_CONTINUE;
 935         else {
 936                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 937
 938                 if (gpa == UNMAPPED_GVA)
 939                         return X86EMUL_PROPAGATE_FAULT;
 940                 vcpu->mmio_needed = 1;
 941                 vcpu->mmio_phys_addr = gpa;
 942                 vcpu->mmio_size = bytes;
 943                 vcpu->mmio_is_write = 0;
 944
 945                 return X86EMUL_UNHANDLEABLE;
 946         }
 947 }
 948
 949 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 950                                unsigned long val, int bytes)
 951 {
 952         struct kvm_memory_slot *m;
 953         struct page *page;
 954         void *virt;
 955
 956         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 957                 return 0;
 958         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 959         if (!m)
 960                 return 0;
 961         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 962         kvm_mmu_pre_write(vcpu, gpa, bytes);
 963         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
 964         virt = kmap_atomic(page, KM_USER0);
 965         memcpy(virt + offset_in_page(gpa), &val, bytes);
 966         kunmap_atomic(virt, KM_USER0);
 967         kvm_mmu_post_write(vcpu, gpa, bytes);
 968         return 1;
 969 }
 970
 971 static int emulator_write_emulated(unsigned long addr,
 972                                    unsigned long val,
 973                                    unsigned int bytes,
 974                                    struct x86_emulate_ctxt *ctxt)
 975 {
 976         struct kvm_vcpu *vcpu = ctxt->vcpu;
 977         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 978
 979         if (gpa == UNMAPPED_GVA)
 980                 return X86EMUL_PROPAGATE_FAULT;
 981
 982         if (emulator_write_phys(vcpu, gpa, val, bytes))
 983                 return X86EMUL_CONTINUE;
 984
 985         vcpu->mmio_needed = 1;
 986         vcpu->mmio_phys_addr = gpa;
 987         vcpu->mmio_size = bytes;
 988         vcpu->mmio_is_write = 1;
 989         memcpy(vcpu->mmio_data, &val, bytes);
 990
 991         return X86EMUL_CONTINUE;
 992 }
 993
 994 static int emulator_cmpxchg_emulated(unsigned long addr,
 995                                      unsigned long old,
 996                                      unsigned long new,
 997                                      unsigned int bytes,
 998                                      struct x86_emulate_ctxt *ctxt)
 999 {
1000         static int reported;
1001
1002         if (!reported) {
1003                 reported = 1;
1004                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1005         }
1006         return emulator_write_emulated(addr, new, bytes, ctxt);
1007 }
1008
1009 #ifdef CONFIG_X86_32
1010
1011 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1012                                        unsigned long old_lo,
1013                                        unsigned long old_hi,
1014                                        unsigned long new_lo,
1015                                        unsigned long new_hi,
1016                                        struct x86_emulate_ctxt *ctxt)
1017 {
1018         static int reported;
1019         int r;
1020
1021         if (!reported) {
1022                 reported = 1;
1023                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1024         }
1025         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1026         if (r != X86EMUL_CONTINUE)
1027                 return r;
1028         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1029 }
1030
1031 #endif
1032
1033 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1034 {
1035         return kvm_arch_ops->get_segment_base(vcpu, seg);
1036 }
1037
1038 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1039 {
1040         return X86EMUL_CONTINUE;
1041 }
1042
1043 int emulate_clts(struct kvm_vcpu *vcpu)
1044 {
1045         unsigned long cr0;
1046
1047         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1048         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1049         kvm_arch_ops->set_cr0(vcpu, cr0);
1050         return X86EMUL_CONTINUE;
1051 }
1052
1053 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1054 {
1055         struct kvm_vcpu *vcpu = ctxt->vcpu;
1056
1057         switch (dr) {
1058         case 0 ... 3:
1059                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1060                 return X86EMUL_CONTINUE;
1061         default:
1062                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1063                        __FUNCTION__, dr);
1064                 return X86EMUL_UNHANDLEABLE;
1065         }
1066 }
1067
1068 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1069 {
1070         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1071         int exception;
1072
1073         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1074         if (exception) {
1075                 /* FIXME: better handling */
1076                 return X86EMUL_UNHANDLEABLE;
1077         }
1078         return X86EMUL_CONTINUE;
1079 }
1080
1081 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1082 {
1083         static int reported;
1084         u8 opcodes[4];
1085         unsigned long rip = ctxt->vcpu->rip;
1086         unsigned long rip_linear;
1087
1088         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1089
1090         if (reported)
1091                 return;
1092
1093         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1094
1095         printk(KERN_ERR "emulation failed but !mmio_needed?"
1096                " rip %lx %02x %02x %02x %02x\n",
1097                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1098         reported = 1;
1099 }
1100
1101 struct x86_emulate_ops emulate_ops = {
1102         .read_std            = emulator_read_std,
1103         .write_std           = emulator_write_std,
1104         .read_emulated       = emulator_read_emulated,
1105         .write_emulated      = emulator_write_emulated,
1106         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1107 #ifdef CONFIG_X86_32
1108         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1109 #endif
1110 };
1111
1112 int emulate_instruction(struct kvm_vcpu *vcpu,
1113                         struct kvm_run *run,
1114                         unsigned long cr2,
1115                         u16 error_code)
1116 {
1117         struct x86_emulate_ctxt emulate_ctxt;
1118         int r;
1119         int cs_db, cs_l;
1120
1121         kvm_arch_ops->cache_regs(vcpu);
1122
1123         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1124
1125         emulate_ctxt.vcpu = vcpu;
1126         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1127         emulate_ctxt.cr2 = cr2;
1128         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1129                 ? X86EMUL_MODE_REAL : cs_l
1130                 ? X86EMUL_MODE_PROT64 : cs_db
1131                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1132
1133         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1134                 emulate_ctxt.cs_base = 0;
1135                 emulate_ctxt.ds_base = 0;
1136                 emulate_ctxt.es_base = 0;
1137                 emulate_ctxt.ss_base = 0;
1138         } else {
1139                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1140                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1141                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1142                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1143         }
1144
1145         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1146         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1147
1148         vcpu->mmio_is_write = 0;
1149         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1150
1151         if ((r || vcpu->mmio_is_write) && run) {
1152                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1153                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1154                 run->mmio.len = vcpu->mmio_size;
1155                 run->mmio.is_write = vcpu->mmio_is_write;
1156         }
1157
1158         if (r) {
1159                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1160                         return EMULATE_DONE;
1161                 if (!vcpu->mmio_needed) {
1162                         report_emulation_failure(&emulate_ctxt);
1163                         return EMULATE_FAIL;
1164                 }
1165                 return EMULATE_DO_MMIO;
1166         }
1167
1168         kvm_arch_ops->decache_regs(vcpu);
1169         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1170
1171         if (vcpu->mmio_is_write)
1172                 return EMULATE_DO_MMIO;
1173
1174         return EMULATE_DONE;
1175 }
1176 EXPORT_SYMBOL_GPL(emulate_instruction);
1177
1178 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1179 {
1180         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1181
1182         kvm_arch_ops->cache_regs(vcpu);
1183         ret = -KVM_EINVAL;
1184 #ifdef CONFIG_X86_64
1185         if (is_long_mode(vcpu)) {
1186                 nr = vcpu->regs[VCPU_REGS_RAX];
1187                 a0 = vcpu->regs[VCPU_REGS_RDI];
1188                 a1 = vcpu->regs[VCPU_REGS_RSI];
1189                 a2 = vcpu->regs[VCPU_REGS_RDX];
1190                 a3 = vcpu->regs[VCPU_REGS_RCX];
1191                 a4 = vcpu->regs[VCPU_REGS_R8];
1192                 a5 = vcpu->regs[VCPU_REGS_R9];
1193         } else
1194 #endif
1195         {
1196                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1197                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1198                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1199                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1200                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1201                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1202                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1203         }
1204         switch (nr) {
1205         default:
1206                 run->hypercall.args[0] = a0;
1207                 run->hypercall.args[1] = a1;
1208                 run->hypercall.args[2] = a2;
1209                 run->hypercall.args[3] = a3;
1210                 run->hypercall.args[4] = a4;
1211                 run->hypercall.args[5] = a5;
1212                 run->hypercall.ret = ret;
1213                 run->hypercall.longmode = is_long_mode(vcpu);
1214                 kvm_arch_ops->decache_regs(vcpu);
1215                 return 0;
1216         }
1217         vcpu->regs[VCPU_REGS_RAX] = ret;
1218         kvm_arch_ops->decache_regs(vcpu);
1219         return 1;
1220 }
1221 EXPORT_SYMBOL_GPL(kvm_hypercall);
1222
1223 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1224 {
1225         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1226 }
1227
1228 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1229 {
1230         struct descriptor_table dt = { limit, base };
1231
1232         kvm_arch_ops->set_gdt(vcpu, &dt);
1233 }
1234
1235 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1236 {
1237         struct descriptor_table dt = { limit, base };
1238
1239         kvm_arch_ops->set_idt(vcpu, &dt);
1240 }
1241
1242 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1243                    unsigned long *rflags)
1244 {
1245         lmsw(vcpu, msw);
1246         *rflags = kvm_arch_ops->get_rflags(vcpu);
1247 }
1248
1249 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1250 {
1251         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1252         switch (cr) {
1253         case 0:
1254                 return vcpu->cr0;
1255         case 2:
1256                 return vcpu->cr2;
1257         case 3:
1258                 return vcpu->cr3;
1259         case 4:
1260                 return vcpu->cr4;
1261         default:
1262                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1263                 return 0;
1264         }
1265 }
1266
1267 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1268                      unsigned long *rflags)
1269 {
1270         switch (cr) {
1271         case 0:
1272                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1273                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1274                 break;
1275         case 2:
1276                 vcpu->cr2 = val;
1277                 break;
1278         case 3:
1279                 set_cr3(vcpu, val);
1280                 break;
1281         case 4:
1282                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1283                 break;
1284         default:
1285                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1286         }
1287 }
1288
1289 /*
1290  * Register the para guest with the host:
1291  */
1292 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1293 {
1294         struct kvm_vcpu_para_state *para_state;
1295         hpa_t para_state_hpa, hypercall_hpa;
1296         struct page *para_state_page;
1297         unsigned char *hypercall;
1298         gpa_t hypercall_gpa;
1299
1300         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1301         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1302
1303         /*
1304          * Needs to be page aligned:
1305          */
1306         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1307                 goto err_gp;
1308
1309         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1310         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1311         if (is_error_hpa(para_state_hpa))
1312                 goto err_gp;
1313
1314         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1315         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1316         para_state = kmap_atomic(para_state_page, KM_USER0);
1317
1318         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1319         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1320
1321         para_state->host_version = KVM_PARA_API_VERSION;
1322         /*
1323          * We cannot support guests that try to register themselves
1324          * with a newer API version than the host supports:
1325          */
1326         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1327                 para_state->ret = -KVM_EINVAL;
1328                 goto err_kunmap_skip;
1329         }
1330
1331         hypercall_gpa = para_state->hypercall_gpa;
1332         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1333         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1334         if (is_error_hpa(hypercall_hpa)) {
1335                 para_state->ret = -KVM_EINVAL;
1336                 goto err_kunmap_skip;
1337         }
1338
1339         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1340         vcpu->para_state_page = para_state_page;
1341         vcpu->para_state_gpa = para_state_gpa;
1342         vcpu->hypercall_gpa = hypercall_gpa;
1343
1344         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1345         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1346                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1347         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1348         kunmap_atomic(hypercall, KM_USER1);
1349
1350         para_state->ret = 0;
1351 err_kunmap_skip:
1352         kunmap_atomic(para_state, KM_USER0);
1353         return 0;
1354 err_gp:
1355         return 1;
1356 }
1357
1358 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1359 {
1360         u64 data;
1361
1362         switch (msr) {
1363         case 0xc0010010: /* SYSCFG */
1364         case 0xc0010015: /* HWCR */
1365         case MSR_IA32_PLATFORM_ID:
1366         case MSR_IA32_P5_MC_ADDR:
1367         case MSR_IA32_P5_MC_TYPE:
1368         case MSR_IA32_MC0_CTL:
1369         case MSR_IA32_MCG_STATUS:
1370         case MSR_IA32_MCG_CAP:
1371         case MSR_IA32_MC0_MISC:
1372         case MSR_IA32_MC0_MISC+4:
1373         case MSR_IA32_MC0_MISC+8:
1374         case MSR_IA32_MC0_MISC+12:
1375         case MSR_IA32_MC0_MISC+16:
1376         case MSR_IA32_UCODE_REV:
1377         case MSR_IA32_PERF_STATUS:
1378                 /* MTRR registers */
1379         case 0xfe:
1380         case 0x200 ... 0x2ff:
1381                 data = 0;
1382                 break;
1383         case 0xcd: /* fsb frequency */
1384                 data = 3;
1385                 break;
1386         case MSR_IA32_APICBASE:
1387                 data = vcpu->apic_base;
1388                 break;
1389         case MSR_IA32_MISC_ENABLE:
1390                 data = vcpu->ia32_misc_enable_msr;
1391                 break;
1392 #ifdef CONFIG_X86_64
1393         case MSR_EFER:
1394                 data = vcpu->shadow_efer;
1395                 break;
1396 #endif
1397         default:
1398                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1399                 return 1;
1400         }
1401         *pdata = data;
1402         return 0;
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1405
1406 /*
1407  * Reads an msr value (of 'msr_index') into 'pdata'.
1408  * Returns 0 on success, non-0 otherwise.
1409  * Assumes vcpu_load() was already called.
1410  */
1411 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1412 {
1413         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1414 }
1415
1416 #ifdef CONFIG_X86_64
1417
1418 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1419 {
1420         if (efer & EFER_RESERVED_BITS) {
1421                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1422                        efer);
1423                 inject_gp(vcpu);
1424                 return;
1425         }
1426
1427         if (is_paging(vcpu)
1428             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1429                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1430                 inject_gp(vcpu);
1431                 return;
1432         }
1433
1434         kvm_arch_ops->set_efer(vcpu, efer);
1435
1436         efer &= ~EFER_LMA;
1437         efer |= vcpu->shadow_efer & EFER_LMA;
1438
1439         vcpu->shadow_efer = efer;
1440 }
1441
1442 #endif
1443
1444 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1445 {
1446         switch (msr) {
1447 #ifdef CONFIG_X86_64
1448         case MSR_EFER:
1449                 set_efer(vcpu, data);
1450                 break;
1451 #endif
1452         case MSR_IA32_MC0_STATUS:
1453                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1454                        __FUNCTION__, data);
1455                 break;
1456         case MSR_IA32_UCODE_REV:
1457         case MSR_IA32_UCODE_WRITE:
1458         case 0x200 ... 0x2ff: /* MTRRs */
1459                 break;
1460         case MSR_IA32_APICBASE:
1461                 vcpu->apic_base = data;
1462                 break;
1463         case MSR_IA32_MISC_ENABLE:
1464                 vcpu->ia32_misc_enable_msr = data;
1465                 break;
1466         /*
1467          * This is the 'probe whether the host is KVM' logic:
1468          */
1469         case MSR_KVM_API_MAGIC:
1470                 return vcpu_register_para(vcpu, data);
1471
1472         default:
1473                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1474                 return 1;
1475         }
1476         return 0;
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1479
1480 /*
1481  * Writes msr value into into the appropriate "register".
1482  * Returns 0 on success, non-0 otherwise.
1483  * Assumes vcpu_load() was already called.
1484  */
1485 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1486 {
1487         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1488 }
1489
1490 void kvm_resched(struct kvm_vcpu *vcpu)
1491 {
1492         vcpu_put(vcpu);
1493         cond_resched();
1494         vcpu_load(vcpu);
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_resched);
1497
1498 void load_msrs(struct vmx_msr_entry *e, int n)
1499 {
1500         int i;
1501
1502         for (i = 0; i < n; ++i)
1503                 wrmsrl(e[i].index, e[i].data);
1504 }
1505 EXPORT_SYMBOL_GPL(load_msrs);
1506
1507 void save_msrs(struct vmx_msr_entry *e, int n)
1508 {
1509         int i;
1510
1511         for (i = 0; i < n; ++i)
1512                 rdmsrl(e[i].index, e[i].data);
1513 }
1514 EXPORT_SYMBOL_GPL(save_msrs);
1515
1516 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1517 {
1518         int i;
1519         u32 function;
1520         struct kvm_cpuid_entry *e, *best;
1521
1522         kvm_arch_ops->cache_regs(vcpu);
1523         function = vcpu->regs[VCPU_REGS_RAX];
1524         vcpu->regs[VCPU_REGS_RAX] = 0;
1525         vcpu->regs[VCPU_REGS_RBX] = 0;
1526         vcpu->regs[VCPU_REGS_RCX] = 0;
1527         vcpu->regs[VCPU_REGS_RDX] = 0;
1528         best = NULL;
1529         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1530                 e = &vcpu->cpuid_entries[i];
1531                 if (e->function == function) {
1532                         best = e;
1533                         break;
1534                 }
1535                 /*
1536                  * Both basic or both extended?
1537                  */
1538                 if (((e->function ^ function) & 0x80000000) == 0)
1539                         if (!best || e->function > best->function)
1540                                 best = e;
1541         }
1542         if (best) {
1543                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1544                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1545                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1546                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1547         }
1548         kvm_arch_ops->decache_regs(vcpu);
1549         kvm_arch_ops->skip_emulated_instruction(vcpu);
1550 }
1551 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1552
1553 static void complete_pio(struct kvm_vcpu *vcpu)
1554 {
1555         struct kvm_io *io = &vcpu->run->io;
1556         long delta;
1557
1558         kvm_arch_ops->cache_regs(vcpu);
1559
1560         if (!io->string) {
1561                 if (io->direction == KVM_EXIT_IO_IN)
1562                         memcpy(&vcpu->regs[VCPU_REGS_RAX], &io->value,
1563                                io->size);
1564         } else {
1565                 delta = 1;
1566                 if (io->rep) {
1567                         delta *= io->count;
1568                         /*
1569                          * The size of the register should really depend on
1570                          * current address size.
1571                          */
1572                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1573                 }
1574                 if (io->string_down)
1575                         delta = -delta;
1576                 delta *= io->size;
1577                 if (io->direction == KVM_EXIT_IO_IN)
1578                         vcpu->regs[VCPU_REGS_RDI] += delta;
1579                 else
1580                         vcpu->regs[VCPU_REGS_RSI] += delta;
1581         }
1582
1583         vcpu->pio_pending = 0;
1584         vcpu->run->io_completed = 0;
1585
1586         kvm_arch_ops->decache_regs(vcpu);
1587
1588         kvm_arch_ops->skip_emulated_instruction(vcpu);
1589 }
1590
1591 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1592 {
1593         int r;
1594         sigset_t sigsaved;
1595
1596         vcpu_load(vcpu);
1597
1598         if (vcpu->sigset_active)
1599                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1600
1601         /* re-sync apic's tpr */
1602         vcpu->cr8 = kvm_run->cr8;
1603
1604         if (kvm_run->io_completed) {
1605                 if (vcpu->pio_pending)
1606                         complete_pio(vcpu);
1607                 else {
1608                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1609                         vcpu->mmio_read_completed = 1;
1610                 }
1611         }
1612
1613         vcpu->mmio_needed = 0;
1614
1615         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1616                 kvm_arch_ops->cache_regs(vcpu);
1617                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1618                 kvm_arch_ops->decache_regs(vcpu);
1619         }
1620
1621         r = kvm_arch_ops->run(vcpu, kvm_run);
1622
1623         if (vcpu->sigset_active)
1624                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1625
1626         vcpu_put(vcpu);
1627         return r;
1628 }
1629
1630 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1631                                    struct kvm_regs *regs)
1632 {
1633         vcpu_load(vcpu);
1634
1635         kvm_arch_ops->cache_regs(vcpu);
1636
1637         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1638         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1639         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1640         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1641         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1642         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1643         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1644         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1645 #ifdef CONFIG_X86_64
1646         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1647         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1648         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1649         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1650         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1651         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1652         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1653         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1654 #endif
1655
1656         regs->rip = vcpu->rip;
1657         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1658
1659         /*
1660          * Don't leak debug flags in case they were set for guest debugging
1661          */
1662         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1663                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1664
1665         vcpu_put(vcpu);
1666
1667         return 0;
1668 }
1669
1670 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1671                                    struct kvm_regs *regs)
1672 {
1673         vcpu_load(vcpu);
1674
1675         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1676         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1677         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1678         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1679         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1680         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1681         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1682         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1683 #ifdef CONFIG_X86_64
1684         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1685         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1686         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1687         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1688         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1689         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1690         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1691         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1692 #endif
1693
1694         vcpu->rip = regs->rip;
1695         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1696
1697         kvm_arch_ops->decache_regs(vcpu);
1698
1699         vcpu_put(vcpu);
1700
1701         return 0;
1702 }
1703
1704 static void get_segment(struct kvm_vcpu *vcpu,
1705                         struct kvm_segment *var, int seg)
1706 {
1707         return kvm_arch_ops->get_segment(vcpu, var, seg);
1708 }
1709
1710 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1711                                     struct kvm_sregs *sregs)
1712 {
1713         struct descriptor_table dt;
1714
1715         vcpu_load(vcpu);
1716
1717         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1718         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1719         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1720         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1721         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1722         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1723
1724         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1725         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1726
1727         kvm_arch_ops->get_idt(vcpu, &dt);
1728         sregs->idt.limit = dt.limit;
1729         sregs->idt.base = dt.base;
1730         kvm_arch_ops->get_gdt(vcpu, &dt);
1731         sregs->gdt.limit = dt.limit;
1732         sregs->gdt.base = dt.base;
1733
1734         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1735         sregs->cr0 = vcpu->cr0;
1736         sregs->cr2 = vcpu->cr2;
1737         sregs->cr3 = vcpu->cr3;
1738         sregs->cr4 = vcpu->cr4;
1739         sregs->cr8 = vcpu->cr8;
1740         sregs->efer = vcpu->shadow_efer;
1741         sregs->apic_base = vcpu->apic_base;
1742
1743         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1744                sizeof sregs->interrupt_bitmap);
1745
1746         vcpu_put(vcpu);
1747
1748         return 0;
1749 }
1750
1751 static void set_segment(struct kvm_vcpu *vcpu,
1752                         struct kvm_segment *var, int seg)
1753 {
1754         return kvm_arch_ops->set_segment(vcpu, var, seg);
1755 }
1756
1757 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1758                                     struct kvm_sregs *sregs)
1759 {
1760         int mmu_reset_needed = 0;
1761         int i;
1762         struct descriptor_table dt;
1763
1764         vcpu_load(vcpu);
1765
1766         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1767         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1768         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1769         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1770         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1771         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1772
1773         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1774         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1775
1776         dt.limit = sregs->idt.limit;
1777         dt.base = sregs->idt.base;
1778         kvm_arch_ops->set_idt(vcpu, &dt);
1779         dt.limit = sregs->gdt.limit;
1780         dt.base = sregs->gdt.base;
1781         kvm_arch_ops->set_gdt(vcpu, &dt);
1782
1783         vcpu->cr2 = sregs->cr2;
1784         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1785         vcpu->cr3 = sregs->cr3;
1786
1787         vcpu->cr8 = sregs->cr8;
1788
1789         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1790 #ifdef CONFIG_X86_64
1791         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1792 #endif
1793         vcpu->apic_base = sregs->apic_base;
1794
1795         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1796
1797         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1798         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1799
1800         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1801         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1802         if (!is_long_mode(vcpu) && is_pae(vcpu))
1803                 load_pdptrs(vcpu, vcpu->cr3);
1804
1805         if (mmu_reset_needed)
1806                 kvm_mmu_reset_context(vcpu);
1807
1808         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1809                sizeof vcpu->irq_pending);
1810         vcpu->irq_summary = 0;
1811         for (i = 0; i < NR_IRQ_WORDS; ++i)
1812                 if (vcpu->irq_pending[i])
1813                         __set_bit(i, &vcpu->irq_summary);
1814
1815         vcpu_put(vcpu);
1816
1817         return 0;
1818 }
1819
1820 /*
1821  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1822  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1823  *
1824  * This list is modified at module load time to reflect the
1825  * capabilities of the host cpu.
1826  */
1827 static u32 msrs_to_save[] = {
1828         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1829         MSR_K6_STAR,
1830 #ifdef CONFIG_X86_64
1831         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1832 #endif
1833         MSR_IA32_TIME_STAMP_COUNTER,
1834 };
1835
1836 static unsigned num_msrs_to_save;
1837
1838 static u32 emulated_msrs[] = {
1839         MSR_IA32_MISC_ENABLE,
1840 };
1841
1842 static __init void kvm_init_msr_list(void)
1843 {
1844         u32 dummy[2];
1845         unsigned i, j;
1846
1847         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1848                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1849                         continue;
1850                 if (j < i)
1851                         msrs_to_save[j] = msrs_to_save[i];
1852                 j++;
1853         }
1854         num_msrs_to_save = j;
1855 }
1856
1857 /*
1858  * Adapt set_msr() to msr_io()'s calling convention
1859  */
1860 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1861 {
1862         return set_msr(vcpu, index, *data);
1863 }
1864
1865 /*
1866  * Read or write a bunch of msrs. All parameters are kernel addresses.
1867  *
1868  * @return number of msrs set successfully.
1869  */
1870 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1871                     struct kvm_msr_entry *entries,
1872                     int (*do_msr)(struct kvm_vcpu *vcpu,
1873                                   unsigned index, u64 *data))
1874 {
1875         int i;
1876
1877         vcpu_load(vcpu);
1878
1879         for (i = 0; i < msrs->nmsrs; ++i)
1880                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1881                         break;
1882
1883         vcpu_put(vcpu);
1884
1885         return i;
1886 }
1887
1888 /*
1889  * Read or write a bunch of msrs. Parameters are user addresses.
1890  *
1891  * @return number of msrs set successfully.
1892  */
1893 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1894                   int (*do_msr)(struct kvm_vcpu *vcpu,
1895                                 unsigned index, u64 *data),
1896                   int writeback)
1897 {
1898         struct kvm_msrs msrs;
1899         struct kvm_msr_entry *entries;
1900         int r, n;
1901         unsigned size;
1902
1903         r = -EFAULT;
1904         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1905                 goto out;
1906
1907         r = -E2BIG;
1908         if (msrs.nmsrs >= MAX_IO_MSRS)
1909                 goto out;
1910
1911         r = -ENOMEM;
1912         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1913         entries = vmalloc(size);
1914         if (!entries)
1915                 goto out;
1916
1917         r = -EFAULT;
1918         if (copy_from_user(entries, user_msrs->entries, size))
1919                 goto out_free;
1920
1921         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1922         if (r < 0)
1923                 goto out_free;
1924
1925         r = -EFAULT;
1926         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1927                 goto out_free;
1928
1929         r = n;
1930
1931 out_free:
1932         vfree(entries);
1933 out:
1934         return r;
1935 }
1936
1937 /*
1938  * Translate a guest virtual address to a guest physical address.
1939  */
1940 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1941                                     struct kvm_translation *tr)
1942 {
1943         unsigned long vaddr = tr->linear_address;
1944         gpa_t gpa;
1945
1946         vcpu_load(vcpu);
1947         spin_lock(&vcpu->kvm->lock);
1948         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1949         tr->physical_address = gpa;
1950         tr->valid = gpa != UNMAPPED_GVA;
1951         tr->writeable = 1;
1952         tr->usermode = 0;
1953         spin_unlock(&vcpu->kvm->lock);
1954         vcpu_put(vcpu);
1955
1956         return 0;
1957 }
1958
1959 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1960                                     struct kvm_interrupt *irq)
1961 {
1962         if (irq->irq < 0 || irq->irq >= 256)
1963                 return -EINVAL;
1964         vcpu_load(vcpu);
1965
1966         set_bit(irq->irq, vcpu->irq_pending);
1967         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1968
1969         vcpu_put(vcpu);
1970
1971         return 0;
1972 }
1973
1974 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1975                                       struct kvm_debug_guest *dbg)
1976 {
1977         int r;
1978
1979         vcpu_load(vcpu);
1980
1981         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1982
1983         vcpu_put(vcpu);
1984
1985         return r;
1986 }
1987
1988 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1989                                     unsigned long address,
1990                                     int *type)
1991 {
1992         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1993         unsigned long pgoff;
1994         struct page *page;
1995
1996         *type = VM_FAULT_MINOR;
1997         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1998         if (pgoff != 0)
1999                 return NOPAGE_SIGBUS;
2000         page = virt_to_page(vcpu->run);
2001         get_page(page);
2002         return page;
2003 }
2004
2005 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2006         .nopage = kvm_vcpu_nopage,
2007 };
2008
2009 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2010 {
2011         vma->vm_ops = &kvm_vcpu_vm_ops;
2012         return 0;
2013 }
2014
2015 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2016 {
2017         struct kvm_vcpu *vcpu = filp->private_data;
2018
2019         fput(vcpu->kvm->filp);
2020         return 0;
2021 }
2022
2023 static struct file_operations kvm_vcpu_fops = {
2024         .release        = kvm_vcpu_release,
2025         .unlocked_ioctl = kvm_vcpu_ioctl,
2026         .compat_ioctl   = kvm_vcpu_ioctl,
2027         .mmap           = kvm_vcpu_mmap,
2028 };
2029
2030 /*
2031  * Allocates an inode for the vcpu.
2032  */
2033 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2034 {
2035         int fd, r;
2036         struct inode *inode;
2037         struct file *file;
2038
2039         atomic_inc(&vcpu->kvm->filp->f_count);
2040         inode = kvmfs_inode(&kvm_vcpu_fops);
2041         if (IS_ERR(inode)) {
2042                 r = PTR_ERR(inode);
2043                 goto out1;
2044         }
2045
2046         file = kvmfs_file(inode, vcpu);
2047         if (IS_ERR(file)) {
2048                 r = PTR_ERR(file);
2049                 goto out2;
2050         }
2051
2052         r = get_unused_fd();
2053         if (r < 0)
2054                 goto out3;
2055         fd = r;
2056         fd_install(fd, file);
2057
2058         return fd;
2059
2060 out3:
2061         fput(file);
2062 out2:
2063         iput(inode);
2064 out1:
2065         fput(vcpu->kvm->filp);
2066         return r;
2067 }
2068
2069 /*
2070  * Creates some virtual cpus.  Good luck creating more than one.
2071  */
2072 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2073 {
2074         int r;
2075         struct kvm_vcpu *vcpu;
2076         struct page *page;
2077
2078         r = -EINVAL;
2079         if (!valid_vcpu(n))
2080                 goto out;
2081
2082         vcpu = &kvm->vcpus[n];
2083
2084         mutex_lock(&vcpu->mutex);
2085
2086         if (vcpu->vmcs) {
2087                 mutex_unlock(&vcpu->mutex);
2088                 return -EEXIST;
2089         }
2090
2091         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2092         r = -ENOMEM;
2093         if (!page)
2094                 goto out_unlock;
2095         vcpu->run = page_address(page);
2096
2097         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2098                                            FX_IMAGE_ALIGN);
2099         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2100
2101         r = kvm_arch_ops->vcpu_create(vcpu);
2102         if (r < 0)
2103                 goto out_free_vcpus;
2104
2105         r = kvm_mmu_create(vcpu);
2106         if (r < 0)
2107                 goto out_free_vcpus;
2108
2109         kvm_arch_ops->vcpu_load(vcpu);
2110         r = kvm_mmu_setup(vcpu);
2111         if (r >= 0)
2112                 r = kvm_arch_ops->vcpu_setup(vcpu);
2113         vcpu_put(vcpu);
2114
2115         if (r < 0)
2116                 goto out_free_vcpus;
2117
2118         r = create_vcpu_fd(vcpu);
2119         if (r < 0)
2120                 goto out_free_vcpus;
2121
2122         return r;
2123
2124 out_free_vcpus:
2125         kvm_free_vcpu(vcpu);
2126 out_unlock:
2127         mutex_unlock(&vcpu->mutex);
2128 out:
2129         return r;
2130 }
2131
2132 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2133                                     struct kvm_cpuid *cpuid,
2134                                     struct kvm_cpuid_entry __user *entries)
2135 {
2136         int r;
2137
2138         r = -E2BIG;
2139         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2140                 goto out;
2141         r = -EFAULT;
2142         if (copy_from_user(&vcpu->cpuid_entries, entries,
2143                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2144                 goto out;
2145         vcpu->cpuid_nent = cpuid->nent;
2146         return 0;
2147
2148 out:
2149         return r;
2150 }
2151
2152 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2153 {
2154         if (sigset) {
2155                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2156                 vcpu->sigset_active = 1;
2157                 vcpu->sigset = *sigset;
2158         } else
2159                 vcpu->sigset_active = 0;
2160         return 0;
2161 }
2162
2163 static long kvm_vcpu_ioctl(struct file *filp,
2164                            unsigned int ioctl, unsigned long arg)
2165 {
2166         struct kvm_vcpu *vcpu = filp->private_data;
2167         void __user *argp = (void __user *)arg;
2168         int r = -EINVAL;
2169
2170         switch (ioctl) {
2171         case KVM_RUN:
2172                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2173                 break;
2174         case KVM_GET_REGS: {
2175                 struct kvm_regs kvm_regs;
2176
2177                 memset(&kvm_regs, 0, sizeof kvm_regs);
2178                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2179                 if (r)
2180                         goto out;
2181                 r = -EFAULT;
2182                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2183                         goto out;
2184                 r = 0;
2185                 break;
2186         }
2187         case KVM_SET_REGS: {
2188                 struct kvm_regs kvm_regs;
2189
2190                 r = -EFAULT;
2191                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2192                         goto out;
2193                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2194                 if (r)
2195                         goto out;
2196                 r = 0;
2197                 break;
2198         }
2199         case KVM_GET_SREGS: {
2200                 struct kvm_sregs kvm_sregs;
2201
2202                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2203                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2204                 if (r)
2205                         goto out;
2206                 r = -EFAULT;
2207                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2208                         goto out;
2209                 r = 0;
2210                 break;
2211         }
2212         case KVM_SET_SREGS: {
2213                 struct kvm_sregs kvm_sregs;
2214
2215                 r = -EFAULT;
2216                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2217                         goto out;
2218                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2219                 if (r)
2220                         goto out;
2221                 r = 0;
2222                 break;
2223         }
2224         case KVM_TRANSLATE: {
2225                 struct kvm_translation tr;
2226
2227                 r = -EFAULT;
2228                 if (copy_from_user(&tr, argp, sizeof tr))
2229                         goto out;
2230                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2231                 if (r)
2232                         goto out;
2233                 r = -EFAULT;
2234                 if (copy_to_user(argp, &tr, sizeof tr))
2235                         goto out;
2236                 r = 0;
2237                 break;
2238         }
2239         case KVM_INTERRUPT: {
2240                 struct kvm_interrupt irq;
2241
2242                 r = -EFAULT;
2243                 if (copy_from_user(&irq, argp, sizeof irq))
2244                         goto out;
2245                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2246                 if (r)
2247                         goto out;
2248                 r = 0;
2249                 break;
2250         }
2251         case KVM_DEBUG_GUEST: {
2252                 struct kvm_debug_guest dbg;
2253
2254                 r = -EFAULT;
2255                 if (copy_from_user(&dbg, argp, sizeof dbg))
2256                         goto out;
2257                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2258                 if (r)
2259                         goto out;
2260                 r = 0;
2261                 break;
2262         }
2263         case KVM_GET_MSRS:
2264                 r = msr_io(vcpu, argp, get_msr, 1);
2265                 break;
2266         case KVM_SET_MSRS:
2267                 r = msr_io(vcpu, argp, do_set_msr, 0);
2268                 break;
2269         case KVM_SET_CPUID: {
2270                 struct kvm_cpuid __user *cpuid_arg = argp;
2271                 struct kvm_cpuid cpuid;
2272
2273                 r = -EFAULT;
2274                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2275                         goto out;
2276                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2277                 if (r)
2278                         goto out;
2279                 break;
2280         }
2281         case KVM_SET_SIGNAL_MASK: {
2282                 struct kvm_signal_mask __user *sigmask_arg = argp;
2283                 struct kvm_signal_mask kvm_sigmask;
2284                 sigset_t sigset, *p;
2285
2286                 p = NULL;
2287                 if (argp) {
2288                         r = -EFAULT;
2289                         if (copy_from_user(&kvm_sigmask, argp,
2290                                            sizeof kvm_sigmask))
2291                                 goto out;
2292                         r = -EINVAL;
2293                         if (kvm_sigmask.len != sizeof sigset)
2294                                 goto out;
2295                         r = -EFAULT;
2296                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2297                                            sizeof sigset))
2298                                 goto out;
2299                         p = &sigset;
2300                 }
2301                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2302                 break;
2303         }
2304         default:
2305                 ;
2306         }
2307 out:
2308         return r;
2309 }
2310
2311 static long kvm_vm_ioctl(struct file *filp,
2312                            unsigned int ioctl, unsigned long arg)
2313 {
2314         struct kvm *kvm = filp->private_data;
2315         void __user *argp = (void __user *)arg;
2316         int r = -EINVAL;
2317
2318         switch (ioctl) {
2319         case KVM_CREATE_VCPU:
2320                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2321                 if (r < 0)
2322                         goto out;
2323                 break;
2324         case KVM_SET_MEMORY_REGION: {
2325                 struct kvm_memory_region kvm_mem;
2326
2327                 r = -EFAULT;
2328                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2329                         goto out;
2330                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2331                 if (r)
2332                         goto out;
2333                 break;
2334         }
2335         case KVM_GET_DIRTY_LOG: {
2336                 struct kvm_dirty_log log;
2337
2338                 r = -EFAULT;
2339                 if (copy_from_user(&log, argp, sizeof log))
2340                         goto out;
2341                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2342                 if (r)
2343                         goto out;
2344                 break;
2345         }
2346         default:
2347                 ;
2348         }
2349 out:
2350         return r;
2351 }
2352
2353 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2354                                   unsigned long address,
2355                                   int *type)
2356 {
2357         struct kvm *kvm = vma->vm_file->private_data;
2358         unsigned long pgoff;
2359         struct kvm_memory_slot *slot;
2360         struct page *page;
2361
2362         *type = VM_FAULT_MINOR;
2363         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2364         slot = gfn_to_memslot(kvm, pgoff);
2365         if (!slot)
2366                 return NOPAGE_SIGBUS;
2367         page = gfn_to_page(slot, pgoff);
2368         if (!page)
2369                 return NOPAGE_SIGBUS;
2370         get_page(page);
2371         return page;
2372 }
2373
2374 static struct vm_operations_struct kvm_vm_vm_ops = {
2375         .nopage = kvm_vm_nopage,
2376 };
2377
2378 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2379 {
2380         vma->vm_ops = &kvm_vm_vm_ops;
2381         return 0;
2382 }
2383
2384 static struct file_operations kvm_vm_fops = {
2385         .release        = kvm_vm_release,
2386         .unlocked_ioctl = kvm_vm_ioctl,
2387         .compat_ioctl   = kvm_vm_ioctl,
2388         .mmap           = kvm_vm_mmap,
2389 };
2390
2391 static int kvm_dev_ioctl_create_vm(void)
2392 {
2393         int fd, r;
2394         struct inode *inode;
2395         struct file *file;
2396         struct kvm *kvm;
2397
2398         inode = kvmfs_inode(&kvm_vm_fops);
2399         if (IS_ERR(inode)) {
2400                 r = PTR_ERR(inode);
2401                 goto out1;
2402         }
2403
2404         kvm = kvm_create_vm();
2405         if (IS_ERR(kvm)) {
2406                 r = PTR_ERR(kvm);
2407                 goto out2;
2408         }
2409
2410         file = kvmfs_file(inode, kvm);
2411         if (IS_ERR(file)) {
2412                 r = PTR_ERR(file);
2413                 goto out3;
2414         }
2415         kvm->filp = file;
2416
2417         r = get_unused_fd();
2418         if (r < 0)
2419                 goto out4;
2420         fd = r;
2421         fd_install(fd, file);
2422
2423         return fd;
2424
2425 out4:
2426         fput(file);
2427 out3:
2428         kvm_destroy_vm(kvm);
2429 out2:
2430         iput(inode);
2431 out1:
2432         return r;
2433 }
2434
2435 static long kvm_dev_ioctl(struct file *filp,
2436                           unsigned int ioctl, unsigned long arg)
2437 {
2438         void __user *argp = (void __user *)arg;
2439         long r = -EINVAL;
2440
2441         switch (ioctl) {
2442         case KVM_GET_API_VERSION:
2443                 r = KVM_API_VERSION;
2444                 break;
2445         case KVM_CREATE_VM:
2446                 r = kvm_dev_ioctl_create_vm();
2447                 break;
2448         case KVM_GET_MSR_INDEX_LIST: {
2449                 struct kvm_msr_list __user *user_msr_list = argp;
2450                 struct kvm_msr_list msr_list;
2451                 unsigned n;
2452
2453                 r = -EFAULT;
2454                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2455                         goto out;
2456                 n = msr_list.nmsrs;
2457                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2458                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2459                         goto out;
2460                 r = -E2BIG;
2461                 if (n < num_msrs_to_save)
2462                         goto out;
2463                 r = -EFAULT;
2464                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2465                                  num_msrs_to_save * sizeof(u32)))
2466                         goto out;
2467                 if (copy_to_user(user_msr_list->indices
2468                                  + num_msrs_to_save * sizeof(u32),
2469                                  &emulated_msrs,
2470                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2471                         goto out;
2472                 r = 0;
2473                 break;
2474         }
2475         case KVM_CHECK_EXTENSION:
2476                 /*
2477                  * No extensions defined at present.
2478                  */
2479                 r = 0;
2480                 break;
2481         case KVM_GET_VCPU_MMAP_SIZE:
2482                 r = -EINVAL;
2483                 if (arg)
2484                         goto out;
2485                 r = PAGE_SIZE;
2486                 break;
2487         default:
2488                 ;
2489         }
2490 out:
2491         return r;
2492 }
2493
2494 static struct file_operations kvm_chardev_ops = {
2495         .open           = kvm_dev_open,
2496         .release        = kvm_dev_release,
2497         .unlocked_ioctl = kvm_dev_ioctl,
2498         .compat_ioctl   = kvm_dev_ioctl,
2499 };
2500
2501 static struct miscdevice kvm_dev = {
2502         KVM_MINOR,
2503         "kvm",
2504         &kvm_chardev_ops,
2505 };
2506
2507 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2508                        void *v)
2509 {
2510         if (val == SYS_RESTART) {
2511                 /*
2512                  * Some (well, at least mine) BIOSes hang on reboot if
2513                  * in vmx root mode.
2514                  */
2515                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2516                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2517         }
2518         return NOTIFY_OK;
2519 }
2520
2521 static struct notifier_block kvm_reboot_notifier = {
2522         .notifier_call = kvm_reboot,
2523         .priority = 0,
2524 };
2525
2526 /*
2527  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2528  * cached on it.
2529  */
2530 static void decache_vcpus_on_cpu(int cpu)
2531 {
2532         struct kvm *vm;
2533         struct kvm_vcpu *vcpu;
2534         int i;
2535
2536         spin_lock(&kvm_lock);
2537         list_for_each_entry(vm, &vm_list, vm_list)
2538                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2539                         vcpu = &vm->vcpus[i];
2540                         /*
2541                          * If the vcpu is locked, then it is running on some
2542                          * other cpu and therefore it is not cached on the
2543                          * cpu in question.
2544                          *
2545                          * If it's not locked, check the last cpu it executed
2546                          * on.
2547                          */
2548                         if (mutex_trylock(&vcpu->mutex)) {
2549                                 if (vcpu->cpu == cpu) {
2550                                         kvm_arch_ops->vcpu_decache(vcpu);
2551                                         vcpu->cpu = -1;
2552                                 }
2553                                 mutex_unlock(&vcpu->mutex);
2554                         }
2555                 }
2556         spin_unlock(&kvm_lock);
2557 }
2558
2559 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2560                            void *v)
2561 {
2562         int cpu = (long)v;
2563
2564         switch (val) {
2565         case CPU_DOWN_PREPARE:
2566         case CPU_UP_CANCELED:
2567                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2568                        cpu);
2569                 decache_vcpus_on_cpu(cpu);
2570                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2571                                          NULL, 0, 1);
2572                 break;
2573         case CPU_ONLINE:
2574                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2575                        cpu);
2576                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2577                                          NULL, 0, 1);
2578                 break;
2579         }
2580         return NOTIFY_OK;
2581 }
2582
2583 static struct notifier_block kvm_cpu_notifier = {
2584         .notifier_call = kvm_cpu_hotplug,
2585         .priority = 20, /* must be > scheduler priority */
2586 };
2587
2588 static __init void kvm_init_debug(void)
2589 {
2590         struct kvm_stats_debugfs_item *p;
2591
2592         debugfs_dir = debugfs_create_dir("kvm", NULL);
2593         for (p = debugfs_entries; p->name; ++p)
2594                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2595                                                p->data);
2596 }
2597
2598 static void kvm_exit_debug(void)
2599 {
2600         struct kvm_stats_debugfs_item *p;
2601
2602         for (p = debugfs_entries; p->name; ++p)
2603                 debugfs_remove(p->dentry);
2604         debugfs_remove(debugfs_dir);
2605 }
2606
2607 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2608 {
2609         decache_vcpus_on_cpu(raw_smp_processor_id());
2610         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2611         return 0;
2612 }
2613
2614 static int kvm_resume(struct sys_device *dev)
2615 {
2616         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2617         return 0;
2618 }
2619
2620 static struct sysdev_class kvm_sysdev_class = {
2621         set_kset_name("kvm"),
2622         .suspend = kvm_suspend,
2623         .resume = kvm_resume,
2624 };
2625
2626 static struct sys_device kvm_sysdev = {
2627         .id = 0,
2628         .cls = &kvm_sysdev_class,
2629 };
2630
2631 hpa_t bad_page_address;
2632
2633 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2634                         const char *dev_name, void *data, struct vfsmount *mnt)
2635 {
2636         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2637 }
2638
2639 static struct file_system_type kvm_fs_type = {
2640         .name           = "kvmfs",
2641         .get_sb         = kvmfs_get_sb,
2642         .kill_sb        = kill_anon_super,
2643 };
2644
2645 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2646 {
2647         int r;
2648
2649         if (kvm_arch_ops) {
2650                 printk(KERN_ERR "kvm: already loaded the other module\n");
2651                 return -EEXIST;
2652         }
2653
2654         if (!ops->cpu_has_kvm_support()) {
2655                 printk(KERN_ERR "kvm: no hardware support\n");
2656                 return -EOPNOTSUPP;
2657         }
2658         if (ops->disabled_by_bios()) {
2659                 printk(KERN_ERR "kvm: disabled by bios\n");
2660                 return -EOPNOTSUPP;
2661         }
2662
2663         kvm_arch_ops = ops;
2664
2665         r = kvm_arch_ops->hardware_setup();
2666         if (r < 0)
2667                 goto out;
2668
2669         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2670         r = register_cpu_notifier(&kvm_cpu_notifier);
2671         if (r)
2672                 goto out_free_1;
2673         register_reboot_notifier(&kvm_reboot_notifier);
2674
2675         r = sysdev_class_register(&kvm_sysdev_class);
2676         if (r)
2677                 goto out_free_2;
2678
2679         r = sysdev_register(&kvm_sysdev);
2680         if (r)
2681                 goto out_free_3;
2682
2683         kvm_chardev_ops.owner = module;
2684
2685         r = misc_register(&kvm_dev);
2686         if (r) {
2687                 printk (KERN_ERR "kvm: misc device register failed\n");
2688                 goto out_free;
2689         }
2690
2691         return r;
2692
2693 out_free:
2694         sysdev_unregister(&kvm_sysdev);
2695 out_free_3:
2696         sysdev_class_unregister(&kvm_sysdev_class);
2697 out_free_2:
2698         unregister_reboot_notifier(&kvm_reboot_notifier);
2699         unregister_cpu_notifier(&kvm_cpu_notifier);
2700 out_free_1:
2701         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2702         kvm_arch_ops->hardware_unsetup();
2703 out:
2704         kvm_arch_ops = NULL;
2705         return r;
2706 }
2707
2708 void kvm_exit_arch(void)
2709 {
2710         misc_deregister(&kvm_dev);
2711         sysdev_unregister(&kvm_sysdev);
2712         sysdev_class_unregister(&kvm_sysdev_class);
2713         unregister_reboot_notifier(&kvm_reboot_notifier);
2714         unregister_cpu_notifier(&kvm_cpu_notifier);
2715         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2716         kvm_arch_ops->hardware_unsetup();
2717         kvm_arch_ops = NULL;
2718 }
2719
2720 static __init int kvm_init(void)
2721 {
2722         static struct page *bad_page;
2723         int r;
2724
2725         r = register_filesystem(&kvm_fs_type);
2726         if (r)
2727                 goto out3;
2728
2729         kvmfs_mnt = kern_mount(&kvm_fs_type);
2730         r = PTR_ERR(kvmfs_mnt);
2731         if (IS_ERR(kvmfs_mnt))
2732                 goto out2;
2733         kvm_init_debug();
2734
2735         kvm_init_msr_list();
2736
2737         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2738                 r = -ENOMEM;
2739                 goto out;
2740         }
2741
2742         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2743         memset(__va(bad_page_address), 0, PAGE_SIZE);
2744
2745         return 0;
2746
2747 out:
2748         kvm_exit_debug();
2749         mntput(kvmfs_mnt);
2750 out2:
2751         unregister_filesystem(&kvm_fs_type);
2752 out3:
2753         return r;
2754 }
2755
2756 static __exit void kvm_exit(void)
2757 {
2758         kvm_exit_debug();
2759         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2760         mntput(kvmfs_mnt);
2761         unregister_filesystem(&kvm_fs_type);
2762 }
2763
2764 module_init(kvm_init)
2765 module_exit(kvm_exit)
2766
2767 EXPORT_SYMBOL_GPL(kvm_init_arch);
2768 EXPORT_SYMBOL_GPL(kvm_exit_arch);