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 page *page;
 424
 425         spin_lock(&vcpu->kvm->lock);
 426         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 427         /* FIXME: !page - emulate? 0xff? */
 428         pdpt = kmap_atomic(page, 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 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 865 {
 866         struct kvm_memory_slot *slot;
 867
 868         slot = gfn_to_memslot(kvm, gfn);
 869         if (!slot)
 870                 return NULL;
 871         return slot->phys_mem[gfn - slot->base_gfn];
 872 }
 873 EXPORT_SYMBOL_GPL(gfn_to_page);
 874
 875 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 876 {
 877         int i;
 878         struct kvm_memory_slot *memslot = NULL;
 879         unsigned long rel_gfn;
 880
 881         for (i = 0; i < kvm->nmemslots; ++i) {
 882                 memslot = &kvm->memslots[i];
 883
 884                 if (gfn >= memslot->base_gfn
 885                     && gfn < memslot->base_gfn + memslot->npages) {
 886
 887                         if (!memslot || !memslot->dirty_bitmap)
 888                                 return;
 889
 890                         rel_gfn = gfn - memslot->base_gfn;
 891
 892                         /* avoid RMW */
 893                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 894                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 895                         return;
 896                 }
 897         }
 898 }
 899
 900 static int emulator_read_std(unsigned long addr,
 901                              unsigned long *val,
 902                              unsigned int bytes,
 903                              struct x86_emulate_ctxt *ctxt)
 904 {
 905         struct kvm_vcpu *vcpu = ctxt->vcpu;
 906         void *data = val;
 907
 908         while (bytes) {
 909                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 910                 unsigned offset = addr & (PAGE_SIZE-1);
 911                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 912                 unsigned long pfn;
 913                 struct page *page;
 914                 void *page_virt;
 915
 916                 if (gpa == UNMAPPED_GVA)
 917                         return X86EMUL_PROPAGATE_FAULT;
 918                 pfn = gpa >> PAGE_SHIFT;
 919                 page = gfn_to_page(vcpu->kvm, pfn);
 920                 if (!page)
 921                         return X86EMUL_UNHANDLEABLE;
 922                 page_virt = kmap_atomic(page, KM_USER0);
 923
 924                 memcpy(data, page_virt + offset, tocopy);
 925
 926                 kunmap_atomic(page_virt, KM_USER0);
 927
 928                 bytes -= tocopy;
 929                 data += tocopy;
 930                 addr += tocopy;
 931         }
 932
 933         return X86EMUL_CONTINUE;
 934 }
 935
 936 static int emulator_write_std(unsigned long addr,
 937                               unsigned long val,
 938                               unsigned int bytes,
 939                               struct x86_emulate_ctxt *ctxt)
 940 {
 941         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 942                addr, bytes);
 943         return X86EMUL_UNHANDLEABLE;
 944 }
 945
 946 static int emulator_read_emulated(unsigned long addr,
 947                                   unsigned long *val,
 948                                   unsigned int bytes,
 949                                   struct x86_emulate_ctxt *ctxt)
 950 {
 951         struct kvm_vcpu *vcpu = ctxt->vcpu;
 952
 953         if (vcpu->mmio_read_completed) {
 954                 memcpy(val, vcpu->mmio_data, bytes);
 955                 vcpu->mmio_read_completed = 0;
 956                 return X86EMUL_CONTINUE;
 957         } else if (emulator_read_std(addr, val, bytes, ctxt)
 958                    == X86EMUL_CONTINUE)
 959                 return X86EMUL_CONTINUE;
 960         else {
 961                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 962
 963                 if (gpa == UNMAPPED_GVA)
 964                         return X86EMUL_PROPAGATE_FAULT;
 965                 vcpu->mmio_needed = 1;
 966                 vcpu->mmio_phys_addr = gpa;
 967                 vcpu->mmio_size = bytes;
 968                 vcpu->mmio_is_write = 0;
 969
 970                 return X86EMUL_UNHANDLEABLE;
 971         }
 972 }
 973
 974 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 975                                unsigned long val, int bytes)
 976 {
 977         struct page *page;
 978         void *virt;
 979
 980         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
 981                 return 0;
 982         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
 983         if (!page)
 984                 return 0;
 985         kvm_mmu_pre_write(vcpu, gpa, bytes);
 986         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
 987         virt = kmap_atomic(page, KM_USER0);
 988         memcpy(virt + offset_in_page(gpa), &val, bytes);
 989         kunmap_atomic(virt, KM_USER0);
 990         kvm_mmu_post_write(vcpu, gpa, bytes);
 991         return 1;
 992 }
 993
 994 static int emulator_write_emulated(unsigned long addr,
 995                                    unsigned long val,
 996                                    unsigned int bytes,
 997                                    struct x86_emulate_ctxt *ctxt)
 998 {
 999         struct kvm_vcpu *vcpu = ctxt->vcpu;
1000         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1001
1002         if (gpa == UNMAPPED_GVA)
1003                 return X86EMUL_PROPAGATE_FAULT;
1004
1005         if (emulator_write_phys(vcpu, gpa, val, bytes))
1006                 return X86EMUL_CONTINUE;
1007
1008         vcpu->mmio_needed = 1;
1009         vcpu->mmio_phys_addr = gpa;
1010         vcpu->mmio_size = bytes;
1011         vcpu->mmio_is_write = 1;
1012         memcpy(vcpu->mmio_data, &val, bytes);
1013
1014         return X86EMUL_CONTINUE;
1015 }
1016
1017 static int emulator_cmpxchg_emulated(unsigned long addr,
1018                                      unsigned long old,
1019                                      unsigned long new,
1020                                      unsigned int bytes,
1021                                      struct x86_emulate_ctxt *ctxt)
1022 {
1023         static int reported;
1024
1025         if (!reported) {
1026                 reported = 1;
1027                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1028         }
1029         return emulator_write_emulated(addr, new, bytes, ctxt);
1030 }
1031
1032 #ifdef CONFIG_X86_32
1033
1034 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1035                                        unsigned long old_lo,
1036                                        unsigned long old_hi,
1037                                        unsigned long new_lo,
1038                                        unsigned long new_hi,
1039                                        struct x86_emulate_ctxt *ctxt)
1040 {
1041         static int reported;
1042         int r;
1043
1044         if (!reported) {
1045                 reported = 1;
1046                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1047         }
1048         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1049         if (r != X86EMUL_CONTINUE)
1050                 return r;
1051         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1052 }
1053
1054 #endif
1055
1056 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1057 {
1058         return kvm_arch_ops->get_segment_base(vcpu, seg);
1059 }
1060
1061 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1062 {
1063         return X86EMUL_CONTINUE;
1064 }
1065
1066 int emulate_clts(struct kvm_vcpu *vcpu)
1067 {
1068         unsigned long cr0;
1069
1070         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1071         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1072         kvm_arch_ops->set_cr0(vcpu, cr0);
1073         return X86EMUL_CONTINUE;
1074 }
1075
1076 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1077 {
1078         struct kvm_vcpu *vcpu = ctxt->vcpu;
1079
1080         switch (dr) {
1081         case 0 ... 3:
1082                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1083                 return X86EMUL_CONTINUE;
1084         default:
1085                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1086                        __FUNCTION__, dr);
1087                 return X86EMUL_UNHANDLEABLE;
1088         }
1089 }
1090
1091 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1092 {
1093         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1094         int exception;
1095
1096         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1097         if (exception) {
1098                 /* FIXME: better handling */
1099                 return X86EMUL_UNHANDLEABLE;
1100         }
1101         return X86EMUL_CONTINUE;
1102 }
1103
1104 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1105 {
1106         static int reported;
1107         u8 opcodes[4];
1108         unsigned long rip = ctxt->vcpu->rip;
1109         unsigned long rip_linear;
1110
1111         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1112
1113         if (reported)
1114                 return;
1115
1116         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1117
1118         printk(KERN_ERR "emulation failed but !mmio_needed?"
1119                " rip %lx %02x %02x %02x %02x\n",
1120                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1121         reported = 1;
1122 }
1123
1124 struct x86_emulate_ops emulate_ops = {
1125         .read_std            = emulator_read_std,
1126         .write_std           = emulator_write_std,
1127         .read_emulated       = emulator_read_emulated,
1128         .write_emulated      = emulator_write_emulated,
1129         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1130 #ifdef CONFIG_X86_32
1131         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1132 #endif
1133 };
1134
1135 int emulate_instruction(struct kvm_vcpu *vcpu,
1136                         struct kvm_run *run,
1137                         unsigned long cr2,
1138                         u16 error_code)
1139 {
1140         struct x86_emulate_ctxt emulate_ctxt;
1141         int r;
1142         int cs_db, cs_l;
1143
1144         kvm_arch_ops->cache_regs(vcpu);
1145
1146         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1147
1148         emulate_ctxt.vcpu = vcpu;
1149         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1150         emulate_ctxt.cr2 = cr2;
1151         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1152                 ? X86EMUL_MODE_REAL : cs_l
1153                 ? X86EMUL_MODE_PROT64 : cs_db
1154                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1155
1156         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1157                 emulate_ctxt.cs_base = 0;
1158                 emulate_ctxt.ds_base = 0;
1159                 emulate_ctxt.es_base = 0;
1160                 emulate_ctxt.ss_base = 0;
1161         } else {
1162                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1163                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1164                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1165                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1166         }
1167
1168         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1169         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1170
1171         vcpu->mmio_is_write = 0;
1172         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1173
1174         if ((r || vcpu->mmio_is_write) && run) {
1175                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1176                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1177                 run->mmio.len = vcpu->mmio_size;
1178                 run->mmio.is_write = vcpu->mmio_is_write;
1179         }
1180
1181         if (r) {
1182                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1183                         return EMULATE_DONE;
1184                 if (!vcpu->mmio_needed) {
1185                         report_emulation_failure(&emulate_ctxt);
1186                         return EMULATE_FAIL;
1187                 }
1188                 return EMULATE_DO_MMIO;
1189         }
1190
1191         kvm_arch_ops->decache_regs(vcpu);
1192         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1193
1194         if (vcpu->mmio_is_write)
1195                 return EMULATE_DO_MMIO;
1196
1197         return EMULATE_DONE;
1198 }
1199 EXPORT_SYMBOL_GPL(emulate_instruction);
1200
1201 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1202 {
1203         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1204
1205         kvm_arch_ops->cache_regs(vcpu);
1206         ret = -KVM_EINVAL;
1207 #ifdef CONFIG_X86_64
1208         if (is_long_mode(vcpu)) {
1209                 nr = vcpu->regs[VCPU_REGS_RAX];
1210                 a0 = vcpu->regs[VCPU_REGS_RDI];
1211                 a1 = vcpu->regs[VCPU_REGS_RSI];
1212                 a2 = vcpu->regs[VCPU_REGS_RDX];
1213                 a3 = vcpu->regs[VCPU_REGS_RCX];
1214                 a4 = vcpu->regs[VCPU_REGS_R8];
1215                 a5 = vcpu->regs[VCPU_REGS_R9];
1216         } else
1217 #endif
1218         {
1219                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1220                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1221                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1222                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1223                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1224                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1225                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1226         }
1227         switch (nr) {
1228         default:
1229                 run->hypercall.args[0] = a0;
1230                 run->hypercall.args[1] = a1;
1231                 run->hypercall.args[2] = a2;
1232                 run->hypercall.args[3] = a3;
1233                 run->hypercall.args[4] = a4;
1234                 run->hypercall.args[5] = a5;
1235                 run->hypercall.ret = ret;
1236                 run->hypercall.longmode = is_long_mode(vcpu);
1237                 kvm_arch_ops->decache_regs(vcpu);
1238                 return 0;
1239         }
1240         vcpu->regs[VCPU_REGS_RAX] = ret;
1241         kvm_arch_ops->decache_regs(vcpu);
1242         return 1;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_hypercall);
1245
1246 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1247 {
1248         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1249 }
1250
1251 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1252 {
1253         struct descriptor_table dt = { limit, base };
1254
1255         kvm_arch_ops->set_gdt(vcpu, &dt);
1256 }
1257
1258 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1259 {
1260         struct descriptor_table dt = { limit, base };
1261
1262         kvm_arch_ops->set_idt(vcpu, &dt);
1263 }
1264
1265 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1266                    unsigned long *rflags)
1267 {
1268         lmsw(vcpu, msw);
1269         *rflags = kvm_arch_ops->get_rflags(vcpu);
1270 }
1271
1272 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1273 {
1274         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1275         switch (cr) {
1276         case 0:
1277                 return vcpu->cr0;
1278         case 2:
1279                 return vcpu->cr2;
1280         case 3:
1281                 return vcpu->cr3;
1282         case 4:
1283                 return vcpu->cr4;
1284         default:
1285                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1286                 return 0;
1287         }
1288 }
1289
1290 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1291                      unsigned long *rflags)
1292 {
1293         switch (cr) {
1294         case 0:
1295                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1296                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1297                 break;
1298         case 2:
1299                 vcpu->cr2 = val;
1300                 break;
1301         case 3:
1302                 set_cr3(vcpu, val);
1303                 break;
1304         case 4:
1305                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1306                 break;
1307         default:
1308                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1309         }
1310 }
1311
1312 /*
1313  * Register the para guest with the host:
1314  */
1315 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1316 {
1317         struct kvm_vcpu_para_state *para_state;
1318         hpa_t para_state_hpa, hypercall_hpa;
1319         struct page *para_state_page;
1320         unsigned char *hypercall;
1321         gpa_t hypercall_gpa;
1322
1323         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1324         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1325
1326         /*
1327          * Needs to be page aligned:
1328          */
1329         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1330                 goto err_gp;
1331
1332         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1333         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1334         if (is_error_hpa(para_state_hpa))
1335                 goto err_gp;
1336
1337         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1338         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1339         para_state = kmap_atomic(para_state_page, KM_USER0);
1340
1341         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1342         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1343
1344         para_state->host_version = KVM_PARA_API_VERSION;
1345         /*
1346          * We cannot support guests that try to register themselves
1347          * with a newer API version than the host supports:
1348          */
1349         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1350                 para_state->ret = -KVM_EINVAL;
1351                 goto err_kunmap_skip;
1352         }
1353
1354         hypercall_gpa = para_state->hypercall_gpa;
1355         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1356         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1357         if (is_error_hpa(hypercall_hpa)) {
1358                 para_state->ret = -KVM_EINVAL;
1359                 goto err_kunmap_skip;
1360         }
1361
1362         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1363         vcpu->para_state_page = para_state_page;
1364         vcpu->para_state_gpa = para_state_gpa;
1365         vcpu->hypercall_gpa = hypercall_gpa;
1366
1367         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1368         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1369                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1370         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1371         kunmap_atomic(hypercall, KM_USER1);
1372
1373         para_state->ret = 0;
1374 err_kunmap_skip:
1375         kunmap_atomic(para_state, KM_USER0);
1376         return 0;
1377 err_gp:
1378         return 1;
1379 }
1380
1381 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1382 {
1383         u64 data;
1384
1385         switch (msr) {
1386         case 0xc0010010: /* SYSCFG */
1387         case 0xc0010015: /* HWCR */
1388         case MSR_IA32_PLATFORM_ID:
1389         case MSR_IA32_P5_MC_ADDR:
1390         case MSR_IA32_P5_MC_TYPE:
1391         case MSR_IA32_MC0_CTL:
1392         case MSR_IA32_MCG_STATUS:
1393         case MSR_IA32_MCG_CAP:
1394         case MSR_IA32_MC0_MISC:
1395         case MSR_IA32_MC0_MISC+4:
1396         case MSR_IA32_MC0_MISC+8:
1397         case MSR_IA32_MC0_MISC+12:
1398         case MSR_IA32_MC0_MISC+16:
1399         case MSR_IA32_UCODE_REV:
1400         case MSR_IA32_PERF_STATUS:
1401                 /* MTRR registers */
1402         case 0xfe:
1403         case 0x200 ... 0x2ff:
1404                 data = 0;
1405                 break;
1406         case 0xcd: /* fsb frequency */
1407                 data = 3;
1408                 break;
1409         case MSR_IA32_APICBASE:
1410                 data = vcpu->apic_base;
1411                 break;
1412         case MSR_IA32_MISC_ENABLE:
1413                 data = vcpu->ia32_misc_enable_msr;
1414                 break;
1415 #ifdef CONFIG_X86_64
1416         case MSR_EFER:
1417                 data = vcpu->shadow_efer;
1418                 break;
1419 #endif
1420         default:
1421                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1422                 return 1;
1423         }
1424         *pdata = data;
1425         return 0;
1426 }
1427 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1428
1429 /*
1430  * Reads an msr value (of 'msr_index') into 'pdata'.
1431  * Returns 0 on success, non-0 otherwise.
1432  * Assumes vcpu_load() was already called.
1433  */
1434 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1435 {
1436         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1437 }
1438
1439 #ifdef CONFIG_X86_64
1440
1441 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1442 {
1443         if (efer & EFER_RESERVED_BITS) {
1444                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1445                        efer);
1446                 inject_gp(vcpu);
1447                 return;
1448         }
1449
1450         if (is_paging(vcpu)
1451             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1452                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1453                 inject_gp(vcpu);
1454                 return;
1455         }
1456
1457         kvm_arch_ops->set_efer(vcpu, efer);
1458
1459         efer &= ~EFER_LMA;
1460         efer |= vcpu->shadow_efer & EFER_LMA;
1461
1462         vcpu->shadow_efer = efer;
1463 }
1464
1465 #endif
1466
1467 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1468 {
1469         switch (msr) {
1470 #ifdef CONFIG_X86_64
1471         case MSR_EFER:
1472                 set_efer(vcpu, data);
1473                 break;
1474 #endif
1475         case MSR_IA32_MC0_STATUS:
1476                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1477                        __FUNCTION__, data);
1478                 break;
1479         case MSR_IA32_MCG_STATUS:
1480                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1481                         __FUNCTION__, data);
1482                 break;
1483         case MSR_IA32_UCODE_REV:
1484         case MSR_IA32_UCODE_WRITE:
1485         case 0x200 ... 0x2ff: /* MTRRs */
1486                 break;
1487         case MSR_IA32_APICBASE:
1488                 vcpu->apic_base = data;
1489                 break;
1490         case MSR_IA32_MISC_ENABLE:
1491                 vcpu->ia32_misc_enable_msr = data;
1492                 break;
1493         /*
1494          * This is the 'probe whether the host is KVM' logic:
1495          */
1496         case MSR_KVM_API_MAGIC:
1497                 return vcpu_register_para(vcpu, data);
1498
1499         default:
1500                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1501                 return 1;
1502         }
1503         return 0;
1504 }
1505 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1506
1507 /*
1508  * Writes msr value into into the appropriate "register".
1509  * Returns 0 on success, non-0 otherwise.
1510  * Assumes vcpu_load() was already called.
1511  */
1512 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1513 {
1514         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1515 }
1516
1517 void kvm_resched(struct kvm_vcpu *vcpu)
1518 {
1519         vcpu_put(vcpu);
1520         cond_resched();
1521         vcpu_load(vcpu);
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_resched);
1524
1525 void load_msrs(struct vmx_msr_entry *e, int n)
1526 {
1527         int i;
1528
1529         for (i = 0; i < n; ++i)
1530                 wrmsrl(e[i].index, e[i].data);
1531 }
1532 EXPORT_SYMBOL_GPL(load_msrs);
1533
1534 void save_msrs(struct vmx_msr_entry *e, int n)
1535 {
1536         int i;
1537
1538         for (i = 0; i < n; ++i)
1539                 rdmsrl(e[i].index, e[i].data);
1540 }
1541 EXPORT_SYMBOL_GPL(save_msrs);
1542
1543 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1544 {
1545         int i;
1546         u32 function;
1547         struct kvm_cpuid_entry *e, *best;
1548
1549         kvm_arch_ops->cache_regs(vcpu);
1550         function = vcpu->regs[VCPU_REGS_RAX];
1551         vcpu->regs[VCPU_REGS_RAX] = 0;
1552         vcpu->regs[VCPU_REGS_RBX] = 0;
1553         vcpu->regs[VCPU_REGS_RCX] = 0;
1554         vcpu->regs[VCPU_REGS_RDX] = 0;
1555         best = NULL;
1556         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1557                 e = &vcpu->cpuid_entries[i];
1558                 if (e->function == function) {
1559                         best = e;
1560                         break;
1561                 }
1562                 /*
1563                  * Both basic or both extended?
1564                  */
1565                 if (((e->function ^ function) & 0x80000000) == 0)
1566                         if (!best || e->function > best->function)
1567                                 best = e;
1568         }
1569         if (best) {
1570                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1571                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1572                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1573                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1574         }
1575         kvm_arch_ops->decache_regs(vcpu);
1576         kvm_arch_ops->skip_emulated_instruction(vcpu);
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1579
1580 static int pio_copy_data(struct kvm_vcpu *vcpu)
1581 {
1582         void *p = vcpu->pio_data;
1583         void *q;
1584         unsigned bytes;
1585         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1586
1587         kvm_arch_ops->vcpu_put(vcpu);
1588         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1589                  PAGE_KERNEL);
1590         if (!q) {
1591                 kvm_arch_ops->vcpu_load(vcpu);
1592                 free_pio_guest_pages(vcpu);
1593                 return -ENOMEM;
1594         }
1595         q += vcpu->pio.guest_page_offset;
1596         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1597         if (vcpu->pio.in)
1598                 memcpy(q, p, bytes);
1599         else
1600                 memcpy(p, q, bytes);
1601         q -= vcpu->pio.guest_page_offset;
1602         vunmap(q);
1603         kvm_arch_ops->vcpu_load(vcpu);
1604         free_pio_guest_pages(vcpu);
1605         return 0;
1606 }
1607
1608 static int complete_pio(struct kvm_vcpu *vcpu)
1609 {
1610         struct kvm_pio_request *io = &vcpu->pio;
1611         long delta;
1612         int r;
1613
1614         kvm_arch_ops->cache_regs(vcpu);
1615
1616         if (!io->string) {
1617                 if (io->in)
1618                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1619                                io->size);
1620         } else {
1621                 if (io->in) {
1622                         r = pio_copy_data(vcpu);
1623                         if (r) {
1624                                 kvm_arch_ops->cache_regs(vcpu);
1625                                 return r;
1626                         }
1627                 }
1628
1629                 delta = 1;
1630                 if (io->rep) {
1631                         delta *= io->cur_count;
1632                         /*
1633                          * The size of the register should really depend on
1634                          * current address size.
1635                          */
1636                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1637                 }
1638                 if (io->down)
1639                         delta = -delta;
1640                 delta *= io->size;
1641                 if (io->in)
1642                         vcpu->regs[VCPU_REGS_RDI] += delta;
1643                 else
1644                         vcpu->regs[VCPU_REGS_RSI] += delta;
1645         }
1646
1647         vcpu->run->io_completed = 0;
1648
1649         kvm_arch_ops->decache_regs(vcpu);
1650
1651         io->count -= io->cur_count;
1652         io->cur_count = 0;
1653
1654         if (!io->count)
1655                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1656         return 0;
1657 }
1658
1659 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1660                   int size, unsigned long count, int string, int down,
1661                   gva_t address, int rep, unsigned port)
1662 {
1663         unsigned now, in_page;
1664         int i;
1665         int nr_pages = 1;
1666         struct page *page;
1667
1668         vcpu->run->exit_reason = KVM_EXIT_IO;
1669         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1670         vcpu->run->io.size = size;
1671         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1672         vcpu->run->io.count = count;
1673         vcpu->run->io.port = port;
1674         vcpu->pio.count = count;
1675         vcpu->pio.cur_count = count;
1676         vcpu->pio.size = size;
1677         vcpu->pio.in = in;
1678         vcpu->pio.string = string;
1679         vcpu->pio.down = down;
1680         vcpu->pio.guest_page_offset = offset_in_page(address);
1681         vcpu->pio.rep = rep;
1682
1683         if (!string) {
1684                 kvm_arch_ops->cache_regs(vcpu);
1685                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1686                 kvm_arch_ops->decache_regs(vcpu);
1687                 return 0;
1688         }
1689
1690         if (!count) {
1691                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1692                 return 1;
1693         }
1694
1695         now = min(count, PAGE_SIZE / size);
1696
1697         if (!down)
1698                 in_page = PAGE_SIZE - offset_in_page(address);
1699         else
1700                 in_page = offset_in_page(address) + size;
1701         now = min(count, (unsigned long)in_page / size);
1702         if (!now) {
1703                 /*
1704                  * String I/O straddles page boundary.  Pin two guest pages
1705                  * so that we satisfy atomicity constraints.  Do just one
1706                  * transaction to avoid complexity.
1707                  */
1708                 nr_pages = 2;
1709                 now = 1;
1710         }
1711         if (down) {
1712                 /*
1713                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1714                  */
1715                 printk(KERN_ERR "kvm: guest string pio down\n");
1716                 inject_gp(vcpu);
1717                 return 1;
1718         }
1719         vcpu->run->io.count = now;
1720         vcpu->pio.cur_count = now;
1721
1722         for (i = 0; i < nr_pages; ++i) {
1723                 spin_lock(&vcpu->kvm->lock);
1724                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1725                 if (page)
1726                         get_page(page);
1727                 vcpu->pio.guest_pages[i] = page;
1728                 spin_unlock(&vcpu->kvm->lock);
1729                 if (!page) {
1730                         inject_gp(vcpu);
1731                         free_pio_guest_pages(vcpu);
1732                         return 1;
1733                 }
1734         }
1735
1736         if (!vcpu->pio.in)
1737                 return pio_copy_data(vcpu);
1738         return 0;
1739 }
1740 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1741
1742 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1743 {
1744         int r;
1745         sigset_t sigsaved;
1746
1747         vcpu_load(vcpu);
1748
1749         if (vcpu->sigset_active)
1750                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1751
1752         /* re-sync apic's tpr */
1753         vcpu->cr8 = kvm_run->cr8;
1754
1755         if (kvm_run->io_completed) {
1756                 if (vcpu->pio.cur_count) {
1757                         r = complete_pio(vcpu);
1758                         if (r)
1759                                 goto out;
1760                 } else {
1761                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1762                         vcpu->mmio_read_completed = 1;
1763                 }
1764         }
1765
1766         vcpu->mmio_needed = 0;
1767
1768         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1769                 kvm_arch_ops->cache_regs(vcpu);
1770                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1771                 kvm_arch_ops->decache_regs(vcpu);
1772         }
1773
1774         r = kvm_arch_ops->run(vcpu, kvm_run);
1775
1776 out:
1777         if (vcpu->sigset_active)
1778                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1779
1780         vcpu_put(vcpu);
1781         return r;
1782 }
1783
1784 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1785                                    struct kvm_regs *regs)
1786 {
1787         vcpu_load(vcpu);
1788
1789         kvm_arch_ops->cache_regs(vcpu);
1790
1791         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1792         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1793         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1794         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1795         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1796         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1797         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1798         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1799 #ifdef CONFIG_X86_64
1800         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1801         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1802         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1803         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1804         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1805         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1806         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1807         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1808 #endif
1809
1810         regs->rip = vcpu->rip;
1811         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1812
1813         /*
1814          * Don't leak debug flags in case they were set for guest debugging
1815          */
1816         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1817                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1818
1819         vcpu_put(vcpu);
1820
1821         return 0;
1822 }
1823
1824 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1825                                    struct kvm_regs *regs)
1826 {
1827         vcpu_load(vcpu);
1828
1829         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1830         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1831         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1832         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1833         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1834         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1835         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1836         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1837 #ifdef CONFIG_X86_64
1838         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1839         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1840         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1841         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1842         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1843         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1844         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1845         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1846 #endif
1847
1848         vcpu->rip = regs->rip;
1849         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1850
1851         kvm_arch_ops->decache_regs(vcpu);
1852
1853         vcpu_put(vcpu);
1854
1855         return 0;
1856 }
1857
1858 static void get_segment(struct kvm_vcpu *vcpu,
1859                         struct kvm_segment *var, int seg)
1860 {
1861         return kvm_arch_ops->get_segment(vcpu, var, seg);
1862 }
1863
1864 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1865                                     struct kvm_sregs *sregs)
1866 {
1867         struct descriptor_table dt;
1868
1869         vcpu_load(vcpu);
1870
1871         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1872         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1873         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1874         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1875         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1876         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1877
1878         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1879         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1880
1881         kvm_arch_ops->get_idt(vcpu, &dt);
1882         sregs->idt.limit = dt.limit;
1883         sregs->idt.base = dt.base;
1884         kvm_arch_ops->get_gdt(vcpu, &dt);
1885         sregs->gdt.limit = dt.limit;
1886         sregs->gdt.base = dt.base;
1887
1888         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1889         sregs->cr0 = vcpu->cr0;
1890         sregs->cr2 = vcpu->cr2;
1891         sregs->cr3 = vcpu->cr3;
1892         sregs->cr4 = vcpu->cr4;
1893         sregs->cr8 = vcpu->cr8;
1894         sregs->efer = vcpu->shadow_efer;
1895         sregs->apic_base = vcpu->apic_base;
1896
1897         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1898                sizeof sregs->interrupt_bitmap);
1899
1900         vcpu_put(vcpu);
1901
1902         return 0;
1903 }
1904
1905 static void set_segment(struct kvm_vcpu *vcpu,
1906                         struct kvm_segment *var, int seg)
1907 {
1908         return kvm_arch_ops->set_segment(vcpu, var, seg);
1909 }
1910
1911 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1912                                     struct kvm_sregs *sregs)
1913 {
1914         int mmu_reset_needed = 0;
1915         int i;
1916         struct descriptor_table dt;
1917
1918         vcpu_load(vcpu);
1919
1920         dt.limit = sregs->idt.limit;
1921         dt.base = sregs->idt.base;
1922         kvm_arch_ops->set_idt(vcpu, &dt);
1923         dt.limit = sregs->gdt.limit;
1924         dt.base = sregs->gdt.base;
1925         kvm_arch_ops->set_gdt(vcpu, &dt);
1926
1927         vcpu->cr2 = sregs->cr2;
1928         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1929         vcpu->cr3 = sregs->cr3;
1930
1931         vcpu->cr8 = sregs->cr8;
1932
1933         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1934 #ifdef CONFIG_X86_64
1935         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1936 #endif
1937         vcpu->apic_base = sregs->apic_base;
1938
1939         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1940
1941         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1942         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
1943
1944         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1945         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1946         if (!is_long_mode(vcpu) && is_pae(vcpu))
1947                 load_pdptrs(vcpu, vcpu->cr3);
1948
1949         if (mmu_reset_needed)
1950                 kvm_mmu_reset_context(vcpu);
1951
1952         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1953                sizeof vcpu->irq_pending);
1954         vcpu->irq_summary = 0;
1955         for (i = 0; i < NR_IRQ_WORDS; ++i)
1956                 if (vcpu->irq_pending[i])
1957                         __set_bit(i, &vcpu->irq_summary);
1958
1959         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1960         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1961         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1962         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1963         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1964         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1965
1966         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1967         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1968
1969         vcpu_put(vcpu);
1970
1971         return 0;
1972 }
1973
1974 /*
1975  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1976  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1977  *
1978  * This list is modified at module load time to reflect the
1979  * capabilities of the host cpu.
1980  */
1981 static u32 msrs_to_save[] = {
1982         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1983         MSR_K6_STAR,
1984 #ifdef CONFIG_X86_64
1985         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1986 #endif
1987         MSR_IA32_TIME_STAMP_COUNTER,
1988 };
1989
1990 static unsigned num_msrs_to_save;
1991
1992 static u32 emulated_msrs[] = {
1993         MSR_IA32_MISC_ENABLE,
1994 };
1995
1996 static __init void kvm_init_msr_list(void)
1997 {
1998         u32 dummy[2];
1999         unsigned i, j;
2000
2001         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2002                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2003                         continue;
2004                 if (j < i)
2005                         msrs_to_save[j] = msrs_to_save[i];
2006                 j++;
2007         }
2008         num_msrs_to_save = j;
2009 }
2010
2011 /*
2012  * Adapt set_msr() to msr_io()'s calling convention
2013  */
2014 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2015 {
2016         return set_msr(vcpu, index, *data);
2017 }
2018
2019 /*
2020  * Read or write a bunch of msrs. All parameters are kernel addresses.
2021  *
2022  * @return number of msrs set successfully.
2023  */
2024 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2025                     struct kvm_msr_entry *entries,
2026                     int (*do_msr)(struct kvm_vcpu *vcpu,
2027                                   unsigned index, u64 *data))
2028 {
2029         int i;
2030
2031         vcpu_load(vcpu);
2032
2033         for (i = 0; i < msrs->nmsrs; ++i)
2034                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2035                         break;
2036
2037         vcpu_put(vcpu);
2038
2039         return i;
2040 }
2041
2042 /*
2043  * Read or write a bunch of msrs. Parameters are user addresses.
2044  *
2045  * @return number of msrs set successfully.
2046  */
2047 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2048                   int (*do_msr)(struct kvm_vcpu *vcpu,
2049                                 unsigned index, u64 *data),
2050                   int writeback)
2051 {
2052         struct kvm_msrs msrs;
2053         struct kvm_msr_entry *entries;
2054         int r, n;
2055         unsigned size;
2056
2057         r = -EFAULT;
2058         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2059                 goto out;
2060
2061         r = -E2BIG;
2062         if (msrs.nmsrs >= MAX_IO_MSRS)
2063                 goto out;
2064
2065         r = -ENOMEM;
2066         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2067         entries = vmalloc(size);
2068         if (!entries)
2069                 goto out;
2070
2071         r = -EFAULT;
2072         if (copy_from_user(entries, user_msrs->entries, size))
2073                 goto out_free;
2074
2075         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2076         if (r < 0)
2077                 goto out_free;
2078
2079         r = -EFAULT;
2080         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2081                 goto out_free;
2082
2083         r = n;
2084
2085 out_free:
2086         vfree(entries);
2087 out:
2088         return r;
2089 }
2090
2091 /*
2092  * Translate a guest virtual address to a guest physical address.
2093  */
2094 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2095                                     struct kvm_translation *tr)
2096 {
2097         unsigned long vaddr = tr->linear_address;
2098         gpa_t gpa;
2099
2100         vcpu_load(vcpu);
2101         spin_lock(&vcpu->kvm->lock);
2102         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2103         tr->physical_address = gpa;
2104         tr->valid = gpa != UNMAPPED_GVA;
2105         tr->writeable = 1;
2106         tr->usermode = 0;
2107         spin_unlock(&vcpu->kvm->lock);
2108         vcpu_put(vcpu);
2109
2110         return 0;
2111 }
2112
2113 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2114                                     struct kvm_interrupt *irq)
2115 {
2116         if (irq->irq < 0 || irq->irq >= 256)
2117                 return -EINVAL;
2118         vcpu_load(vcpu);
2119
2120         set_bit(irq->irq, vcpu->irq_pending);
2121         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2122
2123         vcpu_put(vcpu);
2124
2125         return 0;
2126 }
2127
2128 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2129                                       struct kvm_debug_guest *dbg)
2130 {
2131         int r;
2132
2133         vcpu_load(vcpu);
2134
2135         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2136
2137         vcpu_put(vcpu);
2138
2139         return r;
2140 }
2141
2142 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2143                                     unsigned long address,
2144                                     int *type)
2145 {
2146         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2147         unsigned long pgoff;
2148         struct page *page;
2149
2150         *type = VM_FAULT_MINOR;
2151         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2152         if (pgoff == 0)
2153                 page = virt_to_page(vcpu->run);
2154         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2155                 page = virt_to_page(vcpu->pio_data);
2156         else
2157                 return NOPAGE_SIGBUS;
2158         get_page(page);
2159         return page;
2160 }
2161
2162 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2163         .nopage = kvm_vcpu_nopage,
2164 };
2165
2166 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2167 {
2168         vma->vm_ops = &kvm_vcpu_vm_ops;
2169         return 0;
2170 }
2171
2172 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2173 {
2174         struct kvm_vcpu *vcpu = filp->private_data;
2175
2176         fput(vcpu->kvm->filp);
2177         return 0;
2178 }
2179
2180 static struct file_operations kvm_vcpu_fops = {
2181         .release        = kvm_vcpu_release,
2182         .unlocked_ioctl = kvm_vcpu_ioctl,
2183         .compat_ioctl   = kvm_vcpu_ioctl,
2184         .mmap           = kvm_vcpu_mmap,
2185 };
2186
2187 /*
2188  * Allocates an inode for the vcpu.
2189  */
2190 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2191 {
2192         int fd, r;
2193         struct inode *inode;
2194         struct file *file;
2195
2196         atomic_inc(&vcpu->kvm->filp->f_count);
2197         inode = kvmfs_inode(&kvm_vcpu_fops);
2198         if (IS_ERR(inode)) {
2199                 r = PTR_ERR(inode);
2200                 goto out1;
2201         }
2202
2203         file = kvmfs_file(inode, vcpu);
2204         if (IS_ERR(file)) {
2205                 r = PTR_ERR(file);
2206                 goto out2;
2207         }
2208
2209         r = get_unused_fd();
2210         if (r < 0)
2211                 goto out3;
2212         fd = r;
2213         fd_install(fd, file);
2214
2215         return fd;
2216
2217 out3:
2218         fput(file);
2219 out2:
2220         iput(inode);
2221 out1:
2222         fput(vcpu->kvm->filp);
2223         return r;
2224 }
2225
2226 /*
2227  * Creates some virtual cpus.  Good luck creating more than one.
2228  */
2229 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2230 {
2231         int r;
2232         struct kvm_vcpu *vcpu;
2233         struct page *page;
2234
2235         r = -EINVAL;
2236         if (!valid_vcpu(n))
2237                 goto out;
2238
2239         vcpu = &kvm->vcpus[n];
2240
2241         mutex_lock(&vcpu->mutex);
2242
2243         if (vcpu->vmcs) {
2244                 mutex_unlock(&vcpu->mutex);
2245                 return -EEXIST;
2246         }
2247
2248         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2249         r = -ENOMEM;
2250         if (!page)
2251                 goto out_unlock;
2252         vcpu->run = page_address(page);
2253
2254         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2255         r = -ENOMEM;
2256         if (!page)
2257                 goto out_free_run;
2258         vcpu->pio_data = page_address(page);
2259
2260         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2261                                            FX_IMAGE_ALIGN);
2262         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2263
2264         r = kvm_arch_ops->vcpu_create(vcpu);
2265         if (r < 0)
2266                 goto out_free_vcpus;
2267
2268         r = kvm_mmu_create(vcpu);
2269         if (r < 0)
2270                 goto out_free_vcpus;
2271
2272         kvm_arch_ops->vcpu_load(vcpu);
2273         r = kvm_mmu_setup(vcpu);
2274         if (r >= 0)
2275                 r = kvm_arch_ops->vcpu_setup(vcpu);
2276         vcpu_put(vcpu);
2277
2278         if (r < 0)
2279                 goto out_free_vcpus;
2280
2281         r = create_vcpu_fd(vcpu);
2282         if (r < 0)
2283                 goto out_free_vcpus;
2284
2285         return r;
2286
2287 out_free_vcpus:
2288         kvm_free_vcpu(vcpu);
2289 out_free_run:
2290         free_page((unsigned long)vcpu->run);
2291         vcpu->run = NULL;
2292 out_unlock:
2293         mutex_unlock(&vcpu->mutex);
2294 out:
2295         return r;
2296 }
2297
2298 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2299                                     struct kvm_cpuid *cpuid,
2300                                     struct kvm_cpuid_entry __user *entries)
2301 {
2302         int r;
2303
2304         r = -E2BIG;
2305         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2306                 goto out;
2307         r = -EFAULT;
2308         if (copy_from_user(&vcpu->cpuid_entries, entries,
2309                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2310                 goto out;
2311         vcpu->cpuid_nent = cpuid->nent;
2312         return 0;
2313
2314 out:
2315         return r;
2316 }
2317
2318 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2319 {
2320         if (sigset) {
2321                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2322                 vcpu->sigset_active = 1;
2323                 vcpu->sigset = *sigset;
2324         } else
2325                 vcpu->sigset_active = 0;
2326         return 0;
2327 }
2328
2329 static long kvm_vcpu_ioctl(struct file *filp,
2330                            unsigned int ioctl, unsigned long arg)
2331 {
2332         struct kvm_vcpu *vcpu = filp->private_data;
2333         void __user *argp = (void __user *)arg;
2334         int r = -EINVAL;
2335
2336         switch (ioctl) {
2337         case KVM_RUN:
2338                 r = -EINVAL;
2339                 if (arg)
2340                         goto out;
2341                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2342                 break;
2343         case KVM_GET_REGS: {
2344                 struct kvm_regs kvm_regs;
2345
2346                 memset(&kvm_regs, 0, sizeof kvm_regs);
2347                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2348                 if (r)
2349                         goto out;
2350                 r = -EFAULT;
2351                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2352                         goto out;
2353                 r = 0;
2354                 break;
2355         }
2356         case KVM_SET_REGS: {
2357                 struct kvm_regs kvm_regs;
2358
2359                 r = -EFAULT;
2360                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2361                         goto out;
2362                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2363                 if (r)
2364                         goto out;
2365                 r = 0;
2366                 break;
2367         }
2368         case KVM_GET_SREGS: {
2369                 struct kvm_sregs kvm_sregs;
2370
2371                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2372                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2373                 if (r)
2374                         goto out;
2375                 r = -EFAULT;
2376                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2377                         goto out;
2378                 r = 0;
2379                 break;
2380         }
2381         case KVM_SET_SREGS: {
2382                 struct kvm_sregs kvm_sregs;
2383
2384                 r = -EFAULT;
2385                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2386                         goto out;
2387                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2388                 if (r)
2389                         goto out;
2390                 r = 0;
2391                 break;
2392         }
2393         case KVM_TRANSLATE: {
2394                 struct kvm_translation tr;
2395
2396                 r = -EFAULT;
2397                 if (copy_from_user(&tr, argp, sizeof tr))
2398                         goto out;
2399                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2400                 if (r)
2401                         goto out;
2402                 r = -EFAULT;
2403                 if (copy_to_user(argp, &tr, sizeof tr))
2404                         goto out;
2405                 r = 0;
2406                 break;
2407         }
2408         case KVM_INTERRUPT: {
2409                 struct kvm_interrupt irq;
2410
2411                 r = -EFAULT;
2412                 if (copy_from_user(&irq, argp, sizeof irq))
2413                         goto out;
2414                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2415                 if (r)
2416                         goto out;
2417                 r = 0;
2418                 break;
2419         }
2420         case KVM_DEBUG_GUEST: {
2421                 struct kvm_debug_guest dbg;
2422
2423                 r = -EFAULT;
2424                 if (copy_from_user(&dbg, argp, sizeof dbg))
2425                         goto out;
2426                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2427                 if (r)
2428                         goto out;
2429                 r = 0;
2430                 break;
2431         }
2432         case KVM_GET_MSRS:
2433                 r = msr_io(vcpu, argp, get_msr, 1);
2434                 break;
2435         case KVM_SET_MSRS:
2436                 r = msr_io(vcpu, argp, do_set_msr, 0);
2437                 break;
2438         case KVM_SET_CPUID: {
2439                 struct kvm_cpuid __user *cpuid_arg = argp;
2440                 struct kvm_cpuid cpuid;
2441
2442                 r = -EFAULT;
2443                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2444                         goto out;
2445                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2446                 if (r)
2447                         goto out;
2448                 break;
2449         }
2450         case KVM_SET_SIGNAL_MASK: {
2451                 struct kvm_signal_mask __user *sigmask_arg = argp;
2452                 struct kvm_signal_mask kvm_sigmask;
2453                 sigset_t sigset, *p;
2454
2455                 p = NULL;
2456                 if (argp) {
2457                         r = -EFAULT;
2458                         if (copy_from_user(&kvm_sigmask, argp,
2459                                            sizeof kvm_sigmask))
2460                                 goto out;
2461                         r = -EINVAL;
2462                         if (kvm_sigmask.len != sizeof sigset)
2463                                 goto out;
2464                         r = -EFAULT;
2465                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2466                                            sizeof sigset))
2467                                 goto out;
2468                         p = &sigset;
2469                 }
2470                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2471                 break;
2472         }
2473         default:
2474                 ;
2475         }
2476 out:
2477         return r;
2478 }
2479
2480 static long kvm_vm_ioctl(struct file *filp,
2481                            unsigned int ioctl, unsigned long arg)
2482 {
2483         struct kvm *kvm = filp->private_data;
2484         void __user *argp = (void __user *)arg;
2485         int r = -EINVAL;
2486
2487         switch (ioctl) {
2488         case KVM_CREATE_VCPU:
2489                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2490                 if (r < 0)
2491                         goto out;
2492                 break;
2493         case KVM_SET_MEMORY_REGION: {
2494                 struct kvm_memory_region kvm_mem;
2495
2496                 r = -EFAULT;
2497                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2498                         goto out;
2499                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2500                 if (r)
2501                         goto out;
2502                 break;
2503         }
2504         case KVM_GET_DIRTY_LOG: {
2505                 struct kvm_dirty_log log;
2506
2507                 r = -EFAULT;
2508                 if (copy_from_user(&log, argp, sizeof log))
2509                         goto out;
2510                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2511                 if (r)
2512                         goto out;
2513                 break;
2514         }
2515         default:
2516                 ;
2517         }
2518 out:
2519         return r;
2520 }
2521
2522 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2523                                   unsigned long address,
2524                                   int *type)
2525 {
2526         struct kvm *kvm = vma->vm_file->private_data;
2527         unsigned long pgoff;
2528         struct page *page;
2529
2530         *type = VM_FAULT_MINOR;
2531         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2532         page = gfn_to_page(kvm, pgoff);
2533         if (!page)
2534                 return NOPAGE_SIGBUS;
2535         get_page(page);
2536         return page;
2537 }
2538
2539 static struct vm_operations_struct kvm_vm_vm_ops = {
2540         .nopage = kvm_vm_nopage,
2541 };
2542
2543 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2544 {
2545         vma->vm_ops = &kvm_vm_vm_ops;
2546         return 0;
2547 }
2548
2549 static struct file_operations kvm_vm_fops = {
2550         .release        = kvm_vm_release,
2551         .unlocked_ioctl = kvm_vm_ioctl,
2552         .compat_ioctl   = kvm_vm_ioctl,
2553         .mmap           = kvm_vm_mmap,
2554 };
2555
2556 static int kvm_dev_ioctl_create_vm(void)
2557 {
2558         int fd, r;
2559         struct inode *inode;
2560         struct file *file;
2561         struct kvm *kvm;
2562
2563         inode = kvmfs_inode(&kvm_vm_fops);
2564         if (IS_ERR(inode)) {
2565                 r = PTR_ERR(inode);
2566                 goto out1;
2567         }
2568
2569         kvm = kvm_create_vm();
2570         if (IS_ERR(kvm)) {
2571                 r = PTR_ERR(kvm);
2572                 goto out2;
2573         }
2574
2575         file = kvmfs_file(inode, kvm);
2576         if (IS_ERR(file)) {
2577                 r = PTR_ERR(file);
2578                 goto out3;
2579         }
2580         kvm->filp = file;
2581
2582         r = get_unused_fd();
2583         if (r < 0)
2584                 goto out4;
2585         fd = r;
2586         fd_install(fd, file);
2587
2588         return fd;
2589
2590 out4:
2591         fput(file);
2592 out3:
2593         kvm_destroy_vm(kvm);
2594 out2:
2595         iput(inode);
2596 out1:
2597         return r;
2598 }
2599
2600 static long kvm_dev_ioctl(struct file *filp,
2601                           unsigned int ioctl, unsigned long arg)
2602 {
2603         void __user *argp = (void __user *)arg;
2604         long r = -EINVAL;
2605
2606         switch (ioctl) {
2607         case KVM_GET_API_VERSION:
2608                 r = -EINVAL;
2609                 if (arg)
2610                         goto out;
2611                 r = KVM_API_VERSION;
2612                 break;
2613         case KVM_CREATE_VM:
2614                 r = -EINVAL;
2615                 if (arg)
2616                         goto out;
2617                 r = kvm_dev_ioctl_create_vm();
2618                 break;
2619         case KVM_GET_MSR_INDEX_LIST: {
2620                 struct kvm_msr_list __user *user_msr_list = argp;
2621                 struct kvm_msr_list msr_list;
2622                 unsigned n;
2623
2624                 r = -EFAULT;
2625                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2626                         goto out;
2627                 n = msr_list.nmsrs;
2628                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2629                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2630                         goto out;
2631                 r = -E2BIG;
2632                 if (n < num_msrs_to_save)
2633                         goto out;
2634                 r = -EFAULT;
2635                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2636                                  num_msrs_to_save * sizeof(u32)))
2637                         goto out;
2638                 if (copy_to_user(user_msr_list->indices
2639                                  + num_msrs_to_save * sizeof(u32),
2640                                  &emulated_msrs,
2641                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2642                         goto out;
2643                 r = 0;
2644                 break;
2645         }
2646         case KVM_CHECK_EXTENSION:
2647                 /*
2648                  * No extensions defined at present.
2649                  */
2650                 r = 0;
2651                 break;
2652         case KVM_GET_VCPU_MMAP_SIZE:
2653                 r = -EINVAL;
2654                 if (arg)
2655                         goto out;
2656                 r = 2 * PAGE_SIZE;
2657                 break;
2658         default:
2659                 ;
2660         }
2661 out:
2662         return r;
2663 }
2664
2665 static struct file_operations kvm_chardev_ops = {
2666         .open           = kvm_dev_open,
2667         .release        = kvm_dev_release,
2668         .unlocked_ioctl = kvm_dev_ioctl,
2669         .compat_ioctl   = kvm_dev_ioctl,
2670 };
2671
2672 static struct miscdevice kvm_dev = {
2673         KVM_MINOR,
2674         "kvm",
2675         &kvm_chardev_ops,
2676 };
2677
2678 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2679                        void *v)
2680 {
2681         if (val == SYS_RESTART) {
2682                 /*
2683                  * Some (well, at least mine) BIOSes hang on reboot if
2684                  * in vmx root mode.
2685                  */
2686                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2687                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2688         }
2689         return NOTIFY_OK;
2690 }
2691
2692 static struct notifier_block kvm_reboot_notifier = {
2693         .notifier_call = kvm_reboot,
2694         .priority = 0,
2695 };
2696
2697 /*
2698  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2699  * cached on it.
2700  */
2701 static void decache_vcpus_on_cpu(int cpu)
2702 {
2703         struct kvm *vm;
2704         struct kvm_vcpu *vcpu;
2705         int i;
2706
2707         spin_lock(&kvm_lock);
2708         list_for_each_entry(vm, &vm_list, vm_list)
2709                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2710                         vcpu = &vm->vcpus[i];
2711                         /*
2712                          * If the vcpu is locked, then it is running on some
2713                          * other cpu and therefore it is not cached on the
2714                          * cpu in question.
2715                          *
2716                          * If it's not locked, check the last cpu it executed
2717                          * on.
2718                          */
2719                         if (mutex_trylock(&vcpu->mutex)) {
2720                                 if (vcpu->cpu == cpu) {
2721                                         kvm_arch_ops->vcpu_decache(vcpu);
2722                                         vcpu->cpu = -1;
2723                                 }
2724                                 mutex_unlock(&vcpu->mutex);
2725                         }
2726                 }
2727         spin_unlock(&kvm_lock);
2728 }
2729
2730 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2731                            void *v)
2732 {
2733         int cpu = (long)v;
2734
2735         switch (val) {
2736         case CPU_DOWN_PREPARE:
2737         case CPU_UP_CANCELED:
2738                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2739                        cpu);
2740                 decache_vcpus_on_cpu(cpu);
2741                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2742                                          NULL, 0, 1);
2743                 break;
2744         case CPU_ONLINE:
2745                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2746                        cpu);
2747                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2748                                          NULL, 0, 1);
2749                 break;
2750         }
2751         return NOTIFY_OK;
2752 }
2753
2754 static struct notifier_block kvm_cpu_notifier = {
2755         .notifier_call = kvm_cpu_hotplug,
2756         .priority = 20, /* must be > scheduler priority */
2757 };
2758
2759 static __init void kvm_init_debug(void)
2760 {
2761         struct kvm_stats_debugfs_item *p;
2762
2763         debugfs_dir = debugfs_create_dir("kvm", NULL);
2764         for (p = debugfs_entries; p->name; ++p)
2765                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2766                                                p->data);
2767 }
2768
2769 static void kvm_exit_debug(void)
2770 {
2771         struct kvm_stats_debugfs_item *p;
2772
2773         for (p = debugfs_entries; p->name; ++p)
2774                 debugfs_remove(p->dentry);
2775         debugfs_remove(debugfs_dir);
2776 }
2777
2778 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2779 {
2780         decache_vcpus_on_cpu(raw_smp_processor_id());
2781         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2782         return 0;
2783 }
2784
2785 static int kvm_resume(struct sys_device *dev)
2786 {
2787         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2788         return 0;
2789 }
2790
2791 static struct sysdev_class kvm_sysdev_class = {
2792         set_kset_name("kvm"),
2793         .suspend = kvm_suspend,
2794         .resume = kvm_resume,
2795 };
2796
2797 static struct sys_device kvm_sysdev = {
2798         .id = 0,
2799         .cls = &kvm_sysdev_class,
2800 };
2801
2802 hpa_t bad_page_address;
2803
2804 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2805                         const char *dev_name, void *data, struct vfsmount *mnt)
2806 {
2807         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2808 }
2809
2810 static struct file_system_type kvm_fs_type = {
2811         .name           = "kvmfs",
2812         .get_sb         = kvmfs_get_sb,
2813         .kill_sb        = kill_anon_super,
2814 };
2815
2816 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2817 {
2818         int r;
2819
2820         if (kvm_arch_ops) {
2821                 printk(KERN_ERR "kvm: already loaded the other module\n");
2822                 return -EEXIST;
2823         }
2824
2825         if (!ops->cpu_has_kvm_support()) {
2826                 printk(KERN_ERR "kvm: no hardware support\n");
2827                 return -EOPNOTSUPP;
2828         }
2829         if (ops->disabled_by_bios()) {
2830                 printk(KERN_ERR "kvm: disabled by bios\n");
2831                 return -EOPNOTSUPP;
2832         }
2833
2834         kvm_arch_ops = ops;
2835
2836         r = kvm_arch_ops->hardware_setup();
2837         if (r < 0)
2838                 goto out;
2839
2840         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2841         r = register_cpu_notifier(&kvm_cpu_notifier);
2842         if (r)
2843                 goto out_free_1;
2844         register_reboot_notifier(&kvm_reboot_notifier);
2845
2846         r = sysdev_class_register(&kvm_sysdev_class);
2847         if (r)
2848                 goto out_free_2;
2849
2850         r = sysdev_register(&kvm_sysdev);
2851         if (r)
2852                 goto out_free_3;
2853
2854         kvm_chardev_ops.owner = module;
2855
2856         r = misc_register(&kvm_dev);
2857         if (r) {
2858                 printk (KERN_ERR "kvm: misc device register failed\n");
2859                 goto out_free;
2860         }
2861
2862         return r;
2863
2864 out_free:
2865         sysdev_unregister(&kvm_sysdev);
2866 out_free_3:
2867         sysdev_class_unregister(&kvm_sysdev_class);
2868 out_free_2:
2869         unregister_reboot_notifier(&kvm_reboot_notifier);
2870         unregister_cpu_notifier(&kvm_cpu_notifier);
2871 out_free_1:
2872         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2873         kvm_arch_ops->hardware_unsetup();
2874 out:
2875         kvm_arch_ops = NULL;
2876         return r;
2877 }
2878
2879 void kvm_exit_arch(void)
2880 {
2881         misc_deregister(&kvm_dev);
2882         sysdev_unregister(&kvm_sysdev);
2883         sysdev_class_unregister(&kvm_sysdev_class);
2884         unregister_reboot_notifier(&kvm_reboot_notifier);
2885         unregister_cpu_notifier(&kvm_cpu_notifier);
2886         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2887         kvm_arch_ops->hardware_unsetup();
2888         kvm_arch_ops = NULL;
2889 }
2890
2891 static __init int kvm_init(void)
2892 {
2893         static struct page *bad_page;
2894         int r;
2895
2896         r = register_filesystem(&kvm_fs_type);
2897         if (r)
2898                 goto out3;
2899
2900         kvmfs_mnt = kern_mount(&kvm_fs_type);
2901         r = PTR_ERR(kvmfs_mnt);
2902         if (IS_ERR(kvmfs_mnt))
2903                 goto out2;
2904         kvm_init_debug();
2905
2906         kvm_init_msr_list();
2907
2908         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2909                 r = -ENOMEM;
2910                 goto out;
2911         }
2912
2913         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2914         memset(__va(bad_page_address), 0, PAGE_SIZE);
2915
2916         return 0;
2917
2918 out:
2919         kvm_exit_debug();
2920         mntput(kvmfs_mnt);
2921 out2:
2922         unregister_filesystem(&kvm_fs_type);
2923 out3:
2924         return r;
2925 }
2926
2927 static __exit void kvm_exit(void)
2928 {
2929         kvm_exit_debug();
2930         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2931         mntput(kvmfs_mnt);
2932         unregister_filesystem(&kvm_fs_type);
2933 }
2934
2935 module_init(kvm_init)
2936 module_exit(kvm_exit)
2937
2938 EXPORT_SYMBOL_GPL(kvm_init_arch);
2939 EXPORT_SYMBOL_GPL(kvm_exit_arch);