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