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 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40
  41 #include <asm/processor.h>
  42 #include <asm/msr.h>
  43 #include <asm/io.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/desc.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 static cpumask_t cpus_hardware_enabled;
  54
  55 struct kvm_arch_ops *kvm_arch_ops;
  56
  57 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  58
  59 static struct kvm_stats_debugfs_item {
  60         const char *name;
  61         int offset;
  62         struct dentry *dentry;
  63 } debugfs_entries[] = {
  64         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  65         { "pf_guest", STAT_OFFSET(pf_guest) },
  66         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  67         { "invlpg", STAT_OFFSET(invlpg) },
  68         { "exits", STAT_OFFSET(exits) },
  69         { "io_exits", STAT_OFFSET(io_exits) },
  70         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  71         { "signal_exits", STAT_OFFSET(signal_exits) },
  72         { "irq_window", STAT_OFFSET(irq_window_exits) },
  73         { "halt_exits", STAT_OFFSET(halt_exits) },
  74         { "request_irq", STAT_OFFSET(request_irq_exits) },
  75         { "irq_exits", STAT_OFFSET(irq_exits) },
  76         { "light_exits", STAT_OFFSET(light_exits) },
  77         { "efer_reload", STAT_OFFSET(efer_reload) },
  78         { NULL }
  79 };
  80
  81 static struct dentry *debugfs_dir;
  82
  83 #define MAX_IO_MSRS 256
  84
  85 #define CR0_RESERVED_BITS                                               \
  86         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  87                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  88                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  89 #define CR4_RESERVED_BITS                                               \
  90         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  91                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  92                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  93                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  94
  95 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  96 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  97
  98 #ifdef CONFIG_X86_64
  99 // LDT or TSS descriptor in the GDT. 16 bytes.
 100 struct segment_descriptor_64 {
 101         struct segment_descriptor s;
 102         u32 base_higher;
 103         u32 pad_zero;
 104 };
 105
 106 #endif
 107
 108 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 109                            unsigned long arg);
 110
 111 unsigned long segment_base(u16 selector)
 112 {
 113         struct descriptor_table gdt;
 114         struct segment_descriptor *d;
 115         unsigned long table_base;
 116         typedef unsigned long ul;
 117         unsigned long v;
 118
 119         if (selector == 0)
 120                 return 0;
 121
 122         asm ("sgdt %0" : "=m"(gdt));
 123         table_base = gdt.base;
 124
 125         if (selector & 4) {           /* from ldt */
 126                 u16 ldt_selector;
 127
 128                 asm ("sldt %0" : "=g"(ldt_selector));
 129                 table_base = segment_base(ldt_selector);
 130         }
 131         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 132         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 133 #ifdef CONFIG_X86_64
 134         if (d->system == 0
 135             && (d->type == 2 || d->type == 9 || d->type == 11))
 136                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 137 #endif
 138         return v;
 139 }
 140 EXPORT_SYMBOL_GPL(segment_base);
 141
 142 static inline int valid_vcpu(int n)
 143 {
 144         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 145 }
 146
 147 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 148                    void *dest)
 149 {
 150         unsigned char *host_buf = dest;
 151         unsigned long req_size = size;
 152
 153         while (size) {
 154                 hpa_t paddr;
 155                 unsigned now;
 156                 unsigned offset;
 157                 hva_t guest_buf;
 158
 159                 paddr = gva_to_hpa(vcpu, addr);
 160
 161                 if (is_error_hpa(paddr))
 162                         break;
 163
 164                 guest_buf = (hva_t)kmap_atomic(
 165                                         pfn_to_page(paddr >> PAGE_SHIFT),
 166                                         KM_USER0);
 167                 offset = addr & ~PAGE_MASK;
 168                 guest_buf |= offset;
 169                 now = min(size, PAGE_SIZE - offset);
 170                 memcpy(host_buf, (void*)guest_buf, now);
 171                 host_buf += now;
 172                 addr += now;
 173                 size -= now;
 174                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 175         }
 176         return req_size - size;
 177 }
 178 EXPORT_SYMBOL_GPL(kvm_read_guest);
 179
 180 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 181                     void *data)
 182 {
 183         unsigned char *host_buf = data;
 184         unsigned long req_size = size;
 185
 186         while (size) {
 187                 hpa_t paddr;
 188                 unsigned now;
 189                 unsigned offset;
 190                 hva_t guest_buf;
 191                 gfn_t gfn;
 192
 193                 paddr = gva_to_hpa(vcpu, addr);
 194
 195                 if (is_error_hpa(paddr))
 196                         break;
 197
 198                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 199                 mark_page_dirty(vcpu->kvm, gfn);
 200                 guest_buf = (hva_t)kmap_atomic(
 201                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 202                 offset = addr & ~PAGE_MASK;
 203                 guest_buf |= offset;
 204                 now = min(size, PAGE_SIZE - offset);
 205                 memcpy((void*)guest_buf, host_buf, now);
 206                 host_buf += now;
 207                 addr += now;
 208                 size -= now;
 209                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 210         }
 211         return req_size - size;
 212 }
 213 EXPORT_SYMBOL_GPL(kvm_write_guest);
 214
 215 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 216 {
 217         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 218                 return;
 219
 220         vcpu->guest_fpu_loaded = 1;
 221         fx_save(vcpu->host_fx_image);
 222         fx_restore(vcpu->guest_fx_image);
 223 }
 224 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 225
 226 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 227 {
 228         if (!vcpu->guest_fpu_loaded)
 229                 return;
 230
 231         vcpu->guest_fpu_loaded = 0;
 232         fx_save(vcpu->guest_fx_image);
 233         fx_restore(vcpu->host_fx_image);
 234 }
 235 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 236
 237 /*
 238  * Switches to specified vcpu, until a matching vcpu_put()
 239  */
 240 static void vcpu_load(struct kvm_vcpu *vcpu)
 241 {
 242         mutex_lock(&vcpu->mutex);
 243         kvm_arch_ops->vcpu_load(vcpu);
 244 }
 245
 246 static void vcpu_put(struct kvm_vcpu *vcpu)
 247 {
 248         kvm_arch_ops->vcpu_put(vcpu);
 249         mutex_unlock(&vcpu->mutex);
 250 }
 251
 252 static void ack_flush(void *_completed)
 253 {
 254         atomic_t *completed = _completed;
 255
 256         atomic_inc(completed);
 257 }
 258
 259 void kvm_flush_remote_tlbs(struct kvm *kvm)
 260 {
 261         int i, cpu, needed;
 262         cpumask_t cpus;
 263         struct kvm_vcpu *vcpu;
 264         atomic_t completed;
 265
 266         atomic_set(&completed, 0);
 267         cpus_clear(cpus);
 268         needed = 0;
 269         for (i = 0; i < kvm->nvcpus; ++i) {
 270                 vcpu = &kvm->vcpus[i];
 271                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 272                         continue;
 273                 cpu = vcpu->cpu;
 274                 if (cpu != -1 && cpu != raw_smp_processor_id())
 275                         if (!cpu_isset(cpu, cpus)) {
 276                                 cpu_set(cpu, cpus);
 277                                 ++needed;
 278                         }
 279         }
 280
 281         /*
 282          * We really want smp_call_function_mask() here.  But that's not
 283          * available, so ipi all cpus in parallel and wait for them
 284          * to complete.
 285          */
 286         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 287                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 288         while (atomic_read(&completed) != needed) {
 289                 cpu_relax();
 290                 barrier();
 291         }
 292 }
 293
 294 static struct kvm *kvm_create_vm(void)
 295 {
 296         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 297         int i;
 298
 299         if (!kvm)
 300                 return ERR_PTR(-ENOMEM);
 301
 302         kvm_io_bus_init(&kvm->pio_bus);
 303         spin_lock_init(&kvm->lock);
 304         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 305         kvm_io_bus_init(&kvm->mmio_bus);
 306         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 307                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 308
 309                 mutex_init(&vcpu->mutex);
 310                 vcpu->cpu = -1;
 311                 vcpu->kvm = kvm;
 312                 vcpu->mmu.root_hpa = INVALID_PAGE;
 313         }
 314         spin_lock(&kvm_lock);
 315         list_add(&kvm->vm_list, &vm_list);
 316         spin_unlock(&kvm_lock);
 317         return kvm;
 318 }
 319
 320 static int kvm_dev_open(struct inode *inode, struct file *filp)
 321 {
 322         return 0;
 323 }
 324
 325 /*
 326  * Free any memory in @free but not in @dont.
 327  */
 328 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 329                                   struct kvm_memory_slot *dont)
 330 {
 331         int i;
 332
 333         if (!dont || free->phys_mem != dont->phys_mem)
 334                 if (free->phys_mem) {
 335                         for (i = 0; i < free->npages; ++i)
 336                                 if (free->phys_mem[i])
 337                                         __free_page(free->phys_mem[i]);
 338                         vfree(free->phys_mem);
 339                 }
 340
 341         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 342                 vfree(free->dirty_bitmap);
 343
 344         free->phys_mem = NULL;
 345         free->npages = 0;
 346         free->dirty_bitmap = NULL;
 347 }
 348
 349 static void kvm_free_physmem(struct kvm *kvm)
 350 {
 351         int i;
 352
 353         for (i = 0; i < kvm->nmemslots; ++i)
 354                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 355 }
 356
 357 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 358 {
 359         int i;
 360
 361         for (i = 0; i < 2; ++i)
 362                 if (vcpu->pio.guest_pages[i]) {
 363                         __free_page(vcpu->pio.guest_pages[i]);
 364                         vcpu->pio.guest_pages[i] = NULL;
 365                 }
 366 }
 367
 368 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 369 {
 370         if (!vcpu->vmcs)
 371                 return;
 372
 373         vcpu_load(vcpu);
 374         kvm_mmu_unload(vcpu);
 375         vcpu_put(vcpu);
 376 }
 377
 378 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 379 {
 380         if (!vcpu->vmcs)
 381                 return;
 382
 383         vcpu_load(vcpu);
 384         kvm_mmu_destroy(vcpu);
 385         vcpu_put(vcpu);
 386         kvm_arch_ops->vcpu_free(vcpu);
 387         free_page((unsigned long)vcpu->run);
 388         vcpu->run = NULL;
 389         free_page((unsigned long)vcpu->pio_data);
 390         vcpu->pio_data = NULL;
 391         free_pio_guest_pages(vcpu);
 392 }
 393
 394 static void kvm_free_vcpus(struct kvm *kvm)
 395 {
 396         unsigned int i;
 397
 398         /*
 399          * Unpin any mmu pages first.
 400          */
 401         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 402                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
 403         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 404                 kvm_free_vcpu(&kvm->vcpus[i]);
 405 }
 406
 407 static int kvm_dev_release(struct inode *inode, struct file *filp)
 408 {
 409         return 0;
 410 }
 411
 412 static void kvm_destroy_vm(struct kvm *kvm)
 413 {
 414         spin_lock(&kvm_lock);
 415         list_del(&kvm->vm_list);
 416         spin_unlock(&kvm_lock);
 417         kvm_io_bus_destroy(&kvm->pio_bus);
 418         kvm_io_bus_destroy(&kvm->mmio_bus);
 419         kvm_free_vcpus(kvm);
 420         kvm_free_physmem(kvm);
 421         kfree(kvm);
 422 }
 423
 424 static int kvm_vm_release(struct inode *inode, struct file *filp)
 425 {
 426         struct kvm *kvm = filp->private_data;
 427
 428         kvm_destroy_vm(kvm);
 429         return 0;
 430 }
 431
 432 static void inject_gp(struct kvm_vcpu *vcpu)
 433 {
 434         kvm_arch_ops->inject_gp(vcpu, 0);
 435 }
 436
 437 /*
 438  * Load the pae pdptrs.  Return true is they are all valid.
 439  */
 440 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 441 {
 442         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 443         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 444         int i;
 445         u64 pdpte;
 446         u64 *pdpt;
 447         int ret;
 448         struct page *page;
 449
 450         spin_lock(&vcpu->kvm->lock);
 451         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 452         /* FIXME: !page - emulate? 0xff? */
 453         pdpt = kmap_atomic(page, KM_USER0);
 454
 455         ret = 1;
 456         for (i = 0; i < 4; ++i) {
 457                 pdpte = pdpt[offset + i];
 458                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 459                         ret = 0;
 460                         goto out;
 461                 }
 462         }
 463
 464         for (i = 0; i < 4; ++i)
 465                 vcpu->pdptrs[i] = pdpt[offset + i];
 466
 467 out:
 468         kunmap_atomic(pdpt, KM_USER0);
 469         spin_unlock(&vcpu->kvm->lock);
 470
 471         return ret;
 472 }
 473
 474 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 475 {
 476         if (cr0 & CR0_RESERVED_BITS) {
 477                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 478                        cr0, vcpu->cr0);
 479                 inject_gp(vcpu);
 480                 return;
 481         }
 482
 483         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 484                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 485                 inject_gp(vcpu);
 486                 return;
 487         }
 488
 489         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 490                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 491                        "and a clear PE flag\n");
 492                 inject_gp(vcpu);
 493                 return;
 494         }
 495
 496         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 497 #ifdef CONFIG_X86_64
 498                 if ((vcpu->shadow_efer & EFER_LME)) {
 499                         int cs_db, cs_l;
 500
 501                         if (!is_pae(vcpu)) {
 502                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 503                                        "in long mode while PAE is disabled\n");
 504                                 inject_gp(vcpu);
 505                                 return;
 506                         }
 507                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 508                         if (cs_l) {
 509                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 510                                        "in long mode while CS.L == 1\n");
 511                                 inject_gp(vcpu);
 512                                 return;
 513
 514                         }
 515                 } else
 516 #endif
 517                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 518                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 519                                "reserved bits\n");
 520                         inject_gp(vcpu);
 521                         return;
 522                 }
 523
 524         }
 525
 526         kvm_arch_ops->set_cr0(vcpu, cr0);
 527         vcpu->cr0 = cr0;
 528
 529         spin_lock(&vcpu->kvm->lock);
 530         kvm_mmu_reset_context(vcpu);
 531         spin_unlock(&vcpu->kvm->lock);
 532         return;
 533 }
 534 EXPORT_SYMBOL_GPL(set_cr0);
 535
 536 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 537 {
 538         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 539 }
 540 EXPORT_SYMBOL_GPL(lmsw);
 541
 542 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 543 {
 544         if (cr4 & CR4_RESERVED_BITS) {
 545                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 546                 inject_gp(vcpu);
 547                 return;
 548         }
 549
 550         if (is_long_mode(vcpu)) {
 551                 if (!(cr4 & X86_CR4_PAE)) {
 552                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 553                                "in long mode\n");
 554                         inject_gp(vcpu);
 555                         return;
 556                 }
 557         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 558                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 559                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 560                 inject_gp(vcpu);
 561                 return;
 562         }
 563
 564         if (cr4 & X86_CR4_VMXE) {
 565                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 566                 inject_gp(vcpu);
 567                 return;
 568         }
 569         kvm_arch_ops->set_cr4(vcpu, cr4);
 570         spin_lock(&vcpu->kvm->lock);
 571         kvm_mmu_reset_context(vcpu);
 572         spin_unlock(&vcpu->kvm->lock);
 573 }
 574 EXPORT_SYMBOL_GPL(set_cr4);
 575
 576 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 577 {
 578         if (is_long_mode(vcpu)) {
 579                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 580                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 581                         inject_gp(vcpu);
 582                         return;
 583                 }
 584         } else {
 585                 if (is_pae(vcpu)) {
 586                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 587                                 printk(KERN_DEBUG
 588                                        "set_cr3: #GP, reserved bits\n");
 589                                 inject_gp(vcpu);
 590                                 return;
 591                         }
 592                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 593                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 594                                        "reserved bits\n");
 595                                 inject_gp(vcpu);
 596                                 return;
 597                         }
 598                 } else {
 599                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 600                                 printk(KERN_DEBUG
 601                                        "set_cr3: #GP, reserved bits\n");
 602                                 inject_gp(vcpu);
 603                                 return;
 604                         }
 605                 }
 606         }
 607
 608         vcpu->cr3 = cr3;
 609         spin_lock(&vcpu->kvm->lock);
 610         /*
 611          * Does the new cr3 value map to physical memory? (Note, we
 612          * catch an invalid cr3 even in real-mode, because it would
 613          * cause trouble later on when we turn on paging anyway.)
 614          *
 615          * A real CPU would silently accept an invalid cr3 and would
 616          * attempt to use it - with largely undefined (and often hard
 617          * to debug) behavior on the guest side.
 618          */
 619         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 620                 inject_gp(vcpu);
 621         else
 622                 vcpu->mmu.new_cr3(vcpu);
 623         spin_unlock(&vcpu->kvm->lock);
 624 }
 625 EXPORT_SYMBOL_GPL(set_cr3);
 626
 627 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 628 {
 629         if (cr8 & CR8_RESERVED_BITS) {
 630                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 631                 inject_gp(vcpu);
 632                 return;
 633         }
 634         vcpu->cr8 = cr8;
 635 }
 636 EXPORT_SYMBOL_GPL(set_cr8);
 637
 638 void fx_init(struct kvm_vcpu *vcpu)
 639 {
 640         struct __attribute__ ((__packed__)) fx_image_s {
 641                 u16 control; //fcw
 642                 u16 status; //fsw
 643                 u16 tag; // ftw
 644                 u16 opcode; //fop
 645                 u64 ip; // fpu ip
 646                 u64 operand;// fpu dp
 647                 u32 mxcsr;
 648                 u32 mxcsr_mask;
 649
 650         } *fx_image;
 651
 652         fx_save(vcpu->host_fx_image);
 653         fpu_init();
 654         fx_save(vcpu->guest_fx_image);
 655         fx_restore(vcpu->host_fx_image);
 656
 657         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 658         fx_image->mxcsr = 0x1f80;
 659         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 660                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 661 }
 662 EXPORT_SYMBOL_GPL(fx_init);
 663
 664 /*
 665  * Allocate some memory and give it an address in the guest physical address
 666  * space.
 667  *
 668  * Discontiguous memory is allowed, mostly for framebuffers.
 669  */
 670 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 671                                           struct kvm_memory_region *mem)
 672 {
 673         int r;
 674         gfn_t base_gfn;
 675         unsigned long npages;
 676         unsigned long i;
 677         struct kvm_memory_slot *memslot;
 678         struct kvm_memory_slot old, new;
 679         int memory_config_version;
 680
 681         r = -EINVAL;
 682         /* General sanity checks */
 683         if (mem->memory_size & (PAGE_SIZE - 1))
 684                 goto out;
 685         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 686                 goto out;
 687         if (mem->slot >= KVM_MEMORY_SLOTS)
 688                 goto out;
 689         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 690                 goto out;
 691
 692         memslot = &kvm->memslots[mem->slot];
 693         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 694         npages = mem->memory_size >> PAGE_SHIFT;
 695
 696         if (!npages)
 697                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 698
 699 raced:
 700         spin_lock(&kvm->lock);
 701
 702         memory_config_version = kvm->memory_config_version;
 703         new = old = *memslot;
 704
 705         new.base_gfn = base_gfn;
 706         new.npages = npages;
 707         new.flags = mem->flags;
 708
 709         /* Disallow changing a memory slot's size. */
 710         r = -EINVAL;
 711         if (npages && old.npages && npages != old.npages)
 712                 goto out_unlock;
 713
 714         /* Check for overlaps */
 715         r = -EEXIST;
 716         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 717                 struct kvm_memory_slot *s = &kvm->memslots[i];
 718
 719                 if (s == memslot)
 720                         continue;
 721                 if (!((base_gfn + npages <= s->base_gfn) ||
 722                       (base_gfn >= s->base_gfn + s->npages)))
 723                         goto out_unlock;
 724         }
 725         /*
 726          * Do memory allocations outside lock.  memory_config_version will
 727          * detect any races.
 728          */
 729         spin_unlock(&kvm->lock);
 730
 731         /* Deallocate if slot is being removed */
 732         if (!npages)
 733                 new.phys_mem = NULL;
 734
 735         /* Free page dirty bitmap if unneeded */
 736         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 737                 new.dirty_bitmap = NULL;
 738
 739         r = -ENOMEM;
 740
 741         /* Allocate if a slot is being created */
 742         if (npages && !new.phys_mem) {
 743                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 744
 745                 if (!new.phys_mem)
 746                         goto out_free;
 747
 748                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 749                 for (i = 0; i < npages; ++i) {
 750                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 751                                                      | __GFP_ZERO);
 752                         if (!new.phys_mem[i])
 753                                 goto out_free;
 754                         set_page_private(new.phys_mem[i],0);
 755                 }
 756         }
 757
 758         /* Allocate page dirty bitmap if needed */
 759         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 760                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 761
 762                 new.dirty_bitmap = vmalloc(dirty_bytes);
 763                 if (!new.dirty_bitmap)
 764                         goto out_free;
 765                 memset(new.dirty_bitmap, 0, dirty_bytes);
 766         }
 767
 768         spin_lock(&kvm->lock);
 769
 770         if (memory_config_version != kvm->memory_config_version) {
 771                 spin_unlock(&kvm->lock);
 772                 kvm_free_physmem_slot(&new, &old);
 773                 goto raced;
 774         }
 775
 776         r = -EAGAIN;
 777         if (kvm->busy)
 778                 goto out_unlock;
 779
 780         if (mem->slot >= kvm->nmemslots)
 781                 kvm->nmemslots = mem->slot + 1;
 782
 783         *memslot = new;
 784         ++kvm->memory_config_version;
 785
 786         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 787         kvm_flush_remote_tlbs(kvm);
 788
 789         spin_unlock(&kvm->lock);
 790
 791         kvm_free_physmem_slot(&old, &new);
 792         return 0;
 793
 794 out_unlock:
 795         spin_unlock(&kvm->lock);
 796 out_free:
 797         kvm_free_physmem_slot(&new, &old);
 798 out:
 799         return r;
 800 }
 801
 802 /*
 803  * Get (and clear) the dirty memory log for a memory slot.
 804  */
 805 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 806                                       struct kvm_dirty_log *log)
 807 {
 808         struct kvm_memory_slot *memslot;
 809         int r, i;
 810         int n;
 811         unsigned long any = 0;
 812
 813         spin_lock(&kvm->lock);
 814
 815         /*
 816          * Prevent changes to guest memory configuration even while the lock
 817          * is not taken.
 818          */
 819         ++kvm->busy;
 820         spin_unlock(&kvm->lock);
 821         r = -EINVAL;
 822         if (log->slot >= KVM_MEMORY_SLOTS)
 823                 goto out;
 824
 825         memslot = &kvm->memslots[log->slot];
 826         r = -ENOENT;
 827         if (!memslot->dirty_bitmap)
 828                 goto out;
 829
 830         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 831
 832         for (i = 0; !any && i < n/sizeof(long); ++i)
 833                 any = memslot->dirty_bitmap[i];
 834
 835         r = -EFAULT;
 836         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 837                 goto out;
 838
 839         spin_lock(&kvm->lock);
 840         kvm_mmu_slot_remove_write_access(kvm, log->slot);
 841         kvm_flush_remote_tlbs(kvm);
 842         memset(memslot->dirty_bitmap, 0, n);
 843         spin_unlock(&kvm->lock);
 844
 845         r = 0;
 846
 847 out:
 848         spin_lock(&kvm->lock);
 849         --kvm->busy;
 850         spin_unlock(&kvm->lock);
 851         return r;
 852 }
 853
 854 /*
 855  * Set a new alias region.  Aliases map a portion of physical memory into
 856  * another portion.  This is useful for memory windows, for example the PC
 857  * VGA region.
 858  */
 859 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 860                                          struct kvm_memory_alias *alias)
 861 {
 862         int r, n;
 863         struct kvm_mem_alias *p;
 864
 865         r = -EINVAL;
 866         /* General sanity checks */
 867         if (alias->memory_size & (PAGE_SIZE - 1))
 868                 goto out;
 869         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 870                 goto out;
 871         if (alias->slot >= KVM_ALIAS_SLOTS)
 872                 goto out;
 873         if (alias->guest_phys_addr + alias->memory_size
 874             < alias->guest_phys_addr)
 875                 goto out;
 876         if (alias->target_phys_addr + alias->memory_size
 877             < alias->target_phys_addr)
 878                 goto out;
 879
 880         spin_lock(&kvm->lock);
 881
 882         p = &kvm->aliases[alias->slot];
 883         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 884         p->npages = alias->memory_size >> PAGE_SHIFT;
 885         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 886
 887         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 888                 if (kvm->aliases[n - 1].npages)
 889                         break;
 890         kvm->naliases = n;
 891
 892         kvm_mmu_zap_all(kvm);
 893
 894         spin_unlock(&kvm->lock);
 895
 896         return 0;
 897
 898 out:
 899         return r;
 900 }
 901
 902 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 903 {
 904         int i;
 905         struct kvm_mem_alias *alias;
 906
 907         for (i = 0; i < kvm->naliases; ++i) {
 908                 alias = &kvm->aliases[i];
 909                 if (gfn >= alias->base_gfn
 910                     && gfn < alias->base_gfn + alias->npages)
 911                         return alias->target_gfn + gfn - alias->base_gfn;
 912         }
 913         return gfn;
 914 }
 915
 916 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 917 {
 918         int i;
 919
 920         for (i = 0; i < kvm->nmemslots; ++i) {
 921                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 922
 923                 if (gfn >= memslot->base_gfn
 924                     && gfn < memslot->base_gfn + memslot->npages)
 925                         return memslot;
 926         }
 927         return NULL;
 928 }
 929
 930 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 931 {
 932         gfn = unalias_gfn(kvm, gfn);
 933         return __gfn_to_memslot(kvm, gfn);
 934 }
 935
 936 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 937 {
 938         struct kvm_memory_slot *slot;
 939
 940         gfn = unalias_gfn(kvm, gfn);
 941         slot = __gfn_to_memslot(kvm, gfn);
 942         if (!slot)
 943                 return NULL;
 944         return slot->phys_mem[gfn - slot->base_gfn];
 945 }
 946 EXPORT_SYMBOL_GPL(gfn_to_page);
 947
 948 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 949 {
 950         int i;
 951         struct kvm_memory_slot *memslot;
 952         unsigned long rel_gfn;
 953
 954         for (i = 0; i < kvm->nmemslots; ++i) {
 955                 memslot = &kvm->memslots[i];
 956
 957                 if (gfn >= memslot->base_gfn
 958                     && gfn < memslot->base_gfn + memslot->npages) {
 959
 960                         if (!memslot->dirty_bitmap)
 961                                 return;
 962
 963                         rel_gfn = gfn - memslot->base_gfn;
 964
 965                         /* avoid RMW */
 966                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 967                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 968                         return;
 969                 }
 970         }
 971 }
 972
 973 static int emulator_read_std(unsigned long addr,
 974                              void *val,
 975                              unsigned int bytes,
 976                              struct x86_emulate_ctxt *ctxt)
 977 {
 978         struct kvm_vcpu *vcpu = ctxt->vcpu;
 979         void *data = val;
 980
 981         while (bytes) {
 982                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 983                 unsigned offset = addr & (PAGE_SIZE-1);
 984                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 985                 unsigned long pfn;
 986                 struct page *page;
 987                 void *page_virt;
 988
 989                 if (gpa == UNMAPPED_GVA)
 990                         return X86EMUL_PROPAGATE_FAULT;
 991                 pfn = gpa >> PAGE_SHIFT;
 992                 page = gfn_to_page(vcpu->kvm, pfn);
 993                 if (!page)
 994                         return X86EMUL_UNHANDLEABLE;
 995                 page_virt = kmap_atomic(page, KM_USER0);
 996
 997                 memcpy(data, page_virt + offset, tocopy);
 998
 999                 kunmap_atomic(page_virt, KM_USER0);
1000
1001                 bytes -= tocopy;
1002                 data += tocopy;
1003                 addr += tocopy;
1004         }
1005
1006         return X86EMUL_CONTINUE;
1007 }
1008
1009 static int emulator_write_std(unsigned long addr,
1010                               const void *val,
1011                               unsigned int bytes,
1012                               struct x86_emulate_ctxt *ctxt)
1013 {
1014         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1015                addr, bytes);
1016         return X86EMUL_UNHANDLEABLE;
1017 }
1018
1019 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1020                                                 gpa_t addr)
1021 {
1022         /*
1023          * Note that its important to have this wrapper function because
1024          * in the very near future we will be checking for MMIOs against
1025          * the LAPIC as well as the general MMIO bus
1026          */
1027         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1028 }
1029
1030 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1031                                                gpa_t addr)
1032 {
1033         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1034 }
1035
1036 static int emulator_read_emulated(unsigned long addr,
1037                                   void *val,
1038                                   unsigned int bytes,
1039                                   struct x86_emulate_ctxt *ctxt)
1040 {
1041         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1042         struct kvm_io_device *mmio_dev;
1043         gpa_t                 gpa;
1044
1045         if (vcpu->mmio_read_completed) {
1046                 memcpy(val, vcpu->mmio_data, bytes);
1047                 vcpu->mmio_read_completed = 0;
1048                 return X86EMUL_CONTINUE;
1049         } else if (emulator_read_std(addr, val, bytes, ctxt)
1050                    == X86EMUL_CONTINUE)
1051                 return X86EMUL_CONTINUE;
1052
1053         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1054         if (gpa == UNMAPPED_GVA)
1055                 return X86EMUL_PROPAGATE_FAULT;
1056
1057         /*
1058          * Is this MMIO handled locally?
1059          */
1060         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1061         if (mmio_dev) {
1062                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1063                 return X86EMUL_CONTINUE;
1064         }
1065
1066         vcpu->mmio_needed = 1;
1067         vcpu->mmio_phys_addr = gpa;
1068         vcpu->mmio_size = bytes;
1069         vcpu->mmio_is_write = 0;
1070
1071         return X86EMUL_UNHANDLEABLE;
1072 }
1073
1074 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1075                                const void *val, int bytes)
1076 {
1077         struct page *page;
1078         void *virt;
1079         unsigned offset = offset_in_page(gpa);
1080
1081         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1082                 return 0;
1083         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1084         if (!page)
1085                 return 0;
1086         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1087         virt = kmap_atomic(page, KM_USER0);
1088         kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1089         memcpy(virt + offset_in_page(gpa), val, bytes);
1090         kunmap_atomic(virt, KM_USER0);
1091         return 1;
1092 }
1093
1094 static int emulator_write_emulated_onepage(unsigned long addr,
1095                                            const void *val,
1096                                            unsigned int bytes,
1097                                            struct x86_emulate_ctxt *ctxt)
1098 {
1099         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1100         struct kvm_io_device *mmio_dev;
1101         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1102
1103         if (gpa == UNMAPPED_GVA) {
1104                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1105                 return X86EMUL_PROPAGATE_FAULT;
1106         }
1107
1108         if (emulator_write_phys(vcpu, gpa, val, bytes))
1109                 return X86EMUL_CONTINUE;
1110
1111         /*
1112          * Is this MMIO handled locally?
1113          */
1114         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1115         if (mmio_dev) {
1116                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1117                 return X86EMUL_CONTINUE;
1118         }
1119
1120         vcpu->mmio_needed = 1;
1121         vcpu->mmio_phys_addr = gpa;
1122         vcpu->mmio_size = bytes;
1123         vcpu->mmio_is_write = 1;
1124         memcpy(vcpu->mmio_data, val, bytes);
1125
1126         return X86EMUL_CONTINUE;
1127 }
1128
1129 static int emulator_write_emulated(unsigned long addr,
1130                                    const void *val,
1131                                    unsigned int bytes,
1132                                    struct x86_emulate_ctxt *ctxt)
1133 {
1134         /* Crossing a page boundary? */
1135         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1136                 int rc, now;
1137
1138                 now = -addr & ~PAGE_MASK;
1139                 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1140                 if (rc != X86EMUL_CONTINUE)
1141                         return rc;
1142                 addr += now;
1143                 val += now;
1144                 bytes -= now;
1145         }
1146         return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1147 }
1148
1149 static int emulator_cmpxchg_emulated(unsigned long addr,
1150                                      const void *old,
1151                                      const void *new,
1152                                      unsigned int bytes,
1153                                      struct x86_emulate_ctxt *ctxt)
1154 {
1155         static int reported;
1156
1157         if (!reported) {
1158                 reported = 1;
1159                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1160         }
1161         return emulator_write_emulated(addr, new, bytes, ctxt);
1162 }
1163
1164 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1165 {
1166         return kvm_arch_ops->get_segment_base(vcpu, seg);
1167 }
1168
1169 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1170 {
1171         return X86EMUL_CONTINUE;
1172 }
1173
1174 int emulate_clts(struct kvm_vcpu *vcpu)
1175 {
1176         unsigned long cr0;
1177
1178         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1179         kvm_arch_ops->set_cr0(vcpu, cr0);
1180         return X86EMUL_CONTINUE;
1181 }
1182
1183 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1184 {
1185         struct kvm_vcpu *vcpu = ctxt->vcpu;
1186
1187         switch (dr) {
1188         case 0 ... 3:
1189                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1190                 return X86EMUL_CONTINUE;
1191         default:
1192                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1193                        __FUNCTION__, dr);
1194                 return X86EMUL_UNHANDLEABLE;
1195         }
1196 }
1197
1198 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1199 {
1200         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1201         int exception;
1202
1203         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1204         if (exception) {
1205                 /* FIXME: better handling */
1206                 return X86EMUL_UNHANDLEABLE;
1207         }
1208         return X86EMUL_CONTINUE;
1209 }
1210
1211 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1212 {
1213         static int reported;
1214         u8 opcodes[4];
1215         unsigned long rip = ctxt->vcpu->rip;
1216         unsigned long rip_linear;
1217
1218         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1219
1220         if (reported)
1221                 return;
1222
1223         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1224
1225         printk(KERN_ERR "emulation failed but !mmio_needed?"
1226                " rip %lx %02x %02x %02x %02x\n",
1227                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1228         reported = 1;
1229 }
1230
1231 struct x86_emulate_ops emulate_ops = {
1232         .read_std            = emulator_read_std,
1233         .write_std           = emulator_write_std,
1234         .read_emulated       = emulator_read_emulated,
1235         .write_emulated      = emulator_write_emulated,
1236         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1237 };
1238
1239 int emulate_instruction(struct kvm_vcpu *vcpu,
1240                         struct kvm_run *run,
1241                         unsigned long cr2,
1242                         u16 error_code)
1243 {
1244         struct x86_emulate_ctxt emulate_ctxt;
1245         int r;
1246         int cs_db, cs_l;
1247
1248         vcpu->mmio_fault_cr2 = cr2;
1249         kvm_arch_ops->cache_regs(vcpu);
1250
1251         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1252
1253         emulate_ctxt.vcpu = vcpu;
1254         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1255         emulate_ctxt.cr2 = cr2;
1256         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1257                 ? X86EMUL_MODE_REAL : cs_l
1258                 ? X86EMUL_MODE_PROT64 : cs_db
1259                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1260
1261         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1262                 emulate_ctxt.cs_base = 0;
1263                 emulate_ctxt.ds_base = 0;
1264                 emulate_ctxt.es_base = 0;
1265                 emulate_ctxt.ss_base = 0;
1266         } else {
1267                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1268                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1269                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1270                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1271         }
1272
1273         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1274         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1275
1276         vcpu->mmio_is_write = 0;
1277         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1278
1279         if ((r || vcpu->mmio_is_write) && run) {
1280                 run->exit_reason = KVM_EXIT_MMIO;
1281                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1282                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1283                 run->mmio.len = vcpu->mmio_size;
1284                 run->mmio.is_write = vcpu->mmio_is_write;
1285         }
1286
1287         if (r) {
1288                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1289                         return EMULATE_DONE;
1290                 if (!vcpu->mmio_needed) {
1291                         report_emulation_failure(&emulate_ctxt);
1292                         return EMULATE_FAIL;
1293                 }
1294                 return EMULATE_DO_MMIO;
1295         }
1296
1297         kvm_arch_ops->decache_regs(vcpu);
1298         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1299
1300         if (vcpu->mmio_is_write) {
1301                 vcpu->mmio_needed = 0;
1302                 return EMULATE_DO_MMIO;
1303         }
1304
1305         return EMULATE_DONE;
1306 }
1307 EXPORT_SYMBOL_GPL(emulate_instruction);
1308
1309 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1310 {
1311         if (vcpu->irq_summary)
1312                 return 1;
1313
1314         vcpu->run->exit_reason = KVM_EXIT_HLT;
1315         ++vcpu->stat.halt_exits;
1316         return 0;
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1319
1320 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1321 {
1322         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1323
1324         kvm_arch_ops->cache_regs(vcpu);
1325         ret = -KVM_EINVAL;
1326 #ifdef CONFIG_X86_64
1327         if (is_long_mode(vcpu)) {
1328                 nr = vcpu->regs[VCPU_REGS_RAX];
1329                 a0 = vcpu->regs[VCPU_REGS_RDI];
1330                 a1 = vcpu->regs[VCPU_REGS_RSI];
1331                 a2 = vcpu->regs[VCPU_REGS_RDX];
1332                 a3 = vcpu->regs[VCPU_REGS_RCX];
1333                 a4 = vcpu->regs[VCPU_REGS_R8];
1334                 a5 = vcpu->regs[VCPU_REGS_R9];
1335         } else
1336 #endif
1337         {
1338                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1339                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1340                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1341                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1342                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1343                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1344                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1345         }
1346         switch (nr) {
1347         default:
1348                 run->hypercall.args[0] = a0;
1349                 run->hypercall.args[1] = a1;
1350                 run->hypercall.args[2] = a2;
1351                 run->hypercall.args[3] = a3;
1352                 run->hypercall.args[4] = a4;
1353                 run->hypercall.args[5] = a5;
1354                 run->hypercall.ret = ret;
1355                 run->hypercall.longmode = is_long_mode(vcpu);
1356                 kvm_arch_ops->decache_regs(vcpu);
1357                 return 0;
1358         }
1359         vcpu->regs[VCPU_REGS_RAX] = ret;
1360         kvm_arch_ops->decache_regs(vcpu);
1361         return 1;
1362 }
1363 EXPORT_SYMBOL_GPL(kvm_hypercall);
1364
1365 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1366 {
1367         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1368 }
1369
1370 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1371 {
1372         struct descriptor_table dt = { limit, base };
1373
1374         kvm_arch_ops->set_gdt(vcpu, &dt);
1375 }
1376
1377 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1378 {
1379         struct descriptor_table dt = { limit, base };
1380
1381         kvm_arch_ops->set_idt(vcpu, &dt);
1382 }
1383
1384 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1385                    unsigned long *rflags)
1386 {
1387         lmsw(vcpu, msw);
1388         *rflags = kvm_arch_ops->get_rflags(vcpu);
1389 }
1390
1391 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1392 {
1393         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1394         switch (cr) {
1395         case 0:
1396                 return vcpu->cr0;
1397         case 2:
1398                 return vcpu->cr2;
1399         case 3:
1400                 return vcpu->cr3;
1401         case 4:
1402                 return vcpu->cr4;
1403         default:
1404                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1405                 return 0;
1406         }
1407 }
1408
1409 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1410                      unsigned long *rflags)
1411 {
1412         switch (cr) {
1413         case 0:
1414                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1415                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1416                 break;
1417         case 2:
1418                 vcpu->cr2 = val;
1419                 break;
1420         case 3:
1421                 set_cr3(vcpu, val);
1422                 break;
1423         case 4:
1424                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1425                 break;
1426         default:
1427                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1428         }
1429 }
1430
1431 /*
1432  * Register the para guest with the host:
1433  */
1434 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1435 {
1436         struct kvm_vcpu_para_state *para_state;
1437         hpa_t para_state_hpa, hypercall_hpa;
1438         struct page *para_state_page;
1439         unsigned char *hypercall;
1440         gpa_t hypercall_gpa;
1441
1442         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1443         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1444
1445         /*
1446          * Needs to be page aligned:
1447          */
1448         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1449                 goto err_gp;
1450
1451         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1452         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1453         if (is_error_hpa(para_state_hpa))
1454                 goto err_gp;
1455
1456         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1457         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1458         para_state = kmap_atomic(para_state_page, KM_USER0);
1459
1460         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1461         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1462
1463         para_state->host_version = KVM_PARA_API_VERSION;
1464         /*
1465          * We cannot support guests that try to register themselves
1466          * with a newer API version than the host supports:
1467          */
1468         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1469                 para_state->ret = -KVM_EINVAL;
1470                 goto err_kunmap_skip;
1471         }
1472
1473         hypercall_gpa = para_state->hypercall_gpa;
1474         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1475         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1476         if (is_error_hpa(hypercall_hpa)) {
1477                 para_state->ret = -KVM_EINVAL;
1478                 goto err_kunmap_skip;
1479         }
1480
1481         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1482         vcpu->para_state_page = para_state_page;
1483         vcpu->para_state_gpa = para_state_gpa;
1484         vcpu->hypercall_gpa = hypercall_gpa;
1485
1486         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1487         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1488                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1489         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1490         kunmap_atomic(hypercall, KM_USER1);
1491
1492         para_state->ret = 0;
1493 err_kunmap_skip:
1494         kunmap_atomic(para_state, KM_USER0);
1495         return 0;
1496 err_gp:
1497         return 1;
1498 }
1499
1500 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1501 {
1502         u64 data;
1503
1504         switch (msr) {
1505         case 0xc0010010: /* SYSCFG */
1506         case 0xc0010015: /* HWCR */
1507         case MSR_IA32_PLATFORM_ID:
1508         case MSR_IA32_P5_MC_ADDR:
1509         case MSR_IA32_P5_MC_TYPE:
1510         case MSR_IA32_MC0_CTL:
1511         case MSR_IA32_MCG_STATUS:
1512         case MSR_IA32_MCG_CAP:
1513         case MSR_IA32_MC0_MISC:
1514         case MSR_IA32_MC0_MISC+4:
1515         case MSR_IA32_MC0_MISC+8:
1516         case MSR_IA32_MC0_MISC+12:
1517         case MSR_IA32_MC0_MISC+16:
1518         case MSR_IA32_UCODE_REV:
1519         case MSR_IA32_PERF_STATUS:
1520         case MSR_IA32_EBL_CR_POWERON:
1521                 /* MTRR registers */
1522         case 0xfe:
1523         case 0x200 ... 0x2ff:
1524                 data = 0;
1525                 break;
1526         case 0xcd: /* fsb frequency */
1527                 data = 3;
1528                 break;
1529         case MSR_IA32_APICBASE:
1530                 data = vcpu->apic_base;
1531                 break;
1532         case MSR_IA32_MISC_ENABLE:
1533                 data = vcpu->ia32_misc_enable_msr;
1534                 break;
1535 #ifdef CONFIG_X86_64
1536         case MSR_EFER:
1537                 data = vcpu->shadow_efer;
1538                 break;
1539 #endif
1540         default:
1541                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1542                 return 1;
1543         }
1544         *pdata = data;
1545         return 0;
1546 }
1547 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1548
1549 /*
1550  * Reads an msr value (of 'msr_index') into 'pdata'.
1551  * Returns 0 on success, non-0 otherwise.
1552  * Assumes vcpu_load() was already called.
1553  */
1554 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1555 {
1556         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1557 }
1558
1559 #ifdef CONFIG_X86_64
1560
1561 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1562 {
1563         if (efer & EFER_RESERVED_BITS) {
1564                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1565                        efer);
1566                 inject_gp(vcpu);
1567                 return;
1568         }
1569
1570         if (is_paging(vcpu)
1571             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1572                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1573                 inject_gp(vcpu);
1574                 return;
1575         }
1576
1577         kvm_arch_ops->set_efer(vcpu, efer);
1578
1579         efer &= ~EFER_LMA;
1580         efer |= vcpu->shadow_efer & EFER_LMA;
1581
1582         vcpu->shadow_efer = efer;
1583 }
1584
1585 #endif
1586
1587 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1588 {
1589         switch (msr) {
1590 #ifdef CONFIG_X86_64
1591         case MSR_EFER:
1592                 set_efer(vcpu, data);
1593                 break;
1594 #endif
1595         case MSR_IA32_MC0_STATUS:
1596                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1597                        __FUNCTION__, data);
1598                 break;
1599         case MSR_IA32_MCG_STATUS:
1600                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1601                         __FUNCTION__, data);
1602                 break;
1603         case MSR_IA32_UCODE_REV:
1604         case MSR_IA32_UCODE_WRITE:
1605         case 0x200 ... 0x2ff: /* MTRRs */
1606                 break;
1607         case MSR_IA32_APICBASE:
1608                 vcpu->apic_base = data;
1609                 break;
1610         case MSR_IA32_MISC_ENABLE:
1611                 vcpu->ia32_misc_enable_msr = data;
1612                 break;
1613         /*
1614          * This is the 'probe whether the host is KVM' logic:
1615          */
1616         case MSR_KVM_API_MAGIC:
1617                 return vcpu_register_para(vcpu, data);
1618
1619         default:
1620                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1621                 return 1;
1622         }
1623         return 0;
1624 }
1625 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1626
1627 /*
1628  * Writes msr value into into the appropriate "register".
1629  * Returns 0 on success, non-0 otherwise.
1630  * Assumes vcpu_load() was already called.
1631  */
1632 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1633 {
1634         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1635 }
1636
1637 void kvm_resched(struct kvm_vcpu *vcpu)
1638 {
1639         if (!need_resched())
1640                 return;
1641         vcpu_put(vcpu);
1642         cond_resched();
1643         vcpu_load(vcpu);
1644 }
1645 EXPORT_SYMBOL_GPL(kvm_resched);
1646
1647 void load_msrs(struct vmx_msr_entry *e, int n)
1648 {
1649         int i;
1650
1651         for (i = 0; i < n; ++i)
1652                 wrmsrl(e[i].index, e[i].data);
1653 }
1654 EXPORT_SYMBOL_GPL(load_msrs);
1655
1656 void save_msrs(struct vmx_msr_entry *e, int n)
1657 {
1658         int i;
1659
1660         for (i = 0; i < n; ++i)
1661                 rdmsrl(e[i].index, e[i].data);
1662 }
1663 EXPORT_SYMBOL_GPL(save_msrs);
1664
1665 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1666 {
1667         int i;
1668         u32 function;
1669         struct kvm_cpuid_entry *e, *best;
1670
1671         kvm_arch_ops->cache_regs(vcpu);
1672         function = vcpu->regs[VCPU_REGS_RAX];
1673         vcpu->regs[VCPU_REGS_RAX] = 0;
1674         vcpu->regs[VCPU_REGS_RBX] = 0;
1675         vcpu->regs[VCPU_REGS_RCX] = 0;
1676         vcpu->regs[VCPU_REGS_RDX] = 0;
1677         best = NULL;
1678         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1679                 e = &vcpu->cpuid_entries[i];
1680                 if (e->function == function) {
1681                         best = e;
1682                         break;
1683                 }
1684                 /*
1685                  * Both basic or both extended?
1686                  */
1687                 if (((e->function ^ function) & 0x80000000) == 0)
1688                         if (!best || e->function > best->function)
1689                                 best = e;
1690         }
1691         if (best) {
1692                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1693                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1694                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1695                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1696         }
1697         kvm_arch_ops->decache_regs(vcpu);
1698         kvm_arch_ops->skip_emulated_instruction(vcpu);
1699 }
1700 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1701
1702 static int pio_copy_data(struct kvm_vcpu *vcpu)
1703 {
1704         void *p = vcpu->pio_data;
1705         void *q;
1706         unsigned bytes;
1707         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1708
1709         kvm_arch_ops->vcpu_put(vcpu);
1710         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1711                  PAGE_KERNEL);
1712         if (!q) {
1713                 kvm_arch_ops->vcpu_load(vcpu);
1714                 free_pio_guest_pages(vcpu);
1715                 return -ENOMEM;
1716         }
1717         q += vcpu->pio.guest_page_offset;
1718         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1719         if (vcpu->pio.in)
1720                 memcpy(q, p, bytes);
1721         else
1722                 memcpy(p, q, bytes);
1723         q -= vcpu->pio.guest_page_offset;
1724         vunmap(q);
1725         kvm_arch_ops->vcpu_load(vcpu);
1726         free_pio_guest_pages(vcpu);
1727         return 0;
1728 }
1729
1730 static int complete_pio(struct kvm_vcpu *vcpu)
1731 {
1732         struct kvm_pio_request *io = &vcpu->pio;
1733         long delta;
1734         int r;
1735
1736         kvm_arch_ops->cache_regs(vcpu);
1737
1738         if (!io->string) {
1739                 if (io->in)
1740                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1741                                io->size);
1742         } else {
1743                 if (io->in) {
1744                         r = pio_copy_data(vcpu);
1745                         if (r) {
1746                                 kvm_arch_ops->cache_regs(vcpu);
1747                                 return r;
1748                         }
1749                 }
1750
1751                 delta = 1;
1752                 if (io->rep) {
1753                         delta *= io->cur_count;
1754                         /*
1755                          * The size of the register should really depend on
1756                          * current address size.
1757                          */
1758                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1759                 }
1760                 if (io->down)
1761                         delta = -delta;
1762                 delta *= io->size;
1763                 if (io->in)
1764                         vcpu->regs[VCPU_REGS_RDI] += delta;
1765                 else
1766                         vcpu->regs[VCPU_REGS_RSI] += delta;
1767         }
1768
1769         kvm_arch_ops->decache_regs(vcpu);
1770
1771         io->count -= io->cur_count;
1772         io->cur_count = 0;
1773
1774         if (!io->count)
1775                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1776         return 0;
1777 }
1778
1779 static void kernel_pio(struct kvm_io_device *pio_dev,
1780                        struct kvm_vcpu *vcpu,
1781                        void *pd)
1782 {
1783         /* TODO: String I/O for in kernel device */
1784
1785         if (vcpu->pio.in)
1786                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1787                                   vcpu->pio.size,
1788                                   pd);
1789         else
1790                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1791                                    vcpu->pio.size,
1792                                    pd);
1793 }
1794
1795 static void pio_string_write(struct kvm_io_device *pio_dev,
1796                              struct kvm_vcpu *vcpu)
1797 {
1798         struct kvm_pio_request *io = &vcpu->pio;
1799         void *pd = vcpu->pio_data;
1800         int i;
1801
1802         for (i = 0; i < io->cur_count; i++) {
1803                 kvm_iodevice_write(pio_dev, io->port,
1804                                    io->size,
1805                                    pd);
1806                 pd += io->size;
1807         }
1808 }
1809
1810 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1811                   int size, unsigned long count, int string, int down,
1812                   gva_t address, int rep, unsigned port)
1813 {
1814         unsigned now, in_page;
1815         int i, ret = 0;
1816         int nr_pages = 1;
1817         struct page *page;
1818         struct kvm_io_device *pio_dev;
1819
1820         vcpu->run->exit_reason = KVM_EXIT_IO;
1821         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1822         vcpu->run->io.size = size;
1823         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1824         vcpu->run->io.count = count;
1825         vcpu->run->io.port = port;
1826         vcpu->pio.count = count;
1827         vcpu->pio.cur_count = count;
1828         vcpu->pio.size = size;
1829         vcpu->pio.in = in;
1830         vcpu->pio.port = port;
1831         vcpu->pio.string = string;
1832         vcpu->pio.down = down;
1833         vcpu->pio.guest_page_offset = offset_in_page(address);
1834         vcpu->pio.rep = rep;
1835
1836         pio_dev = vcpu_find_pio_dev(vcpu, port);
1837         if (!string) {
1838                 kvm_arch_ops->cache_regs(vcpu);
1839                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1840                 kvm_arch_ops->decache_regs(vcpu);
1841                 if (pio_dev) {
1842                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1843                         complete_pio(vcpu);
1844                         return 1;
1845                 }
1846                 return 0;
1847         }
1848
1849         if (!count) {
1850                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1851                 return 1;
1852         }
1853
1854         now = min(count, PAGE_SIZE / size);
1855
1856         if (!down)
1857                 in_page = PAGE_SIZE - offset_in_page(address);
1858         else
1859                 in_page = offset_in_page(address) + size;
1860         now = min(count, (unsigned long)in_page / size);
1861         if (!now) {
1862                 /*
1863                  * String I/O straddles page boundary.  Pin two guest pages
1864                  * so that we satisfy atomicity constraints.  Do just one
1865                  * transaction to avoid complexity.
1866                  */
1867                 nr_pages = 2;
1868                 now = 1;
1869         }
1870         if (down) {
1871                 /*
1872                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1873                  */
1874                 printk(KERN_ERR "kvm: guest string pio down\n");
1875                 inject_gp(vcpu);
1876                 return 1;
1877         }
1878         vcpu->run->io.count = now;
1879         vcpu->pio.cur_count = now;
1880
1881         for (i = 0; i < nr_pages; ++i) {
1882                 spin_lock(&vcpu->kvm->lock);
1883                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1884                 if (page)
1885                         get_page(page);
1886                 vcpu->pio.guest_pages[i] = page;
1887                 spin_unlock(&vcpu->kvm->lock);
1888                 if (!page) {
1889                         inject_gp(vcpu);
1890                         free_pio_guest_pages(vcpu);
1891                         return 1;
1892                 }
1893         }
1894
1895         if (!vcpu->pio.in) {
1896                 /* string PIO write */
1897                 ret = pio_copy_data(vcpu);
1898                 if (ret >= 0 && pio_dev) {
1899                         pio_string_write(pio_dev, vcpu);
1900                         complete_pio(vcpu);
1901                         if (vcpu->pio.count == 0)
1902                                 ret = 1;
1903                 }
1904         } else if (pio_dev)
1905                 printk(KERN_ERR "no string pio read support yet, "
1906                        "port %x size %d count %ld\n",
1907                         port, size, count);
1908
1909         return ret;
1910 }
1911 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1912
1913 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1914 {
1915         int r;
1916         sigset_t sigsaved;
1917
1918         vcpu_load(vcpu);
1919
1920         if (vcpu->sigset_active)
1921                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1922
1923         /* re-sync apic's tpr */
1924         vcpu->cr8 = kvm_run->cr8;
1925
1926         if (vcpu->pio.cur_count) {
1927                 r = complete_pio(vcpu);
1928                 if (r)
1929                         goto out;
1930         }
1931
1932         if (vcpu->mmio_needed) {
1933                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1934                 vcpu->mmio_read_completed = 1;
1935                 vcpu->mmio_needed = 0;
1936                 r = emulate_instruction(vcpu, kvm_run,
1937                                         vcpu->mmio_fault_cr2, 0);
1938                 if (r == EMULATE_DO_MMIO) {
1939                         /*
1940                          * Read-modify-write.  Back to userspace.
1941                          */
1942                         r = 0;
1943                         goto out;
1944                 }
1945         }
1946
1947         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1948                 kvm_arch_ops->cache_regs(vcpu);
1949                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1950                 kvm_arch_ops->decache_regs(vcpu);
1951         }
1952
1953         r = kvm_arch_ops->run(vcpu, kvm_run);
1954
1955 out:
1956         if (vcpu->sigset_active)
1957                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1958
1959         vcpu_put(vcpu);
1960         return r;
1961 }
1962
1963 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1964                                    struct kvm_regs *regs)
1965 {
1966         vcpu_load(vcpu);
1967
1968         kvm_arch_ops->cache_regs(vcpu);
1969
1970         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1971         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1972         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1973         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1974         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1975         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1976         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1977         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1978 #ifdef CONFIG_X86_64
1979         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1980         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1981         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1982         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1983         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1984         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1985         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1986         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1987 #endif
1988
1989         regs->rip = vcpu->rip;
1990         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1991
1992         /*
1993          * Don't leak debug flags in case they were set for guest debugging
1994          */
1995         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1996                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1997
1998         vcpu_put(vcpu);
1999
2000         return 0;
2001 }
2002
2003 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2004                                    struct kvm_regs *regs)
2005 {
2006         vcpu_load(vcpu);
2007
2008         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2009         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2010         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2011         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2012         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2013         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2014         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2015         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2016 #ifdef CONFIG_X86_64
2017         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2018         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2019         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2020         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2021         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2022         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2023         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2024         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2025 #endif
2026
2027         vcpu->rip = regs->rip;
2028         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2029
2030         kvm_arch_ops->decache_regs(vcpu);
2031
2032         vcpu_put(vcpu);
2033
2034         return 0;
2035 }
2036
2037 static void get_segment(struct kvm_vcpu *vcpu,
2038                         struct kvm_segment *var, int seg)
2039 {
2040         return kvm_arch_ops->get_segment(vcpu, var, seg);
2041 }
2042
2043 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2044                                     struct kvm_sregs *sregs)
2045 {
2046         struct descriptor_table dt;
2047
2048         vcpu_load(vcpu);
2049
2050         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2051         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2052         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2053         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2054         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2055         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2056
2057         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2058         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2059
2060         kvm_arch_ops->get_idt(vcpu, &dt);
2061         sregs->idt.limit = dt.limit;
2062         sregs->idt.base = dt.base;
2063         kvm_arch_ops->get_gdt(vcpu, &dt);
2064         sregs->gdt.limit = dt.limit;
2065         sregs->gdt.base = dt.base;
2066
2067         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2068         sregs->cr0 = vcpu->cr0;
2069         sregs->cr2 = vcpu->cr2;
2070         sregs->cr3 = vcpu->cr3;
2071         sregs->cr4 = vcpu->cr4;
2072         sregs->cr8 = vcpu->cr8;
2073         sregs->efer = vcpu->shadow_efer;
2074         sregs->apic_base = vcpu->apic_base;
2075
2076         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2077                sizeof sregs->interrupt_bitmap);
2078
2079         vcpu_put(vcpu);
2080
2081         return 0;
2082 }
2083
2084 static void set_segment(struct kvm_vcpu *vcpu,
2085                         struct kvm_segment *var, int seg)
2086 {
2087         return kvm_arch_ops->set_segment(vcpu, var, seg);
2088 }
2089
2090 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2091                                     struct kvm_sregs *sregs)
2092 {
2093         int mmu_reset_needed = 0;
2094         int i;
2095         struct descriptor_table dt;
2096
2097         vcpu_load(vcpu);
2098
2099         dt.limit = sregs->idt.limit;
2100         dt.base = sregs->idt.base;
2101         kvm_arch_ops->set_idt(vcpu, &dt);
2102         dt.limit = sregs->gdt.limit;
2103         dt.base = sregs->gdt.base;
2104         kvm_arch_ops->set_gdt(vcpu, &dt);
2105
2106         vcpu->cr2 = sregs->cr2;
2107         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2108         vcpu->cr3 = sregs->cr3;
2109
2110         vcpu->cr8 = sregs->cr8;
2111
2112         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2113 #ifdef CONFIG_X86_64
2114         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2115 #endif
2116         vcpu->apic_base = sregs->apic_base;
2117
2118         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2119
2120         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2121         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2122
2123         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2124         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2125         if (!is_long_mode(vcpu) && is_pae(vcpu))
2126                 load_pdptrs(vcpu, vcpu->cr3);
2127
2128         if (mmu_reset_needed)
2129                 kvm_mmu_reset_context(vcpu);
2130
2131         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2132                sizeof vcpu->irq_pending);
2133         vcpu->irq_summary = 0;
2134         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2135                 if (vcpu->irq_pending[i])
2136                         __set_bit(i, &vcpu->irq_summary);
2137
2138         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2139         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2140         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2141         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2142         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2143         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2144
2145         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2146         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2147
2148         vcpu_put(vcpu);
2149
2150         return 0;
2151 }
2152
2153 /*
2154  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2155  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2156  *
2157  * This list is modified at module load time to reflect the
2158  * capabilities of the host cpu.
2159  */
2160 static u32 msrs_to_save[] = {
2161         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2162         MSR_K6_STAR,
2163 #ifdef CONFIG_X86_64
2164         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2165 #endif
2166         MSR_IA32_TIME_STAMP_COUNTER,
2167 };
2168
2169 static unsigned num_msrs_to_save;
2170
2171 static u32 emulated_msrs[] = {
2172         MSR_IA32_MISC_ENABLE,
2173 };
2174
2175 static __init void kvm_init_msr_list(void)
2176 {
2177         u32 dummy[2];
2178         unsigned i, j;
2179
2180         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2181                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2182                         continue;
2183                 if (j < i)
2184                         msrs_to_save[j] = msrs_to_save[i];
2185                 j++;
2186         }
2187         num_msrs_to_save = j;
2188 }
2189
2190 /*
2191  * Adapt set_msr() to msr_io()'s calling convention
2192  */
2193 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2194 {
2195         return kvm_set_msr(vcpu, index, *data);
2196 }
2197
2198 /*
2199  * Read or write a bunch of msrs. All parameters are kernel addresses.
2200  *
2201  * @return number of msrs set successfully.
2202  */
2203 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2204                     struct kvm_msr_entry *entries,
2205                     int (*do_msr)(struct kvm_vcpu *vcpu,
2206                                   unsigned index, u64 *data))
2207 {
2208         int i;
2209
2210         vcpu_load(vcpu);
2211
2212         for (i = 0; i < msrs->nmsrs; ++i)
2213                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2214                         break;
2215
2216         vcpu_put(vcpu);
2217
2218         return i;
2219 }
2220
2221 /*
2222  * Read or write a bunch of msrs. Parameters are user addresses.
2223  *
2224  * @return number of msrs set successfully.
2225  */
2226 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2227                   int (*do_msr)(struct kvm_vcpu *vcpu,
2228                                 unsigned index, u64 *data),
2229                   int writeback)
2230 {
2231         struct kvm_msrs msrs;
2232         struct kvm_msr_entry *entries;
2233         int r, n;
2234         unsigned size;
2235
2236         r = -EFAULT;
2237         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2238                 goto out;
2239
2240         r = -E2BIG;
2241         if (msrs.nmsrs >= MAX_IO_MSRS)
2242                 goto out;
2243
2244         r = -ENOMEM;
2245         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2246         entries = vmalloc(size);
2247         if (!entries)
2248                 goto out;
2249
2250         r = -EFAULT;
2251         if (copy_from_user(entries, user_msrs->entries, size))
2252                 goto out_free;
2253
2254         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2255         if (r < 0)
2256                 goto out_free;
2257
2258         r = -EFAULT;
2259         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2260                 goto out_free;
2261
2262         r = n;
2263
2264 out_free:
2265         vfree(entries);
2266 out:
2267         return r;
2268 }
2269
2270 /*
2271  * Translate a guest virtual address to a guest physical address.
2272  */
2273 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2274                                     struct kvm_translation *tr)
2275 {
2276         unsigned long vaddr = tr->linear_address;
2277         gpa_t gpa;
2278
2279         vcpu_load(vcpu);
2280         spin_lock(&vcpu->kvm->lock);
2281         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2282         tr->physical_address = gpa;
2283         tr->valid = gpa != UNMAPPED_GVA;
2284         tr->writeable = 1;
2285         tr->usermode = 0;
2286         spin_unlock(&vcpu->kvm->lock);
2287         vcpu_put(vcpu);
2288
2289         return 0;
2290 }
2291
2292 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2293                                     struct kvm_interrupt *irq)
2294 {
2295         if (irq->irq < 0 || irq->irq >= 256)
2296                 return -EINVAL;
2297         vcpu_load(vcpu);
2298
2299         set_bit(irq->irq, vcpu->irq_pending);
2300         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2301
2302         vcpu_put(vcpu);
2303
2304         return 0;
2305 }
2306
2307 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2308                                       struct kvm_debug_guest *dbg)
2309 {
2310         int r;
2311
2312         vcpu_load(vcpu);
2313
2314         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2315
2316         vcpu_put(vcpu);
2317
2318         return r;
2319 }
2320
2321 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2322                                     unsigned long address,
2323                                     int *type)
2324 {
2325         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2326         unsigned long pgoff;
2327         struct page *page;
2328
2329         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2330         if (pgoff == 0)
2331                 page = virt_to_page(vcpu->run);
2332         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2333                 page = virt_to_page(vcpu->pio_data);
2334         else
2335                 return NOPAGE_SIGBUS;
2336         get_page(page);
2337         if (type != NULL)
2338                 *type = VM_FAULT_MINOR;
2339
2340         return page;
2341 }
2342
2343 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2344         .nopage = kvm_vcpu_nopage,
2345 };
2346
2347 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2348 {
2349         vma->vm_ops = &kvm_vcpu_vm_ops;
2350         return 0;
2351 }
2352
2353 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2354 {
2355         struct kvm_vcpu *vcpu = filp->private_data;
2356
2357         fput(vcpu->kvm->filp);
2358         return 0;
2359 }
2360
2361 static struct file_operations kvm_vcpu_fops = {
2362         .release        = kvm_vcpu_release,
2363         .unlocked_ioctl = kvm_vcpu_ioctl,
2364         .compat_ioctl   = kvm_vcpu_ioctl,
2365         .mmap           = kvm_vcpu_mmap,
2366 };
2367
2368 /*
2369  * Allocates an inode for the vcpu.
2370  */
2371 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2372 {
2373         int fd, r;
2374         struct inode *inode;
2375         struct file *file;
2376
2377         r = anon_inode_getfd(&fd, &inode, &file,
2378                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2379         if (r)
2380                 return r;
2381         atomic_inc(&vcpu->kvm->filp->f_count);
2382         return fd;
2383 }
2384
2385 /*
2386  * Creates some virtual cpus.  Good luck creating more than one.
2387  */
2388 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2389 {
2390         int r;
2391         struct kvm_vcpu *vcpu;
2392         struct page *page;
2393
2394         r = -EINVAL;
2395         if (!valid_vcpu(n))
2396                 goto out;
2397
2398         vcpu = &kvm->vcpus[n];
2399         vcpu->vcpu_id = n;
2400
2401         mutex_lock(&vcpu->mutex);
2402
2403         if (vcpu->vmcs) {
2404                 mutex_unlock(&vcpu->mutex);
2405                 return -EEXIST;
2406         }
2407
2408         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2409         r = -ENOMEM;
2410         if (!page)
2411                 goto out_unlock;
2412         vcpu->run = page_address(page);
2413
2414         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2415         r = -ENOMEM;
2416         if (!page)
2417                 goto out_free_run;
2418         vcpu->pio_data = page_address(page);
2419
2420         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2421                                            FX_IMAGE_ALIGN);
2422         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2423         vcpu->cr0 = 0x10;
2424
2425         r = kvm_arch_ops->vcpu_create(vcpu);
2426         if (r < 0)
2427                 goto out_free_vcpus;
2428
2429         r = kvm_mmu_create(vcpu);
2430         if (r < 0)
2431                 goto out_free_vcpus;
2432
2433         kvm_arch_ops->vcpu_load(vcpu);
2434         r = kvm_mmu_setup(vcpu);
2435         if (r >= 0)
2436                 r = kvm_arch_ops->vcpu_setup(vcpu);
2437         vcpu_put(vcpu);
2438
2439         if (r < 0)
2440                 goto out_free_vcpus;
2441
2442         r = create_vcpu_fd(vcpu);
2443         if (r < 0)
2444                 goto out_free_vcpus;
2445
2446         spin_lock(&kvm_lock);
2447         if (n >= kvm->nvcpus)
2448                 kvm->nvcpus = n + 1;
2449         spin_unlock(&kvm_lock);
2450
2451         return r;
2452
2453 out_free_vcpus:
2454         kvm_free_vcpu(vcpu);
2455 out_free_run:
2456         free_page((unsigned long)vcpu->run);
2457         vcpu->run = NULL;
2458 out_unlock:
2459         mutex_unlock(&vcpu->mutex);
2460 out:
2461         return r;
2462 }
2463
2464 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2465 {
2466         u64 efer;
2467         int i;
2468         struct kvm_cpuid_entry *e, *entry;
2469
2470         rdmsrl(MSR_EFER, efer);
2471         entry = NULL;
2472         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2473                 e = &vcpu->cpuid_entries[i];
2474                 if (e->function == 0x80000001) {
2475                         entry = e;
2476                         break;
2477                 }
2478         }
2479         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2480                 entry->edx &= ~(1 << 20);
2481                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2482         }
2483 }
2484
2485 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2486                                     struct kvm_cpuid *cpuid,
2487                                     struct kvm_cpuid_entry __user *entries)
2488 {
2489         int r;
2490
2491         r = -E2BIG;
2492         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2493                 goto out;
2494         r = -EFAULT;
2495         if (copy_from_user(&vcpu->cpuid_entries, entries,
2496                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2497                 goto out;
2498         vcpu->cpuid_nent = cpuid->nent;
2499         cpuid_fix_nx_cap(vcpu);
2500         return 0;
2501
2502 out:
2503         return r;
2504 }
2505
2506 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2507 {
2508         if (sigset) {
2509                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2510                 vcpu->sigset_active = 1;
2511                 vcpu->sigset = *sigset;
2512         } else
2513                 vcpu->sigset_active = 0;
2514         return 0;
2515 }
2516
2517 /*
2518  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2519  * we have asm/x86/processor.h
2520  */
2521 struct fxsave {
2522         u16     cwd;
2523         u16     swd;
2524         u16     twd;
2525         u16     fop;
2526         u64     rip;
2527         u64     rdp;
2528         u32     mxcsr;
2529         u32     mxcsr_mask;
2530         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2531 #ifdef CONFIG_X86_64
2532         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2533 #else
2534         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2535 #endif
2536 };
2537
2538 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2539 {
2540         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2541
2542         vcpu_load(vcpu);
2543
2544         memcpy(fpu->fpr, fxsave->st_space, 128);
2545         fpu->fcw = fxsave->cwd;
2546         fpu->fsw = fxsave->swd;
2547         fpu->ftwx = fxsave->twd;
2548         fpu->last_opcode = fxsave->fop;
2549         fpu->last_ip = fxsave->rip;
2550         fpu->last_dp = fxsave->rdp;
2551         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2552
2553         vcpu_put(vcpu);
2554
2555         return 0;
2556 }
2557
2558 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2559 {
2560         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2561
2562         vcpu_load(vcpu);
2563
2564         memcpy(fxsave->st_space, fpu->fpr, 128);
2565         fxsave->cwd = fpu->fcw;
2566         fxsave->swd = fpu->fsw;
2567         fxsave->twd = fpu->ftwx;
2568         fxsave->fop = fpu->last_opcode;
2569         fxsave->rip = fpu->last_ip;
2570         fxsave->rdp = fpu->last_dp;
2571         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2572
2573         vcpu_put(vcpu);
2574
2575         return 0;
2576 }
2577
2578 static long kvm_vcpu_ioctl(struct file *filp,
2579                            unsigned int ioctl, unsigned long arg)
2580 {
2581         struct kvm_vcpu *vcpu = filp->private_data;
2582         void __user *argp = (void __user *)arg;
2583         int r = -EINVAL;
2584
2585         switch (ioctl) {
2586         case KVM_RUN:
2587                 r = -EINVAL;
2588                 if (arg)
2589                         goto out;
2590                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2591                 break;
2592         case KVM_GET_REGS: {
2593                 struct kvm_regs kvm_regs;
2594
2595                 memset(&kvm_regs, 0, sizeof kvm_regs);
2596                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2597                 if (r)
2598                         goto out;
2599                 r = -EFAULT;
2600                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2601                         goto out;
2602                 r = 0;
2603                 break;
2604         }
2605         case KVM_SET_REGS: {
2606                 struct kvm_regs kvm_regs;
2607
2608                 r = -EFAULT;
2609                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2610                         goto out;
2611                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2612                 if (r)
2613                         goto out;
2614                 r = 0;
2615                 break;
2616         }
2617         case KVM_GET_SREGS: {
2618                 struct kvm_sregs kvm_sregs;
2619
2620                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2621                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2622                 if (r)
2623                         goto out;
2624                 r = -EFAULT;
2625                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2626                         goto out;
2627                 r = 0;
2628                 break;
2629         }
2630         case KVM_SET_SREGS: {
2631                 struct kvm_sregs kvm_sregs;
2632
2633                 r = -EFAULT;
2634                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2635                         goto out;
2636                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2637                 if (r)
2638                         goto out;
2639                 r = 0;
2640                 break;
2641         }
2642         case KVM_TRANSLATE: {
2643                 struct kvm_translation tr;
2644
2645                 r = -EFAULT;
2646                 if (copy_from_user(&tr, argp, sizeof tr))
2647                         goto out;
2648                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2649                 if (r)
2650                         goto out;
2651                 r = -EFAULT;
2652                 if (copy_to_user(argp, &tr, sizeof tr))
2653                         goto out;
2654                 r = 0;
2655                 break;
2656         }
2657         case KVM_INTERRUPT: {
2658                 struct kvm_interrupt irq;
2659
2660                 r = -EFAULT;
2661                 if (copy_from_user(&irq, argp, sizeof irq))
2662                         goto out;
2663                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2664                 if (r)
2665                         goto out;
2666                 r = 0;
2667                 break;
2668         }
2669         case KVM_DEBUG_GUEST: {
2670                 struct kvm_debug_guest dbg;
2671
2672                 r = -EFAULT;
2673                 if (copy_from_user(&dbg, argp, sizeof dbg))
2674                         goto out;
2675                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2676                 if (r)
2677                         goto out;
2678                 r = 0;
2679                 break;
2680         }
2681         case KVM_GET_MSRS:
2682                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2683                 break;
2684         case KVM_SET_MSRS:
2685                 r = msr_io(vcpu, argp, do_set_msr, 0);
2686                 break;
2687         case KVM_SET_CPUID: {
2688                 struct kvm_cpuid __user *cpuid_arg = argp;
2689                 struct kvm_cpuid cpuid;
2690
2691                 r = -EFAULT;
2692                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2693                         goto out;
2694                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2695                 if (r)
2696                         goto out;
2697                 break;
2698         }
2699         case KVM_SET_SIGNAL_MASK: {
2700                 struct kvm_signal_mask __user *sigmask_arg = argp;
2701                 struct kvm_signal_mask kvm_sigmask;
2702                 sigset_t sigset, *p;
2703
2704                 p = NULL;
2705                 if (argp) {
2706                         r = -EFAULT;
2707                         if (copy_from_user(&kvm_sigmask, argp,
2708                                            sizeof kvm_sigmask))
2709                                 goto out;
2710                         r = -EINVAL;
2711                         if (kvm_sigmask.len != sizeof sigset)
2712                                 goto out;
2713                         r = -EFAULT;
2714                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2715                                            sizeof sigset))
2716                                 goto out;
2717                         p = &sigset;
2718                 }
2719                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2720                 break;
2721         }
2722         case KVM_GET_FPU: {
2723                 struct kvm_fpu fpu;
2724
2725                 memset(&fpu, 0, sizeof fpu);
2726                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2727                 if (r)
2728                         goto out;
2729                 r = -EFAULT;
2730                 if (copy_to_user(argp, &fpu, sizeof fpu))
2731                         goto out;
2732                 r = 0;
2733                 break;
2734         }
2735         case KVM_SET_FPU: {
2736                 struct kvm_fpu fpu;
2737
2738                 r = -EFAULT;
2739                 if (copy_from_user(&fpu, argp, sizeof fpu))
2740                         goto out;
2741                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2742                 if (r)
2743                         goto out;
2744                 r = 0;
2745                 break;
2746         }
2747         default:
2748                 ;
2749         }
2750 out:
2751         return r;
2752 }
2753
2754 static long kvm_vm_ioctl(struct file *filp,
2755                            unsigned int ioctl, unsigned long arg)
2756 {
2757         struct kvm *kvm = filp->private_data;
2758         void __user *argp = (void __user *)arg;
2759         int r = -EINVAL;
2760
2761         switch (ioctl) {
2762         case KVM_CREATE_VCPU:
2763                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2764                 if (r < 0)
2765                         goto out;
2766                 break;
2767         case KVM_SET_MEMORY_REGION: {
2768                 struct kvm_memory_region kvm_mem;
2769
2770                 r = -EFAULT;
2771                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2772                         goto out;
2773                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2774                 if (r)
2775                         goto out;
2776                 break;
2777         }
2778         case KVM_GET_DIRTY_LOG: {
2779                 struct kvm_dirty_log log;
2780
2781                 r = -EFAULT;
2782                 if (copy_from_user(&log, argp, sizeof log))
2783                         goto out;
2784                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2785                 if (r)
2786                         goto out;
2787                 break;
2788         }
2789         case KVM_SET_MEMORY_ALIAS: {
2790                 struct kvm_memory_alias alias;
2791
2792                 r = -EFAULT;
2793                 if (copy_from_user(&alias, argp, sizeof alias))
2794                         goto out;
2795                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2796                 if (r)
2797                         goto out;
2798                 break;
2799         }
2800         default:
2801                 ;
2802         }
2803 out:
2804         return r;
2805 }
2806
2807 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2808                                   unsigned long address,
2809                                   int *type)
2810 {
2811         struct kvm *kvm = vma->vm_file->private_data;
2812         unsigned long pgoff;
2813         struct page *page;
2814
2815         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2816         page = gfn_to_page(kvm, pgoff);
2817         if (!page)
2818                 return NOPAGE_SIGBUS;
2819         get_page(page);
2820         if (type != NULL)
2821                 *type = VM_FAULT_MINOR;
2822
2823         return page;
2824 }
2825
2826 static struct vm_operations_struct kvm_vm_vm_ops = {
2827         .nopage = kvm_vm_nopage,
2828 };
2829
2830 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2831 {
2832         vma->vm_ops = &kvm_vm_vm_ops;
2833         return 0;
2834 }
2835
2836 static struct file_operations kvm_vm_fops = {
2837         .release        = kvm_vm_release,
2838         .unlocked_ioctl = kvm_vm_ioctl,
2839         .compat_ioctl   = kvm_vm_ioctl,
2840         .mmap           = kvm_vm_mmap,
2841 };
2842
2843 static int kvm_dev_ioctl_create_vm(void)
2844 {
2845         int fd, r;
2846         struct inode *inode;
2847         struct file *file;
2848         struct kvm *kvm;
2849
2850         kvm = kvm_create_vm();
2851         if (IS_ERR(kvm))
2852                 return PTR_ERR(kvm);
2853         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2854         if (r) {
2855                 kvm_destroy_vm(kvm);
2856                 return r;
2857         }
2858
2859         kvm->filp = file;
2860
2861         return fd;
2862 }
2863
2864 static long kvm_dev_ioctl(struct file *filp,
2865                           unsigned int ioctl, unsigned long arg)
2866 {
2867         void __user *argp = (void __user *)arg;
2868         long r = -EINVAL;
2869
2870         switch (ioctl) {
2871         case KVM_GET_API_VERSION:
2872                 r = -EINVAL;
2873                 if (arg)
2874                         goto out;
2875                 r = KVM_API_VERSION;
2876                 break;
2877         case KVM_CREATE_VM:
2878                 r = -EINVAL;
2879                 if (arg)
2880                         goto out;
2881                 r = kvm_dev_ioctl_create_vm();
2882                 break;
2883         case KVM_GET_MSR_INDEX_LIST: {
2884                 struct kvm_msr_list __user *user_msr_list = argp;
2885                 struct kvm_msr_list msr_list;
2886                 unsigned n;
2887
2888                 r = -EFAULT;
2889                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2890                         goto out;
2891                 n = msr_list.nmsrs;
2892                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2893                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2894                         goto out;
2895                 r = -E2BIG;
2896                 if (n < num_msrs_to_save)
2897                         goto out;
2898                 r = -EFAULT;
2899                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2900                                  num_msrs_to_save * sizeof(u32)))
2901                         goto out;
2902                 if (copy_to_user(user_msr_list->indices
2903                                  + num_msrs_to_save * sizeof(u32),
2904                                  &emulated_msrs,
2905                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2906                         goto out;
2907                 r = 0;
2908                 break;
2909         }
2910         case KVM_CHECK_EXTENSION:
2911                 /*
2912                  * No extensions defined at present.
2913                  */
2914                 r = 0;
2915                 break;
2916         case KVM_GET_VCPU_MMAP_SIZE:
2917                 r = -EINVAL;
2918                 if (arg)
2919                         goto out;
2920                 r = 2 * PAGE_SIZE;
2921                 break;
2922         default:
2923                 ;
2924         }
2925 out:
2926         return r;
2927 }
2928
2929 static struct file_operations kvm_chardev_ops = {
2930         .open           = kvm_dev_open,
2931         .release        = kvm_dev_release,
2932         .unlocked_ioctl = kvm_dev_ioctl,
2933         .compat_ioctl   = kvm_dev_ioctl,
2934 };
2935
2936 static struct miscdevice kvm_dev = {
2937         KVM_MINOR,
2938         "kvm",
2939         &kvm_chardev_ops,
2940 };
2941
2942 /*
2943  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2944  * cached on it.
2945  */
2946 static void decache_vcpus_on_cpu(int cpu)
2947 {
2948         struct kvm *vm;
2949         struct kvm_vcpu *vcpu;
2950         int i;
2951
2952         spin_lock(&kvm_lock);
2953         list_for_each_entry(vm, &vm_list, vm_list)
2954                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2955                         vcpu = &vm->vcpus[i];
2956                         /*
2957                          * If the vcpu is locked, then it is running on some
2958                          * other cpu and therefore it is not cached on the
2959                          * cpu in question.
2960                          *
2961                          * If it's not locked, check the last cpu it executed
2962                          * on.
2963                          */
2964                         if (mutex_trylock(&vcpu->mutex)) {
2965                                 if (vcpu->cpu == cpu) {
2966                                         kvm_arch_ops->vcpu_decache(vcpu);
2967                                         vcpu->cpu = -1;
2968                                 }
2969                                 mutex_unlock(&vcpu->mutex);
2970                         }
2971                 }
2972         spin_unlock(&kvm_lock);
2973 }
2974
2975 static void hardware_enable(void *junk)
2976 {
2977         int cpu = raw_smp_processor_id();
2978
2979         if (cpu_isset(cpu, cpus_hardware_enabled))
2980                 return;
2981         cpu_set(cpu, cpus_hardware_enabled);
2982         kvm_arch_ops->hardware_enable(NULL);
2983 }
2984
2985 static void hardware_disable(void *junk)
2986 {
2987         int cpu = raw_smp_processor_id();
2988
2989         if (!cpu_isset(cpu, cpus_hardware_enabled))
2990                 return;
2991         cpu_clear(cpu, cpus_hardware_enabled);
2992         decache_vcpus_on_cpu(cpu);
2993         kvm_arch_ops->hardware_disable(NULL);
2994 }
2995
2996 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2997                            void *v)
2998 {
2999         int cpu = (long)v;
3000
3001         switch (val) {
3002         case CPU_DYING:
3003         case CPU_DYING_FROZEN:
3004                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3005                        cpu);
3006                 hardware_disable(NULL);
3007                 break;
3008         case CPU_UP_CANCELED:
3009         case CPU_UP_CANCELED_FROZEN:
3010                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3011                        cpu);
3012                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3013                 break;
3014         case CPU_ONLINE:
3015         case CPU_ONLINE_FROZEN:
3016                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3017                        cpu);
3018                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3019                 break;
3020         }
3021         return NOTIFY_OK;
3022 }
3023
3024 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3025                        void *v)
3026 {
3027         if (val == SYS_RESTART) {
3028                 /*
3029                  * Some (well, at least mine) BIOSes hang on reboot if
3030                  * in vmx root mode.
3031                  */
3032                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3033                 on_each_cpu(hardware_disable, NULL, 0, 1);
3034         }
3035         return NOTIFY_OK;
3036 }
3037
3038 static struct notifier_block kvm_reboot_notifier = {
3039         .notifier_call = kvm_reboot,
3040         .priority = 0,
3041 };
3042
3043 void kvm_io_bus_init(struct kvm_io_bus *bus)
3044 {
3045         memset(bus, 0, sizeof(*bus));
3046 }
3047
3048 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3049 {
3050         int i;
3051
3052         for (i = 0; i < bus->dev_count; i++) {
3053                 struct kvm_io_device *pos = bus->devs[i];
3054
3055                 kvm_iodevice_destructor(pos);
3056         }
3057 }
3058
3059 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3060 {
3061         int i;
3062
3063         for (i = 0; i < bus->dev_count; i++) {
3064                 struct kvm_io_device *pos = bus->devs[i];
3065
3066                 if (pos->in_range(pos, addr))
3067                         return pos;
3068         }
3069
3070         return NULL;
3071 }
3072
3073 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3074 {
3075         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3076
3077         bus->devs[bus->dev_count++] = dev;
3078 }
3079
3080 static struct notifier_block kvm_cpu_notifier = {
3081         .notifier_call = kvm_cpu_hotplug,
3082         .priority = 20, /* must be > scheduler priority */
3083 };
3084
3085 static u64 stat_get(void *_offset)
3086 {
3087         unsigned offset = (long)_offset;
3088         u64 total = 0;
3089         struct kvm *kvm;
3090         struct kvm_vcpu *vcpu;
3091         int i;
3092
3093         spin_lock(&kvm_lock);
3094         list_for_each_entry(kvm, &vm_list, vm_list)
3095                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3096                         vcpu = &kvm->vcpus[i];
3097                         total += *(u32 *)((void *)vcpu + offset);
3098                 }
3099         spin_unlock(&kvm_lock);
3100         return total;
3101 }
3102
3103 static void stat_set(void *offset, u64 val)
3104 {
3105 }
3106
3107 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3108
3109 static __init void kvm_init_debug(void)
3110 {
3111         struct kvm_stats_debugfs_item *p;
3112
3113         debugfs_dir = debugfs_create_dir("kvm", NULL);
3114         for (p = debugfs_entries; p->name; ++p)
3115                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3116                                                 (void *)(long)p->offset,
3117                                                 &stat_fops);
3118 }
3119
3120 static void kvm_exit_debug(void)
3121 {
3122         struct kvm_stats_debugfs_item *p;
3123
3124         for (p = debugfs_entries; p->name; ++p)
3125                 debugfs_remove(p->dentry);
3126         debugfs_remove(debugfs_dir);
3127 }
3128
3129 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3130 {
3131         hardware_disable(NULL);
3132         return 0;
3133 }
3134
3135 static int kvm_resume(struct sys_device *dev)
3136 {
3137         hardware_enable(NULL);
3138         return 0;
3139 }
3140
3141 static struct sysdev_class kvm_sysdev_class = {
3142         set_kset_name("kvm"),
3143         .suspend = kvm_suspend,
3144         .resume = kvm_resume,
3145 };
3146
3147 static struct sys_device kvm_sysdev = {
3148         .id = 0,
3149         .cls = &kvm_sysdev_class,
3150 };
3151
3152 hpa_t bad_page_address;
3153
3154 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3155 {
3156         int r;
3157
3158         if (kvm_arch_ops) {
3159                 printk(KERN_ERR "kvm: already loaded the other module\n");
3160                 return -EEXIST;
3161         }
3162
3163         if (!ops->cpu_has_kvm_support()) {
3164                 printk(KERN_ERR "kvm: no hardware support\n");
3165                 return -EOPNOTSUPP;
3166         }
3167         if (ops->disabled_by_bios()) {
3168                 printk(KERN_ERR "kvm: disabled by bios\n");
3169                 return -EOPNOTSUPP;
3170         }
3171
3172         kvm_arch_ops = ops;
3173
3174         r = kvm_arch_ops->hardware_setup();
3175         if (r < 0)
3176                 goto out;
3177
3178         on_each_cpu(hardware_enable, NULL, 0, 1);
3179         r = register_cpu_notifier(&kvm_cpu_notifier);
3180         if (r)
3181                 goto out_free_1;
3182         register_reboot_notifier(&kvm_reboot_notifier);
3183
3184         r = sysdev_class_register(&kvm_sysdev_class);
3185         if (r)
3186                 goto out_free_2;
3187
3188         r = sysdev_register(&kvm_sysdev);
3189         if (r)
3190                 goto out_free_3;
3191
3192         kvm_chardev_ops.owner = module;
3193
3194         r = misc_register(&kvm_dev);
3195         if (r) {
3196                 printk (KERN_ERR "kvm: misc device register failed\n");
3197                 goto out_free;
3198         }
3199
3200         return r;
3201
3202 out_free:
3203         sysdev_unregister(&kvm_sysdev);
3204 out_free_3:
3205         sysdev_class_unregister(&kvm_sysdev_class);
3206 out_free_2:
3207         unregister_reboot_notifier(&kvm_reboot_notifier);
3208         unregister_cpu_notifier(&kvm_cpu_notifier);
3209 out_free_1:
3210         on_each_cpu(hardware_disable, NULL, 0, 1);
3211         kvm_arch_ops->hardware_unsetup();
3212 out:
3213         kvm_arch_ops = NULL;
3214         return r;
3215 }
3216
3217 void kvm_exit_arch(void)
3218 {
3219         misc_deregister(&kvm_dev);
3220         sysdev_unregister(&kvm_sysdev);
3221         sysdev_class_unregister(&kvm_sysdev_class);
3222         unregister_reboot_notifier(&kvm_reboot_notifier);
3223         unregister_cpu_notifier(&kvm_cpu_notifier);
3224         on_each_cpu(hardware_disable, NULL, 0, 1);
3225         kvm_arch_ops->hardware_unsetup();
3226         kvm_arch_ops = NULL;
3227 }
3228
3229 static __init int kvm_init(void)
3230 {
3231         static struct page *bad_page;
3232         int r;
3233
3234         r = kvm_mmu_module_init();
3235         if (r)
3236                 goto out4;
3237
3238         kvm_init_debug();
3239
3240         kvm_init_msr_list();
3241
3242         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3243                 r = -ENOMEM;
3244                 goto out;
3245         }
3246
3247         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3248         memset(__va(bad_page_address), 0, PAGE_SIZE);
3249
3250         return 0;
3251
3252 out:
3253         kvm_exit_debug();
3254         kvm_mmu_module_exit();
3255 out4:
3256         return r;
3257 }
3258
3259 static __exit void kvm_exit(void)
3260 {
3261         kvm_exit_debug();
3262         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3263         kvm_mmu_module_exit();
3264 }
3265
3266 module_init(kvm_init)
3267 module_exit(kvm_exit)
3268
3269 EXPORT_SYMBOL_GPL(kvm_init_arch);
3270 EXPORT_SYMBOL_GPL(kvm_exit_arch);