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