SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <linux/magic.h>
  24 #include <asm/processor.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <asm/msr.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/reboot.h>
  33 #include <asm/io.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/highmem.h>
  36 #include <linux/file.h>
  37 #include <asm/desc.h>
  38 #include <linux/sysdev.h>
  39 #include <linux/cpu.h>
  40 #include <linux/file.h>
  41 #include <linux/fs.h>
  42 #include <linux/mount.h>
  43
  44 #include "x86_emulate.h"
  45 #include "segment_descriptor.h"
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 struct kvm_arch_ops *kvm_arch_ops;
  54 struct kvm_stat kvm_stat;
  55 EXPORT_SYMBOL_GPL(kvm_stat);
  56
  57 static struct kvm_stats_debugfs_item {
  58         const char *name;
  59         u32 *data;
  60         struct dentry *dentry;
  61 } debugfs_entries[] = {
  62         { "pf_fixed", &kvm_stat.pf_fixed },
  63         { "pf_guest", &kvm_stat.pf_guest },
  64         { "tlb_flush", &kvm_stat.tlb_flush },
  65         { "invlpg", &kvm_stat.invlpg },
  66         { "exits", &kvm_stat.exits },
  67         { "io_exits", &kvm_stat.io_exits },
  68         { "mmio_exits", &kvm_stat.mmio_exits },
  69         { "signal_exits", &kvm_stat.signal_exits },
  70         { "irq_window", &kvm_stat.irq_window_exits },
  71         { "halt_exits", &kvm_stat.halt_exits },
  72         { "request_irq", &kvm_stat.request_irq_exits },
  73         { "irq_exits", &kvm_stat.irq_exits },
  74         { NULL, NULL }
  75 };
  76
  77 static struct dentry *debugfs_dir;
  78
  79 struct vfsmount *kvmfs_mnt;
  80
  81 #define MAX_IO_MSRS 256
  82
  83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  84 #define LMSW_GUEST_MASK 0x0eULL
  85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  86 #define CR8_RESEVED_BITS (~0x0fULL)
  87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  88
  89 #ifdef CONFIG_X86_64
  90 // LDT or TSS descriptor in the GDT. 16 bytes.
  91 struct segment_descriptor_64 {
  92         struct segment_descriptor s;
  93         u32 base_higher;
  94         u32 pad_zero;
  95 };
  96
  97 #endif
  98
  99 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 100                            unsigned long arg);
 101
 102 static struct inode *kvmfs_inode(struct file_operations *fops)
 103 {
 104         int error = -ENOMEM;
 105         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 106
 107         if (!inode)
 108                 goto eexit_1;
 109
 110         inode->i_fop = fops;
 111
 112         /*
 113          * Mark the inode dirty from the very beginning,
 114          * that way it will never be moved to the dirty
 115          * list because mark_inode_dirty() will think
 116          * that it already _is_ on the dirty list.
 117          */
 118         inode->i_state = I_DIRTY;
 119         inode->i_mode = S_IRUSR | S_IWUSR;
 120         inode->i_uid = current->fsuid;
 121         inode->i_gid = current->fsgid;
 122         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 123         return inode;
 124
 125 eexit_1:
 126         return ERR_PTR(error);
 127 }
 128
 129 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 130 {
 131         struct file *file = get_empty_filp();
 132
 133         if (!file)
 134                 return ERR_PTR(-ENFILE);
 135
 136         file->f_path.mnt = mntget(kvmfs_mnt);
 137         file->f_path.dentry = d_alloc_anon(inode);
 138         if (!file->f_path.dentry)
 139                 return ERR_PTR(-ENOMEM);
 140         file->f_mapping = inode->i_mapping;
 141
 142         file->f_pos = 0;
 143         file->f_flags = O_RDWR;
 144         file->f_op = inode->i_fop;
 145         file->f_mode = FMODE_READ | FMODE_WRITE;
 146         file->f_version = 0;
 147         file->private_data = private_data;
 148         return file;
 149 }
 150
 151 unsigned long segment_base(u16 selector)
 152 {
 153         struct descriptor_table gdt;
 154         struct segment_descriptor *d;
 155         unsigned long table_base;
 156         typedef unsigned long ul;
 157         unsigned long v;
 158
 159         if (selector == 0)
 160                 return 0;
 161
 162         asm ("sgdt %0" : "=m"(gdt));
 163         table_base = gdt.base;
 164
 165         if (selector & 4) {           /* from ldt */
 166                 u16 ldt_selector;
 167
 168                 asm ("sldt %0" : "=g"(ldt_selector));
 169                 table_base = segment_base(ldt_selector);
 170         }
 171         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 172         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 173 #ifdef CONFIG_X86_64
 174         if (d->system == 0
 175             && (d->type == 2 || d->type == 9 || d->type == 11))
 176                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 177 #endif
 178         return v;
 179 }
 180 EXPORT_SYMBOL_GPL(segment_base);
 181
 182 static inline int valid_vcpu(int n)
 183 {
 184         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 185 }
 186
 187 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 188                    void *dest)
 189 {
 190         unsigned char *host_buf = dest;
 191         unsigned long req_size = size;
 192
 193         while (size) {
 194                 hpa_t paddr;
 195                 unsigned now;
 196                 unsigned offset;
 197                 hva_t guest_buf;
 198
 199                 paddr = gva_to_hpa(vcpu, addr);
 200
 201                 if (is_error_hpa(paddr))
 202                         break;
 203
 204                 guest_buf = (hva_t)kmap_atomic(
 205                                         pfn_to_page(paddr >> PAGE_SHIFT),
 206                                         KM_USER0);
 207                 offset = addr & ~PAGE_MASK;
 208                 guest_buf |= offset;
 209                 now = min(size, PAGE_SIZE - offset);
 210                 memcpy(host_buf, (void*)guest_buf, now);
 211                 host_buf += now;
 212                 addr += now;
 213                 size -= now;
 214                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 215         }
 216         return req_size - size;
 217 }
 218 EXPORT_SYMBOL_GPL(kvm_read_guest);
 219
 220 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 221                     void *data)
 222 {
 223         unsigned char *host_buf = data;
 224         unsigned long req_size = size;
 225
 226         while (size) {
 227                 hpa_t paddr;
 228                 unsigned now;
 229                 unsigned offset;
 230                 hva_t guest_buf;
 231                 gfn_t gfn;
 232
 233                 paddr = gva_to_hpa(vcpu, addr);
 234
 235                 if (is_error_hpa(paddr))
 236                         break;
 237
 238                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 239                 mark_page_dirty(vcpu->kvm, gfn);
 240                 guest_buf = (hva_t)kmap_atomic(
 241                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 242                 offset = addr & ~PAGE_MASK;
 243                 guest_buf |= offset;
 244                 now = min(size, PAGE_SIZE - offset);
 245                 memcpy((void*)guest_buf, host_buf, now);
 246                 host_buf += now;
 247                 addr += now;
 248                 size -= now;
 249                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 250         }
 251         return req_size - size;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_write_guest);
 254
 255 /*
 256  * Switches to specified vcpu, until a matching vcpu_put()
 257  */
 258 static void vcpu_load(struct kvm_vcpu *vcpu)
 259 {
 260         mutex_lock(&vcpu->mutex);
 261         kvm_arch_ops->vcpu_load(vcpu);
 262 }
 263
 264 /*
 265  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 266  * if the slot is not populated.
 267  */
 268 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 269 {
 270         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 271
 272         mutex_lock(&vcpu->mutex);
 273         if (!vcpu->vmcs) {
 274                 mutex_unlock(&vcpu->mutex);
 275                 return NULL;
 276         }
 277         kvm_arch_ops->vcpu_load(vcpu);
 278         return vcpu;
 279 }
 280
 281 static void vcpu_put(struct kvm_vcpu *vcpu)
 282 {
 283         kvm_arch_ops->vcpu_put(vcpu);
 284         mutex_unlock(&vcpu->mutex);
 285 }
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290         int i;
 291
 292         if (!kvm)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         spin_lock_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 298                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 299
 300                 mutex_init(&vcpu->mutex);
 301                 vcpu->cpu = -1;
 302                 vcpu->kvm = kvm;
 303                 vcpu->mmu.root_hpa = INVALID_PAGE;
 304                 INIT_LIST_HEAD(&vcpu->free_pages);
 305                 spin_lock(&kvm_lock);
 306                 list_add(&kvm->vm_list, &vm_list);
 307                 spin_unlock(&kvm_lock);
 308         }
 309         return kvm;
 310 }
 311
 312 static int kvm_dev_open(struct inode *inode, struct file *filp)
 313 {
 314         return 0;
 315 }
 316
 317 /*
 318  * Free any memory in @free but not in @dont.
 319  */
 320 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 321                                   struct kvm_memory_slot *dont)
 322 {
 323         int i;
 324
 325         if (!dont || free->phys_mem != dont->phys_mem)
 326                 if (free->phys_mem) {
 327                         for (i = 0; i < free->npages; ++i)
 328                                 if (free->phys_mem[i])
 329                                         __free_page(free->phys_mem[i]);
 330                         vfree(free->phys_mem);
 331                 }
 332
 333         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 334                 vfree(free->dirty_bitmap);
 335
 336         free->phys_mem = NULL;
 337         free->npages = 0;
 338         free->dirty_bitmap = NULL;
 339 }
 340
 341 static void kvm_free_physmem(struct kvm *kvm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < kvm->nmemslots; ++i)
 346                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 347 }
 348
 349 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 350 {
 351         int i;
 352
 353         for (i = 0; i < 2; ++i)
 354                 if (vcpu->pio.guest_pages[i]) {
 355                         __free_page(vcpu->pio.guest_pages[i]);
 356                         vcpu->pio.guest_pages[i] = NULL;
 357                 }
 358 }
 359
 360 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 361 {
 362         if (!vcpu->vmcs)
 363                 return;
 364
 365         vcpu_load(vcpu);
 366         kvm_mmu_destroy(vcpu);
 367         vcpu_put(vcpu);
 368         kvm_arch_ops->vcpu_free(vcpu);
 369         free_page((unsigned long)vcpu->run);
 370         vcpu->run = NULL;
 371         free_page((unsigned long)vcpu->pio_data);
 372         vcpu->pio_data = NULL;
 373         free_pio_guest_pages(vcpu);
 374 }
 375
 376 static void kvm_free_vcpus(struct kvm *kvm)
 377 {
 378         unsigned int i;
 379
 380         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 381                 kvm_free_vcpu(&kvm->vcpus[i]);
 382 }
 383
 384 static int kvm_dev_release(struct inode *inode, struct file *filp)
 385 {
 386         return 0;
 387 }
 388
 389 static void kvm_destroy_vm(struct kvm *kvm)
 390 {
 391         spin_lock(&kvm_lock);
 392         list_del(&kvm->vm_list);
 393         spin_unlock(&kvm_lock);
 394         kvm_free_vcpus(kvm);
 395         kvm_free_physmem(kvm);
 396         kfree(kvm);
 397 }
 398
 399 static int kvm_vm_release(struct inode *inode, struct file *filp)
 400 {
 401         struct kvm *kvm = filp->private_data;
 402
 403         kvm_destroy_vm(kvm);
 404         return 0;
 405 }
 406
 407 static void inject_gp(struct kvm_vcpu *vcpu)
 408 {
 409         kvm_arch_ops->inject_gp(vcpu, 0);
 410 }
 411
 412 /*
 413  * Load the pae pdptrs.  Return true is they are all valid.
 414  */
 415 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 416 {
 417         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 418         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 419         int i;
 420         u64 pdpte;
 421         u64 *pdpt;
 422         int ret;
 423         struct kvm_memory_slot *memslot;
 424
 425         spin_lock(&vcpu->kvm->lock);
 426         memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
 427         /* FIXME: !memslot - emulate? 0xff? */
 428         pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
 429
 430         ret = 1;
 431         for (i = 0; i < 4; ++i) {
 432                 pdpte = pdpt[offset + i];
 433                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 434                         ret = 0;
 435                         goto out;
 436                 }
 437         }
 438
 439         for (i = 0; i < 4; ++i)
 440                 vcpu->pdptrs[i] = pdpt[offset + i];
 441
 442 out:
 443         kunmap_atomic(pdpt, KM_USER0);
 444         spin_unlock(&vcpu->kvm->lock);
 445
 446         return ret;
 447 }
 448
 449 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 450 {
 451         if (cr0 & CR0_RESEVED_BITS) {
 452                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 453                        cr0, vcpu->cr0);
 454                 inject_gp(vcpu);
 455                 return;
 456         }
 457
 458         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 459                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 460                 inject_gp(vcpu);
 461                 return;
 462         }
 463
 464         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 465                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 466                        "and a clear PE flag\n");
 467                 inject_gp(vcpu);
 468                 return;
 469         }
 470
 471         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 472 #ifdef CONFIG_X86_64
 473                 if ((vcpu->shadow_efer & EFER_LME)) {
 474                         int cs_db, cs_l;
 475
 476                         if (!is_pae(vcpu)) {
 477                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 478                                        "in long mode while PAE is disabled\n");
 479                                 inject_gp(vcpu);
 480                                 return;
 481                         }
 482                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 483                         if (cs_l) {
 484                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 485                                        "in long mode while CS.L == 1\n");
 486                                 inject_gp(vcpu);
 487                                 return;
 488
 489                         }
 490                 } else
 491 #endif
 492                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 493                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 494                                "reserved bits\n");
 495                         inject_gp(vcpu);
 496                         return;
 497                 }
 498
 499         }
 500
 501         kvm_arch_ops->set_cr0(vcpu, cr0);
 502         vcpu->cr0 = cr0;
 503
 504         spin_lock(&vcpu->kvm->lock);
 505         kvm_mmu_reset_context(vcpu);
 506         spin_unlock(&vcpu->kvm->lock);
 507         return;
 508 }
 509 EXPORT_SYMBOL_GPL(set_cr0);
 510
 511 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 512 {
 513         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 514         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 515 }
 516 EXPORT_SYMBOL_GPL(lmsw);
 517
 518 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 519 {
 520         if (cr4 & CR4_RESEVED_BITS) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 522                 inject_gp(vcpu);
 523                 return;
 524         }
 525
 526         if (is_long_mode(vcpu)) {
 527                 if (!(cr4 & CR4_PAE_MASK)) {
 528                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 529                                "in long mode\n");
 530                         inject_gp(vcpu);
 531                         return;
 532                 }
 533         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 534                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 535                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 536                 inject_gp(vcpu);
 537         }
 538
 539         if (cr4 & CR4_VMXE_MASK) {
 540                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 541                 inject_gp(vcpu);
 542                 return;
 543         }
 544         kvm_arch_ops->set_cr4(vcpu, cr4);
 545         spin_lock(&vcpu->kvm->lock);
 546         kvm_mmu_reset_context(vcpu);
 547         spin_unlock(&vcpu->kvm->lock);
 548 }
 549 EXPORT_SYMBOL_GPL(set_cr4);
 550
 551 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 552 {
 553         if (is_long_mode(vcpu)) {
 554                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 555                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 556                         inject_gp(vcpu);
 557                         return;
 558                 }
 559         } else {
 560                 if (cr3 & CR3_RESEVED_BITS) {
 561                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 562                         inject_gp(vcpu);
 563                         return;
 564                 }
 565                 if (is_paging(vcpu) && is_pae(vcpu) &&
 566                     !load_pdptrs(vcpu, cr3)) {
 567                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 568                                "reserved bits\n");
 569                         inject_gp(vcpu);
 570                         return;
 571                 }
 572         }
 573
 574         vcpu->cr3 = cr3;
 575         spin_lock(&vcpu->kvm->lock);
 576         /*
 577          * Does the new cr3 value map to physical memory? (Note, we
 578          * catch an invalid cr3 even in real-mode, because it would
 579          * cause trouble later on when we turn on paging anyway.)
 580          *
 581          * A real CPU would silently accept an invalid cr3 and would
 582          * attempt to use it - with largely undefined (and often hard
 583          * to debug) behavior on the guest side.
 584          */
 585         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 586                 inject_gp(vcpu);
 587         else
 588                 vcpu->mmu.new_cr3(vcpu);
 589         spin_unlock(&vcpu->kvm->lock);
 590 }
 591 EXPORT_SYMBOL_GPL(set_cr3);
 592
 593 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 594 {
 595         if ( cr8 & CR8_RESEVED_BITS) {
 596                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 597                 inject_gp(vcpu);
 598                 return;
 599         }
 600         vcpu->cr8 = cr8;
 601 }
 602 EXPORT_SYMBOL_GPL(set_cr8);
 603
 604 void fx_init(struct kvm_vcpu *vcpu)
 605 {
 606         struct __attribute__ ((__packed__)) fx_image_s {
 607                 u16 control; //fcw
 608                 u16 status; //fsw
 609                 u16 tag; // ftw
 610                 u16 opcode; //fop
 611                 u64 ip; // fpu ip
 612                 u64 operand;// fpu dp
 613                 u32 mxcsr;
 614                 u32 mxcsr_mask;
 615
 616         } *fx_image;
 617
 618         fx_save(vcpu->host_fx_image);
 619         fpu_init();
 620         fx_save(vcpu->guest_fx_image);
 621         fx_restore(vcpu->host_fx_image);
 622
 623         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 624         fx_image->mxcsr = 0x1f80;
 625         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 626                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 627 }
 628 EXPORT_SYMBOL_GPL(fx_init);
 629
 630 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 631 {
 632         spin_lock(&vcpu->kvm->lock);
 633         kvm_mmu_slot_remove_write_access(vcpu, slot);
 634         spin_unlock(&vcpu->kvm->lock);
 635 }
 636
 637 /*
 638  * Allocate some memory and give it an address in the guest physical address
 639  * space.
 640  *
 641  * Discontiguous memory is allowed, mostly for framebuffers.
 642  */
 643 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 644                                           struct kvm_memory_region *mem)
 645 {
 646         int r;
 647         gfn_t base_gfn;
 648         unsigned long npages;
 649         unsigned long i;
 650         struct kvm_memory_slot *memslot;
 651         struct kvm_memory_slot old, new;
 652         int memory_config_version;
 653
 654         r = -EINVAL;
 655         /* General sanity checks */
 656         if (mem->memory_size & (PAGE_SIZE - 1))
 657                 goto out;
 658         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 659                 goto out;
 660         if (mem->slot >= KVM_MEMORY_SLOTS)
 661                 goto out;
 662         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 663                 goto out;
 664
 665         memslot = &kvm->memslots[mem->slot];
 666         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 667         npages = mem->memory_size >> PAGE_SHIFT;
 668
 669         if (!npages)
 670                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 671
 672 raced:
 673         spin_lock(&kvm->lock);
 674
 675         memory_config_version = kvm->memory_config_version;
 676         new = old = *memslot;
 677
 678         new.base_gfn = base_gfn;
 679         new.npages = npages;
 680         new.flags = mem->flags;
 681
 682         /* Disallow changing a memory slot's size. */
 683         r = -EINVAL;
 684         if (npages && old.npages && npages != old.npages)
 685                 goto out_unlock;
 686
 687         /* Check for overlaps */
 688         r = -EEXIST;
 689         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 690                 struct kvm_memory_slot *s = &kvm->memslots[i];
 691
 692                 if (s == memslot)
 693                         continue;
 694                 if (!((base_gfn + npages <= s->base_gfn) ||
 695                       (base_gfn >= s->base_gfn + s->npages)))
 696                         goto out_unlock;
 697         }
 698         /*
 699          * Do memory allocations outside lock.  memory_config_version will
 700          * detect any races.
 701          */
 702         spin_unlock(&kvm->lock);
 703
 704         /* Deallocate if slot is being removed */
 705         if (!npages)
 706                 new.phys_mem = NULL;
 707
 708         /* Free page dirty bitmap if unneeded */
 709         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 710                 new.dirty_bitmap = NULL;
 711
 712         r = -ENOMEM;
 713
 714         /* Allocate if a slot is being created */
 715         if (npages && !new.phys_mem) {
 716                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 717
 718                 if (!new.phys_mem)
 719                         goto out_free;
 720
 721                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 722                 for (i = 0; i < npages; ++i) {
 723                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 724                                                      | __GFP_ZERO);
 725                         if (!new.phys_mem[i])
 726                                 goto out_free;
 727                         set_page_private(new.phys_mem[i],0);
 728                 }
 729         }
 730
 731         /* Allocate page dirty bitmap if needed */
 732         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 733                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 734
 735                 new.dirty_bitmap = vmalloc(dirty_bytes);
 736                 if (!new.dirty_bitmap)
 737                         goto out_free;
 738                 memset(new.dirty_bitmap, 0, dirty_bytes);
 739         }
 740
 741         spin_lock(&kvm->lock);
 742
 743         if (memory_config_version != kvm->memory_config_version) {
 744                 spin_unlock(&kvm->lock);
 745                 kvm_free_physmem_slot(&new, &old);
 746                 goto raced;
 747         }
 748
 749         r = -EAGAIN;
 750         if (kvm->busy)
 751                 goto out_unlock;
 752
 753         if (mem->slot >= kvm->nmemslots)
 754                 kvm->nmemslots = mem->slot + 1;
 755
 756         *memslot = new;
 757         ++kvm->memory_config_version;
 758
 759         spin_unlock(&kvm->lock);
 760
 761         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 762                 struct kvm_vcpu *vcpu;
 763
 764                 vcpu = vcpu_load_slot(kvm, i);
 765                 if (!vcpu)
 766                         continue;
 767                 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
 768                         do_remove_write_access(vcpu, mem->slot);
 769                 kvm_mmu_reset_context(vcpu);
 770                 vcpu_put(vcpu);
 771         }
 772
 773         kvm_free_physmem_slot(&old, &new);
 774         return 0;
 775
 776 out_unlock:
 777         spin_unlock(&kvm->lock);
 778 out_free:
 779         kvm_free_physmem_slot(&new, &old);
 780 out:
 781         return r;
 782 }
 783
 784 /*
 785  * Get (and clear) the dirty memory log for a memory slot.
 786  */
 787 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 788                                       struct kvm_dirty_log *log)
 789 {
 790         struct kvm_memory_slot *memslot;
 791         int r, i;
 792         int n;
 793         int cleared;
 794         unsigned long any = 0;
 795
 796         spin_lock(&kvm->lock);
 797
 798         /*
 799          * Prevent changes to guest memory configuration even while the lock
 800          * is not taken.
 801          */
 802         ++kvm->busy;
 803         spin_unlock(&kvm->lock);
 804         r = -EINVAL;
 805         if (log->slot >= KVM_MEMORY_SLOTS)
 806                 goto out;
 807
 808         memslot = &kvm->memslots[log->slot];
 809         r = -ENOENT;
 810         if (!memslot->dirty_bitmap)
 811                 goto out;
 812
 813         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 814
 815         for (i = 0; !any && i < n/sizeof(long); ++i)
 816                 any = memslot->dirty_bitmap[i];
 817
 818         r = -EFAULT;
 819         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 820                 goto out;
 821
 822         if (any) {
 823                 cleared = 0;
 824                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 825                         struct kvm_vcpu *vcpu;
 826
 827                         vcpu = vcpu_load_slot(kvm, i);
 828                         if (!vcpu)
 829                                 continue;
 830                         if (!cleared) {
 831                                 do_remove_write_access(vcpu, log->slot);
 832                                 memset(memslot->dirty_bitmap, 0, n);
 833                                 cleared = 1;
 834                         }
 835                         kvm_arch_ops->tlb_flush(vcpu);
 836                         vcpu_put(vcpu);
 837                 }
 838         }
 839
 840         r = 0;
 841
 842 out:
 843         spin_lock(&kvm->lock);
 844         --kvm->busy;
 845         spin_unlock(&kvm->lock);
 846         return r;
 847 }
 848
 849 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 850 {
 851         int i;
 852
 853         for (i = 0; i < kvm->nmemslots; ++i) {
 854                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 855
 856                 if (gfn >= memslot->base_gfn
 857                     && gfn < memslot->base_gfn + memslot->npages)
 858                         return memslot;
 859         }
 860         return NULL;
 861 }
 862 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 863
 864 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 865 {
 866         int i;
 867         struct kvm_memory_slot *memslot = NULL;
 868         unsigned long rel_gfn;
 869
 870         for (i = 0; i < kvm->nmemslots; ++i) {
 871                 memslot = &kvm->memslots[i];
 872
 873                 if (gfn >= memslot->base_gfn
 874                     && gfn < memslot->base_gfn + memslot->npages) {
 875
 876                         if (!memslot || !memslot->dirty_bitmap)
 877                                 return;
 878
 879                         rel_gfn = gfn - memslot->base_gfn;
 880
 881                         /* avoid RMW */
 882                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 883                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 884                         return;
 885                 }
 886         }
 887 }
 888
 889 static int emulator_read_std(unsigned long addr,
 890                              unsigned long *val,
 891                              unsigned int bytes,
 892                              struct x86_emulate_ctxt *ctxt)
 893 {
 894         struct kvm_vcpu *vcpu = ctxt->vcpu;
 895         void *data = val;
 896
 897         while (bytes) {
 898                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 899                 unsigned offset = addr & (PAGE_SIZE-1);
 900                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 901                 unsigned long pfn;
 902                 struct kvm_memory_slot *memslot;
 903                 void *page;
 904
 905                 if (gpa == UNMAPPED_GVA)
 906                         return X86EMUL_PROPAGATE_FAULT;
 907                 pfn = gpa >> PAGE_SHIFT;
 908                 memslot = gfn_to_memslot(vcpu->kvm, pfn);
 909                 if (!memslot)
 910                         return X86EMUL_UNHANDLEABLE;
 911                 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
 912
 913                 memcpy(data, page + offset, tocopy);
 914
 915                 kunmap_atomic(page, KM_USER0);
 916
 917                 bytes -= tocopy;
 918                 data += tocopy;
 919                 addr += tocopy;
 920         }
 921
 922         return X86EMUL_CONTINUE;
 923 }
 924
 925 static int emulator_write_std(unsigned long addr,
 926                               unsigned long val,
 927                               unsigned int bytes,
 928                               struct x86_emulate_ctxt *ctxt)
 929 {
 930         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 931                addr, bytes);
 932         return X86EMUL_UNHANDLEABLE;
 933 }
 934
 935 static int emulator_read_emulated(unsigned long addr,
 936                                   unsigned long *val,
 937                                   unsigned int bytes,
 938                                   struct x86_emulate_ctxt *ctxt)
 939 {
 940         struct kvm_vcpu *vcpu = ctxt->vcpu;
 941
 942         if (vcpu->mmio_read_completed) {
 943                 memcpy(val, vcpu->mmio_data, bytes);
 944                 vcpu->mmio_read_completed = 0;
 945                 return X86EMUL_CONTINUE;
 946         } else if (emulator_read_std(addr, val, bytes, ctxt)
 947                    == X86EMUL_CONTINUE)
 948                 return X86EMUL_CONTINUE;
 949         else {
 950                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 951
 952                 if (gpa == UNMAPPED_GVA)
 953                         return X86EMUL_PROPAGATE_FAULT;
 954                 vcpu->mmio_needed = 1;
 955                 vcpu->mmio_phys_addr = gpa;
 956                 vcpu->mmio_size = bytes;
 957                 vcpu->mmio_is_write = 0;
 958
 959                 return X86EMUL_UNHANDLEABLE;
 960         }
 961 }
 962
 963 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 964                                unsigned long val, int bytes)
 965 {
 966         struct kvm_memory_slot *m;
 967         struct page *page;
 968         void *virt;
 969
 970         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 971                 return 0;
 972         m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
 973         if (!m)
 974                 return 0;
 975         page = gfn_to_page(m, gpa >> PAGE_SHIFT);
 976         kvm_mmu_pre_write(vcpu, gpa, bytes);
 977         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
 978         virt = kmap_atomic(page, KM_USER0);
 979         memcpy(virt + offset_in_page(gpa), &val, bytes);
 980         kunmap_atomic(virt, KM_USER0);
 981         kvm_mmu_post_write(vcpu, gpa, bytes);
 982         return 1;
 983 }
 984
 985 static int emulator_write_emulated(unsigned long addr,
 986                                    unsigned long val,
 987                                    unsigned int bytes,
 988                                    struct x86_emulate_ctxt *ctxt)
 989 {
 990         struct kvm_vcpu *vcpu = ctxt->vcpu;
 991         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 992
 993         if (gpa == UNMAPPED_GVA)
 994                 return X86EMUL_PROPAGATE_FAULT;
 995
 996         if (emulator_write_phys(vcpu, gpa, val, bytes))
 997                 return X86EMUL_CONTINUE;
 998
 999         vcpu->mmio_needed = 1;
1000         vcpu->mmio_phys_addr = gpa;
1001         vcpu->mmio_size = bytes;
1002         vcpu->mmio_is_write = 1;
1003         memcpy(vcpu->mmio_data, &val, bytes);
1004
1005         return X86EMUL_CONTINUE;
1006 }
1007
1008 static int emulator_cmpxchg_emulated(unsigned long addr,
1009                                      unsigned long old,
1010                                      unsigned long new,
1011                                      unsigned int bytes,
1012                                      struct x86_emulate_ctxt *ctxt)
1013 {
1014         static int reported;
1015
1016         if (!reported) {
1017                 reported = 1;
1018                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1019         }
1020         return emulator_write_emulated(addr, new, bytes, ctxt);
1021 }
1022
1023 #ifdef CONFIG_X86_32
1024
1025 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1026                                        unsigned long old_lo,
1027                                        unsigned long old_hi,
1028                                        unsigned long new_lo,
1029                                        unsigned long new_hi,
1030                                        struct x86_emulate_ctxt *ctxt)
1031 {
1032         static int reported;
1033         int r;
1034
1035         if (!reported) {
1036                 reported = 1;
1037                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1038         }
1039         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1040         if (r != X86EMUL_CONTINUE)
1041                 return r;
1042         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1043 }
1044
1045 #endif
1046
1047 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1048 {
1049         return kvm_arch_ops->get_segment_base(vcpu, seg);
1050 }
1051
1052 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1053 {
1054         return X86EMUL_CONTINUE;
1055 }
1056
1057 int emulate_clts(struct kvm_vcpu *vcpu)
1058 {
1059         unsigned long cr0;
1060
1061         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1062         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1063         kvm_arch_ops->set_cr0(vcpu, cr0);
1064         return X86EMUL_CONTINUE;
1065 }
1066
1067 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1068 {
1069         struct kvm_vcpu *vcpu = ctxt->vcpu;
1070
1071         switch (dr) {
1072         case 0 ... 3:
1073                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1074                 return X86EMUL_CONTINUE;
1075         default:
1076                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1077                        __FUNCTION__, dr);
1078                 return X86EMUL_UNHANDLEABLE;
1079         }
1080 }
1081
1082 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1083 {
1084         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1085         int exception;
1086
1087         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1088         if (exception) {
1089                 /* FIXME: better handling */
1090                 return X86EMUL_UNHANDLEABLE;
1091         }
1092         return X86EMUL_CONTINUE;
1093 }
1094
1095 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1096 {
1097         static int reported;
1098         u8 opcodes[4];
1099         unsigned long rip = ctxt->vcpu->rip;
1100         unsigned long rip_linear;
1101
1102         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1103
1104         if (reported)
1105                 return;
1106
1107         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1108
1109         printk(KERN_ERR "emulation failed but !mmio_needed?"
1110                " rip %lx %02x %02x %02x %02x\n",
1111                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1112         reported = 1;
1113 }
1114
1115 struct x86_emulate_ops emulate_ops = {
1116         .read_std            = emulator_read_std,
1117         .write_std           = emulator_write_std,
1118         .read_emulated       = emulator_read_emulated,
1119         .write_emulated      = emulator_write_emulated,
1120         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1121 #ifdef CONFIG_X86_32
1122         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1123 #endif
1124 };
1125
1126 int emulate_instruction(struct kvm_vcpu *vcpu,
1127                         struct kvm_run *run,
1128                         unsigned long cr2,
1129                         u16 error_code)
1130 {
1131         struct x86_emulate_ctxt emulate_ctxt;
1132         int r;
1133         int cs_db, cs_l;
1134
1135         kvm_arch_ops->cache_regs(vcpu);
1136
1137         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1138
1139         emulate_ctxt.vcpu = vcpu;
1140         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1141         emulate_ctxt.cr2 = cr2;
1142         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1143                 ? X86EMUL_MODE_REAL : cs_l
1144                 ? X86EMUL_MODE_PROT64 : cs_db
1145                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1146
1147         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1148                 emulate_ctxt.cs_base = 0;
1149                 emulate_ctxt.ds_base = 0;
1150                 emulate_ctxt.es_base = 0;
1151                 emulate_ctxt.ss_base = 0;
1152         } else {
1153                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1154                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1155                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1156                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1157         }
1158
1159         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1160         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1161
1162         vcpu->mmio_is_write = 0;
1163         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1164
1165         if ((r || vcpu->mmio_is_write) && run) {
1166                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1167                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1168                 run->mmio.len = vcpu->mmio_size;
1169                 run->mmio.is_write = vcpu->mmio_is_write;
1170         }
1171
1172         if (r) {
1173                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1174                         return EMULATE_DONE;
1175                 if (!vcpu->mmio_needed) {
1176                         report_emulation_failure(&emulate_ctxt);
1177                         return EMULATE_FAIL;
1178                 }
1179                 return EMULATE_DO_MMIO;
1180         }
1181
1182         kvm_arch_ops->decache_regs(vcpu);
1183         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1184
1185         if (vcpu->mmio_is_write)
1186                 return EMULATE_DO_MMIO;
1187
1188         return EMULATE_DONE;
1189 }
1190 EXPORT_SYMBOL_GPL(emulate_instruction);
1191
1192 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1193 {
1194         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1195
1196         kvm_arch_ops->cache_regs(vcpu);
1197         ret = -KVM_EINVAL;
1198 #ifdef CONFIG_X86_64
1199         if (is_long_mode(vcpu)) {
1200                 nr = vcpu->regs[VCPU_REGS_RAX];
1201                 a0 = vcpu->regs[VCPU_REGS_RDI];
1202                 a1 = vcpu->regs[VCPU_REGS_RSI];
1203                 a2 = vcpu->regs[VCPU_REGS_RDX];
1204                 a3 = vcpu->regs[VCPU_REGS_RCX];
1205                 a4 = vcpu->regs[VCPU_REGS_R8];
1206                 a5 = vcpu->regs[VCPU_REGS_R9];
1207         } else
1208 #endif
1209         {
1210                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1211                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1212                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1213                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1214                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1215                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1216                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1217         }
1218         switch (nr) {
1219         default:
1220                 run->hypercall.args[0] = a0;
1221                 run->hypercall.args[1] = a1;
1222                 run->hypercall.args[2] = a2;
1223                 run->hypercall.args[3] = a3;
1224                 run->hypercall.args[4] = a4;
1225                 run->hypercall.args[5] = a5;
1226                 run->hypercall.ret = ret;
1227                 run->hypercall.longmode = is_long_mode(vcpu);
1228                 kvm_arch_ops->decache_regs(vcpu);
1229                 return 0;
1230         }
1231         vcpu->regs[VCPU_REGS_RAX] = ret;
1232         kvm_arch_ops->decache_regs(vcpu);
1233         return 1;
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_hypercall);
1236
1237 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1238 {
1239         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1240 }
1241
1242 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1243 {
1244         struct descriptor_table dt = { limit, base };
1245
1246         kvm_arch_ops->set_gdt(vcpu, &dt);
1247 }
1248
1249 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1250 {
1251         struct descriptor_table dt = { limit, base };
1252
1253         kvm_arch_ops->set_idt(vcpu, &dt);
1254 }
1255
1256 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1257                    unsigned long *rflags)
1258 {
1259         lmsw(vcpu, msw);
1260         *rflags = kvm_arch_ops->get_rflags(vcpu);
1261 }
1262
1263 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1264 {
1265         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1266         switch (cr) {
1267         case 0:
1268                 return vcpu->cr0;
1269         case 2:
1270                 return vcpu->cr2;
1271         case 3:
1272                 return vcpu->cr3;
1273         case 4:
1274                 return vcpu->cr4;
1275         default:
1276                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1277                 return 0;
1278         }
1279 }
1280
1281 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1282                      unsigned long *rflags)
1283 {
1284         switch (cr) {
1285         case 0:
1286                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1287                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1288                 break;
1289         case 2:
1290                 vcpu->cr2 = val;
1291                 break;
1292         case 3:
1293                 set_cr3(vcpu, val);
1294                 break;
1295         case 4:
1296                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1297                 break;
1298         default:
1299                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1300         }
1301 }
1302
1303 /*
1304  * Register the para guest with the host:
1305  */
1306 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1307 {
1308         struct kvm_vcpu_para_state *para_state;
1309         hpa_t para_state_hpa, hypercall_hpa;
1310         struct page *para_state_page;
1311         unsigned char *hypercall;
1312         gpa_t hypercall_gpa;
1313
1314         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1315         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1316
1317         /*
1318          * Needs to be page aligned:
1319          */
1320         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1321                 goto err_gp;
1322
1323         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1324         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1325         if (is_error_hpa(para_state_hpa))
1326                 goto err_gp;
1327
1328         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1329         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1330         para_state = kmap_atomic(para_state_page, KM_USER0);
1331
1332         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1333         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1334
1335         para_state->host_version = KVM_PARA_API_VERSION;
1336         /*
1337          * We cannot support guests that try to register themselves
1338          * with a newer API version than the host supports:
1339          */
1340         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1341                 para_state->ret = -KVM_EINVAL;
1342                 goto err_kunmap_skip;
1343         }
1344
1345         hypercall_gpa = para_state->hypercall_gpa;
1346         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1347         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1348         if (is_error_hpa(hypercall_hpa)) {
1349                 para_state->ret = -KVM_EINVAL;
1350                 goto err_kunmap_skip;
1351         }
1352
1353         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1354         vcpu->para_state_page = para_state_page;
1355         vcpu->para_state_gpa = para_state_gpa;
1356         vcpu->hypercall_gpa = hypercall_gpa;
1357
1358         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1359         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1360                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1361         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1362         kunmap_atomic(hypercall, KM_USER1);
1363
1364         para_state->ret = 0;
1365 err_kunmap_skip:
1366         kunmap_atomic(para_state, KM_USER0);
1367         return 0;
1368 err_gp:
1369         return 1;
1370 }
1371
1372 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1373 {
1374         u64 data;
1375
1376         switch (msr) {
1377         case 0xc0010010: /* SYSCFG */
1378         case 0xc0010015: /* HWCR */
1379         case MSR_IA32_PLATFORM_ID:
1380         case MSR_IA32_P5_MC_ADDR:
1381         case MSR_IA32_P5_MC_TYPE:
1382         case MSR_IA32_MC0_CTL:
1383         case MSR_IA32_MCG_STATUS:
1384         case MSR_IA32_MCG_CAP:
1385         case MSR_IA32_MC0_MISC:
1386         case MSR_IA32_MC0_MISC+4:
1387         case MSR_IA32_MC0_MISC+8:
1388         case MSR_IA32_MC0_MISC+12:
1389         case MSR_IA32_MC0_MISC+16:
1390         case MSR_IA32_UCODE_REV:
1391         case MSR_IA32_PERF_STATUS:
1392                 /* MTRR registers */
1393         case 0xfe:
1394         case 0x200 ... 0x2ff:
1395                 data = 0;
1396                 break;
1397         case 0xcd: /* fsb frequency */
1398                 data = 3;
1399                 break;
1400         case MSR_IA32_APICBASE:
1401                 data = vcpu->apic_base;
1402                 break;
1403         case MSR_IA32_MISC_ENABLE:
1404                 data = vcpu->ia32_misc_enable_msr;
1405                 break;
1406 #ifdef CONFIG_X86_64
1407         case MSR_EFER:
1408                 data = vcpu->shadow_efer;
1409                 break;
1410 #endif
1411         default:
1412                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1413                 return 1;
1414         }
1415         *pdata = data;
1416         return 0;
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1419
1420 /*
1421  * Reads an msr value (of 'msr_index') into 'pdata'.
1422  * Returns 0 on success, non-0 otherwise.
1423  * Assumes vcpu_load() was already called.
1424  */
1425 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1426 {
1427         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1428 }
1429
1430 #ifdef CONFIG_X86_64
1431
1432 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1433 {
1434         if (efer & EFER_RESERVED_BITS) {
1435                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1436                        efer);
1437                 inject_gp(vcpu);
1438                 return;
1439         }
1440
1441         if (is_paging(vcpu)
1442             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1443                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1444                 inject_gp(vcpu);
1445                 return;
1446         }
1447
1448         kvm_arch_ops->set_efer(vcpu, efer);
1449
1450         efer &= ~EFER_LMA;
1451         efer |= vcpu->shadow_efer & EFER_LMA;
1452
1453         vcpu->shadow_efer = efer;
1454 }
1455
1456 #endif
1457
1458 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1459 {
1460         switch (msr) {
1461 #ifdef CONFIG_X86_64
1462         case MSR_EFER:
1463                 set_efer(vcpu, data);
1464                 break;
1465 #endif
1466         case MSR_IA32_MC0_STATUS:
1467                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1468                        __FUNCTION__, data);
1469                 break;
1470         case MSR_IA32_UCODE_REV:
1471         case MSR_IA32_UCODE_WRITE:
1472         case 0x200 ... 0x2ff: /* MTRRs */
1473                 break;
1474         case MSR_IA32_APICBASE:
1475                 vcpu->apic_base = data;
1476                 break;
1477         case MSR_IA32_MISC_ENABLE:
1478                 vcpu->ia32_misc_enable_msr = data;
1479                 break;
1480         /*
1481          * This is the 'probe whether the host is KVM' logic:
1482          */
1483         case MSR_KVM_API_MAGIC:
1484                 return vcpu_register_para(vcpu, data);
1485
1486         default:
1487                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1488                 return 1;
1489         }
1490         return 0;
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1493
1494 /*
1495  * Writes msr value into into the appropriate "register".
1496  * Returns 0 on success, non-0 otherwise.
1497  * Assumes vcpu_load() was already called.
1498  */
1499 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1500 {
1501         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1502 }
1503
1504 void kvm_resched(struct kvm_vcpu *vcpu)
1505 {
1506         vcpu_put(vcpu);
1507         cond_resched();
1508         vcpu_load(vcpu);
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_resched);
1511
1512 void load_msrs(struct vmx_msr_entry *e, int n)
1513 {
1514         int i;
1515
1516         for (i = 0; i < n; ++i)
1517                 wrmsrl(e[i].index, e[i].data);
1518 }
1519 EXPORT_SYMBOL_GPL(load_msrs);
1520
1521 void save_msrs(struct vmx_msr_entry *e, int n)
1522 {
1523         int i;
1524
1525         for (i = 0; i < n; ++i)
1526                 rdmsrl(e[i].index, e[i].data);
1527 }
1528 EXPORT_SYMBOL_GPL(save_msrs);
1529
1530 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1531 {
1532         int i;
1533         u32 function;
1534         struct kvm_cpuid_entry *e, *best;
1535
1536         kvm_arch_ops->cache_regs(vcpu);
1537         function = vcpu->regs[VCPU_REGS_RAX];
1538         vcpu->regs[VCPU_REGS_RAX] = 0;
1539         vcpu->regs[VCPU_REGS_RBX] = 0;
1540         vcpu->regs[VCPU_REGS_RCX] = 0;
1541         vcpu->regs[VCPU_REGS_RDX] = 0;
1542         best = NULL;
1543         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1544                 e = &vcpu->cpuid_entries[i];
1545                 if (e->function == function) {
1546                         best = e;
1547                         break;
1548                 }
1549                 /*
1550                  * Both basic or both extended?
1551                  */
1552                 if (((e->function ^ function) & 0x80000000) == 0)
1553                         if (!best || e->function > best->function)
1554                                 best = e;
1555         }
1556         if (best) {
1557                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1558                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1559                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1560                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1561         }
1562         kvm_arch_ops->decache_regs(vcpu);
1563         kvm_arch_ops->skip_emulated_instruction(vcpu);
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1566
1567 static int pio_copy_data(struct kvm_vcpu *vcpu)
1568 {
1569         void *p = vcpu->pio_data;
1570         void *q;
1571         unsigned bytes;
1572         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1573
1574         kvm_arch_ops->vcpu_put(vcpu);
1575         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1576                  PAGE_KERNEL);
1577         if (!q) {
1578                 kvm_arch_ops->vcpu_load(vcpu);
1579                 free_pio_guest_pages(vcpu);
1580                 return -ENOMEM;
1581         }
1582         q += vcpu->pio.guest_page_offset;
1583         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1584         if (vcpu->pio.in)
1585                 memcpy(q, p, bytes);
1586         else
1587                 memcpy(p, q, bytes);
1588         q -= vcpu->pio.guest_page_offset;
1589         vunmap(q);
1590         kvm_arch_ops->vcpu_load(vcpu);
1591         free_pio_guest_pages(vcpu);
1592         return 0;
1593 }
1594
1595 static int complete_pio(struct kvm_vcpu *vcpu)
1596 {
1597         struct kvm_pio_request *io = &vcpu->pio;
1598         long delta;
1599         int r;
1600
1601         kvm_arch_ops->cache_regs(vcpu);
1602
1603         if (!io->string) {
1604                 if (io->in)
1605                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1606                                io->size);
1607         } else {
1608                 if (io->in) {
1609                         r = pio_copy_data(vcpu);
1610                         if (r) {
1611                                 kvm_arch_ops->cache_regs(vcpu);
1612                                 return r;
1613                         }
1614                 }
1615
1616                 delta = 1;
1617                 if (io->rep) {
1618                         delta *= io->cur_count;
1619                         /*
1620                          * The size of the register should really depend on
1621                          * current address size.
1622                          */
1623                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1624                 }
1625                 if (io->down)
1626                         delta = -delta;
1627                 delta *= io->size;
1628                 if (io->in)
1629                         vcpu->regs[VCPU_REGS_RDI] += delta;
1630                 else
1631                         vcpu->regs[VCPU_REGS_RSI] += delta;
1632         }
1633
1634         vcpu->run->io_completed = 0;
1635
1636         kvm_arch_ops->decache_regs(vcpu);
1637
1638         io->count -= io->cur_count;
1639         io->cur_count = 0;
1640
1641         if (!io->count)
1642                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1643         return 0;
1644 }
1645
1646 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1647                   int size, unsigned long count, int string, int down,
1648                   gva_t address, int rep, unsigned port)
1649 {
1650         unsigned now, in_page;
1651         int i;
1652         int nr_pages = 1;
1653         struct page *page;
1654
1655         vcpu->run->exit_reason = KVM_EXIT_IO;
1656         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1657         vcpu->run->io.size = size;
1658         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1659         vcpu->run->io.count = count;
1660         vcpu->run->io.port = port;
1661         vcpu->pio.count = count;
1662         vcpu->pio.cur_count = count;
1663         vcpu->pio.size = size;
1664         vcpu->pio.in = in;
1665         vcpu->pio.string = string;
1666         vcpu->pio.down = down;
1667         vcpu->pio.guest_page_offset = offset_in_page(address);
1668         vcpu->pio.rep = rep;
1669
1670         if (!string) {
1671                 kvm_arch_ops->cache_regs(vcpu);
1672                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1673                 kvm_arch_ops->decache_regs(vcpu);
1674                 return 0;
1675         }
1676
1677         if (!count) {
1678                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1679                 return 1;
1680         }
1681
1682         now = min(count, PAGE_SIZE / size);
1683
1684         if (!down)
1685                 in_page = PAGE_SIZE - offset_in_page(address);
1686         else
1687                 in_page = offset_in_page(address) + size;
1688         now = min(count, (unsigned long)in_page / size);
1689         if (!now) {
1690                 /*
1691                  * String I/O straddles page boundary.  Pin two guest pages
1692                  * so that we satisfy atomicity constraints.  Do just one
1693                  * transaction to avoid complexity.
1694                  */
1695                 nr_pages = 2;
1696                 now = 1;
1697         }
1698         if (down) {
1699                 /*
1700                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1701                  */
1702                 printk(KERN_ERR "kvm: guest string pio down\n");
1703                 inject_gp(vcpu);
1704                 return 1;
1705         }
1706         vcpu->run->io.count = now;
1707         vcpu->pio.cur_count = now;
1708
1709         for (i = 0; i < nr_pages; ++i) {
1710                 spin_lock(&vcpu->kvm->lock);
1711                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1712                 if (page)
1713                         get_page(page);
1714                 vcpu->pio.guest_pages[i] = page;
1715                 spin_unlock(&vcpu->kvm->lock);
1716                 if (!page) {
1717                         inject_gp(vcpu);
1718                         free_pio_guest_pages(vcpu);
1719                         return 1;
1720                 }
1721         }
1722
1723         if (!vcpu->pio.in)
1724                 return pio_copy_data(vcpu);
1725         return 0;
1726 }
1727 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1728
1729 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1730 {
1731         int r;
1732         sigset_t sigsaved;
1733
1734         vcpu_load(vcpu);
1735
1736         if (vcpu->sigset_active)
1737                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1738
1739         /* re-sync apic's tpr */
1740         vcpu->cr8 = kvm_run->cr8;
1741
1742         if (kvm_run->io_completed) {
1743                 if (vcpu->pio.cur_count) {
1744                         r = complete_pio(vcpu);
1745                         if (r)
1746                                 goto out;
1747                 } else {
1748                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1749                         vcpu->mmio_read_completed = 1;
1750                 }
1751         }
1752
1753         vcpu->mmio_needed = 0;
1754
1755         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1756                 kvm_arch_ops->cache_regs(vcpu);
1757                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1758                 kvm_arch_ops->decache_regs(vcpu);
1759         }
1760
1761         r = kvm_arch_ops->run(vcpu, kvm_run);
1762
1763 out:
1764         if (vcpu->sigset_active)
1765                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1766
1767         vcpu_put(vcpu);
1768         return r;
1769 }
1770
1771 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1772                                    struct kvm_regs *regs)
1773 {
1774         vcpu_load(vcpu);
1775
1776         kvm_arch_ops->cache_regs(vcpu);
1777
1778         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1779         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1780         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1781         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1782         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1783         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1784         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1785         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1786 #ifdef CONFIG_X86_64
1787         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1788         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1789         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1790         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1791         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1792         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1793         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1794         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1795 #endif
1796
1797         regs->rip = vcpu->rip;
1798         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1799
1800         /*
1801          * Don't leak debug flags in case they were set for guest debugging
1802          */
1803         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1804                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1805
1806         vcpu_put(vcpu);
1807
1808         return 0;
1809 }
1810
1811 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1812                                    struct kvm_regs *regs)
1813 {
1814         vcpu_load(vcpu);
1815
1816         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1817         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1818         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1819         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1820         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1821         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1822         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1823         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1824 #ifdef CONFIG_X86_64
1825         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1826         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1827         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1828         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1829         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1830         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1831         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1832         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1833 #endif
1834
1835         vcpu->rip = regs->rip;
1836         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1837
1838         kvm_arch_ops->decache_regs(vcpu);
1839
1840         vcpu_put(vcpu);
1841
1842         return 0;
1843 }
1844
1845 static void get_segment(struct kvm_vcpu *vcpu,
1846                         struct kvm_segment *var, int seg)
1847 {
1848         return kvm_arch_ops->get_segment(vcpu, var, seg);
1849 }
1850
1851 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1852                                     struct kvm_sregs *sregs)
1853 {
1854         struct descriptor_table dt;
1855
1856         vcpu_load(vcpu);
1857
1858         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1859         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1860         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1861         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1862         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1863         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1864
1865         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1866         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1867
1868         kvm_arch_ops->get_idt(vcpu, &dt);
1869         sregs->idt.limit = dt.limit;
1870         sregs->idt.base = dt.base;
1871         kvm_arch_ops->get_gdt(vcpu, &dt);
1872         sregs->gdt.limit = dt.limit;
1873         sregs->gdt.base = dt.base;
1874
1875         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1876         sregs->cr0 = vcpu->cr0;
1877         sregs->cr2 = vcpu->cr2;
1878         sregs->cr3 = vcpu->cr3;
1879         sregs->cr4 = vcpu->cr4;
1880         sregs->cr8 = vcpu->cr8;
1881         sregs->efer = vcpu->shadow_efer;
1882         sregs->apic_base = vcpu->apic_base;
1883
1884         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1885                sizeof sregs->interrupt_bitmap);
1886
1887         vcpu_put(vcpu);
1888
1889         return 0;
1890 }
1891
1892 static void set_segment(struct kvm_vcpu *vcpu,
1893                         struct kvm_segment *var, int seg)
1894 {
1895         return kvm_arch_ops->set_segment(vcpu, var, seg);
1896 }
1897
1898 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1899                                     struct kvm_sregs *sregs)
1900 {
1901         int mmu_reset_needed = 0;
1902         int i;
1903         struct descriptor_table dt;
1904
1905         vcpu_load(vcpu);
1906
1907         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1908         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1909         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1910         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1911         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1912         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1913
1914         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1915         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1916
1917         dt.limit = sregs->idt.limit;
1918         dt.base = sregs->idt.base;
1919         kvm_arch_ops->set_idt(vcpu, &dt);
1920         dt.limit = sregs->gdt.limit;
1921         dt.base = sregs->gdt.base;
1922         kvm_arch_ops->set_gdt(vcpu, &dt);
1923
1924         vcpu->cr2 = sregs->cr2;
1925         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1926         vcpu->cr3 = sregs->cr3;
1927
1928         vcpu->cr8 = sregs->cr8;
1929
1930         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1931 #ifdef CONFIG_X86_64
1932         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1933 #endif
1934         vcpu->apic_base = sregs->apic_base;
1935
1936         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1937
1938         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1939         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
1940
1941         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1942         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1943         if (!is_long_mode(vcpu) && is_pae(vcpu))
1944                 load_pdptrs(vcpu, vcpu->cr3);
1945
1946         if (mmu_reset_needed)
1947                 kvm_mmu_reset_context(vcpu);
1948
1949         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1950                sizeof vcpu->irq_pending);
1951         vcpu->irq_summary = 0;
1952         for (i = 0; i < NR_IRQ_WORDS; ++i)
1953                 if (vcpu->irq_pending[i])
1954                         __set_bit(i, &vcpu->irq_summary);
1955
1956         vcpu_put(vcpu);
1957
1958         return 0;
1959 }
1960
1961 /*
1962  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1963  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1964  *
1965  * This list is modified at module load time to reflect the
1966  * capabilities of the host cpu.
1967  */
1968 static u32 msrs_to_save[] = {
1969         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1970         MSR_K6_STAR,
1971 #ifdef CONFIG_X86_64
1972         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1973 #endif
1974         MSR_IA32_TIME_STAMP_COUNTER,
1975 };
1976
1977 static unsigned num_msrs_to_save;
1978
1979 static u32 emulated_msrs[] = {
1980         MSR_IA32_MISC_ENABLE,
1981 };
1982
1983 static __init void kvm_init_msr_list(void)
1984 {
1985         u32 dummy[2];
1986         unsigned i, j;
1987
1988         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1989                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1990                         continue;
1991                 if (j < i)
1992                         msrs_to_save[j] = msrs_to_save[i];
1993                 j++;
1994         }
1995         num_msrs_to_save = j;
1996 }
1997
1998 /*
1999  * Adapt set_msr() to msr_io()'s calling convention
2000  */
2001 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2002 {
2003         return set_msr(vcpu, index, *data);
2004 }
2005
2006 /*
2007  * Read or write a bunch of msrs. All parameters are kernel addresses.
2008  *
2009  * @return number of msrs set successfully.
2010  */
2011 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2012                     struct kvm_msr_entry *entries,
2013                     int (*do_msr)(struct kvm_vcpu *vcpu,
2014                                   unsigned index, u64 *data))
2015 {
2016         int i;
2017
2018         vcpu_load(vcpu);
2019
2020         for (i = 0; i < msrs->nmsrs; ++i)
2021                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2022                         break;
2023
2024         vcpu_put(vcpu);
2025
2026         return i;
2027 }
2028
2029 /*
2030  * Read or write a bunch of msrs. Parameters are user addresses.
2031  *
2032  * @return number of msrs set successfully.
2033  */
2034 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2035                   int (*do_msr)(struct kvm_vcpu *vcpu,
2036                                 unsigned index, u64 *data),
2037                   int writeback)
2038 {
2039         struct kvm_msrs msrs;
2040         struct kvm_msr_entry *entries;
2041         int r, n;
2042         unsigned size;
2043
2044         r = -EFAULT;
2045         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2046                 goto out;
2047
2048         r = -E2BIG;
2049         if (msrs.nmsrs >= MAX_IO_MSRS)
2050                 goto out;
2051
2052         r = -ENOMEM;
2053         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2054         entries = vmalloc(size);
2055         if (!entries)
2056                 goto out;
2057
2058         r = -EFAULT;
2059         if (copy_from_user(entries, user_msrs->entries, size))
2060                 goto out_free;
2061
2062         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2063         if (r < 0)
2064                 goto out_free;
2065
2066         r = -EFAULT;
2067         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2068                 goto out_free;
2069
2070         r = n;
2071
2072 out_free:
2073         vfree(entries);
2074 out:
2075         return r;
2076 }
2077
2078 /*
2079  * Translate a guest virtual address to a guest physical address.
2080  */
2081 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2082                                     struct kvm_translation *tr)
2083 {
2084         unsigned long vaddr = tr->linear_address;
2085         gpa_t gpa;
2086
2087         vcpu_load(vcpu);
2088         spin_lock(&vcpu->kvm->lock);
2089         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2090         tr->physical_address = gpa;
2091         tr->valid = gpa != UNMAPPED_GVA;
2092         tr->writeable = 1;
2093         tr->usermode = 0;
2094         spin_unlock(&vcpu->kvm->lock);
2095         vcpu_put(vcpu);
2096
2097         return 0;
2098 }
2099
2100 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2101                                     struct kvm_interrupt *irq)
2102 {
2103         if (irq->irq < 0 || irq->irq >= 256)
2104                 return -EINVAL;
2105         vcpu_load(vcpu);
2106
2107         set_bit(irq->irq, vcpu->irq_pending);
2108         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2109
2110         vcpu_put(vcpu);
2111
2112         return 0;
2113 }
2114
2115 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2116                                       struct kvm_debug_guest *dbg)
2117 {
2118         int r;
2119
2120         vcpu_load(vcpu);
2121
2122         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2123
2124         vcpu_put(vcpu);
2125
2126         return r;
2127 }
2128
2129 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2130                                     unsigned long address,
2131                                     int *type)
2132 {
2133         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2134         unsigned long pgoff;
2135         struct page *page;
2136
2137         *type = VM_FAULT_MINOR;
2138         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2139         if (pgoff == 0)
2140                 page = virt_to_page(vcpu->run);
2141         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2142                 page = virt_to_page(vcpu->pio_data);
2143         else
2144                 return NOPAGE_SIGBUS;
2145         get_page(page);
2146         return page;
2147 }
2148
2149 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2150         .nopage = kvm_vcpu_nopage,
2151 };
2152
2153 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2154 {
2155         vma->vm_ops = &kvm_vcpu_vm_ops;
2156         return 0;
2157 }
2158
2159 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2160 {
2161         struct kvm_vcpu *vcpu = filp->private_data;
2162
2163         fput(vcpu->kvm->filp);
2164         return 0;
2165 }
2166
2167 static struct file_operations kvm_vcpu_fops = {
2168         .release        = kvm_vcpu_release,
2169         .unlocked_ioctl = kvm_vcpu_ioctl,
2170         .compat_ioctl   = kvm_vcpu_ioctl,
2171         .mmap           = kvm_vcpu_mmap,
2172 };
2173
2174 /*
2175  * Allocates an inode for the vcpu.
2176  */
2177 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2178 {
2179         int fd, r;
2180         struct inode *inode;
2181         struct file *file;
2182
2183         atomic_inc(&vcpu->kvm->filp->f_count);
2184         inode = kvmfs_inode(&kvm_vcpu_fops);
2185         if (IS_ERR(inode)) {
2186                 r = PTR_ERR(inode);
2187                 goto out1;
2188         }
2189
2190         file = kvmfs_file(inode, vcpu);
2191         if (IS_ERR(file)) {
2192                 r = PTR_ERR(file);
2193                 goto out2;
2194         }
2195
2196         r = get_unused_fd();
2197         if (r < 0)
2198                 goto out3;
2199         fd = r;
2200         fd_install(fd, file);
2201
2202         return fd;
2203
2204 out3:
2205         fput(file);
2206 out2:
2207         iput(inode);
2208 out1:
2209         fput(vcpu->kvm->filp);
2210         return r;
2211 }
2212
2213 /*
2214  * Creates some virtual cpus.  Good luck creating more than one.
2215  */
2216 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2217 {
2218         int r;
2219         struct kvm_vcpu *vcpu;
2220         struct page *page;
2221
2222         r = -EINVAL;
2223         if (!valid_vcpu(n))
2224                 goto out;
2225
2226         vcpu = &kvm->vcpus[n];
2227
2228         mutex_lock(&vcpu->mutex);
2229
2230         if (vcpu->vmcs) {
2231                 mutex_unlock(&vcpu->mutex);
2232                 return -EEXIST;
2233         }
2234
2235         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2236         r = -ENOMEM;
2237         if (!page)
2238                 goto out_unlock;
2239         vcpu->run = page_address(page);
2240
2241         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2242         r = -ENOMEM;
2243         if (!page)
2244                 goto out_free_run;
2245         vcpu->pio_data = page_address(page);
2246
2247         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2248                                            FX_IMAGE_ALIGN);
2249         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2250
2251         r = kvm_arch_ops->vcpu_create(vcpu);
2252         if (r < 0)
2253                 goto out_free_vcpus;
2254
2255         r = kvm_mmu_create(vcpu);
2256         if (r < 0)
2257                 goto out_free_vcpus;
2258
2259         kvm_arch_ops->vcpu_load(vcpu);
2260         r = kvm_mmu_setup(vcpu);
2261         if (r >= 0)
2262                 r = kvm_arch_ops->vcpu_setup(vcpu);
2263         vcpu_put(vcpu);
2264
2265         if (r < 0)
2266                 goto out_free_vcpus;
2267
2268         r = create_vcpu_fd(vcpu);
2269         if (r < 0)
2270                 goto out_free_vcpus;
2271
2272         return r;
2273
2274 out_free_vcpus:
2275         kvm_free_vcpu(vcpu);
2276 out_free_run:
2277         free_page((unsigned long)vcpu->run);
2278         vcpu->run = NULL;
2279 out_unlock:
2280         mutex_unlock(&vcpu->mutex);
2281 out:
2282         return r;
2283 }
2284
2285 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2286                                     struct kvm_cpuid *cpuid,
2287                                     struct kvm_cpuid_entry __user *entries)
2288 {
2289         int r;
2290
2291         r = -E2BIG;
2292         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2293                 goto out;
2294         r = -EFAULT;
2295         if (copy_from_user(&vcpu->cpuid_entries, entries,
2296                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2297                 goto out;
2298         vcpu->cpuid_nent = cpuid->nent;
2299         return 0;
2300
2301 out:
2302         return r;
2303 }
2304
2305 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2306 {
2307         if (sigset) {
2308                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2309                 vcpu->sigset_active = 1;
2310                 vcpu->sigset = *sigset;
2311         } else
2312                 vcpu->sigset_active = 0;
2313         return 0;
2314 }
2315
2316 static long kvm_vcpu_ioctl(struct file *filp,
2317                            unsigned int ioctl, unsigned long arg)
2318 {
2319         struct kvm_vcpu *vcpu = filp->private_data;
2320         void __user *argp = (void __user *)arg;
2321         int r = -EINVAL;
2322
2323         switch (ioctl) {
2324         case KVM_RUN:
2325                 r = -EINVAL;
2326                 if (arg)
2327                         goto out;
2328                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2329                 break;
2330         case KVM_GET_REGS: {
2331                 struct kvm_regs kvm_regs;
2332
2333                 memset(&kvm_regs, 0, sizeof kvm_regs);
2334                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2335                 if (r)
2336                         goto out;
2337                 r = -EFAULT;
2338                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2339                         goto out;
2340                 r = 0;
2341                 break;
2342         }
2343         case KVM_SET_REGS: {
2344                 struct kvm_regs kvm_regs;
2345
2346                 r = -EFAULT;
2347                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2348                         goto out;
2349                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2350                 if (r)
2351                         goto out;
2352                 r = 0;
2353                 break;
2354         }
2355         case KVM_GET_SREGS: {
2356                 struct kvm_sregs kvm_sregs;
2357
2358                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2359                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2360                 if (r)
2361                         goto out;
2362                 r = -EFAULT;
2363                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2364                         goto out;
2365                 r = 0;
2366                 break;
2367         }
2368         case KVM_SET_SREGS: {
2369                 struct kvm_sregs kvm_sregs;
2370
2371                 r = -EFAULT;
2372                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2373                         goto out;
2374                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2375                 if (r)
2376                         goto out;
2377                 r = 0;
2378                 break;
2379         }
2380         case KVM_TRANSLATE: {
2381                 struct kvm_translation tr;
2382
2383                 r = -EFAULT;
2384                 if (copy_from_user(&tr, argp, sizeof tr))
2385                         goto out;
2386                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2387                 if (r)
2388                         goto out;
2389                 r = -EFAULT;
2390                 if (copy_to_user(argp, &tr, sizeof tr))
2391                         goto out;
2392                 r = 0;
2393                 break;
2394         }
2395         case KVM_INTERRUPT: {
2396                 struct kvm_interrupt irq;
2397
2398                 r = -EFAULT;
2399                 if (copy_from_user(&irq, argp, sizeof irq))
2400                         goto out;
2401                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2402                 if (r)
2403                         goto out;
2404                 r = 0;
2405                 break;
2406         }
2407         case KVM_DEBUG_GUEST: {
2408                 struct kvm_debug_guest dbg;
2409
2410                 r = -EFAULT;
2411                 if (copy_from_user(&dbg, argp, sizeof dbg))
2412                         goto out;
2413                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2414                 if (r)
2415                         goto out;
2416                 r = 0;
2417                 break;
2418         }
2419         case KVM_GET_MSRS:
2420                 r = msr_io(vcpu, argp, get_msr, 1);
2421                 break;
2422         case KVM_SET_MSRS:
2423                 r = msr_io(vcpu, argp, do_set_msr, 0);
2424                 break;
2425         case KVM_SET_CPUID: {
2426                 struct kvm_cpuid __user *cpuid_arg = argp;
2427                 struct kvm_cpuid cpuid;
2428
2429                 r = -EFAULT;
2430                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2431                         goto out;
2432                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2433                 if (r)
2434                         goto out;
2435                 break;
2436         }
2437         case KVM_SET_SIGNAL_MASK: {
2438                 struct kvm_signal_mask __user *sigmask_arg = argp;
2439                 struct kvm_signal_mask kvm_sigmask;
2440                 sigset_t sigset, *p;
2441
2442                 p = NULL;
2443                 if (argp) {
2444                         r = -EFAULT;
2445                         if (copy_from_user(&kvm_sigmask, argp,
2446                                            sizeof kvm_sigmask))
2447                                 goto out;
2448                         r = -EINVAL;
2449                         if (kvm_sigmask.len != sizeof sigset)
2450                                 goto out;
2451                         r = -EFAULT;
2452                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2453                                            sizeof sigset))
2454                                 goto out;
2455                         p = &sigset;
2456                 }
2457                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2458                 break;
2459         }
2460         default:
2461                 ;
2462         }
2463 out:
2464         return r;
2465 }
2466
2467 static long kvm_vm_ioctl(struct file *filp,
2468                            unsigned int ioctl, unsigned long arg)
2469 {
2470         struct kvm *kvm = filp->private_data;
2471         void __user *argp = (void __user *)arg;
2472         int r = -EINVAL;
2473
2474         switch (ioctl) {
2475         case KVM_CREATE_VCPU:
2476                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2477                 if (r < 0)
2478                         goto out;
2479                 break;
2480         case KVM_SET_MEMORY_REGION: {
2481                 struct kvm_memory_region kvm_mem;
2482
2483                 r = -EFAULT;
2484                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2485                         goto out;
2486                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2487                 if (r)
2488                         goto out;
2489                 break;
2490         }
2491         case KVM_GET_DIRTY_LOG: {
2492                 struct kvm_dirty_log log;
2493
2494                 r = -EFAULT;
2495                 if (copy_from_user(&log, argp, sizeof log))
2496                         goto out;
2497                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2498                 if (r)
2499                         goto out;
2500                 break;
2501         }
2502         default:
2503                 ;
2504         }
2505 out:
2506         return r;
2507 }
2508
2509 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2510                                   unsigned long address,
2511                                   int *type)
2512 {
2513         struct kvm *kvm = vma->vm_file->private_data;
2514         unsigned long pgoff;
2515         struct kvm_memory_slot *slot;
2516         struct page *page;
2517
2518         *type = VM_FAULT_MINOR;
2519         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2520         slot = gfn_to_memslot(kvm, pgoff);
2521         if (!slot)
2522                 return NOPAGE_SIGBUS;
2523         page = gfn_to_page(slot, pgoff);
2524         if (!page)
2525                 return NOPAGE_SIGBUS;
2526         get_page(page);
2527         return page;
2528 }
2529
2530 static struct vm_operations_struct kvm_vm_vm_ops = {
2531         .nopage = kvm_vm_nopage,
2532 };
2533
2534 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2535 {
2536         vma->vm_ops = &kvm_vm_vm_ops;
2537         return 0;
2538 }
2539
2540 static struct file_operations kvm_vm_fops = {
2541         .release        = kvm_vm_release,
2542         .unlocked_ioctl = kvm_vm_ioctl,
2543         .compat_ioctl   = kvm_vm_ioctl,
2544         .mmap           = kvm_vm_mmap,
2545 };
2546
2547 static int kvm_dev_ioctl_create_vm(void)
2548 {
2549         int fd, r;
2550         struct inode *inode;
2551         struct file *file;
2552         struct kvm *kvm;
2553
2554         inode = kvmfs_inode(&kvm_vm_fops);
2555         if (IS_ERR(inode)) {
2556                 r = PTR_ERR(inode);
2557                 goto out1;
2558         }
2559
2560         kvm = kvm_create_vm();
2561         if (IS_ERR(kvm)) {
2562                 r = PTR_ERR(kvm);
2563                 goto out2;
2564         }
2565
2566         file = kvmfs_file(inode, kvm);
2567         if (IS_ERR(file)) {
2568                 r = PTR_ERR(file);
2569                 goto out3;
2570         }
2571         kvm->filp = file;
2572
2573         r = get_unused_fd();
2574         if (r < 0)
2575                 goto out4;
2576         fd = r;
2577         fd_install(fd, file);
2578
2579         return fd;
2580
2581 out4:
2582         fput(file);
2583 out3:
2584         kvm_destroy_vm(kvm);
2585 out2:
2586         iput(inode);
2587 out1:
2588         return r;
2589 }
2590
2591 static long kvm_dev_ioctl(struct file *filp,
2592                           unsigned int ioctl, unsigned long arg)
2593 {
2594         void __user *argp = (void __user *)arg;
2595         long r = -EINVAL;
2596
2597         switch (ioctl) {
2598         case KVM_GET_API_VERSION:
2599                 r = -EINVAL;
2600                 if (arg)
2601                         goto out;
2602                 r = KVM_API_VERSION;
2603                 break;
2604         case KVM_CREATE_VM:
2605                 r = -EINVAL;
2606                 if (arg)
2607                         goto out;
2608                 r = kvm_dev_ioctl_create_vm();
2609                 break;
2610         case KVM_GET_MSR_INDEX_LIST: {
2611                 struct kvm_msr_list __user *user_msr_list = argp;
2612                 struct kvm_msr_list msr_list;
2613                 unsigned n;
2614
2615                 r = -EFAULT;
2616                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2617                         goto out;
2618                 n = msr_list.nmsrs;
2619                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2620                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2621                         goto out;
2622                 r = -E2BIG;
2623                 if (n < num_msrs_to_save)
2624                         goto out;
2625                 r = -EFAULT;
2626                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2627                                  num_msrs_to_save * sizeof(u32)))
2628                         goto out;
2629                 if (copy_to_user(user_msr_list->indices
2630                                  + num_msrs_to_save * sizeof(u32),
2631                                  &emulated_msrs,
2632                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2633                         goto out;
2634                 r = 0;
2635                 break;
2636         }
2637         case KVM_CHECK_EXTENSION:
2638                 /*
2639                  * No extensions defined at present.
2640                  */
2641                 r = 0;
2642                 break;
2643         case KVM_GET_VCPU_MMAP_SIZE:
2644                 r = -EINVAL;
2645                 if (arg)
2646                         goto out;
2647                 r = 2 * PAGE_SIZE;
2648                 break;
2649         default:
2650                 ;
2651         }
2652 out:
2653         return r;
2654 }
2655
2656 static struct file_operations kvm_chardev_ops = {
2657         .open           = kvm_dev_open,
2658         .release        = kvm_dev_release,
2659         .unlocked_ioctl = kvm_dev_ioctl,
2660         .compat_ioctl   = kvm_dev_ioctl,
2661 };
2662
2663 static struct miscdevice kvm_dev = {
2664         KVM_MINOR,
2665         "kvm",
2666         &kvm_chardev_ops,
2667 };
2668
2669 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2670                        void *v)
2671 {
2672         if (val == SYS_RESTART) {
2673                 /*
2674                  * Some (well, at least mine) BIOSes hang on reboot if
2675                  * in vmx root mode.
2676                  */
2677                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2678                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2679         }
2680         return NOTIFY_OK;
2681 }
2682
2683 static struct notifier_block kvm_reboot_notifier = {
2684         .notifier_call = kvm_reboot,
2685         .priority = 0,
2686 };
2687
2688 /*
2689  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2690  * cached on it.
2691  */
2692 static void decache_vcpus_on_cpu(int cpu)
2693 {
2694         struct kvm *vm;
2695         struct kvm_vcpu *vcpu;
2696         int i;
2697
2698         spin_lock(&kvm_lock);
2699         list_for_each_entry(vm, &vm_list, vm_list)
2700                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2701                         vcpu = &vm->vcpus[i];
2702                         /*
2703                          * If the vcpu is locked, then it is running on some
2704                          * other cpu and therefore it is not cached on the
2705                          * cpu in question.
2706                          *
2707                          * If it's not locked, check the last cpu it executed
2708                          * on.
2709                          */
2710                         if (mutex_trylock(&vcpu->mutex)) {
2711                                 if (vcpu->cpu == cpu) {
2712                                         kvm_arch_ops->vcpu_decache(vcpu);
2713                                         vcpu->cpu = -1;
2714                                 }
2715                                 mutex_unlock(&vcpu->mutex);
2716                         }
2717                 }
2718         spin_unlock(&kvm_lock);
2719 }
2720
2721 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2722                            void *v)
2723 {
2724         int cpu = (long)v;
2725
2726         switch (val) {
2727         case CPU_DOWN_PREPARE:
2728         case CPU_UP_CANCELED:
2729                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2730                        cpu);
2731                 decache_vcpus_on_cpu(cpu);
2732                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2733                                          NULL, 0, 1);
2734                 break;
2735         case CPU_ONLINE:
2736                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2737                        cpu);
2738                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2739                                          NULL, 0, 1);
2740                 break;
2741         }
2742         return NOTIFY_OK;
2743 }
2744
2745 static struct notifier_block kvm_cpu_notifier = {
2746         .notifier_call = kvm_cpu_hotplug,
2747         .priority = 20, /* must be > scheduler priority */
2748 };
2749
2750 static __init void kvm_init_debug(void)
2751 {
2752         struct kvm_stats_debugfs_item *p;
2753
2754         debugfs_dir = debugfs_create_dir("kvm", NULL);
2755         for (p = debugfs_entries; p->name; ++p)
2756                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2757                                                p->data);
2758 }
2759
2760 static void kvm_exit_debug(void)
2761 {
2762         struct kvm_stats_debugfs_item *p;
2763
2764         for (p = debugfs_entries; p->name; ++p)
2765                 debugfs_remove(p->dentry);
2766         debugfs_remove(debugfs_dir);
2767 }
2768
2769 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2770 {
2771         decache_vcpus_on_cpu(raw_smp_processor_id());
2772         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2773         return 0;
2774 }
2775
2776 static int kvm_resume(struct sys_device *dev)
2777 {
2778         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2779         return 0;
2780 }
2781
2782 static struct sysdev_class kvm_sysdev_class = {
2783         set_kset_name("kvm"),
2784         .suspend = kvm_suspend,
2785         .resume = kvm_resume,
2786 };
2787
2788 static struct sys_device kvm_sysdev = {
2789         .id = 0,
2790         .cls = &kvm_sysdev_class,
2791 };
2792
2793 hpa_t bad_page_address;
2794
2795 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2796                         const char *dev_name, void *data, struct vfsmount *mnt)
2797 {
2798         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2799 }
2800
2801 static struct file_system_type kvm_fs_type = {
2802         .name           = "kvmfs",
2803         .get_sb         = kvmfs_get_sb,
2804         .kill_sb        = kill_anon_super,
2805 };
2806
2807 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2808 {
2809         int r;
2810
2811         if (kvm_arch_ops) {
2812                 printk(KERN_ERR "kvm: already loaded the other module\n");
2813                 return -EEXIST;
2814         }
2815
2816         if (!ops->cpu_has_kvm_support()) {
2817                 printk(KERN_ERR "kvm: no hardware support\n");
2818                 return -EOPNOTSUPP;
2819         }
2820         if (ops->disabled_by_bios()) {
2821                 printk(KERN_ERR "kvm: disabled by bios\n");
2822                 return -EOPNOTSUPP;
2823         }
2824
2825         kvm_arch_ops = ops;
2826
2827         r = kvm_arch_ops->hardware_setup();
2828         if (r < 0)
2829                 goto out;
2830
2831         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2832         r = register_cpu_notifier(&kvm_cpu_notifier);
2833         if (r)
2834                 goto out_free_1;
2835         register_reboot_notifier(&kvm_reboot_notifier);
2836
2837         r = sysdev_class_register(&kvm_sysdev_class);
2838         if (r)
2839                 goto out_free_2;
2840
2841         r = sysdev_register(&kvm_sysdev);
2842         if (r)
2843                 goto out_free_3;
2844
2845         kvm_chardev_ops.owner = module;
2846
2847         r = misc_register(&kvm_dev);
2848         if (r) {
2849                 printk (KERN_ERR "kvm: misc device register failed\n");
2850                 goto out_free;
2851         }
2852
2853         return r;
2854
2855 out_free:
2856         sysdev_unregister(&kvm_sysdev);
2857 out_free_3:
2858         sysdev_class_unregister(&kvm_sysdev_class);
2859 out_free_2:
2860         unregister_reboot_notifier(&kvm_reboot_notifier);
2861         unregister_cpu_notifier(&kvm_cpu_notifier);
2862 out_free_1:
2863         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2864         kvm_arch_ops->hardware_unsetup();
2865 out:
2866         kvm_arch_ops = NULL;
2867         return r;
2868 }
2869
2870 void kvm_exit_arch(void)
2871 {
2872         misc_deregister(&kvm_dev);
2873         sysdev_unregister(&kvm_sysdev);
2874         sysdev_class_unregister(&kvm_sysdev_class);
2875         unregister_reboot_notifier(&kvm_reboot_notifier);
2876         unregister_cpu_notifier(&kvm_cpu_notifier);
2877         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2878         kvm_arch_ops->hardware_unsetup();
2879         kvm_arch_ops = NULL;
2880 }
2881
2882 static __init int kvm_init(void)
2883 {
2884         static struct page *bad_page;
2885         int r;
2886
2887         r = register_filesystem(&kvm_fs_type);
2888         if (r)
2889                 goto out3;
2890
2891         kvmfs_mnt = kern_mount(&kvm_fs_type);
2892         r = PTR_ERR(kvmfs_mnt);
2893         if (IS_ERR(kvmfs_mnt))
2894                 goto out2;
2895         kvm_init_debug();
2896
2897         kvm_init_msr_list();
2898
2899         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2900                 r = -ENOMEM;
2901                 goto out;
2902         }
2903
2904         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2905         memset(__va(bad_page_address), 0, PAGE_SIZE);
2906
2907         return 0;
2908
2909 out:
2910         kvm_exit_debug();
2911         mntput(kvmfs_mnt);
2912 out2:
2913         unregister_filesystem(&kvm_fs_type);
2914 out3:
2915         return r;
2916 }
2917
2918 static __exit void kvm_exit(void)
2919 {
2920         kvm_exit_debug();
2921         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2922         mntput(kvmfs_mnt);
2923         unregister_filesystem(&kvm_fs_type);
2924 }
2925
2926 module_init(kvm_init)
2927 module_exit(kvm_exit)
2928
2929 EXPORT_SYMBOL_GPL(kvm_init_arch);
2930 EXPORT_SYMBOL_GPL(kvm_exit_arch);