drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <asm/processor.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <asm/msr.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/reboot.h>
  32 #include <asm/io.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/highmem.h>
  35 #include <linux/file.h>
  36 #include <asm/desc.h>
  37 #include <linux/sysdev.h>
  38 #include <linux/cpu.h>
  39
  40 #include "x86_emulate.h"
  41 #include "segment_descriptor.h"
  42
  43 MODULE_AUTHOR("Qumranet");
  44 MODULE_LICENSE("GPL");
  45
  46 static DEFINE_SPINLOCK(kvm_lock);
  47 static LIST_HEAD(vm_list);
  48
  49 struct kvm_arch_ops *kvm_arch_ops;
  50 struct kvm_stat kvm_stat;
  51 EXPORT_SYMBOL_GPL(kvm_stat);
  52
  53 static struct kvm_stats_debugfs_item {
  54         const char *name;
  55         u32 *data;
  56         struct dentry *dentry;
  57 } debugfs_entries[] = {
  58         { "pf_fixed", &kvm_stat.pf_fixed },
  59         { "pf_guest", &kvm_stat.pf_guest },
  60         { "tlb_flush", &kvm_stat.tlb_flush },
  61         { "invlpg", &kvm_stat.invlpg },
  62         { "exits", &kvm_stat.exits },
  63         { "io_exits", &kvm_stat.io_exits },
  64         { "mmio_exits", &kvm_stat.mmio_exits },
  65         { "signal_exits", &kvm_stat.signal_exits },
  66         { "irq_window", &kvm_stat.irq_window_exits },
  67         { "halt_exits", &kvm_stat.halt_exits },
  68         { "request_irq", &kvm_stat.request_irq_exits },
  69         { "irq_exits", &kvm_stat.irq_exits },
  70         { NULL, NULL }
  71 };
  72
  73 static struct dentry *debugfs_dir;
  74
  75 #define MAX_IO_MSRS 256
  76
  77 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  78 #define LMSW_GUEST_MASK 0x0eULL
  79 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  80 #define CR8_RESEVED_BITS (~0x0fULL)
  81 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  82
  83 #ifdef CONFIG_X86_64
  84 // LDT or TSS descriptor in the GDT. 16 bytes.
  85 struct segment_descriptor_64 {
  86         struct segment_descriptor s;
  87         u32 base_higher;
  88         u32 pad_zero;
  89 };
  90
  91 #endif
  92
  93 unsigned long segment_base(u16 selector)
  94 {
  95         struct descriptor_table gdt;
  96         struct segment_descriptor *d;
  97         unsigned long table_base;
  98         typedef unsigned long ul;
  99         unsigned long v;
 100
 101         if (selector == 0)
 102                 return 0;
 103
 104         asm ("sgdt %0" : "=m"(gdt));
 105         table_base = gdt.base;
 106
 107         if (selector & 4) {           /* from ldt */
 108                 u16 ldt_selector;
 109
 110                 asm ("sldt %0" : "=g"(ldt_selector));
 111                 table_base = segment_base(ldt_selector);
 112         }
 113         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 114         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 115 #ifdef CONFIG_X86_64
 116         if (d->system == 0
 117             && (d->type == 2 || d->type == 9 || d->type == 11))
 118                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 119 #endif
 120         return v;
 121 }
 122 EXPORT_SYMBOL_GPL(segment_base);
 123
 124 static inline int valid_vcpu(int n)
 125 {
 126         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 127 }
 128
 129 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 130                    void *dest)
 131 {
 132         unsigned char *host_buf = dest;
 133         unsigned long req_size = size;
 134
 135         while (size) {
 136                 hpa_t paddr;
 137                 unsigned now;
 138                 unsigned offset;
 139                 hva_t guest_buf;
 140
 141                 paddr = gva_to_hpa(vcpu, addr);
 142
 143                 if (is_error_hpa(paddr))
 144                         break;
 145
 146                 guest_buf = (hva_t)kmap_atomic(
 147                                         pfn_to_page(paddr >> PAGE_SHIFT),
 148                                         KM_USER0);
 149                 offset = addr & ~PAGE_MASK;
 150                 guest_buf |= offset;
 151                 now = min(size, PAGE_SIZE - offset);
 152                 memcpy(host_buf, (void*)guest_buf, now);
 153                 host_buf += now;
 154                 addr += now;
 155                 size -= now;
 156                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 157         }
 158         return req_size - size;
 159 }
 160 EXPORT_SYMBOL_GPL(kvm_read_guest);
 161
 162 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 163                     void *data)
 164 {
 165         unsigned char *host_buf = data;
 166         unsigned long req_size = size;
 167
 168         while (size) {
 169                 hpa_t paddr;
 170                 unsigned now;
 171                 unsigned offset;
 172                 hva_t guest_buf;
 173
 174                 paddr = gva_to_hpa(vcpu, addr);
 175
 176                 if (is_error_hpa(paddr))
 177                         break;
 178
 179                 guest_buf = (hva_t)kmap_atomic(
 180                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 181                 offset = addr & ~PAGE_MASK;
 182                 guest_buf |= offset;
 183                 now = min(size, PAGE_SIZE - offset);
 184                 memcpy((void*)guest_buf, host_buf, now);
 185                 host_buf += now;
 186                 addr += now;
 187                 size -= now;
 188                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 189         }
 190         return req_size - size;
 191 }
 192 EXPORT_SYMBOL_GPL(kvm_write_guest);
 193
 194 static int vcpu_slot(struct kvm_vcpu *vcpu)
 195 {
 196         return vcpu - vcpu->kvm->vcpus;
 197 }
 198
 199 /*
 200  * Switches to specified vcpu, until a matching vcpu_put()
 201  */
 202 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
 203 {
 204         struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
 205
 206         mutex_lock(&vcpu->mutex);
 207         if (unlikely(!vcpu->vmcs)) {
 208                 mutex_unlock(&vcpu->mutex);
 209                 return NULL;
 210         }
 211         return kvm_arch_ops->vcpu_load(vcpu);
 212 }
 213
 214 static void vcpu_put(struct kvm_vcpu *vcpu)
 215 {
 216         kvm_arch_ops->vcpu_put(vcpu);
 217         mutex_unlock(&vcpu->mutex);
 218 }
 219
 220 static int kvm_dev_open(struct inode *inode, struct file *filp)
 221 {
 222         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 223         int i;
 224
 225         if (!kvm)
 226                 return -ENOMEM;
 227
 228         spin_lock_init(&kvm->lock);
 229         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 230         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 231                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 232
 233                 mutex_init(&vcpu->mutex);
 234                 vcpu->cpu = -1;
 235                 vcpu->kvm = kvm;
 236                 vcpu->mmu.root_hpa = INVALID_PAGE;
 237                 INIT_LIST_HEAD(&vcpu->free_pages);
 238                 spin_lock(&kvm_lock);
 239                 list_add(&kvm->vm_list, &vm_list);
 240                 spin_unlock(&kvm_lock);
 241         }
 242         filp->private_data = kvm;
 243         return 0;
 244 }
 245
 246 /*
 247  * Free any memory in @free but not in @dont.
 248  */
 249 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 250                                   struct kvm_memory_slot *dont)
 251 {
 252         int i;
 253
 254         if (!dont || free->phys_mem != dont->phys_mem)
 255                 if (free->phys_mem) {
 256                         for (i = 0; i < free->npages; ++i)
 257                                 if (free->phys_mem[i])
 258                                         __free_page(free->phys_mem[i]);
 259                         vfree(free->phys_mem);
 260                 }
 261
 262         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 263                 vfree(free->dirty_bitmap);
 264
 265         free->phys_mem = NULL;
 266         free->npages = 0;
 267         free->dirty_bitmap = NULL;
 268 }
 269
 270 static void kvm_free_physmem(struct kvm *kvm)
 271 {
 272         int i;
 273
 274         for (i = 0; i < kvm->nmemslots; ++i)
 275                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 276 }
 277
 278 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 279 {
 280         if (!vcpu_load(vcpu->kvm, vcpu_slot(vcpu)))
 281                 return;
 282
 283         kvm_mmu_destroy(vcpu);
 284         vcpu_put(vcpu);
 285         kvm_arch_ops->vcpu_free(vcpu);
 286 }
 287
 288 static void kvm_free_vcpus(struct kvm *kvm)
 289 {
 290         unsigned int i;
 291
 292         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 293                 kvm_free_vcpu(&kvm->vcpus[i]);
 294 }
 295
 296 static int kvm_dev_release(struct inode *inode, struct file *filp)
 297 {
 298         struct kvm *kvm = filp->private_data;
 299
 300         spin_lock(&kvm_lock);
 301         list_del(&kvm->vm_list);
 302         spin_unlock(&kvm_lock);
 303         kvm_free_vcpus(kvm);
 304         kvm_free_physmem(kvm);
 305         kfree(kvm);
 306         return 0;
 307 }
 308
 309 static void inject_gp(struct kvm_vcpu *vcpu)
 310 {
 311         kvm_arch_ops->inject_gp(vcpu, 0);
 312 }
 313
 314 /*
 315  * Load the pae pdptrs.  Return true is they are all valid.
 316  */
 317 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 318 {
 319         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 320         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 321         int i;
 322         u64 pdpte;
 323         u64 *pdpt;
 324         int ret;
 325         struct kvm_memory_slot *memslot;
 326
 327         spin_lock(&vcpu->kvm->lock);
 328         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 329         /* FIXME: !memslot - emulate? 0xff? */
 330         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 331
 332         ret = 1;
 333         for (i = 0; i < 4; ++i) {
 334                 pdpte = pdpt[offset + i];
 335                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 336                         ret = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         for (i = 0; i < 4; ++i)
 342                 vcpu->pdptrs[i] = pdpt[offset + i];
 343
 344 out:
 345         kunmap_atomic(pdpt, KM_USER0);
 346         spin_unlock(&vcpu->kvm->lock);
 347
 348         return ret;
 349 }
 350
 351 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 352 {
 353         if (cr0 & CR0_RESEVED_BITS) {
 354                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 355                        cr0, vcpu->cr0);
 356                 inject_gp(vcpu);
 357                 return;
 358         }
 359
 360         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 361                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 362                 inject_gp(vcpu);
 363                 return;
 364         }
 365
 366         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 367                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 368                        "and a clear PE flag\n");
 369                 inject_gp(vcpu);
 370                 return;
 371         }
 372
 373         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 374 #ifdef CONFIG_X86_64
 375                 if ((vcpu->shadow_efer & EFER_LME)) {
 376                         int cs_db, cs_l;
 377
 378                         if (!is_pae(vcpu)) {
 379                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 380                                        "in long mode while PAE is disabled\n");
 381                                 inject_gp(vcpu);
 382                                 return;
 383                         }
 384                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 385                         if (cs_l) {
 386                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 387                                        "in long mode while CS.L == 1\n");
 388                                 inject_gp(vcpu);
 389                                 return;
 390
 391                         }
 392                 } else
 393 #endif
 394                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 395                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 396                                "reserved bits\n");
 397                         inject_gp(vcpu);
 398                         return;
 399                 }
 400
 401         }
 402
 403         kvm_arch_ops->set_cr0(vcpu, cr0);
 404         vcpu->cr0 = cr0;
 405
 406         spin_lock(&vcpu->kvm->lock);
 407         kvm_mmu_reset_context(vcpu);
 408         spin_unlock(&vcpu->kvm->lock);
 409         return;
 410 }
 411 EXPORT_SYMBOL_GPL(set_cr0);
 412
 413 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 414 {
 415         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 416         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 417 }
 418 EXPORT_SYMBOL_GPL(lmsw);
 419
 420 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 421 {
 422         if (cr4 & CR4_RESEVED_BITS) {
 423                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 424                 inject_gp(vcpu);
 425                 return;
 426         }
 427
 428         if (is_long_mode(vcpu)) {
 429                 if (!(cr4 & CR4_PAE_MASK)) {
 430                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 431                                "in long mode\n");
 432                         inject_gp(vcpu);
 433                         return;
 434                 }
 435         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 436                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 437                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 438                 inject_gp(vcpu);
 439         }
 440
 441         if (cr4 & CR4_VMXE_MASK) {
 442                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 443                 inject_gp(vcpu);
 444                 return;
 445         }
 446         kvm_arch_ops->set_cr4(vcpu, cr4);
 447         spin_lock(&vcpu->kvm->lock);
 448         kvm_mmu_reset_context(vcpu);
 449         spin_unlock(&vcpu->kvm->lock);
 450 }
 451 EXPORT_SYMBOL_GPL(set_cr4);
 452
 453 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 454 {
 455         if (is_long_mode(vcpu)) {
 456                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 457                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 458                         inject_gp(vcpu);
 459                         return;
 460                 }
 461         } else {
 462                 if (cr3 & CR3_RESEVED_BITS) {
 463                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 464                         inject_gp(vcpu);
 465                         return;
 466                 }
 467                 if (is_paging(vcpu) && is_pae(vcpu) &&
 468                     !load_pdptrs(vcpu, cr3)) {
 469                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 470                                "reserved bits\n");
 471                         inject_gp(vcpu);
 472                         return;
 473                 }
 474         }
 475
 476         vcpu->cr3 = cr3;
 477         spin_lock(&vcpu->kvm->lock);
 478         /*
 479          * Does the new cr3 value map to physical memory? (Note, we
 480          * catch an invalid cr3 even in real-mode, because it would
 481          * cause trouble later on when we turn on paging anyway.)
 482          *
 483          * A real CPU would silently accept an invalid cr3 and would
 484          * attempt to use it - with largely undefined (and often hard
 485          * to debug) behavior on the guest side.
 486          */
 487         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 488                 inject_gp(vcpu);
 489         else
 490                 vcpu->mmu.new_cr3(vcpu);
 491         spin_unlock(&vcpu->kvm->lock);
 492 }
 493 EXPORT_SYMBOL_GPL(set_cr3);
 494
 495 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 496 {
 497         if ( cr8 & CR8_RESEVED_BITS) {
 498                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 499                 inject_gp(vcpu);
 500                 return;
 501         }
 502         vcpu->cr8 = cr8;
 503 }
 504 EXPORT_SYMBOL_GPL(set_cr8);
 505
 506 void fx_init(struct kvm_vcpu *vcpu)
 507 {
 508         struct __attribute__ ((__packed__)) fx_image_s {
 509                 u16 control; //fcw
 510                 u16 status; //fsw
 511                 u16 tag; // ftw
 512                 u16 opcode; //fop
 513                 u64 ip; // fpu ip
 514                 u64 operand;// fpu dp
 515                 u32 mxcsr;
 516                 u32 mxcsr_mask;
 517
 518         } *fx_image;
 519
 520         fx_save(vcpu->host_fx_image);
 521         fpu_init();
 522         fx_save(vcpu->guest_fx_image);
 523         fx_restore(vcpu->host_fx_image);
 524
 525         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 526         fx_image->mxcsr = 0x1f80;
 527         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 528                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 529 }
 530 EXPORT_SYMBOL_GPL(fx_init);
 531
 532 /*
 533  * Creates some virtual cpus.  Good luck creating more than one.
 534  */
 535 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
 536 {
 537         int r;
 538         struct kvm_vcpu *vcpu;
 539
 540         r = -EINVAL;
 541         if (!valid_vcpu(n))
 542                 goto out;
 543
 544         vcpu = &kvm->vcpus[n];
 545
 546         mutex_lock(&vcpu->mutex);
 547
 548         if (vcpu->vmcs) {
 549                 mutex_unlock(&vcpu->mutex);
 550                 return -EEXIST;
 551         }
 552
 553         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 554                                            FX_IMAGE_ALIGN);
 555         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 556
 557         r = kvm_arch_ops->vcpu_create(vcpu);
 558         if (r < 0)
 559                 goto out_free_vcpus;
 560
 561         r = kvm_mmu_create(vcpu);
 562         if (r < 0)
 563                 goto out_free_vcpus;
 564
 565         kvm_arch_ops->vcpu_load(vcpu);
 566         r = kvm_mmu_setup(vcpu);
 567         if (r >= 0)
 568                 r = kvm_arch_ops->vcpu_setup(vcpu);
 569         vcpu_put(vcpu);
 570
 571         if (r < 0)
 572                 goto out_free_vcpus;
 573
 574         return 0;
 575
 576 out_free_vcpus:
 577         kvm_free_vcpu(vcpu);
 578         mutex_unlock(&vcpu->mutex);
 579 out:
 580         return r;
 581 }
 582
 583 /*
 584  * Allocate some memory and give it an address in the guest physical address
 585  * space.
 586  *
 587  * Discontiguous memory is allowed, mostly for framebuffers.
 588  */
 589 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
 590                                            struct kvm_memory_region *mem)
 591 {
 592         int r;
 593         gfn_t base_gfn;
 594         unsigned long npages;
 595         unsigned long i;
 596         struct kvm_memory_slot *memslot;
 597         struct kvm_memory_slot old, new;
 598         int memory_config_version;
 599
 600         r = -EINVAL;
 601         /* General sanity checks */
 602         if (mem->memory_size & (PAGE_SIZE - 1))
 603                 goto out;
 604         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 605                 goto out;
 606         if (mem->slot >= KVM_MEMORY_SLOTS)
 607                 goto out;
 608         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 609                 goto out;
 610
 611         memslot = &kvm->memslots[mem->slot];
 612         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 613         npages = mem->memory_size >> PAGE_SHIFT;
 614
 615         if (!npages)
 616                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 617
 618 raced:
 619         spin_lock(&kvm->lock);
 620
 621         memory_config_version = kvm->memory_config_version;
 622         new = old = *memslot;
 623
 624         new.base_gfn = base_gfn;
 625         new.npages = npages;
 626         new.flags = mem->flags;
 627
 628         /* Disallow changing a memory slot's size. */
 629         r = -EINVAL;
 630         if (npages && old.npages && npages != old.npages)
 631                 goto out_unlock;
 632
 633         /* Check for overlaps */
 634         r = -EEXIST;
 635         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 636                 struct kvm_memory_slot *s = &kvm->memslots[i];
 637
 638                 if (s == memslot)
 639                         continue;
 640                 if (!((base_gfn + npages <= s->base_gfn) ||
 641                       (base_gfn >= s->base_gfn + s->npages)))
 642                         goto out_unlock;
 643         }
 644         /*
 645          * Do memory allocations outside lock.  memory_config_version will
 646          * detect any races.
 647          */
 648         spin_unlock(&kvm->lock);
 649
 650         /* Deallocate if slot is being removed */
 651         if (!npages)
 652                 new.phys_mem = NULL;
 653
 654         /* Free page dirty bitmap if unneeded */
 655         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 656                 new.dirty_bitmap = NULL;
 657
 658         r = -ENOMEM;
 659
 660         /* Allocate if a slot is being created */
 661         if (npages && !new.phys_mem) {
 662                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 663
 664                 if (!new.phys_mem)
 665                         goto out_free;
 666
 667                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 668                 for (i = 0; i < npages; ++i) {
 669                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 670                                                      | __GFP_ZERO);
 671                         if (!new.phys_mem[i])
 672                                 goto out_free;
 673                         set_page_private(new.phys_mem[i],0);
 674                 }
 675         }
 676
 677         /* Allocate page dirty bitmap if needed */
 678         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 679                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 680
 681                 new.dirty_bitmap = vmalloc(dirty_bytes);
 682                 if (!new.dirty_bitmap)
 683                         goto out_free;
 684                 memset(new.dirty_bitmap, 0, dirty_bytes);
 685         }
 686
 687         spin_lock(&kvm->lock);
 688
 689         if (memory_config_version != kvm->memory_config_version) {
 690                 spin_unlock(&kvm->lock);
 691                 kvm_free_physmem_slot(&new, &old);
 692                 goto raced;
 693         }
 694
 695         r = -EAGAIN;
 696         if (kvm->busy)
 697                 goto out_unlock;
 698
 699         if (mem->slot >= kvm->nmemslots)
 700                 kvm->nmemslots = mem->slot + 1;
 701
 702         *memslot = new;
 703         ++kvm->memory_config_version;
 704
 705         spin_unlock(&kvm->lock);
 706
 707         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 708                 struct kvm_vcpu *vcpu;
 709
 710                 vcpu = vcpu_load(kvm, i);
 711                 if (!vcpu)
 712                         continue;
 713                 kvm_mmu_reset_context(vcpu);
 714                 vcpu_put(vcpu);
 715         }
 716
 717         kvm_free_physmem_slot(&old, &new);
 718         return 0;
 719
 720 out_unlock:
 721         spin_unlock(&kvm->lock);
 722 out_free:
 723         kvm_free_physmem_slot(&new, &old);
 724 out:
 725         return r;
 726 }
 727
 728 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 729 {
 730         spin_lock(&vcpu->kvm->lock);
 731         kvm_mmu_slot_remove_write_access(vcpu, slot);
 732         spin_unlock(&vcpu->kvm->lock);
 733 }
 734
 735 /*
 736  * Get (and clear) the dirty memory log for a memory slot.
 737  */
 738 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
 739                                        struct kvm_dirty_log *log)
 740 {
 741         struct kvm_memory_slot *memslot;
 742         int r, i;
 743         int n;
 744         int cleared;
 745         unsigned long any = 0;
 746
 747         spin_lock(&kvm->lock);
 748
 749         /*
 750          * Prevent changes to guest memory configuration even while the lock
 751          * is not taken.
 752          */
 753         ++kvm->busy;
 754         spin_unlock(&kvm->lock);
 755         r = -EINVAL;
 756         if (log->slot >= KVM_MEMORY_SLOTS)
 757                 goto out;
 758
 759         memslot = &kvm->memslots[log->slot];
 760         r = -ENOENT;
 761         if (!memslot->dirty_bitmap)
 762                 goto out;
 763
 764         n = ALIGN(memslot->npages, 8) / 8;
 765
 766         for (i = 0; !any && i < n; ++i)
 767                 any = memslot->dirty_bitmap[i];
 768
 769         r = -EFAULT;
 770         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 771                 goto out;
 772
 773         if (any) {
 774                 cleared = 0;
 775                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 776                         struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
 777
 778                         if (!vcpu)
 779                                 continue;
 780                         if (!cleared) {
 781                                 do_remove_write_access(vcpu, log->slot);
 782                                 memset(memslot->dirty_bitmap, 0, n);
 783                                 cleared = 1;
 784                         }
 785                         kvm_arch_ops->tlb_flush(vcpu);
 786                         vcpu_put(vcpu);
 787                 }
 788         }
 789
 790         r = 0;
 791
 792 out:
 793         spin_lock(&kvm->lock);
 794         --kvm->busy;
 795         spin_unlock(&kvm->lock);
 796         return r;
 797 }
 798
 799 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 800 {
 801         int i;
 802
 803         for (i = 0; i < kvm->nmemslots; ++i) {
 804                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 805
 806                 if (gfn >= memslot->base_gfn
 807                     && gfn < memslot->base_gfn + memslot->npages)
 808                         return memslot;
 809         }
 810         return NULL;
 811 }
 812 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 813
 814 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 815 {
 816         int i;
 817         struct kvm_memory_slot *memslot = NULL;
 818         unsigned long rel_gfn;
 819
 820         for (i = 0; i < kvm->nmemslots; ++i) {
 821                 memslot = &kvm->memslots[i];
 822
 823                 if (gfn >= memslot->base_gfn
 824                     && gfn < memslot->base_gfn + memslot->npages) {
 825
 826                         if (!memslot || !memslot->dirty_bitmap)
 827                                 return;
 828
 829                         rel_gfn = gfn - memslot->base_gfn;
 830
 831                         /* avoid RMW */
 832                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 833                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 834                         return;
 835                 }
 836         }
 837 }
 838
 839 static int emulator_read_std(unsigned long addr,
 840                              unsigned long *val,
 841                              unsigned int bytes,
 842                              struct x86_emulate_ctxt *ctxt)
 843 {
 844         struct kvm_vcpu *vcpu = ctxt->vcpu;
 845         void *data = val;
 846
 847         while (bytes) {
 848                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 849                 unsigned offset = addr & (PAGE_SIZE-1);
 850                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 851                 unsigned long pfn;
 852                 struct kvm_memory_slot *memslot;
 853                 void *page;
 854
 855                 if (gpa == UNMAPPED_GVA)
 856                         return X86EMUL_PROPAGATE_FAULT;
 857                 pfn = gpa >> PAGE_SHIFT;
 858                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 859                 if (!memslot)
 860                         return X86EMUL_UNHANDLEABLE;
 861                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 862
 863                 memcpy(data, page + offset, tocopy);
 864
 865                 kunmap_atomic(page, KM_USER0);
 866
 867                 bytes -= tocopy;
 868                 data += tocopy;
 869                 addr += tocopy;
 870         }
 871
 872         return X86EMUL_CONTINUE;
 873 }
 874
 875 static int emulator_write_std(unsigned long addr,
 876                               unsigned long val,
 877                               unsigned int bytes,
 878                               struct x86_emulate_ctxt *ctxt)
 879 {
 880         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 881                addr, bytes);
 882         return X86EMUL_UNHANDLEABLE;
 883 }
 884
 885 static int emulator_read_emulated(unsigned long addr,
 886                                   unsigned long *val,
 887                                   unsigned int bytes,
 888                                   struct x86_emulate_ctxt *ctxt)
 889 {
 890         struct kvm_vcpu *vcpu = ctxt->vcpu;
 891
 892         if (vcpu->mmio_read_completed) {
 893                 memcpy(val, vcpu->mmio_data, bytes);
 894                 vcpu->mmio_read_completed = 0;
 895                 return X86EMUL_CONTINUE;
 896         } else if (emulator_read_std(addr, val, bytes, ctxt)
 897                    == X86EMUL_CONTINUE)
 898                 return X86EMUL_CONTINUE;
 899         else {
 900                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 901
 902                 if (gpa == UNMAPPED_GVA)
 903                         return X86EMUL_PROPAGATE_FAULT;
 904                 vcpu->mmio_needed = 1;
 905                 vcpu->mmio_phys_addr = gpa;
 906                 vcpu->mmio_size = bytes;
 907                 vcpu->mmio_is_write = 0;
 908
 909                 return X86EMUL_UNHANDLEABLE;
 910         }
 911 }
 912
 913 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 914                                unsigned long val, int bytes)
 915 {
 916         struct kvm_memory_slot *m;
 917         struct page *page;
 918         void *virt;
 919
 920         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 921                 return 0;
 922         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 923         if (!m)
 924                 return 0;
 925         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 926         kvm_mmu_pre_write(vcpu, gpa, bytes);
 927         virt = kmap_atomic(page, KM_USER0);
 928         memcpy(virt + offset_in_page(gpa), &val, bytes);
 929         kunmap_atomic(virt, KM_USER0);
 930         kvm_mmu_post_write(vcpu, gpa, bytes);
 931         return 1;
 932 }
 933
 934 static int emulator_write_emulated(unsigned long addr,
 935                                    unsigned long val,
 936                                    unsigned int bytes,
 937                                    struct x86_emulate_ctxt *ctxt)
 938 {
 939         struct kvm_vcpu *vcpu = ctxt->vcpu;
 940         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 941
 942         if (gpa == UNMAPPED_GVA)
 943                 return X86EMUL_PROPAGATE_FAULT;
 944
 945         if (emulator_write_phys(vcpu, gpa, val, bytes))
 946                 return X86EMUL_CONTINUE;
 947
 948         vcpu->mmio_needed = 1;
 949         vcpu->mmio_phys_addr = gpa;
 950         vcpu->mmio_size = bytes;
 951         vcpu->mmio_is_write = 1;
 952         memcpy(vcpu->mmio_data, &val, bytes);
 953
 954         return X86EMUL_CONTINUE;
 955 }
 956
 957 static int emulator_cmpxchg_emulated(unsigned long addr,
 958                                      unsigned long old,
 959                                      unsigned long new,
 960                                      unsigned int bytes,
 961                                      struct x86_emulate_ctxt *ctxt)
 962 {
 963         static int reported;
 964
 965         if (!reported) {
 966                 reported = 1;
 967                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
 968         }
 969         return emulator_write_emulated(addr, new, bytes, ctxt);
 970 }
 971
 972 #ifdef CONFIG_X86_32
 973
 974 static int emulator_cmpxchg8b_emulated(unsigned long addr,
 975                                        unsigned long old_lo,
 976                                        unsigned long old_hi,
 977                                        unsigned long new_lo,
 978                                        unsigned long new_hi,
 979                                        struct x86_emulate_ctxt *ctxt)
 980 {
 981         static int reported;
 982         int r;
 983
 984         if (!reported) {
 985                 reported = 1;
 986                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
 987         }
 988         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
 989         if (r != X86EMUL_CONTINUE)
 990                 return r;
 991         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
 992 }
 993
 994 #endif
 995
 996 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
 997 {
 998         return kvm_arch_ops->get_segment_base(vcpu, seg);
 999 }
1000
1001 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1002 {
1003         return X86EMUL_CONTINUE;
1004 }
1005
1006 int emulate_clts(struct kvm_vcpu *vcpu)
1007 {
1008         unsigned long cr0;
1009
1010         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1011         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1012         kvm_arch_ops->set_cr0(vcpu, cr0);
1013         return X86EMUL_CONTINUE;
1014 }
1015
1016 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1017 {
1018         struct kvm_vcpu *vcpu = ctxt->vcpu;
1019
1020         switch (dr) {
1021         case 0 ... 3:
1022                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1023                 return X86EMUL_CONTINUE;
1024         default:
1025                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1026                        __FUNCTION__, dr);
1027                 return X86EMUL_UNHANDLEABLE;
1028         }
1029 }
1030
1031 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1032 {
1033         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1034         int exception;
1035
1036         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1037         if (exception) {
1038                 /* FIXME: better handling */
1039                 return X86EMUL_UNHANDLEABLE;
1040         }
1041         return X86EMUL_CONTINUE;
1042 }
1043
1044 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1045 {
1046         static int reported;
1047         u8 opcodes[4];
1048         unsigned long rip = ctxt->vcpu->rip;
1049         unsigned long rip_linear;
1050
1051         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1052
1053         if (reported)
1054                 return;
1055
1056         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1057
1058         printk(KERN_ERR "emulation failed but !mmio_needed?"
1059                " rip %lx %02x %02x %02x %02x\n",
1060                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1061         reported = 1;
1062 }
1063
1064 struct x86_emulate_ops emulate_ops = {
1065         .read_std            = emulator_read_std,
1066         .write_std           = emulator_write_std,
1067         .read_emulated       = emulator_read_emulated,
1068         .write_emulated      = emulator_write_emulated,
1069         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1070 #ifdef CONFIG_X86_32
1071         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1072 #endif
1073 };
1074
1075 int emulate_instruction(struct kvm_vcpu *vcpu,
1076                         struct kvm_run *run,
1077                         unsigned long cr2,
1078                         u16 error_code)
1079 {
1080         struct x86_emulate_ctxt emulate_ctxt;
1081         int r;
1082         int cs_db, cs_l;
1083
1084         kvm_arch_ops->cache_regs(vcpu);
1085
1086         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1087
1088         emulate_ctxt.vcpu = vcpu;
1089         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1090         emulate_ctxt.cr2 = cr2;
1091         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1092                 ? X86EMUL_MODE_REAL : cs_l
1093                 ? X86EMUL_MODE_PROT64 : cs_db
1094                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1095
1096         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1097                 emulate_ctxt.cs_base = 0;
1098                 emulate_ctxt.ds_base = 0;
1099                 emulate_ctxt.es_base = 0;
1100                 emulate_ctxt.ss_base = 0;
1101         } else {
1102                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1103                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1104                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1105                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1106         }
1107
1108         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1109         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1110
1111         vcpu->mmio_is_write = 0;
1112         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1113
1114         if ((r || vcpu->mmio_is_write) && run) {
1115                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1116                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1117                 run->mmio.len = vcpu->mmio_size;
1118                 run->mmio.is_write = vcpu->mmio_is_write;
1119         }
1120
1121         if (r) {
1122                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1123                         return EMULATE_DONE;
1124                 if (!vcpu->mmio_needed) {
1125                         report_emulation_failure(&emulate_ctxt);
1126                         return EMULATE_FAIL;
1127                 }
1128                 return EMULATE_DO_MMIO;
1129         }
1130
1131         kvm_arch_ops->decache_regs(vcpu);
1132         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1133
1134         if (vcpu->mmio_is_write)
1135                 return EMULATE_DO_MMIO;
1136
1137         return EMULATE_DONE;
1138 }
1139 EXPORT_SYMBOL_GPL(emulate_instruction);
1140
1141 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1142 {
1143         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1144 }
1145
1146 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1147 {
1148         struct descriptor_table dt = { limit, base };
1149
1150         kvm_arch_ops->set_gdt(vcpu, &dt);
1151 }
1152
1153 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1154 {
1155         struct descriptor_table dt = { limit, base };
1156
1157         kvm_arch_ops->set_idt(vcpu, &dt);
1158 }
1159
1160 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1161                    unsigned long *rflags)
1162 {
1163         lmsw(vcpu, msw);
1164         *rflags = kvm_arch_ops->get_rflags(vcpu);
1165 }
1166
1167 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1168 {
1169         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1170         switch (cr) {
1171         case 0:
1172                 return vcpu->cr0;
1173         case 2:
1174                 return vcpu->cr2;
1175         case 3:
1176                 return vcpu->cr3;
1177         case 4:
1178                 return vcpu->cr4;
1179         default:
1180                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1181                 return 0;
1182         }
1183 }
1184
1185 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1186                      unsigned long *rflags)
1187 {
1188         switch (cr) {
1189         case 0:
1190                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1191                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1192                 break;
1193         case 2:
1194                 vcpu->cr2 = val;
1195                 break;
1196         case 3:
1197                 set_cr3(vcpu, val);
1198                 break;
1199         case 4:
1200                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1201                 break;
1202         default:
1203                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1204         }
1205 }
1206
1207 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1208 {
1209         u64 data;
1210
1211         switch (msr) {
1212         case 0xc0010010: /* SYSCFG */
1213         case 0xc0010015: /* HWCR */
1214         case MSR_IA32_PLATFORM_ID:
1215         case MSR_IA32_P5_MC_ADDR:
1216         case MSR_IA32_P5_MC_TYPE:
1217         case MSR_IA32_MC0_CTL:
1218         case MSR_IA32_MCG_STATUS:
1219         case MSR_IA32_MCG_CAP:
1220         case MSR_IA32_MC0_MISC:
1221         case MSR_IA32_MC0_MISC+4:
1222         case MSR_IA32_MC0_MISC+8:
1223         case MSR_IA32_MC0_MISC+12:
1224         case MSR_IA32_MC0_MISC+16:
1225         case MSR_IA32_UCODE_REV:
1226         case MSR_IA32_PERF_STATUS:
1227                 /* MTRR registers */
1228         case 0xfe:
1229         case 0x200 ... 0x2ff:
1230                 data = 0;
1231                 break;
1232         case 0xcd: /* fsb frequency */
1233                 data = 3;
1234                 break;
1235         case MSR_IA32_APICBASE:
1236                 data = vcpu->apic_base;
1237                 break;
1238         case MSR_IA32_MISC_ENABLE:
1239                 data = vcpu->ia32_misc_enable_msr;
1240                 break;
1241 #ifdef CONFIG_X86_64
1242         case MSR_EFER:
1243                 data = vcpu->shadow_efer;
1244                 break;
1245 #endif
1246         default:
1247                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1248                 return 1;
1249         }
1250         *pdata = data;
1251         return 0;
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1254
1255 /*
1256  * Reads an msr value (of 'msr_index') into 'pdata'.
1257  * Returns 0 on success, non-0 otherwise.
1258  * Assumes vcpu_load() was already called.
1259  */
1260 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1261 {
1262         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1263 }
1264
1265 #ifdef CONFIG_X86_64
1266
1267 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1268 {
1269         if (efer & EFER_RESERVED_BITS) {
1270                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1271                        efer);
1272                 inject_gp(vcpu);
1273                 return;
1274         }
1275
1276         if (is_paging(vcpu)
1277             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1278                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1279                 inject_gp(vcpu);
1280                 return;
1281         }
1282
1283         kvm_arch_ops->set_efer(vcpu, efer);
1284
1285         efer &= ~EFER_LMA;
1286         efer |= vcpu->shadow_efer & EFER_LMA;
1287
1288         vcpu->shadow_efer = efer;
1289 }
1290
1291 #endif
1292
1293 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1294 {
1295         switch (msr) {
1296 #ifdef CONFIG_X86_64
1297         case MSR_EFER:
1298                 set_efer(vcpu, data);
1299                 break;
1300 #endif
1301         case MSR_IA32_MC0_STATUS:
1302                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1303                        __FUNCTION__, data);
1304                 break;
1305         case MSR_IA32_UCODE_REV:
1306         case MSR_IA32_UCODE_WRITE:
1307         case 0x200 ... 0x2ff: /* MTRRs */
1308                 break;
1309         case MSR_IA32_APICBASE:
1310                 vcpu->apic_base = data;
1311                 break;
1312         case MSR_IA32_MISC_ENABLE:
1313                 vcpu->ia32_misc_enable_msr = data;
1314                 break;
1315         default:
1316                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1317                 return 1;
1318         }
1319         return 0;
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1322
1323 /*
1324  * Writes msr value into into the appropriate "register".
1325  * Returns 0 on success, non-0 otherwise.
1326  * Assumes vcpu_load() was already called.
1327  */
1328 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1329 {
1330         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1331 }
1332
1333 void kvm_resched(struct kvm_vcpu *vcpu)
1334 {
1335         vcpu_put(vcpu);
1336         cond_resched();
1337         /* Cannot fail -  no vcpu unplug yet. */
1338         vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1339 }
1340 EXPORT_SYMBOL_GPL(kvm_resched);
1341
1342 void load_msrs(struct vmx_msr_entry *e, int n)
1343 {
1344         int i;
1345
1346         for (i = 0; i < n; ++i)
1347                 wrmsrl(e[i].index, e[i].data);
1348 }
1349 EXPORT_SYMBOL_GPL(load_msrs);
1350
1351 void save_msrs(struct vmx_msr_entry *e, int n)
1352 {
1353         int i;
1354
1355         for (i = 0; i < n; ++i)
1356                 rdmsrl(e[i].index, e[i].data);
1357 }
1358 EXPORT_SYMBOL_GPL(save_msrs);
1359
1360 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1361 {
1362         struct kvm_vcpu *vcpu;
1363         int r;
1364
1365         if (!valid_vcpu(kvm_run->vcpu))
1366                 return -EINVAL;
1367
1368         vcpu = vcpu_load(kvm, kvm_run->vcpu);
1369         if (!vcpu)
1370                 return -ENOENT;
1371
1372         /* re-sync apic's tpr */
1373         vcpu->cr8 = kvm_run->cr8;
1374
1375         if (kvm_run->emulated) {
1376                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1377                 kvm_run->emulated = 0;
1378         }
1379
1380         if (kvm_run->mmio_completed) {
1381                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1382                 vcpu->mmio_read_completed = 1;
1383         }
1384
1385         vcpu->mmio_needed = 0;
1386
1387         r = kvm_arch_ops->run(vcpu, kvm_run);
1388
1389         vcpu_put(vcpu);
1390         return r;
1391 }
1392
1393 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1394 {
1395         struct kvm_vcpu *vcpu;
1396
1397         if (!valid_vcpu(regs->vcpu))
1398                 return -EINVAL;
1399
1400         vcpu = vcpu_load(kvm, regs->vcpu);
1401         if (!vcpu)
1402                 return -ENOENT;
1403
1404         kvm_arch_ops->cache_regs(vcpu);
1405
1406         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1407         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1408         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1409         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1410         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1411         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1412         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1413         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1414 #ifdef CONFIG_X86_64
1415         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1416         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1417         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1418         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1419         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1420         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1421         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1422         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1423 #endif
1424
1425         regs->rip = vcpu->rip;
1426         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1427
1428         /*
1429          * Don't leak debug flags in case they were set for guest debugging
1430          */
1431         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1432                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1433
1434         vcpu_put(vcpu);
1435
1436         return 0;
1437 }
1438
1439 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1440 {
1441         struct kvm_vcpu *vcpu;
1442
1443         if (!valid_vcpu(regs->vcpu))
1444                 return -EINVAL;
1445
1446         vcpu = vcpu_load(kvm, regs->vcpu);
1447         if (!vcpu)
1448                 return -ENOENT;
1449
1450         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1451         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1452         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1453         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1454         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1455         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1456         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1457         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1458 #ifdef CONFIG_X86_64
1459         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1460         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1461         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1462         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1463         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1464         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1465         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1466         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1467 #endif
1468
1469         vcpu->rip = regs->rip;
1470         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1471
1472         kvm_arch_ops->decache_regs(vcpu);
1473
1474         vcpu_put(vcpu);
1475
1476         return 0;
1477 }
1478
1479 static void get_segment(struct kvm_vcpu *vcpu,
1480                         struct kvm_segment *var, int seg)
1481 {
1482         return kvm_arch_ops->get_segment(vcpu, var, seg);
1483 }
1484
1485 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1486 {
1487         struct kvm_vcpu *vcpu;
1488         struct descriptor_table dt;
1489
1490         if (!valid_vcpu(sregs->vcpu))
1491                 return -EINVAL;
1492         vcpu = vcpu_load(kvm, sregs->vcpu);
1493         if (!vcpu)
1494                 return -ENOENT;
1495
1496         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1497         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1498         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1499         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1500         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1501         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1502
1503         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1504         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1505
1506         kvm_arch_ops->get_idt(vcpu, &dt);
1507         sregs->idt.limit = dt.limit;
1508         sregs->idt.base = dt.base;
1509         kvm_arch_ops->get_gdt(vcpu, &dt);
1510         sregs->gdt.limit = dt.limit;
1511         sregs->gdt.base = dt.base;
1512
1513         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1514         sregs->cr0 = vcpu->cr0;
1515         sregs->cr2 = vcpu->cr2;
1516         sregs->cr3 = vcpu->cr3;
1517         sregs->cr4 = vcpu->cr4;
1518         sregs->cr8 = vcpu->cr8;
1519         sregs->efer = vcpu->shadow_efer;
1520         sregs->apic_base = vcpu->apic_base;
1521
1522         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1523                sizeof sregs->interrupt_bitmap);
1524
1525         vcpu_put(vcpu);
1526
1527         return 0;
1528 }
1529
1530 static void set_segment(struct kvm_vcpu *vcpu,
1531                         struct kvm_segment *var, int seg)
1532 {
1533         return kvm_arch_ops->set_segment(vcpu, var, seg);
1534 }
1535
1536 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1537 {
1538         struct kvm_vcpu *vcpu;
1539         int mmu_reset_needed = 0;
1540         int i;
1541         struct descriptor_table dt;
1542
1543         if (!valid_vcpu(sregs->vcpu))
1544                 return -EINVAL;
1545         vcpu = vcpu_load(kvm, sregs->vcpu);
1546         if (!vcpu)
1547                 return -ENOENT;
1548
1549         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1550         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1551         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1552         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1553         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1554         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1555
1556         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1557         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1558
1559         dt.limit = sregs->idt.limit;
1560         dt.base = sregs->idt.base;
1561         kvm_arch_ops->set_idt(vcpu, &dt);
1562         dt.limit = sregs->gdt.limit;
1563         dt.base = sregs->gdt.base;
1564         kvm_arch_ops->set_gdt(vcpu, &dt);
1565
1566         vcpu->cr2 = sregs->cr2;
1567         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1568         vcpu->cr3 = sregs->cr3;
1569
1570         vcpu->cr8 = sregs->cr8;
1571
1572         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1573 #ifdef CONFIG_X86_64
1574         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1575 #endif
1576         vcpu->apic_base = sregs->apic_base;
1577
1578         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1579
1580         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1581         kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1582
1583         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1584         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1585         if (!is_long_mode(vcpu) && is_pae(vcpu))
1586                 load_pdptrs(vcpu, vcpu->cr3);
1587
1588         if (mmu_reset_needed)
1589                 kvm_mmu_reset_context(vcpu);
1590
1591         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1592                sizeof vcpu->irq_pending);
1593         vcpu->irq_summary = 0;
1594         for (i = 0; i < NR_IRQ_WORDS; ++i)
1595                 if (vcpu->irq_pending[i])
1596                         __set_bit(i, &vcpu->irq_summary);
1597
1598         vcpu_put(vcpu);
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1605  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1606  *
1607  * This list is modified at module load time to reflect the
1608  * capabilities of the host cpu.
1609  */
1610 static u32 msrs_to_save[] = {
1611         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1612         MSR_K6_STAR,
1613 #ifdef CONFIG_X86_64
1614         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1615 #endif
1616         MSR_IA32_TIME_STAMP_COUNTER,
1617 };
1618
1619 static unsigned num_msrs_to_save;
1620
1621 static u32 emulated_msrs[] = {
1622         MSR_IA32_MISC_ENABLE,
1623 };
1624
1625 static __init void kvm_init_msr_list(void)
1626 {
1627         u32 dummy[2];
1628         unsigned i, j;
1629
1630         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1631                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1632                         continue;
1633                 if (j < i)
1634                         msrs_to_save[j] = msrs_to_save[i];
1635                 j++;
1636         }
1637         num_msrs_to_save = j;
1638 }
1639
1640 /*
1641  * Adapt set_msr() to msr_io()'s calling convention
1642  */
1643 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1644 {
1645         return set_msr(vcpu, index, *data);
1646 }
1647
1648 /*
1649  * Read or write a bunch of msrs. All parameters are kernel addresses.
1650  *
1651  * @return number of msrs set successfully.
1652  */
1653 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1654                     struct kvm_msr_entry *entries,
1655                     int (*do_msr)(struct kvm_vcpu *vcpu,
1656                                   unsigned index, u64 *data))
1657 {
1658         struct kvm_vcpu *vcpu;
1659         int i;
1660
1661         if (!valid_vcpu(msrs->vcpu))
1662                 return -EINVAL;
1663
1664         vcpu = vcpu_load(kvm, msrs->vcpu);
1665         if (!vcpu)
1666                 return -ENOENT;
1667
1668         for (i = 0; i < msrs->nmsrs; ++i)
1669                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1670                         break;
1671
1672         vcpu_put(vcpu);
1673
1674         return i;
1675 }
1676
1677 /*
1678  * Read or write a bunch of msrs. Parameters are user addresses.
1679  *
1680  * @return number of msrs set successfully.
1681  */
1682 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1683                   int (*do_msr)(struct kvm_vcpu *vcpu,
1684                                 unsigned index, u64 *data),
1685                   int writeback)
1686 {
1687         struct kvm_msrs msrs;
1688         struct kvm_msr_entry *entries;
1689         int r, n;
1690         unsigned size;
1691
1692         r = -EFAULT;
1693         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1694                 goto out;
1695
1696         r = -E2BIG;
1697         if (msrs.nmsrs >= MAX_IO_MSRS)
1698                 goto out;
1699
1700         r = -ENOMEM;
1701         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1702         entries = vmalloc(size);
1703         if (!entries)
1704                 goto out;
1705
1706         r = -EFAULT;
1707         if (copy_from_user(entries, user_msrs->entries, size))
1708                 goto out_free;
1709
1710         r = n = __msr_io(kvm, &msrs, entries, do_msr);
1711         if (r < 0)
1712                 goto out_free;
1713
1714         r = -EFAULT;
1715         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1716                 goto out_free;
1717
1718         r = n;
1719
1720 out_free:
1721         vfree(entries);
1722 out:
1723         return r;
1724 }
1725
1726 /*
1727  * Translate a guest virtual address to a guest physical address.
1728  */
1729 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1730 {
1731         unsigned long vaddr = tr->linear_address;
1732         struct kvm_vcpu *vcpu;
1733         gpa_t gpa;
1734
1735         vcpu = vcpu_load(kvm, tr->vcpu);
1736         if (!vcpu)
1737                 return -ENOENT;
1738         spin_lock(&kvm->lock);
1739         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1740         tr->physical_address = gpa;
1741         tr->valid = gpa != UNMAPPED_GVA;
1742         tr->writeable = 1;
1743         tr->usermode = 0;
1744         spin_unlock(&kvm->lock);
1745         vcpu_put(vcpu);
1746
1747         return 0;
1748 }
1749
1750 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1751 {
1752         struct kvm_vcpu *vcpu;
1753
1754         if (!valid_vcpu(irq->vcpu))
1755                 return -EINVAL;
1756         if (irq->irq < 0 || irq->irq >= 256)
1757                 return -EINVAL;
1758         vcpu = vcpu_load(kvm, irq->vcpu);
1759         if (!vcpu)
1760                 return -ENOENT;
1761
1762         set_bit(irq->irq, vcpu->irq_pending);
1763         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1764
1765         vcpu_put(vcpu);
1766
1767         return 0;
1768 }
1769
1770 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1771                                      struct kvm_debug_guest *dbg)
1772 {
1773         struct kvm_vcpu *vcpu;
1774         int r;
1775
1776         if (!valid_vcpu(dbg->vcpu))
1777                 return -EINVAL;
1778         vcpu = vcpu_load(kvm, dbg->vcpu);
1779         if (!vcpu)
1780                 return -ENOENT;
1781
1782         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1783
1784         vcpu_put(vcpu);
1785
1786         return r;
1787 }
1788
1789 static long kvm_dev_ioctl(struct file *filp,
1790                           unsigned int ioctl, unsigned long arg)
1791 {
1792         struct kvm *kvm = filp->private_data;
1793         void __user *argp = (void __user *)arg;
1794         int r = -EINVAL;
1795
1796         switch (ioctl) {
1797         case KVM_GET_API_VERSION:
1798                 r = KVM_API_VERSION;
1799                 break;
1800         case KVM_CREATE_VCPU:
1801                 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1802                 if (r)
1803                         goto out;
1804                 break;
1805         case KVM_RUN: {
1806                 struct kvm_run kvm_run;
1807
1808                 r = -EFAULT;
1809                 if (copy_from_user(&kvm_run, argp, sizeof kvm_run))
1810                         goto out;
1811                 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1812                 if (r < 0 &&  r != -EINTR)
1813                         goto out;
1814                 if (copy_to_user(argp, &kvm_run, sizeof kvm_run)) {
1815                         r = -EFAULT;
1816                         goto out;
1817                 }
1818                 break;
1819         }
1820         case KVM_GET_REGS: {
1821                 struct kvm_regs kvm_regs;
1822
1823                 r = -EFAULT;
1824                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1825                         goto out;
1826                 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1827                 if (r)
1828                         goto out;
1829                 r = -EFAULT;
1830                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_SET_REGS: {
1836                 struct kvm_regs kvm_regs;
1837
1838                 r = -EFAULT;
1839                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1840                         goto out;
1841                 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1842                 if (r)
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_GET_SREGS: {
1848                 struct kvm_sregs kvm_sregs;
1849
1850                 r = -EFAULT;
1851                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1852                         goto out;
1853                 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1854                 if (r)
1855                         goto out;
1856                 r = -EFAULT;
1857                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_SET_SREGS: {
1863                 struct kvm_sregs kvm_sregs;
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1867                         goto out;
1868                 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1869                 if (r)
1870                         goto out;
1871                 r = 0;
1872                 break;
1873         }
1874         case KVM_TRANSLATE: {
1875                 struct kvm_translation tr;
1876
1877                 r = -EFAULT;
1878                 if (copy_from_user(&tr, argp, sizeof tr))
1879                         goto out;
1880                 r = kvm_dev_ioctl_translate(kvm, &tr);
1881                 if (r)
1882                         goto out;
1883                 r = -EFAULT;
1884                 if (copy_to_user(argp, &tr, sizeof tr))
1885                         goto out;
1886                 r = 0;
1887                 break;
1888         }
1889         case KVM_INTERRUPT: {
1890                 struct kvm_interrupt irq;
1891
1892                 r = -EFAULT;
1893                 if (copy_from_user(&irq, argp, sizeof irq))
1894                         goto out;
1895                 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1896                 if (r)
1897                         goto out;
1898                 r = 0;
1899                 break;
1900         }
1901         case KVM_DEBUG_GUEST: {
1902                 struct kvm_debug_guest dbg;
1903
1904                 r = -EFAULT;
1905                 if (copy_from_user(&dbg, argp, sizeof dbg))
1906                         goto out;
1907                 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1908                 if (r)
1909                         goto out;
1910                 r = 0;
1911                 break;
1912         }
1913         case KVM_SET_MEMORY_REGION: {
1914                 struct kvm_memory_region kvm_mem;
1915
1916                 r = -EFAULT;
1917                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1918                         goto out;
1919                 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1920                 if (r)
1921                         goto out;
1922                 break;
1923         }
1924         case KVM_GET_DIRTY_LOG: {
1925                 struct kvm_dirty_log log;
1926
1927                 r = -EFAULT;
1928                 if (copy_from_user(&log, argp, sizeof log))
1929                         goto out;
1930                 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1931                 if (r)
1932                         goto out;
1933                 break;
1934         }
1935         case KVM_GET_MSRS:
1936                 r = msr_io(kvm, argp, get_msr, 1);
1937                 break;
1938         case KVM_SET_MSRS:
1939                 r = msr_io(kvm, argp, do_set_msr, 0);
1940                 break;
1941         case KVM_GET_MSR_INDEX_LIST: {
1942                 struct kvm_msr_list __user *user_msr_list = argp;
1943                 struct kvm_msr_list msr_list;
1944                 unsigned n;
1945
1946                 r = -EFAULT;
1947                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1948                         goto out;
1949                 n = msr_list.nmsrs;
1950                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1951                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1952                         goto out;
1953                 r = -E2BIG;
1954                 if (n < num_msrs_to_save)
1955                         goto out;
1956                 r = -EFAULT;
1957                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1958                                  num_msrs_to_save * sizeof(u32)))
1959                         goto out;
1960                 if (copy_to_user(user_msr_list->indices
1961                                  + num_msrs_to_save * sizeof(u32),
1962                                  &emulated_msrs,
1963                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1964                         goto out;
1965                 r = 0;
1966                 break;
1967         }
1968         default:
1969                 ;
1970         }
1971 out:
1972         return r;
1973 }
1974
1975 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1976                                    unsigned long address,
1977                                    int *type)
1978 {
1979         struct kvm *kvm = vma->vm_file->private_data;
1980         unsigned long pgoff;
1981         struct kvm_memory_slot *slot;
1982         struct page *page;
1983
1984         *type = VM_FAULT_MINOR;
1985         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1986         slot = gfn_to_memslot(kvm, pgoff);
1987         if (!slot)
1988                 return NOPAGE_SIGBUS;
1989         page = gfn_to_page(slot, pgoff);
1990         if (!page)
1991                 return NOPAGE_SIGBUS;
1992         get_page(page);
1993         return page;
1994 }
1995
1996 static struct vm_operations_struct kvm_dev_vm_ops = {
1997         .nopage = kvm_dev_nopage,
1998 };
1999
2000 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
2001 {
2002         vma->vm_ops = &kvm_dev_vm_ops;
2003         return 0;
2004 }
2005
2006 static struct file_operations kvm_chardev_ops = {
2007         .open           = kvm_dev_open,
2008         .release        = kvm_dev_release,
2009         .unlocked_ioctl = kvm_dev_ioctl,
2010         .compat_ioctl   = kvm_dev_ioctl,
2011         .mmap           = kvm_dev_mmap,
2012 };
2013
2014 static struct miscdevice kvm_dev = {
2015         MISC_DYNAMIC_MINOR,
2016         "kvm",
2017         &kvm_chardev_ops,
2018 };
2019
2020 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2021                        void *v)
2022 {
2023         if (val == SYS_RESTART) {
2024                 /*
2025                  * Some (well, at least mine) BIOSes hang on reboot if
2026                  * in vmx root mode.
2027                  */
2028                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2029                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2030         }
2031         return NOTIFY_OK;
2032 }
2033
2034 static struct notifier_block kvm_reboot_notifier = {
2035         .notifier_call = kvm_reboot,
2036         .priority = 0,
2037 };
2038
2039 /*
2040  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2041  * cached on it.
2042  */
2043 static void decache_vcpus_on_cpu(int cpu)
2044 {
2045         struct kvm *vm;
2046         struct kvm_vcpu *vcpu;
2047         int i;
2048
2049         spin_lock(&kvm_lock);
2050         list_for_each_entry(vm, &vm_list, vm_list)
2051                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2052                         vcpu = &vm->vcpus[i];
2053                         /*
2054                          * If the vcpu is locked, then it is running on some
2055                          * other cpu and therefore it is not cached on the
2056                          * cpu in question.
2057                          *
2058                          * If it's not locked, check the last cpu it executed
2059                          * on.
2060                          */
2061                         if (mutex_trylock(&vcpu->mutex)) {
2062                                 if (vcpu->cpu == cpu) {
2063                                         kvm_arch_ops->vcpu_decache(vcpu);
2064                                         vcpu->cpu = -1;
2065                                 }
2066                                 mutex_unlock(&vcpu->mutex);
2067                         }
2068                 }
2069         spin_unlock(&kvm_lock);
2070 }
2071
2072 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2073                            void *v)
2074 {
2075         int cpu = (long)v;
2076
2077         switch (val) {
2078         case CPU_DOWN_PREPARE:
2079         case CPU_UP_CANCELED:
2080                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2081                        cpu);
2082                 decache_vcpus_on_cpu(cpu);
2083                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2084                                          NULL, 0, 1);
2085                 break;
2086         case CPU_ONLINE:
2087                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2088                        cpu);
2089                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2090                                          NULL, 0, 1);
2091                 break;
2092         }
2093         return NOTIFY_OK;
2094 }
2095
2096 static struct notifier_block kvm_cpu_notifier = {
2097         .notifier_call = kvm_cpu_hotplug,
2098         .priority = 20, /* must be > scheduler priority */
2099 };
2100
2101 static __init void kvm_init_debug(void)
2102 {
2103         struct kvm_stats_debugfs_item *p;
2104
2105         debugfs_dir = debugfs_create_dir("kvm", NULL);
2106         for (p = debugfs_entries; p->name; ++p)
2107                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2108                                                p->data);
2109 }
2110
2111 static void kvm_exit_debug(void)
2112 {
2113         struct kvm_stats_debugfs_item *p;
2114
2115         for (p = debugfs_entries; p->name; ++p)
2116                 debugfs_remove(p->dentry);
2117         debugfs_remove(debugfs_dir);
2118 }
2119
2120 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2121 {
2122         decache_vcpus_on_cpu(raw_smp_processor_id());
2123         on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2124         return 0;
2125 }
2126
2127 static int kvm_resume(struct sys_device *dev)
2128 {
2129         on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2130         return 0;
2131 }
2132
2133 static struct sysdev_class kvm_sysdev_class = {
2134         set_kset_name("kvm"),
2135         .suspend = kvm_suspend,
2136         .resume = kvm_resume,
2137 };
2138
2139 static struct sys_device kvm_sysdev = {
2140         .id = 0,
2141         .cls = &kvm_sysdev_class,
2142 };
2143
2144 hpa_t bad_page_address;
2145
2146 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2147 {
2148         int r;
2149
2150         if (kvm_arch_ops) {
2151                 printk(KERN_ERR "kvm: already loaded the other module\n");
2152                 return -EEXIST;
2153         }
2154
2155         if (!ops->cpu_has_kvm_support()) {
2156                 printk(KERN_ERR "kvm: no hardware support\n");
2157                 return -EOPNOTSUPP;
2158         }
2159         if (ops->disabled_by_bios()) {
2160                 printk(KERN_ERR "kvm: disabled by bios\n");
2161                 return -EOPNOTSUPP;
2162         }
2163
2164         kvm_arch_ops = ops;
2165
2166         r = kvm_arch_ops->hardware_setup();
2167         if (r < 0)
2168             return r;
2169
2170         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2171         r = register_cpu_notifier(&kvm_cpu_notifier);
2172         if (r)
2173                 goto out_free_1;
2174         register_reboot_notifier(&kvm_reboot_notifier);
2175
2176         r = sysdev_class_register(&kvm_sysdev_class);
2177         if (r)
2178                 goto out_free_2;
2179
2180         r = sysdev_register(&kvm_sysdev);
2181         if (r)
2182                 goto out_free_3;
2183
2184         kvm_chardev_ops.owner = module;
2185
2186         r = misc_register(&kvm_dev);
2187         if (r) {
2188                 printk (KERN_ERR "kvm: misc device register failed\n");
2189                 goto out_free;
2190         }
2191
2192         return r;
2193
2194 out_free:
2195         sysdev_unregister(&kvm_sysdev);
2196 out_free_3:
2197         sysdev_class_unregister(&kvm_sysdev_class);
2198 out_free_2:
2199         unregister_reboot_notifier(&kvm_reboot_notifier);
2200         unregister_cpu_notifier(&kvm_cpu_notifier);
2201 out_free_1:
2202         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2203         kvm_arch_ops->hardware_unsetup();
2204         return r;
2205 }
2206
2207 void kvm_exit_arch(void)
2208 {
2209         misc_deregister(&kvm_dev);
2210         sysdev_unregister(&kvm_sysdev);
2211         sysdev_class_unregister(&kvm_sysdev_class);
2212         unregister_reboot_notifier(&kvm_reboot_notifier);
2213         unregister_cpu_notifier(&kvm_cpu_notifier);
2214         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2215         kvm_arch_ops->hardware_unsetup();
2216         kvm_arch_ops = NULL;
2217 }
2218
2219 static __init int kvm_init(void)
2220 {
2221         static struct page *bad_page;
2222         int r = 0;
2223
2224         kvm_init_debug();
2225
2226         kvm_init_msr_list();
2227
2228         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2229                 r = -ENOMEM;
2230                 goto out;
2231         }
2232
2233         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2234         memset(__va(bad_page_address), 0, PAGE_SIZE);
2235
2236         return r;
2237
2238 out:
2239         kvm_exit_debug();
2240         return r;
2241 }
2242
2243 static __exit void kvm_exit(void)
2244 {
2245         kvm_exit_debug();
2246         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2247 }
2248
2249 module_init(kvm_init)
2250 module_exit(kvm_exit)
2251
2252 EXPORT_SYMBOL_GPL(kvm_init_arch);
2253 EXPORT_SYMBOL_GPL(kvm_exit_arch);