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 #include <linux/sched.h>
  44
  45 #include "x86_emulate.h"
  46 #include "segment_descriptor.h"
  47
  48 MODULE_AUTHOR("Qumranet");
  49 MODULE_LICENSE("GPL");
  50
  51 static DEFINE_SPINLOCK(kvm_lock);
  52 static LIST_HEAD(vm_list);
  53
  54 struct kvm_arch_ops *kvm_arch_ops;
  55
  56 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  57
  58 static struct kvm_stats_debugfs_item {
  59         const char *name;
  60         int offset;
  61         struct dentry *dentry;
  62 } debugfs_entries[] = {
  63         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  64         { "pf_guest", STAT_OFFSET(pf_guest) },
  65         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  66         { "invlpg", STAT_OFFSET(invlpg) },
  67         { "exits", STAT_OFFSET(exits) },
  68         { "io_exits", STAT_OFFSET(io_exits) },
  69         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  70         { "signal_exits", STAT_OFFSET(signal_exits) },
  71         { "irq_window", STAT_OFFSET(irq_window_exits) },
  72         { "halt_exits", STAT_OFFSET(halt_exits) },
  73         { "request_irq", STAT_OFFSET(request_irq_exits) },
  74         { "irq_exits", STAT_OFFSET(irq_exits) },
  75         { "light_exits", STAT_OFFSET(light_exits) },
  76         { "efer_reload", STAT_OFFSET(efer_reload) },
  77         { NULL }
  78 };
  79
  80 static struct dentry *debugfs_dir;
  81
  82 struct vfsmount *kvmfs_mnt;
  83
  84 #define MAX_IO_MSRS 256
  85
  86 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  87 #define LMSW_GUEST_MASK 0x0eULL
  88 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  89 #define CR8_RESEVED_BITS (~0x0fULL)
  90 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  91
  92 #ifdef CONFIG_X86_64
  93 // LDT or TSS descriptor in the GDT. 16 bytes.
  94 struct segment_descriptor_64 {
  95         struct segment_descriptor s;
  96         u32 base_higher;
  97         u32 pad_zero;
  98 };
  99
 100 #endif
 101
 102 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 103                            unsigned long arg);
 104
 105 static struct inode *kvmfs_inode(struct file_operations *fops)
 106 {
 107         int error = -ENOMEM;
 108         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 109
 110         if (!inode)
 111                 goto eexit_1;
 112
 113         inode->i_fop = fops;
 114
 115         /*
 116          * Mark the inode dirty from the very beginning,
 117          * that way it will never be moved to the dirty
 118          * list because mark_inode_dirty() will think
 119          * that it already _is_ on the dirty list.
 120          */
 121         inode->i_state = I_DIRTY;
 122         inode->i_mode = S_IRUSR | S_IWUSR;
 123         inode->i_uid = current->fsuid;
 124         inode->i_gid = current->fsgid;
 125         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 126         return inode;
 127
 128 eexit_1:
 129         return ERR_PTR(error);
 130 }
 131
 132 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 133 {
 134         struct file *file = get_empty_filp();
 135
 136         if (!file)
 137                 return ERR_PTR(-ENFILE);
 138
 139         file->f_path.mnt = mntget(kvmfs_mnt);
 140         file->f_path.dentry = d_alloc_anon(inode);
 141         if (!file->f_path.dentry)
 142                 return ERR_PTR(-ENOMEM);
 143         file->f_mapping = inode->i_mapping;
 144
 145         file->f_pos = 0;
 146         file->f_flags = O_RDWR;
 147         file->f_op = inode->i_fop;
 148         file->f_mode = FMODE_READ | FMODE_WRITE;
 149         file->f_version = 0;
 150         file->private_data = private_data;
 151         return file;
 152 }
 153
 154 unsigned long segment_base(u16 selector)
 155 {
 156         struct descriptor_table gdt;
 157         struct segment_descriptor *d;
 158         unsigned long table_base;
 159         typedef unsigned long ul;
 160         unsigned long v;
 161
 162         if (selector == 0)
 163                 return 0;
 164
 165         asm ("sgdt %0" : "=m"(gdt));
 166         table_base = gdt.base;
 167
 168         if (selector & 4) {           /* from ldt */
 169                 u16 ldt_selector;
 170
 171                 asm ("sldt %0" : "=g"(ldt_selector));
 172                 table_base = segment_base(ldt_selector);
 173         }
 174         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 175         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 176 #ifdef CONFIG_X86_64
 177         if (d->system == 0
 178             && (d->type == 2 || d->type == 9 || d->type == 11))
 179                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 180 #endif
 181         return v;
 182 }
 183 EXPORT_SYMBOL_GPL(segment_base);
 184
 185 static inline int valid_vcpu(int n)
 186 {
 187         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 188 }
 189
 190 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 191                    void *dest)
 192 {
 193         unsigned char *host_buf = dest;
 194         unsigned long req_size = size;
 195
 196         while (size) {
 197                 hpa_t paddr;
 198                 unsigned now;
 199                 unsigned offset;
 200                 hva_t guest_buf;
 201
 202                 paddr = gva_to_hpa(vcpu, addr);
 203
 204                 if (is_error_hpa(paddr))
 205                         break;
 206
 207                 guest_buf = (hva_t)kmap_atomic(
 208                                         pfn_to_page(paddr >> PAGE_SHIFT),
 209                                         KM_USER0);
 210                 offset = addr & ~PAGE_MASK;
 211                 guest_buf |= offset;
 212                 now = min(size, PAGE_SIZE - offset);
 213                 memcpy(host_buf, (void*)guest_buf, now);
 214                 host_buf += now;
 215                 addr += now;
 216                 size -= now;
 217                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 218         }
 219         return req_size - size;
 220 }
 221 EXPORT_SYMBOL_GPL(kvm_read_guest);
 222
 223 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 224                     void *data)
 225 {
 226         unsigned char *host_buf = data;
 227         unsigned long req_size = size;
 228
 229         while (size) {
 230                 hpa_t paddr;
 231                 unsigned now;
 232                 unsigned offset;
 233                 hva_t guest_buf;
 234                 gfn_t gfn;
 235
 236                 paddr = gva_to_hpa(vcpu, addr);
 237
 238                 if (is_error_hpa(paddr))
 239                         break;
 240
 241                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 242                 mark_page_dirty(vcpu->kvm, gfn);
 243                 guest_buf = (hva_t)kmap_atomic(
 244                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 245                 offset = addr & ~PAGE_MASK;
 246                 guest_buf |= offset;
 247                 now = min(size, PAGE_SIZE - offset);
 248                 memcpy((void*)guest_buf, host_buf, now);
 249                 host_buf += now;
 250                 addr += now;
 251                 size -= now;
 252                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 253         }
 254         return req_size - size;
 255 }
 256 EXPORT_SYMBOL_GPL(kvm_write_guest);
 257
 258 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 259 {
 260         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 261                 return;
 262
 263         vcpu->guest_fpu_loaded = 1;
 264         fx_save(vcpu->host_fx_image);
 265         fx_restore(vcpu->guest_fx_image);
 266 }
 267 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 268
 269 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 270 {
 271         if (!vcpu->guest_fpu_loaded)
 272                 return;
 273
 274         vcpu->guest_fpu_loaded = 0;
 275         fx_save(vcpu->guest_fx_image);
 276         fx_restore(vcpu->host_fx_image);
 277 }
 278 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 279
 280 /*
 281  * Switches to specified vcpu, until a matching vcpu_put()
 282  */
 283 static void vcpu_load(struct kvm_vcpu *vcpu)
 284 {
 285         mutex_lock(&vcpu->mutex);
 286         kvm_arch_ops->vcpu_load(vcpu);
 287 }
 288
 289 /*
 290  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 291  * if the slot is not populated.
 292  */
 293 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 294 {
 295         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 296
 297         mutex_lock(&vcpu->mutex);
 298         if (!vcpu->vmcs) {
 299                 mutex_unlock(&vcpu->mutex);
 300                 return NULL;
 301         }
 302         kvm_arch_ops->vcpu_load(vcpu);
 303         return vcpu;
 304 }
 305
 306 static void vcpu_put(struct kvm_vcpu *vcpu)
 307 {
 308         kvm_arch_ops->vcpu_put(vcpu);
 309         mutex_unlock(&vcpu->mutex);
 310 }
 311
 312 static struct kvm *kvm_create_vm(void)
 313 {
 314         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 315         int i;
 316
 317         if (!kvm)
 318                 return ERR_PTR(-ENOMEM);
 319
 320         spin_lock_init(&kvm->lock);
 321         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 322         spin_lock(&kvm_lock);
 323         list_add(&kvm->vm_list, &vm_list);
 324         spin_unlock(&kvm_lock);
 325         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 326                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 327
 328                 mutex_init(&vcpu->mutex);
 329                 vcpu->cpu = -1;
 330                 vcpu->kvm = kvm;
 331                 vcpu->mmu.root_hpa = INVALID_PAGE;
 332         }
 333         return kvm;
 334 }
 335
 336 static int kvm_dev_open(struct inode *inode, struct file *filp)
 337 {
 338         return 0;
 339 }
 340
 341 /*
 342  * Free any memory in @free but not in @dont.
 343  */
 344 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 345                                   struct kvm_memory_slot *dont)
 346 {
 347         int i;
 348
 349         if (!dont || free->phys_mem != dont->phys_mem)
 350                 if (free->phys_mem) {
 351                         for (i = 0; i < free->npages; ++i)
 352                                 if (free->phys_mem[i])
 353                                         __free_page(free->phys_mem[i]);
 354                         vfree(free->phys_mem);
 355                 }
 356
 357         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 358                 vfree(free->dirty_bitmap);
 359
 360         free->phys_mem = NULL;
 361         free->npages = 0;
 362         free->dirty_bitmap = NULL;
 363 }
 364
 365 static void kvm_free_physmem(struct kvm *kvm)
 366 {
 367         int i;
 368
 369         for (i = 0; i < kvm->nmemslots; ++i)
 370                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 371 }
 372
 373 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 374 {
 375         int i;
 376
 377         for (i = 0; i < 2; ++i)
 378                 if (vcpu->pio.guest_pages[i]) {
 379                         __free_page(vcpu->pio.guest_pages[i]);
 380                         vcpu->pio.guest_pages[i] = NULL;
 381                 }
 382 }
 383
 384 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 385 {
 386         if (!vcpu->vmcs)
 387                 return;
 388
 389         vcpu_load(vcpu);
 390         kvm_mmu_unload(vcpu);
 391         vcpu_put(vcpu);
 392 }
 393
 394 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 395 {
 396         if (!vcpu->vmcs)
 397                 return;
 398
 399         vcpu_load(vcpu);
 400         kvm_mmu_destroy(vcpu);
 401         vcpu_put(vcpu);
 402         kvm_arch_ops->vcpu_free(vcpu);
 403         free_page((unsigned long)vcpu->run);
 404         vcpu->run = NULL;
 405         free_page((unsigned long)vcpu->pio_data);
 406         vcpu->pio_data = NULL;
 407         free_pio_guest_pages(vcpu);
 408 }
 409
 410 static void kvm_free_vcpus(struct kvm *kvm)
 411 {
 412         unsigned int i;
 413
 414         /*
 415          * Unpin any mmu pages first.
 416          */
 417         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 418                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
 419         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 420                 kvm_free_vcpu(&kvm->vcpus[i]);
 421 }
 422
 423 static int kvm_dev_release(struct inode *inode, struct file *filp)
 424 {
 425         return 0;
 426 }
 427
 428 static void kvm_destroy_vm(struct kvm *kvm)
 429 {
 430         spin_lock(&kvm_lock);
 431         list_del(&kvm->vm_list);
 432         spin_unlock(&kvm_lock);
 433         kvm_free_vcpus(kvm);
 434         kvm_free_physmem(kvm);
 435         kfree(kvm);
 436 }
 437
 438 static int kvm_vm_release(struct inode *inode, struct file *filp)
 439 {
 440         struct kvm *kvm = filp->private_data;
 441
 442         kvm_destroy_vm(kvm);
 443         return 0;
 444 }
 445
 446 static void inject_gp(struct kvm_vcpu *vcpu)
 447 {
 448         kvm_arch_ops->inject_gp(vcpu, 0);
 449 }
 450
 451 /*
 452  * Load the pae pdptrs.  Return true is they are all valid.
 453  */
 454 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 455 {
 456         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 457         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 458         int i;
 459         u64 pdpte;
 460         u64 *pdpt;
 461         int ret;
 462         struct page *page;
 463
 464         spin_lock(&vcpu->kvm->lock);
 465         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 466         /* FIXME: !page - emulate? 0xff? */
 467         pdpt = kmap_atomic(page, KM_USER0);
 468
 469         ret = 1;
 470         for (i = 0; i < 4; ++i) {
 471                 pdpte = pdpt[offset + i];
 472                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 473                         ret = 0;
 474                         goto out;
 475                 }
 476         }
 477
 478         for (i = 0; i < 4; ++i)
 479                 vcpu->pdptrs[i] = pdpt[offset + i];
 480
 481 out:
 482         kunmap_atomic(pdpt, KM_USER0);
 483         spin_unlock(&vcpu->kvm->lock);
 484
 485         return ret;
 486 }
 487
 488 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 489 {
 490         if (cr0 & CR0_RESEVED_BITS) {
 491                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 492                        cr0, vcpu->cr0);
 493                 inject_gp(vcpu);
 494                 return;
 495         }
 496
 497         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 498                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 499                 inject_gp(vcpu);
 500                 return;
 501         }
 502
 503         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 504                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 505                        "and a clear PE flag\n");
 506                 inject_gp(vcpu);
 507                 return;
 508         }
 509
 510         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 511 #ifdef CONFIG_X86_64
 512                 if ((vcpu->shadow_efer & EFER_LME)) {
 513                         int cs_db, cs_l;
 514
 515                         if (!is_pae(vcpu)) {
 516                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 517                                        "in long mode while PAE is disabled\n");
 518                                 inject_gp(vcpu);
 519                                 return;
 520                         }
 521                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 522                         if (cs_l) {
 523                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 524                                        "in long mode while CS.L == 1\n");
 525                                 inject_gp(vcpu);
 526                                 return;
 527
 528                         }
 529                 } else
 530 #endif
 531                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 532                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 533                                "reserved bits\n");
 534                         inject_gp(vcpu);
 535                         return;
 536                 }
 537
 538         }
 539
 540         kvm_arch_ops->set_cr0(vcpu, cr0);
 541         vcpu->cr0 = cr0;
 542
 543         spin_lock(&vcpu->kvm->lock);
 544         kvm_mmu_reset_context(vcpu);
 545         spin_unlock(&vcpu->kvm->lock);
 546         return;
 547 }
 548 EXPORT_SYMBOL_GPL(set_cr0);
 549
 550 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 551 {
 552         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 553 }
 554 EXPORT_SYMBOL_GPL(lmsw);
 555
 556 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 557 {
 558         if (cr4 & CR4_RESEVED_BITS) {
 559                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 560                 inject_gp(vcpu);
 561                 return;
 562         }
 563
 564         if (is_long_mode(vcpu)) {
 565                 if (!(cr4 & CR4_PAE_MASK)) {
 566                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 567                                "in long mode\n");
 568                         inject_gp(vcpu);
 569                         return;
 570                 }
 571         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 572                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 573                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 574                 inject_gp(vcpu);
 575         }
 576
 577         if (cr4 & CR4_VMXE_MASK) {
 578                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 579                 inject_gp(vcpu);
 580                 return;
 581         }
 582         kvm_arch_ops->set_cr4(vcpu, cr4);
 583         spin_lock(&vcpu->kvm->lock);
 584         kvm_mmu_reset_context(vcpu);
 585         spin_unlock(&vcpu->kvm->lock);
 586 }
 587 EXPORT_SYMBOL_GPL(set_cr4);
 588
 589 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 590 {
 591         if (is_long_mode(vcpu)) {
 592                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 593                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 594                         inject_gp(vcpu);
 595                         return;
 596                 }
 597         } else {
 598                 if (cr3 & CR3_RESEVED_BITS) {
 599                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 600                         inject_gp(vcpu);
 601                         return;
 602                 }
 603                 if (is_paging(vcpu) && is_pae(vcpu) &&
 604                     !load_pdptrs(vcpu, cr3)) {
 605                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 606                                "reserved bits\n");
 607                         inject_gp(vcpu);
 608                         return;
 609                 }
 610         }
 611
 612         vcpu->cr3 = cr3;
 613         spin_lock(&vcpu->kvm->lock);
 614         /*
 615          * Does the new cr3 value map to physical memory? (Note, we
 616          * catch an invalid cr3 even in real-mode, because it would
 617          * cause trouble later on when we turn on paging anyway.)
 618          *
 619          * A real CPU would silently accept an invalid cr3 and would
 620          * attempt to use it - with largely undefined (and often hard
 621          * to debug) behavior on the guest side.
 622          */
 623         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 624                 inject_gp(vcpu);
 625         else
 626                 vcpu->mmu.new_cr3(vcpu);
 627         spin_unlock(&vcpu->kvm->lock);
 628 }
 629 EXPORT_SYMBOL_GPL(set_cr3);
 630
 631 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 632 {
 633         if ( cr8 & CR8_RESEVED_BITS) {
 634                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 635                 inject_gp(vcpu);
 636                 return;
 637         }
 638         vcpu->cr8 = cr8;
 639 }
 640 EXPORT_SYMBOL_GPL(set_cr8);
 641
 642 void fx_init(struct kvm_vcpu *vcpu)
 643 {
 644         struct __attribute__ ((__packed__)) fx_image_s {
 645                 u16 control; //fcw
 646                 u16 status; //fsw
 647                 u16 tag; // ftw
 648                 u16 opcode; //fop
 649                 u64 ip; // fpu ip
 650                 u64 operand;// fpu dp
 651                 u32 mxcsr;
 652                 u32 mxcsr_mask;
 653
 654         } *fx_image;
 655
 656         fx_save(vcpu->host_fx_image);
 657         fpu_init();
 658         fx_save(vcpu->guest_fx_image);
 659         fx_restore(vcpu->host_fx_image);
 660
 661         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 662         fx_image->mxcsr = 0x1f80;
 663         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 664                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 665 }
 666 EXPORT_SYMBOL_GPL(fx_init);
 667
 668 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 669 {
 670         spin_lock(&vcpu->kvm->lock);
 671         kvm_mmu_slot_remove_write_access(vcpu, slot);
 672         spin_unlock(&vcpu->kvm->lock);
 673 }
 674
 675 /*
 676  * Allocate some memory and give it an address in the guest physical address
 677  * space.
 678  *
 679  * Discontiguous memory is allowed, mostly for framebuffers.
 680  */
 681 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 682                                           struct kvm_memory_region *mem)
 683 {
 684         int r;
 685         gfn_t base_gfn;
 686         unsigned long npages;
 687         unsigned long i;
 688         struct kvm_memory_slot *memslot;
 689         struct kvm_memory_slot old, new;
 690         int memory_config_version;
 691
 692         r = -EINVAL;
 693         /* General sanity checks */
 694         if (mem->memory_size & (PAGE_SIZE - 1))
 695                 goto out;
 696         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 697                 goto out;
 698         if (mem->slot >= KVM_MEMORY_SLOTS)
 699                 goto out;
 700         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 701                 goto out;
 702
 703         memslot = &kvm->memslots[mem->slot];
 704         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 705         npages = mem->memory_size >> PAGE_SHIFT;
 706
 707         if (!npages)
 708                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 709
 710 raced:
 711         spin_lock(&kvm->lock);
 712
 713         memory_config_version = kvm->memory_config_version;
 714         new = old = *memslot;
 715
 716         new.base_gfn = base_gfn;
 717         new.npages = npages;
 718         new.flags = mem->flags;
 719
 720         /* Disallow changing a memory slot's size. */
 721         r = -EINVAL;
 722         if (npages && old.npages && npages != old.npages)
 723                 goto out_unlock;
 724
 725         /* Check for overlaps */
 726         r = -EEXIST;
 727         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 728                 struct kvm_memory_slot *s = &kvm->memslots[i];
 729
 730                 if (s == memslot)
 731                         continue;
 732                 if (!((base_gfn + npages <= s->base_gfn) ||
 733                       (base_gfn >= s->base_gfn + s->npages)))
 734                         goto out_unlock;
 735         }
 736         /*
 737          * Do memory allocations outside lock.  memory_config_version will
 738          * detect any races.
 739          */
 740         spin_unlock(&kvm->lock);
 741
 742         /* Deallocate if slot is being removed */
 743         if (!npages)
 744                 new.phys_mem = NULL;
 745
 746         /* Free page dirty bitmap if unneeded */
 747         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 748                 new.dirty_bitmap = NULL;
 749
 750         r = -ENOMEM;
 751
 752         /* Allocate if a slot is being created */
 753         if (npages && !new.phys_mem) {
 754                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 755
 756                 if (!new.phys_mem)
 757                         goto out_free;
 758
 759                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 760                 for (i = 0; i < npages; ++i) {
 761                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 762                                                      | __GFP_ZERO);
 763                         if (!new.phys_mem[i])
 764                                 goto out_free;
 765                         set_page_private(new.phys_mem[i],0);
 766                 }
 767         }
 768
 769         /* Allocate page dirty bitmap if needed */
 770         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 771                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 772
 773                 new.dirty_bitmap = vmalloc(dirty_bytes);
 774                 if (!new.dirty_bitmap)
 775                         goto out_free;
 776                 memset(new.dirty_bitmap, 0, dirty_bytes);
 777         }
 778
 779         spin_lock(&kvm->lock);
 780
 781         if (memory_config_version != kvm->memory_config_version) {
 782                 spin_unlock(&kvm->lock);
 783                 kvm_free_physmem_slot(&new, &old);
 784                 goto raced;
 785         }
 786
 787         r = -EAGAIN;
 788         if (kvm->busy)
 789                 goto out_unlock;
 790
 791         if (mem->slot >= kvm->nmemslots)
 792                 kvm->nmemslots = mem->slot + 1;
 793
 794         *memslot = new;
 795         ++kvm->memory_config_version;
 796
 797         spin_unlock(&kvm->lock);
 798
 799         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 800                 struct kvm_vcpu *vcpu;
 801
 802                 vcpu = vcpu_load_slot(kvm, i);
 803                 if (!vcpu)
 804                         continue;
 805                 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
 806                         do_remove_write_access(vcpu, mem->slot);
 807                 kvm_mmu_reset_context(vcpu);
 808                 vcpu_put(vcpu);
 809         }
 810
 811         kvm_free_physmem_slot(&old, &new);
 812         return 0;
 813
 814 out_unlock:
 815         spin_unlock(&kvm->lock);
 816 out_free:
 817         kvm_free_physmem_slot(&new, &old);
 818 out:
 819         return r;
 820 }
 821
 822 /*
 823  * Get (and clear) the dirty memory log for a memory slot.
 824  */
 825 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 826                                       struct kvm_dirty_log *log)
 827 {
 828         struct kvm_memory_slot *memslot;
 829         int r, i;
 830         int n;
 831         int cleared;
 832         unsigned long any = 0;
 833
 834         spin_lock(&kvm->lock);
 835
 836         /*
 837          * Prevent changes to guest memory configuration even while the lock
 838          * is not taken.
 839          */
 840         ++kvm->busy;
 841         spin_unlock(&kvm->lock);
 842         r = -EINVAL;
 843         if (log->slot >= KVM_MEMORY_SLOTS)
 844                 goto out;
 845
 846         memslot = &kvm->memslots[log->slot];
 847         r = -ENOENT;
 848         if (!memslot->dirty_bitmap)
 849                 goto out;
 850
 851         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 852
 853         for (i = 0; !any && i < n/sizeof(long); ++i)
 854                 any = memslot->dirty_bitmap[i];
 855
 856         r = -EFAULT;
 857         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 858                 goto out;
 859
 860         if (any) {
 861                 cleared = 0;
 862                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 863                         struct kvm_vcpu *vcpu;
 864
 865                         vcpu = vcpu_load_slot(kvm, i);
 866                         if (!vcpu)
 867                                 continue;
 868                         if (!cleared) {
 869                                 do_remove_write_access(vcpu, log->slot);
 870                                 memset(memslot->dirty_bitmap, 0, n);
 871                                 cleared = 1;
 872                         }
 873                         kvm_arch_ops->tlb_flush(vcpu);
 874                         vcpu_put(vcpu);
 875                 }
 876         }
 877
 878         r = 0;
 879
 880 out:
 881         spin_lock(&kvm->lock);
 882         --kvm->busy;
 883         spin_unlock(&kvm->lock);
 884         return r;
 885 }
 886
 887 /*
 888  * Set a new alias region.  Aliases map a portion of physical memory into
 889  * another portion.  This is useful for memory windows, for example the PC
 890  * VGA region.
 891  */
 892 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 893                                          struct kvm_memory_alias *alias)
 894 {
 895         int r, n;
 896         struct kvm_mem_alias *p;
 897
 898         r = -EINVAL;
 899         /* General sanity checks */
 900         if (alias->memory_size & (PAGE_SIZE - 1))
 901                 goto out;
 902         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 903                 goto out;
 904         if (alias->slot >= KVM_ALIAS_SLOTS)
 905                 goto out;
 906         if (alias->guest_phys_addr + alias->memory_size
 907             < alias->guest_phys_addr)
 908                 goto out;
 909         if (alias->target_phys_addr + alias->memory_size
 910             < alias->target_phys_addr)
 911                 goto out;
 912
 913         spin_lock(&kvm->lock);
 914
 915         p = &kvm->aliases[alias->slot];
 916         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 917         p->npages = alias->memory_size >> PAGE_SHIFT;
 918         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 919
 920         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 921                 if (kvm->aliases[n - 1].npages)
 922                         break;
 923         kvm->naliases = n;
 924
 925         spin_unlock(&kvm->lock);
 926
 927         vcpu_load(&kvm->vcpus[0]);
 928         spin_lock(&kvm->lock);
 929         kvm_mmu_zap_all(&kvm->vcpus[0]);
 930         spin_unlock(&kvm->lock);
 931         vcpu_put(&kvm->vcpus[0]);
 932
 933         return 0;
 934
 935 out:
 936         return r;
 937 }
 938
 939 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 940 {
 941         int i;
 942         struct kvm_mem_alias *alias;
 943
 944         for (i = 0; i < kvm->naliases; ++i) {
 945                 alias = &kvm->aliases[i];
 946                 if (gfn >= alias->base_gfn
 947                     && gfn < alias->base_gfn + alias->npages)
 948                         return alias->target_gfn + gfn - alias->base_gfn;
 949         }
 950         return gfn;
 951 }
 952
 953 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 954 {
 955         int i;
 956
 957         for (i = 0; i < kvm->nmemslots; ++i) {
 958                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 959
 960                 if (gfn >= memslot->base_gfn
 961                     && gfn < memslot->base_gfn + memslot->npages)
 962                         return memslot;
 963         }
 964         return NULL;
 965 }
 966
 967 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 968 {
 969         gfn = unalias_gfn(kvm, gfn);
 970         return __gfn_to_memslot(kvm, gfn);
 971 }
 972
 973 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 974 {
 975         struct kvm_memory_slot *slot;
 976
 977         gfn = unalias_gfn(kvm, gfn);
 978         slot = __gfn_to_memslot(kvm, gfn);
 979         if (!slot)
 980                 return NULL;
 981         return slot->phys_mem[gfn - slot->base_gfn];
 982 }
 983 EXPORT_SYMBOL_GPL(gfn_to_page);
 984
 985 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 986 {
 987         int i;
 988         struct kvm_memory_slot *memslot;
 989         unsigned long rel_gfn;
 990
 991         for (i = 0; i < kvm->nmemslots; ++i) {
 992                 memslot = &kvm->memslots[i];
 993
 994                 if (gfn >= memslot->base_gfn
 995                     && gfn < memslot->base_gfn + memslot->npages) {
 996
 997                         if (!memslot->dirty_bitmap)
 998                                 return;
 999
1000                         rel_gfn = gfn - memslot->base_gfn;
1001
1002                         /* avoid RMW */
1003                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1004                                 set_bit(rel_gfn, memslot->dirty_bitmap);
1005                         return;
1006                 }
1007         }
1008 }
1009
1010 static int emulator_read_std(unsigned long addr,
1011                              void *val,
1012                              unsigned int bytes,
1013                              struct x86_emulate_ctxt *ctxt)
1014 {
1015         struct kvm_vcpu *vcpu = ctxt->vcpu;
1016         void *data = val;
1017
1018         while (bytes) {
1019                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1020                 unsigned offset = addr & (PAGE_SIZE-1);
1021                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1022                 unsigned long pfn;
1023                 struct page *page;
1024                 void *page_virt;
1025
1026                 if (gpa == UNMAPPED_GVA)
1027                         return X86EMUL_PROPAGATE_FAULT;
1028                 pfn = gpa >> PAGE_SHIFT;
1029                 page = gfn_to_page(vcpu->kvm, pfn);
1030                 if (!page)
1031                         return X86EMUL_UNHANDLEABLE;
1032                 page_virt = kmap_atomic(page, KM_USER0);
1033
1034                 memcpy(data, page_virt + offset, tocopy);
1035
1036                 kunmap_atomic(page_virt, KM_USER0);
1037
1038                 bytes -= tocopy;
1039                 data += tocopy;
1040                 addr += tocopy;
1041         }
1042
1043         return X86EMUL_CONTINUE;
1044 }
1045
1046 static int emulator_write_std(unsigned long addr,
1047                               const void *val,
1048                               unsigned int bytes,
1049                               struct x86_emulate_ctxt *ctxt)
1050 {
1051         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1052                addr, bytes);
1053         return X86EMUL_UNHANDLEABLE;
1054 }
1055
1056 static int emulator_read_emulated(unsigned long addr,
1057                                   void *val,
1058                                   unsigned int bytes,
1059                                   struct x86_emulate_ctxt *ctxt)
1060 {
1061         struct kvm_vcpu *vcpu = ctxt->vcpu;
1062
1063         if (vcpu->mmio_read_completed) {
1064                 memcpy(val, vcpu->mmio_data, bytes);
1065                 vcpu->mmio_read_completed = 0;
1066                 return X86EMUL_CONTINUE;
1067         } else if (emulator_read_std(addr, val, bytes, ctxt)
1068                    == X86EMUL_CONTINUE)
1069                 return X86EMUL_CONTINUE;
1070         else {
1071                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1072
1073                 if (gpa == UNMAPPED_GVA)
1074                         return X86EMUL_PROPAGATE_FAULT;
1075                 vcpu->mmio_needed = 1;
1076                 vcpu->mmio_phys_addr = gpa;
1077                 vcpu->mmio_size = bytes;
1078                 vcpu->mmio_is_write = 0;
1079
1080                 return X86EMUL_UNHANDLEABLE;
1081         }
1082 }
1083
1084 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1085                                const void *val, int bytes)
1086 {
1087         struct page *page;
1088         void *virt;
1089         unsigned offset = offset_in_page(gpa);
1090
1091         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1092                 return 0;
1093         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1094         if (!page)
1095                 return 0;
1096         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1097         virt = kmap_atomic(page, KM_USER0);
1098         kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1099         memcpy(virt + offset_in_page(gpa), val, bytes);
1100         kunmap_atomic(virt, KM_USER0);
1101         return 1;
1102 }
1103
1104 static int emulator_write_emulated(unsigned long addr,
1105                                    const void *val,
1106                                    unsigned int bytes,
1107                                    struct x86_emulate_ctxt *ctxt)
1108 {
1109         struct kvm_vcpu *vcpu = ctxt->vcpu;
1110         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1111
1112         if (gpa == UNMAPPED_GVA) {
1113                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1114                 return X86EMUL_PROPAGATE_FAULT;
1115         }
1116
1117         if (emulator_write_phys(vcpu, gpa, val, bytes))
1118                 return X86EMUL_CONTINUE;
1119
1120         vcpu->mmio_needed = 1;
1121         vcpu->mmio_phys_addr = gpa;
1122         vcpu->mmio_size = bytes;
1123         vcpu->mmio_is_write = 1;
1124         memcpy(vcpu->mmio_data, val, bytes);
1125
1126         return X86EMUL_CONTINUE;
1127 }
1128
1129 static int emulator_cmpxchg_emulated(unsigned long addr,
1130                                      const void *old,
1131                                      const void *new,
1132                                      unsigned int bytes,
1133                                      struct x86_emulate_ctxt *ctxt)
1134 {
1135         static int reported;
1136
1137         if (!reported) {
1138                 reported = 1;
1139                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1140         }
1141         return emulator_write_emulated(addr, new, bytes, ctxt);
1142 }
1143
1144 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1145 {
1146         return kvm_arch_ops->get_segment_base(vcpu, seg);
1147 }
1148
1149 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1150 {
1151         return X86EMUL_CONTINUE;
1152 }
1153
1154 int emulate_clts(struct kvm_vcpu *vcpu)
1155 {
1156         unsigned long cr0;
1157
1158         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1159         kvm_arch_ops->set_cr0(vcpu, cr0);
1160         return X86EMUL_CONTINUE;
1161 }
1162
1163 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1164 {
1165         struct kvm_vcpu *vcpu = ctxt->vcpu;
1166
1167         switch (dr) {
1168         case 0 ... 3:
1169                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1170                 return X86EMUL_CONTINUE;
1171         default:
1172                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1173                        __FUNCTION__, dr);
1174                 return X86EMUL_UNHANDLEABLE;
1175         }
1176 }
1177
1178 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1179 {
1180         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1181         int exception;
1182
1183         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1184         if (exception) {
1185                 /* FIXME: better handling */
1186                 return X86EMUL_UNHANDLEABLE;
1187         }
1188         return X86EMUL_CONTINUE;
1189 }
1190
1191 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1192 {
1193         static int reported;
1194         u8 opcodes[4];
1195         unsigned long rip = ctxt->vcpu->rip;
1196         unsigned long rip_linear;
1197
1198         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1199
1200         if (reported)
1201                 return;
1202
1203         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1204
1205         printk(KERN_ERR "emulation failed but !mmio_needed?"
1206                " rip %lx %02x %02x %02x %02x\n",
1207                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1208         reported = 1;
1209 }
1210
1211 struct x86_emulate_ops emulate_ops = {
1212         .read_std            = emulator_read_std,
1213         .write_std           = emulator_write_std,
1214         .read_emulated       = emulator_read_emulated,
1215         .write_emulated      = emulator_write_emulated,
1216         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1217 };
1218
1219 int emulate_instruction(struct kvm_vcpu *vcpu,
1220                         struct kvm_run *run,
1221                         unsigned long cr2,
1222                         u16 error_code)
1223 {
1224         struct x86_emulate_ctxt emulate_ctxt;
1225         int r;
1226         int cs_db, cs_l;
1227
1228         vcpu->mmio_fault_cr2 = cr2;
1229         kvm_arch_ops->cache_regs(vcpu);
1230
1231         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1232
1233         emulate_ctxt.vcpu = vcpu;
1234         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1235         emulate_ctxt.cr2 = cr2;
1236         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1237                 ? X86EMUL_MODE_REAL : cs_l
1238                 ? X86EMUL_MODE_PROT64 : cs_db
1239                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1240
1241         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1242                 emulate_ctxt.cs_base = 0;
1243                 emulate_ctxt.ds_base = 0;
1244                 emulate_ctxt.es_base = 0;
1245                 emulate_ctxt.ss_base = 0;
1246         } else {
1247                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1248                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1249                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1250                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1251         }
1252
1253         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1254         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1255
1256         vcpu->mmio_is_write = 0;
1257         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1258
1259         if ((r || vcpu->mmio_is_write) && run) {
1260                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1261                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1262                 run->mmio.len = vcpu->mmio_size;
1263                 run->mmio.is_write = vcpu->mmio_is_write;
1264         }
1265
1266         if (r) {
1267                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1268                         return EMULATE_DONE;
1269                 if (!vcpu->mmio_needed) {
1270                         report_emulation_failure(&emulate_ctxt);
1271                         return EMULATE_FAIL;
1272                 }
1273                 return EMULATE_DO_MMIO;
1274         }
1275
1276         kvm_arch_ops->decache_regs(vcpu);
1277         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1278
1279         if (vcpu->mmio_is_write) {
1280                 vcpu->mmio_needed = 0;
1281                 return EMULATE_DO_MMIO;
1282         }
1283
1284         return EMULATE_DONE;
1285 }
1286 EXPORT_SYMBOL_GPL(emulate_instruction);
1287
1288 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1289 {
1290         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1291
1292         kvm_arch_ops->cache_regs(vcpu);
1293         ret = -KVM_EINVAL;
1294 #ifdef CONFIG_X86_64
1295         if (is_long_mode(vcpu)) {
1296                 nr = vcpu->regs[VCPU_REGS_RAX];
1297                 a0 = vcpu->regs[VCPU_REGS_RDI];
1298                 a1 = vcpu->regs[VCPU_REGS_RSI];
1299                 a2 = vcpu->regs[VCPU_REGS_RDX];
1300                 a3 = vcpu->regs[VCPU_REGS_RCX];
1301                 a4 = vcpu->regs[VCPU_REGS_R8];
1302                 a5 = vcpu->regs[VCPU_REGS_R9];
1303         } else
1304 #endif
1305         {
1306                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1307                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1308                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1309                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1310                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1311                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1312                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1313         }
1314         switch (nr) {
1315         default:
1316                 run->hypercall.args[0] = a0;
1317                 run->hypercall.args[1] = a1;
1318                 run->hypercall.args[2] = a2;
1319                 run->hypercall.args[3] = a3;
1320                 run->hypercall.args[4] = a4;
1321                 run->hypercall.args[5] = a5;
1322                 run->hypercall.ret = ret;
1323                 run->hypercall.longmode = is_long_mode(vcpu);
1324                 kvm_arch_ops->decache_regs(vcpu);
1325                 return 0;
1326         }
1327         vcpu->regs[VCPU_REGS_RAX] = ret;
1328         kvm_arch_ops->decache_regs(vcpu);
1329         return 1;
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_hypercall);
1332
1333 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1334 {
1335         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1336 }
1337
1338 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1339 {
1340         struct descriptor_table dt = { limit, base };
1341
1342         kvm_arch_ops->set_gdt(vcpu, &dt);
1343 }
1344
1345 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1346 {
1347         struct descriptor_table dt = { limit, base };
1348
1349         kvm_arch_ops->set_idt(vcpu, &dt);
1350 }
1351
1352 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1353                    unsigned long *rflags)
1354 {
1355         lmsw(vcpu, msw);
1356         *rflags = kvm_arch_ops->get_rflags(vcpu);
1357 }
1358
1359 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1360 {
1361         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1362         switch (cr) {
1363         case 0:
1364                 return vcpu->cr0;
1365         case 2:
1366                 return vcpu->cr2;
1367         case 3:
1368                 return vcpu->cr3;
1369         case 4:
1370                 return vcpu->cr4;
1371         default:
1372                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1373                 return 0;
1374         }
1375 }
1376
1377 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1378                      unsigned long *rflags)
1379 {
1380         switch (cr) {
1381         case 0:
1382                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1383                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1384                 break;
1385         case 2:
1386                 vcpu->cr2 = val;
1387                 break;
1388         case 3:
1389                 set_cr3(vcpu, val);
1390                 break;
1391         case 4:
1392                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1393                 break;
1394         default:
1395                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1396         }
1397 }
1398
1399 /*
1400  * Register the para guest with the host:
1401  */
1402 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1403 {
1404         struct kvm_vcpu_para_state *para_state;
1405         hpa_t para_state_hpa, hypercall_hpa;
1406         struct page *para_state_page;
1407         unsigned char *hypercall;
1408         gpa_t hypercall_gpa;
1409
1410         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1411         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1412
1413         /*
1414          * Needs to be page aligned:
1415          */
1416         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1417                 goto err_gp;
1418
1419         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1420         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1421         if (is_error_hpa(para_state_hpa))
1422                 goto err_gp;
1423
1424         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1425         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1426         para_state = kmap_atomic(para_state_page, KM_USER0);
1427
1428         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1429         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1430
1431         para_state->host_version = KVM_PARA_API_VERSION;
1432         /*
1433          * We cannot support guests that try to register themselves
1434          * with a newer API version than the host supports:
1435          */
1436         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1437                 para_state->ret = -KVM_EINVAL;
1438                 goto err_kunmap_skip;
1439         }
1440
1441         hypercall_gpa = para_state->hypercall_gpa;
1442         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1443         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1444         if (is_error_hpa(hypercall_hpa)) {
1445                 para_state->ret = -KVM_EINVAL;
1446                 goto err_kunmap_skip;
1447         }
1448
1449         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1450         vcpu->para_state_page = para_state_page;
1451         vcpu->para_state_gpa = para_state_gpa;
1452         vcpu->hypercall_gpa = hypercall_gpa;
1453
1454         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1455         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1456                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1457         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1458         kunmap_atomic(hypercall, KM_USER1);
1459
1460         para_state->ret = 0;
1461 err_kunmap_skip:
1462         kunmap_atomic(para_state, KM_USER0);
1463         return 0;
1464 err_gp:
1465         return 1;
1466 }
1467
1468 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1469 {
1470         u64 data;
1471
1472         switch (msr) {
1473         case 0xc0010010: /* SYSCFG */
1474         case 0xc0010015: /* HWCR */
1475         case MSR_IA32_PLATFORM_ID:
1476         case MSR_IA32_P5_MC_ADDR:
1477         case MSR_IA32_P5_MC_TYPE:
1478         case MSR_IA32_MC0_CTL:
1479         case MSR_IA32_MCG_STATUS:
1480         case MSR_IA32_MCG_CAP:
1481         case MSR_IA32_MC0_MISC:
1482         case MSR_IA32_MC0_MISC+4:
1483         case MSR_IA32_MC0_MISC+8:
1484         case MSR_IA32_MC0_MISC+12:
1485         case MSR_IA32_MC0_MISC+16:
1486         case MSR_IA32_UCODE_REV:
1487         case MSR_IA32_PERF_STATUS:
1488         case MSR_IA32_EBL_CR_POWERON:
1489                 /* MTRR registers */
1490         case 0xfe:
1491         case 0x200 ... 0x2ff:
1492                 data = 0;
1493                 break;
1494         case 0xcd: /* fsb frequency */
1495                 data = 3;
1496                 break;
1497         case MSR_IA32_APICBASE:
1498                 data = vcpu->apic_base;
1499                 break;
1500         case MSR_IA32_MISC_ENABLE:
1501                 data = vcpu->ia32_misc_enable_msr;
1502                 break;
1503 #ifdef CONFIG_X86_64
1504         case MSR_EFER:
1505                 data = vcpu->shadow_efer;
1506                 break;
1507 #endif
1508         default:
1509                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1510                 return 1;
1511         }
1512         *pdata = data;
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1516
1517 /*
1518  * Reads an msr value (of 'msr_index') into 'pdata'.
1519  * Returns 0 on success, non-0 otherwise.
1520  * Assumes vcpu_load() was already called.
1521  */
1522 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1523 {
1524         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1525 }
1526
1527 #ifdef CONFIG_X86_64
1528
1529 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1530 {
1531         if (efer & EFER_RESERVED_BITS) {
1532                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1533                        efer);
1534                 inject_gp(vcpu);
1535                 return;
1536         }
1537
1538         if (is_paging(vcpu)
1539             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1540                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1541                 inject_gp(vcpu);
1542                 return;
1543         }
1544
1545         kvm_arch_ops->set_efer(vcpu, efer);
1546
1547         efer &= ~EFER_LMA;
1548         efer |= vcpu->shadow_efer & EFER_LMA;
1549
1550         vcpu->shadow_efer = efer;
1551 }
1552
1553 #endif
1554
1555 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1556 {
1557         switch (msr) {
1558 #ifdef CONFIG_X86_64
1559         case MSR_EFER:
1560                 set_efer(vcpu, data);
1561                 break;
1562 #endif
1563         case MSR_IA32_MC0_STATUS:
1564                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1565                        __FUNCTION__, data);
1566                 break;
1567         case MSR_IA32_MCG_STATUS:
1568                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1569                         __FUNCTION__, data);
1570                 break;
1571         case MSR_IA32_UCODE_REV:
1572         case MSR_IA32_UCODE_WRITE:
1573         case 0x200 ... 0x2ff: /* MTRRs */
1574                 break;
1575         case MSR_IA32_APICBASE:
1576                 vcpu->apic_base = data;
1577                 break;
1578         case MSR_IA32_MISC_ENABLE:
1579                 vcpu->ia32_misc_enable_msr = data;
1580                 break;
1581         /*
1582          * This is the 'probe whether the host is KVM' logic:
1583          */
1584         case MSR_KVM_API_MAGIC:
1585                 return vcpu_register_para(vcpu, data);
1586
1587         default:
1588                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1589                 return 1;
1590         }
1591         return 0;
1592 }
1593 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1594
1595 /*
1596  * Writes msr value into into the appropriate "register".
1597  * Returns 0 on success, non-0 otherwise.
1598  * Assumes vcpu_load() was already called.
1599  */
1600 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1601 {
1602         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1603 }
1604
1605 void kvm_resched(struct kvm_vcpu *vcpu)
1606 {
1607         if (!need_resched())
1608                 return;
1609         vcpu_put(vcpu);
1610         cond_resched();
1611         vcpu_load(vcpu);
1612 }
1613 EXPORT_SYMBOL_GPL(kvm_resched);
1614
1615 void load_msrs(struct vmx_msr_entry *e, int n)
1616 {
1617         int i;
1618
1619         for (i = 0; i < n; ++i)
1620                 wrmsrl(e[i].index, e[i].data);
1621 }
1622 EXPORT_SYMBOL_GPL(load_msrs);
1623
1624 void save_msrs(struct vmx_msr_entry *e, int n)
1625 {
1626         int i;
1627
1628         for (i = 0; i < n; ++i)
1629                 rdmsrl(e[i].index, e[i].data);
1630 }
1631 EXPORT_SYMBOL_GPL(save_msrs);
1632
1633 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1634 {
1635         int i;
1636         u32 function;
1637         struct kvm_cpuid_entry *e, *best;
1638
1639         kvm_arch_ops->cache_regs(vcpu);
1640         function = vcpu->regs[VCPU_REGS_RAX];
1641         vcpu->regs[VCPU_REGS_RAX] = 0;
1642         vcpu->regs[VCPU_REGS_RBX] = 0;
1643         vcpu->regs[VCPU_REGS_RCX] = 0;
1644         vcpu->regs[VCPU_REGS_RDX] = 0;
1645         best = NULL;
1646         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1647                 e = &vcpu->cpuid_entries[i];
1648                 if (e->function == function) {
1649                         best = e;
1650                         break;
1651                 }
1652                 /*
1653                  * Both basic or both extended?
1654                  */
1655                 if (((e->function ^ function) & 0x80000000) == 0)
1656                         if (!best || e->function > best->function)
1657                                 best = e;
1658         }
1659         if (best) {
1660                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1661                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1662                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1663                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1664         }
1665         kvm_arch_ops->decache_regs(vcpu);
1666         kvm_arch_ops->skip_emulated_instruction(vcpu);
1667 }
1668 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1669
1670 static int pio_copy_data(struct kvm_vcpu *vcpu)
1671 {
1672         void *p = vcpu->pio_data;
1673         void *q;
1674         unsigned bytes;
1675         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1676
1677         kvm_arch_ops->vcpu_put(vcpu);
1678         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1679                  PAGE_KERNEL);
1680         if (!q) {
1681                 kvm_arch_ops->vcpu_load(vcpu);
1682                 free_pio_guest_pages(vcpu);
1683                 return -ENOMEM;
1684         }
1685         q += vcpu->pio.guest_page_offset;
1686         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1687         if (vcpu->pio.in)
1688                 memcpy(q, p, bytes);
1689         else
1690                 memcpy(p, q, bytes);
1691         q -= vcpu->pio.guest_page_offset;
1692         vunmap(q);
1693         kvm_arch_ops->vcpu_load(vcpu);
1694         free_pio_guest_pages(vcpu);
1695         return 0;
1696 }
1697
1698 static int complete_pio(struct kvm_vcpu *vcpu)
1699 {
1700         struct kvm_pio_request *io = &vcpu->pio;
1701         long delta;
1702         int r;
1703
1704         kvm_arch_ops->cache_regs(vcpu);
1705
1706         if (!io->string) {
1707                 if (io->in)
1708                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1709                                io->size);
1710         } else {
1711                 if (io->in) {
1712                         r = pio_copy_data(vcpu);
1713                         if (r) {
1714                                 kvm_arch_ops->cache_regs(vcpu);
1715                                 return r;
1716                         }
1717                 }
1718
1719                 delta = 1;
1720                 if (io->rep) {
1721                         delta *= io->cur_count;
1722                         /*
1723                          * The size of the register should really depend on
1724                          * current address size.
1725                          */
1726                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1727                 }
1728                 if (io->down)
1729                         delta = -delta;
1730                 delta *= io->size;
1731                 if (io->in)
1732                         vcpu->regs[VCPU_REGS_RDI] += delta;
1733                 else
1734                         vcpu->regs[VCPU_REGS_RSI] += delta;
1735         }
1736
1737         kvm_arch_ops->decache_regs(vcpu);
1738
1739         io->count -= io->cur_count;
1740         io->cur_count = 0;
1741
1742         if (!io->count)
1743                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1744         return 0;
1745 }
1746
1747 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1748                   int size, unsigned long count, int string, int down,
1749                   gva_t address, int rep, unsigned port)
1750 {
1751         unsigned now, in_page;
1752         int i;
1753         int nr_pages = 1;
1754         struct page *page;
1755
1756         vcpu->run->exit_reason = KVM_EXIT_IO;
1757         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1758         vcpu->run->io.size = size;
1759         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1760         vcpu->run->io.count = count;
1761         vcpu->run->io.port = port;
1762         vcpu->pio.count = count;
1763         vcpu->pio.cur_count = count;
1764         vcpu->pio.size = size;
1765         vcpu->pio.in = in;
1766         vcpu->pio.string = string;
1767         vcpu->pio.down = down;
1768         vcpu->pio.guest_page_offset = offset_in_page(address);
1769         vcpu->pio.rep = rep;
1770
1771         if (!string) {
1772                 kvm_arch_ops->cache_regs(vcpu);
1773                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1774                 kvm_arch_ops->decache_regs(vcpu);
1775                 return 0;
1776         }
1777
1778         if (!count) {
1779                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1780                 return 1;
1781         }
1782
1783         now = min(count, PAGE_SIZE / size);
1784
1785         if (!down)
1786                 in_page = PAGE_SIZE - offset_in_page(address);
1787         else
1788                 in_page = offset_in_page(address) + size;
1789         now = min(count, (unsigned long)in_page / size);
1790         if (!now) {
1791                 /*
1792                  * String I/O straddles page boundary.  Pin two guest pages
1793                  * so that we satisfy atomicity constraints.  Do just one
1794                  * transaction to avoid complexity.
1795                  */
1796                 nr_pages = 2;
1797                 now = 1;
1798         }
1799         if (down) {
1800                 /*
1801                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1802                  */
1803                 printk(KERN_ERR "kvm: guest string pio down\n");
1804                 inject_gp(vcpu);
1805                 return 1;
1806         }
1807         vcpu->run->io.count = now;
1808         vcpu->pio.cur_count = now;
1809
1810         for (i = 0; i < nr_pages; ++i) {
1811                 spin_lock(&vcpu->kvm->lock);
1812                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1813                 if (page)
1814                         get_page(page);
1815                 vcpu->pio.guest_pages[i] = page;
1816                 spin_unlock(&vcpu->kvm->lock);
1817                 if (!page) {
1818                         inject_gp(vcpu);
1819                         free_pio_guest_pages(vcpu);
1820                         return 1;
1821                 }
1822         }
1823
1824         if (!vcpu->pio.in)
1825                 return pio_copy_data(vcpu);
1826         return 0;
1827 }
1828 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1829
1830 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1831 {
1832         int r;
1833         sigset_t sigsaved;
1834
1835         vcpu_load(vcpu);
1836
1837         if (vcpu->sigset_active)
1838                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1839
1840         /* re-sync apic's tpr */
1841         vcpu->cr8 = kvm_run->cr8;
1842
1843         if (vcpu->pio.cur_count) {
1844                 r = complete_pio(vcpu);
1845                 if (r)
1846                         goto out;
1847         }
1848
1849         if (vcpu->mmio_needed) {
1850                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1851                 vcpu->mmio_read_completed = 1;
1852                 vcpu->mmio_needed = 0;
1853                 r = emulate_instruction(vcpu, kvm_run,
1854                                         vcpu->mmio_fault_cr2, 0);
1855                 if (r == EMULATE_DO_MMIO) {
1856                         /*
1857                          * Read-modify-write.  Back to userspace.
1858                          */
1859                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1860                         r = 0;
1861                         goto out;
1862                 }
1863         }
1864
1865         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1866                 kvm_arch_ops->cache_regs(vcpu);
1867                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1868                 kvm_arch_ops->decache_regs(vcpu);
1869         }
1870
1871         r = kvm_arch_ops->run(vcpu, kvm_run);
1872
1873 out:
1874         if (vcpu->sigset_active)
1875                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1876
1877         vcpu_put(vcpu);
1878         return r;
1879 }
1880
1881 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1882                                    struct kvm_regs *regs)
1883 {
1884         vcpu_load(vcpu);
1885
1886         kvm_arch_ops->cache_regs(vcpu);
1887
1888         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1889         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1890         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1891         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1892         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1893         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1894         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1895         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1896 #ifdef CONFIG_X86_64
1897         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1898         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1899         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1900         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1901         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1902         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1903         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1904         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1905 #endif
1906
1907         regs->rip = vcpu->rip;
1908         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1909
1910         /*
1911          * Don't leak debug flags in case they were set for guest debugging
1912          */
1913         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1914                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1915
1916         vcpu_put(vcpu);
1917
1918         return 0;
1919 }
1920
1921 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1922                                    struct kvm_regs *regs)
1923 {
1924         vcpu_load(vcpu);
1925
1926         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1927         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1928         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1929         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1930         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1931         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1932         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1933         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1934 #ifdef CONFIG_X86_64
1935         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1936         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1937         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1938         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1939         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1940         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1941         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1942         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1943 #endif
1944
1945         vcpu->rip = regs->rip;
1946         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1947
1948         kvm_arch_ops->decache_regs(vcpu);
1949
1950         vcpu_put(vcpu);
1951
1952         return 0;
1953 }
1954
1955 static void get_segment(struct kvm_vcpu *vcpu,
1956                         struct kvm_segment *var, int seg)
1957 {
1958         return kvm_arch_ops->get_segment(vcpu, var, seg);
1959 }
1960
1961 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1962                                     struct kvm_sregs *sregs)
1963 {
1964         struct descriptor_table dt;
1965
1966         vcpu_load(vcpu);
1967
1968         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1969         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1970         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1971         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1972         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1973         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1974
1975         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1976         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1977
1978         kvm_arch_ops->get_idt(vcpu, &dt);
1979         sregs->idt.limit = dt.limit;
1980         sregs->idt.base = dt.base;
1981         kvm_arch_ops->get_gdt(vcpu, &dt);
1982         sregs->gdt.limit = dt.limit;
1983         sregs->gdt.base = dt.base;
1984
1985         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1986         sregs->cr0 = vcpu->cr0;
1987         sregs->cr2 = vcpu->cr2;
1988         sregs->cr3 = vcpu->cr3;
1989         sregs->cr4 = vcpu->cr4;
1990         sregs->cr8 = vcpu->cr8;
1991         sregs->efer = vcpu->shadow_efer;
1992         sregs->apic_base = vcpu->apic_base;
1993
1994         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1995                sizeof sregs->interrupt_bitmap);
1996
1997         vcpu_put(vcpu);
1998
1999         return 0;
2000 }
2001
2002 static void set_segment(struct kvm_vcpu *vcpu,
2003                         struct kvm_segment *var, int seg)
2004 {
2005         return kvm_arch_ops->set_segment(vcpu, var, seg);
2006 }
2007
2008 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2009                                     struct kvm_sregs *sregs)
2010 {
2011         int mmu_reset_needed = 0;
2012         int i;
2013         struct descriptor_table dt;
2014
2015         vcpu_load(vcpu);
2016
2017         dt.limit = sregs->idt.limit;
2018         dt.base = sregs->idt.base;
2019         kvm_arch_ops->set_idt(vcpu, &dt);
2020         dt.limit = sregs->gdt.limit;
2021         dt.base = sregs->gdt.base;
2022         kvm_arch_ops->set_gdt(vcpu, &dt);
2023
2024         vcpu->cr2 = sregs->cr2;
2025         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2026         vcpu->cr3 = sregs->cr3;
2027
2028         vcpu->cr8 = sregs->cr8;
2029
2030         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2031 #ifdef CONFIG_X86_64
2032         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2033 #endif
2034         vcpu->apic_base = sregs->apic_base;
2035
2036         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2037
2038         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2039         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2040
2041         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2042         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2043         if (!is_long_mode(vcpu) && is_pae(vcpu))
2044                 load_pdptrs(vcpu, vcpu->cr3);
2045
2046         if (mmu_reset_needed)
2047                 kvm_mmu_reset_context(vcpu);
2048
2049         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2050                sizeof vcpu->irq_pending);
2051         vcpu->irq_summary = 0;
2052         for (i = 0; i < NR_IRQ_WORDS; ++i)
2053                 if (vcpu->irq_pending[i])
2054                         __set_bit(i, &vcpu->irq_summary);
2055
2056         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2057         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2058         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2059         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2060         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2061         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2062
2063         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2064         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2065
2066         vcpu_put(vcpu);
2067
2068         return 0;
2069 }
2070
2071 /*
2072  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2073  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2074  *
2075  * This list is modified at module load time to reflect the
2076  * capabilities of the host cpu.
2077  */
2078 static u32 msrs_to_save[] = {
2079         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2080         MSR_K6_STAR,
2081 #ifdef CONFIG_X86_64
2082         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2083 #endif
2084         MSR_IA32_TIME_STAMP_COUNTER,
2085 };
2086
2087 static unsigned num_msrs_to_save;
2088
2089 static u32 emulated_msrs[] = {
2090         MSR_IA32_MISC_ENABLE,
2091 };
2092
2093 static __init void kvm_init_msr_list(void)
2094 {
2095         u32 dummy[2];
2096         unsigned i, j;
2097
2098         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2099                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2100                         continue;
2101                 if (j < i)
2102                         msrs_to_save[j] = msrs_to_save[i];
2103                 j++;
2104         }
2105         num_msrs_to_save = j;
2106 }
2107
2108 /*
2109  * Adapt set_msr() to msr_io()'s calling convention
2110  */
2111 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2112 {
2113         return set_msr(vcpu, index, *data);
2114 }
2115
2116 /*
2117  * Read or write a bunch of msrs. All parameters are kernel addresses.
2118  *
2119  * @return number of msrs set successfully.
2120  */
2121 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2122                     struct kvm_msr_entry *entries,
2123                     int (*do_msr)(struct kvm_vcpu *vcpu,
2124                                   unsigned index, u64 *data))
2125 {
2126         int i;
2127
2128         vcpu_load(vcpu);
2129
2130         for (i = 0; i < msrs->nmsrs; ++i)
2131                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2132                         break;
2133
2134         vcpu_put(vcpu);
2135
2136         return i;
2137 }
2138
2139 /*
2140  * Read or write a bunch of msrs. Parameters are user addresses.
2141  *
2142  * @return number of msrs set successfully.
2143  */
2144 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2145                   int (*do_msr)(struct kvm_vcpu *vcpu,
2146                                 unsigned index, u64 *data),
2147                   int writeback)
2148 {
2149         struct kvm_msrs msrs;
2150         struct kvm_msr_entry *entries;
2151         int r, n;
2152         unsigned size;
2153
2154         r = -EFAULT;
2155         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2156                 goto out;
2157
2158         r = -E2BIG;
2159         if (msrs.nmsrs >= MAX_IO_MSRS)
2160                 goto out;
2161
2162         r = -ENOMEM;
2163         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2164         entries = vmalloc(size);
2165         if (!entries)
2166                 goto out;
2167
2168         r = -EFAULT;
2169         if (copy_from_user(entries, user_msrs->entries, size))
2170                 goto out_free;
2171
2172         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2173         if (r < 0)
2174                 goto out_free;
2175
2176         r = -EFAULT;
2177         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2178                 goto out_free;
2179
2180         r = n;
2181
2182 out_free:
2183         vfree(entries);
2184 out:
2185         return r;
2186 }
2187
2188 /*
2189  * Translate a guest virtual address to a guest physical address.
2190  */
2191 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2192                                     struct kvm_translation *tr)
2193 {
2194         unsigned long vaddr = tr->linear_address;
2195         gpa_t gpa;
2196
2197         vcpu_load(vcpu);
2198         spin_lock(&vcpu->kvm->lock);
2199         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2200         tr->physical_address = gpa;
2201         tr->valid = gpa != UNMAPPED_GVA;
2202         tr->writeable = 1;
2203         tr->usermode = 0;
2204         spin_unlock(&vcpu->kvm->lock);
2205         vcpu_put(vcpu);
2206
2207         return 0;
2208 }
2209
2210 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2211                                     struct kvm_interrupt *irq)
2212 {
2213         if (irq->irq < 0 || irq->irq >= 256)
2214                 return -EINVAL;
2215         vcpu_load(vcpu);
2216
2217         set_bit(irq->irq, vcpu->irq_pending);
2218         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2219
2220         vcpu_put(vcpu);
2221
2222         return 0;
2223 }
2224
2225 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2226                                       struct kvm_debug_guest *dbg)
2227 {
2228         int r;
2229
2230         vcpu_load(vcpu);
2231
2232         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2233
2234         vcpu_put(vcpu);
2235
2236         return r;
2237 }
2238
2239 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2240                                     unsigned long address,
2241                                     int *type)
2242 {
2243         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2244         unsigned long pgoff;
2245         struct page *page;
2246
2247         *type = VM_FAULT_MINOR;
2248         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2249         if (pgoff == 0)
2250                 page = virt_to_page(vcpu->run);
2251         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2252                 page = virt_to_page(vcpu->pio_data);
2253         else
2254                 return NOPAGE_SIGBUS;
2255         get_page(page);
2256         return page;
2257 }
2258
2259 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2260         .nopage = kvm_vcpu_nopage,
2261 };
2262
2263 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2264 {
2265         vma->vm_ops = &kvm_vcpu_vm_ops;
2266         return 0;
2267 }
2268
2269 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2270 {
2271         struct kvm_vcpu *vcpu = filp->private_data;
2272
2273         fput(vcpu->kvm->filp);
2274         return 0;
2275 }
2276
2277 static struct file_operations kvm_vcpu_fops = {
2278         .release        = kvm_vcpu_release,
2279         .unlocked_ioctl = kvm_vcpu_ioctl,
2280         .compat_ioctl   = kvm_vcpu_ioctl,
2281         .mmap           = kvm_vcpu_mmap,
2282 };
2283
2284 /*
2285  * Allocates an inode for the vcpu.
2286  */
2287 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2288 {
2289         int fd, r;
2290         struct inode *inode;
2291         struct file *file;
2292
2293         atomic_inc(&vcpu->kvm->filp->f_count);
2294         inode = kvmfs_inode(&kvm_vcpu_fops);
2295         if (IS_ERR(inode)) {
2296                 r = PTR_ERR(inode);
2297                 goto out1;
2298         }
2299
2300         file = kvmfs_file(inode, vcpu);
2301         if (IS_ERR(file)) {
2302                 r = PTR_ERR(file);
2303                 goto out2;
2304         }
2305
2306         r = get_unused_fd();
2307         if (r < 0)
2308                 goto out3;
2309         fd = r;
2310         fd_install(fd, file);
2311
2312         return fd;
2313
2314 out3:
2315         fput(file);
2316 out2:
2317         iput(inode);
2318 out1:
2319         fput(vcpu->kvm->filp);
2320         return r;
2321 }
2322
2323 /*
2324  * Creates some virtual cpus.  Good luck creating more than one.
2325  */
2326 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2327 {
2328         int r;
2329         struct kvm_vcpu *vcpu;
2330         struct page *page;
2331
2332         r = -EINVAL;
2333         if (!valid_vcpu(n))
2334                 goto out;
2335
2336         vcpu = &kvm->vcpus[n];
2337
2338         mutex_lock(&vcpu->mutex);
2339
2340         if (vcpu->vmcs) {
2341                 mutex_unlock(&vcpu->mutex);
2342                 return -EEXIST;
2343         }
2344
2345         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2346         r = -ENOMEM;
2347         if (!page)
2348                 goto out_unlock;
2349         vcpu->run = page_address(page);
2350
2351         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2352         r = -ENOMEM;
2353         if (!page)
2354                 goto out_free_run;
2355         vcpu->pio_data = page_address(page);
2356
2357         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2358                                            FX_IMAGE_ALIGN);
2359         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2360         vcpu->cr0 = 0x10;
2361
2362         r = kvm_arch_ops->vcpu_create(vcpu);
2363         if (r < 0)
2364                 goto out_free_vcpus;
2365
2366         r = kvm_mmu_create(vcpu);
2367         if (r < 0)
2368                 goto out_free_vcpus;
2369
2370         kvm_arch_ops->vcpu_load(vcpu);
2371         r = kvm_mmu_setup(vcpu);
2372         if (r >= 0)
2373                 r = kvm_arch_ops->vcpu_setup(vcpu);
2374         vcpu_put(vcpu);
2375
2376         if (r < 0)
2377                 goto out_free_vcpus;
2378
2379         r = create_vcpu_fd(vcpu);
2380         if (r < 0)
2381                 goto out_free_vcpus;
2382
2383         return r;
2384
2385 out_free_vcpus:
2386         kvm_free_vcpu(vcpu);
2387 out_free_run:
2388         free_page((unsigned long)vcpu->run);
2389         vcpu->run = NULL;
2390 out_unlock:
2391         mutex_unlock(&vcpu->mutex);
2392 out:
2393         return r;
2394 }
2395
2396 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2397 {
2398         u64 efer;
2399         int i;
2400         struct kvm_cpuid_entry *e, *entry;
2401
2402         rdmsrl(MSR_EFER, efer);
2403         entry = NULL;
2404         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2405                 e = &vcpu->cpuid_entries[i];
2406                 if (e->function == 0x80000001) {
2407                         entry = e;
2408                         break;
2409                 }
2410         }
2411         if (entry && (entry->edx & EFER_NX) && !(efer & EFER_NX)) {
2412                 entry->edx &= ~(1 << 20);
2413                 printk(KERN_INFO ": guest NX capability removed\n");
2414         }
2415 }
2416
2417 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2418                                     struct kvm_cpuid *cpuid,
2419                                     struct kvm_cpuid_entry __user *entries)
2420 {
2421         int r;
2422
2423         r = -E2BIG;
2424         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2425                 goto out;
2426         r = -EFAULT;
2427         if (copy_from_user(&vcpu->cpuid_entries, entries,
2428                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2429                 goto out;
2430         vcpu->cpuid_nent = cpuid->nent;
2431         cpuid_fix_nx_cap(vcpu);
2432         return 0;
2433
2434 out:
2435         return r;
2436 }
2437
2438 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2439 {
2440         if (sigset) {
2441                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2442                 vcpu->sigset_active = 1;
2443                 vcpu->sigset = *sigset;
2444         } else
2445                 vcpu->sigset_active = 0;
2446         return 0;
2447 }
2448
2449 /*
2450  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2451  * we have asm/x86/processor.h
2452  */
2453 struct fxsave {
2454         u16     cwd;
2455         u16     swd;
2456         u16     twd;
2457         u16     fop;
2458         u64     rip;
2459         u64     rdp;
2460         u32     mxcsr;
2461         u32     mxcsr_mask;
2462         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2463 #ifdef CONFIG_X86_64
2464         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2465 #else
2466         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2467 #endif
2468 };
2469
2470 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2471 {
2472         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2473
2474         vcpu_load(vcpu);
2475
2476         memcpy(fpu->fpr, fxsave->st_space, 128);
2477         fpu->fcw = fxsave->cwd;
2478         fpu->fsw = fxsave->swd;
2479         fpu->ftwx = fxsave->twd;
2480         fpu->last_opcode = fxsave->fop;
2481         fpu->last_ip = fxsave->rip;
2482         fpu->last_dp = fxsave->rdp;
2483         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2484
2485         vcpu_put(vcpu);
2486
2487         return 0;
2488 }
2489
2490 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2491 {
2492         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2493
2494         vcpu_load(vcpu);
2495
2496         memcpy(fxsave->st_space, fpu->fpr, 128);
2497         fxsave->cwd = fpu->fcw;
2498         fxsave->swd = fpu->fsw;
2499         fxsave->twd = fpu->ftwx;
2500         fxsave->fop = fpu->last_opcode;
2501         fxsave->rip = fpu->last_ip;
2502         fxsave->rdp = fpu->last_dp;
2503         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2504
2505         vcpu_put(vcpu);
2506
2507         return 0;
2508 }
2509
2510 static long kvm_vcpu_ioctl(struct file *filp,
2511                            unsigned int ioctl, unsigned long arg)
2512 {
2513         struct kvm_vcpu *vcpu = filp->private_data;
2514         void __user *argp = (void __user *)arg;
2515         int r = -EINVAL;
2516
2517         switch (ioctl) {
2518         case KVM_RUN:
2519                 r = -EINVAL;
2520                 if (arg)
2521                         goto out;
2522                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2523                 break;
2524         case KVM_GET_REGS: {
2525                 struct kvm_regs kvm_regs;
2526
2527                 memset(&kvm_regs, 0, sizeof kvm_regs);
2528                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2529                 if (r)
2530                         goto out;
2531                 r = -EFAULT;
2532                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2533                         goto out;
2534                 r = 0;
2535                 break;
2536         }
2537         case KVM_SET_REGS: {
2538                 struct kvm_regs kvm_regs;
2539
2540                 r = -EFAULT;
2541                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2542                         goto out;
2543                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2544                 if (r)
2545                         goto out;
2546                 r = 0;
2547                 break;
2548         }
2549         case KVM_GET_SREGS: {
2550                 struct kvm_sregs kvm_sregs;
2551
2552                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2553                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2554                 if (r)
2555                         goto out;
2556                 r = -EFAULT;
2557                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2558                         goto out;
2559                 r = 0;
2560                 break;
2561         }
2562         case KVM_SET_SREGS: {
2563                 struct kvm_sregs kvm_sregs;
2564
2565                 r = -EFAULT;
2566                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2567                         goto out;
2568                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2569                 if (r)
2570                         goto out;
2571                 r = 0;
2572                 break;
2573         }
2574         case KVM_TRANSLATE: {
2575                 struct kvm_translation tr;
2576
2577                 r = -EFAULT;
2578                 if (copy_from_user(&tr, argp, sizeof tr))
2579                         goto out;
2580                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2581                 if (r)
2582                         goto out;
2583                 r = -EFAULT;
2584                 if (copy_to_user(argp, &tr, sizeof tr))
2585                         goto out;
2586                 r = 0;
2587                 break;
2588         }
2589         case KVM_INTERRUPT: {
2590                 struct kvm_interrupt irq;
2591
2592                 r = -EFAULT;
2593                 if (copy_from_user(&irq, argp, sizeof irq))
2594                         goto out;
2595                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2596                 if (r)
2597                         goto out;
2598                 r = 0;
2599                 break;
2600         }
2601         case KVM_DEBUG_GUEST: {
2602                 struct kvm_debug_guest dbg;
2603
2604                 r = -EFAULT;
2605                 if (copy_from_user(&dbg, argp, sizeof dbg))
2606                         goto out;
2607                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2608                 if (r)
2609                         goto out;
2610                 r = 0;
2611                 break;
2612         }
2613         case KVM_GET_MSRS:
2614                 r = msr_io(vcpu, argp, get_msr, 1);
2615                 break;
2616         case KVM_SET_MSRS:
2617                 r = msr_io(vcpu, argp, do_set_msr, 0);
2618                 break;
2619         case KVM_SET_CPUID: {
2620                 struct kvm_cpuid __user *cpuid_arg = argp;
2621                 struct kvm_cpuid cpuid;
2622
2623                 r = -EFAULT;
2624                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2625                         goto out;
2626                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2627                 if (r)
2628                         goto out;
2629                 break;
2630         }
2631         case KVM_SET_SIGNAL_MASK: {
2632                 struct kvm_signal_mask __user *sigmask_arg = argp;
2633                 struct kvm_signal_mask kvm_sigmask;
2634                 sigset_t sigset, *p;
2635
2636                 p = NULL;
2637                 if (argp) {
2638                         r = -EFAULT;
2639                         if (copy_from_user(&kvm_sigmask, argp,
2640                                            sizeof kvm_sigmask))
2641                                 goto out;
2642                         r = -EINVAL;
2643                         if (kvm_sigmask.len != sizeof sigset)
2644                                 goto out;
2645                         r = -EFAULT;
2646                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2647                                            sizeof sigset))
2648                                 goto out;
2649                         p = &sigset;
2650                 }
2651                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2652                 break;
2653         }
2654         case KVM_GET_FPU: {
2655                 struct kvm_fpu fpu;
2656
2657                 memset(&fpu, 0, sizeof fpu);
2658                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2659                 if (r)
2660                         goto out;
2661                 r = -EFAULT;
2662                 if (copy_to_user(argp, &fpu, sizeof fpu))
2663                         goto out;
2664                 r = 0;
2665                 break;
2666         }
2667         case KVM_SET_FPU: {
2668                 struct kvm_fpu fpu;
2669
2670                 r = -EFAULT;
2671                 if (copy_from_user(&fpu, argp, sizeof fpu))
2672                         goto out;
2673                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2674                 if (r)
2675                         goto out;
2676                 r = 0;
2677                 break;
2678         }
2679         default:
2680                 ;
2681         }
2682 out:
2683         return r;
2684 }
2685
2686 static long kvm_vm_ioctl(struct file *filp,
2687                            unsigned int ioctl, unsigned long arg)
2688 {
2689         struct kvm *kvm = filp->private_data;
2690         void __user *argp = (void __user *)arg;
2691         int r = -EINVAL;
2692
2693         switch (ioctl) {
2694         case KVM_CREATE_VCPU:
2695                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2696                 if (r < 0)
2697                         goto out;
2698                 break;
2699         case KVM_SET_MEMORY_REGION: {
2700                 struct kvm_memory_region kvm_mem;
2701
2702                 r = -EFAULT;
2703                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2704                         goto out;
2705                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2706                 if (r)
2707                         goto out;
2708                 break;
2709         }
2710         case KVM_GET_DIRTY_LOG: {
2711                 struct kvm_dirty_log log;
2712
2713                 r = -EFAULT;
2714                 if (copy_from_user(&log, argp, sizeof log))
2715                         goto out;
2716                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2717                 if (r)
2718                         goto out;
2719                 break;
2720         }
2721         case KVM_SET_MEMORY_ALIAS: {
2722                 struct kvm_memory_alias alias;
2723
2724                 r = -EFAULT;
2725                 if (copy_from_user(&alias, argp, sizeof alias))
2726                         goto out;
2727                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2728                 if (r)
2729                         goto out;
2730                 break;
2731         }
2732         default:
2733                 ;
2734         }
2735 out:
2736         return r;
2737 }
2738
2739 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2740                                   unsigned long address,
2741                                   int *type)
2742 {
2743         struct kvm *kvm = vma->vm_file->private_data;
2744         unsigned long pgoff;
2745         struct page *page;
2746
2747         *type = VM_FAULT_MINOR;
2748         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2749         page = gfn_to_page(kvm, pgoff);
2750         if (!page)
2751                 return NOPAGE_SIGBUS;
2752         get_page(page);
2753         return page;
2754 }
2755
2756 static struct vm_operations_struct kvm_vm_vm_ops = {
2757         .nopage = kvm_vm_nopage,
2758 };
2759
2760 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2761 {
2762         vma->vm_ops = &kvm_vm_vm_ops;
2763         return 0;
2764 }
2765
2766 static struct file_operations kvm_vm_fops = {
2767         .release        = kvm_vm_release,
2768         .unlocked_ioctl = kvm_vm_ioctl,
2769         .compat_ioctl   = kvm_vm_ioctl,
2770         .mmap           = kvm_vm_mmap,
2771 };
2772
2773 static int kvm_dev_ioctl_create_vm(void)
2774 {
2775         int fd, r;
2776         struct inode *inode;
2777         struct file *file;
2778         struct kvm *kvm;
2779
2780         inode = kvmfs_inode(&kvm_vm_fops);
2781         if (IS_ERR(inode)) {
2782                 r = PTR_ERR(inode);
2783                 goto out1;
2784         }
2785
2786         kvm = kvm_create_vm();
2787         if (IS_ERR(kvm)) {
2788                 r = PTR_ERR(kvm);
2789                 goto out2;
2790         }
2791
2792         file = kvmfs_file(inode, kvm);
2793         if (IS_ERR(file)) {
2794                 r = PTR_ERR(file);
2795                 goto out3;
2796         }
2797         kvm->filp = file;
2798
2799         r = get_unused_fd();
2800         if (r < 0)
2801                 goto out4;
2802         fd = r;
2803         fd_install(fd, file);
2804
2805         return fd;
2806
2807 out4:
2808         fput(file);
2809 out3:
2810         kvm_destroy_vm(kvm);
2811 out2:
2812         iput(inode);
2813 out1:
2814         return r;
2815 }
2816
2817 static long kvm_dev_ioctl(struct file *filp,
2818                           unsigned int ioctl, unsigned long arg)
2819 {
2820         void __user *argp = (void __user *)arg;
2821         long r = -EINVAL;
2822
2823         switch (ioctl) {
2824         case KVM_GET_API_VERSION:
2825                 r = -EINVAL;
2826                 if (arg)
2827                         goto out;
2828                 r = KVM_API_VERSION;
2829                 break;
2830         case KVM_CREATE_VM:
2831                 r = -EINVAL;
2832                 if (arg)
2833                         goto out;
2834                 r = kvm_dev_ioctl_create_vm();
2835                 break;
2836         case KVM_GET_MSR_INDEX_LIST: {
2837                 struct kvm_msr_list __user *user_msr_list = argp;
2838                 struct kvm_msr_list msr_list;
2839                 unsigned n;
2840
2841                 r = -EFAULT;
2842                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2843                         goto out;
2844                 n = msr_list.nmsrs;
2845                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2846                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2847                         goto out;
2848                 r = -E2BIG;
2849                 if (n < num_msrs_to_save)
2850                         goto out;
2851                 r = -EFAULT;
2852                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2853                                  num_msrs_to_save * sizeof(u32)))
2854                         goto out;
2855                 if (copy_to_user(user_msr_list->indices
2856                                  + num_msrs_to_save * sizeof(u32),
2857                                  &emulated_msrs,
2858                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2859                         goto out;
2860                 r = 0;
2861                 break;
2862         }
2863         case KVM_CHECK_EXTENSION:
2864                 /*
2865                  * No extensions defined at present.
2866                  */
2867                 r = 0;
2868                 break;
2869         case KVM_GET_VCPU_MMAP_SIZE:
2870                 r = -EINVAL;
2871                 if (arg)
2872                         goto out;
2873                 r = 2 * PAGE_SIZE;
2874                 break;
2875         default:
2876                 ;
2877         }
2878 out:
2879         return r;
2880 }
2881
2882 static struct file_operations kvm_chardev_ops = {
2883         .open           = kvm_dev_open,
2884         .release        = kvm_dev_release,
2885         .unlocked_ioctl = kvm_dev_ioctl,
2886         .compat_ioctl   = kvm_dev_ioctl,
2887 };
2888
2889 static struct miscdevice kvm_dev = {
2890         KVM_MINOR,
2891         "kvm",
2892         &kvm_chardev_ops,
2893 };
2894
2895 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2896                        void *v)
2897 {
2898         if (val == SYS_RESTART) {
2899                 /*
2900                  * Some (well, at least mine) BIOSes hang on reboot if
2901                  * in vmx root mode.
2902                  */
2903                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2904                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2905         }
2906         return NOTIFY_OK;
2907 }
2908
2909 static struct notifier_block kvm_reboot_notifier = {
2910         .notifier_call = kvm_reboot,
2911         .priority = 0,
2912 };
2913
2914 /*
2915  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2916  * cached on it.
2917  */
2918 static void decache_vcpus_on_cpu(int cpu)
2919 {
2920         struct kvm *vm;
2921         struct kvm_vcpu *vcpu;
2922         int i;
2923
2924         spin_lock(&kvm_lock);
2925         list_for_each_entry(vm, &vm_list, vm_list)
2926                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2927                         vcpu = &vm->vcpus[i];
2928                         /*
2929                          * If the vcpu is locked, then it is running on some
2930                          * other cpu and therefore it is not cached on the
2931                          * cpu in question.
2932                          *
2933                          * If it's not locked, check the last cpu it executed
2934                          * on.
2935                          */
2936                         if (mutex_trylock(&vcpu->mutex)) {
2937                                 if (vcpu->cpu == cpu) {
2938                                         kvm_arch_ops->vcpu_decache(vcpu);
2939                                         vcpu->cpu = -1;
2940                                 }
2941                                 mutex_unlock(&vcpu->mutex);
2942                         }
2943                 }
2944         spin_unlock(&kvm_lock);
2945 }
2946
2947 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2948                            void *v)
2949 {
2950         int cpu = (long)v;
2951
2952         switch (val) {
2953         case CPU_DOWN_PREPARE:
2954         case CPU_DOWN_PREPARE_FROZEN:
2955         case CPU_UP_CANCELED:
2956         case CPU_UP_CANCELED_FROZEN:
2957                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2958                        cpu);
2959                 decache_vcpus_on_cpu(cpu);
2960                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2961                                          NULL, 0, 1);
2962                 break;
2963         case CPU_ONLINE:
2964         case CPU_ONLINE_FROZEN:
2965                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2966                        cpu);
2967                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2968                                          NULL, 0, 1);
2969                 break;
2970         }
2971         return NOTIFY_OK;
2972 }
2973
2974 static struct notifier_block kvm_cpu_notifier = {
2975         .notifier_call = kvm_cpu_hotplug,
2976         .priority = 20, /* must be > scheduler priority */
2977 };
2978
2979 static u64 stat_get(void *_offset)
2980 {
2981         unsigned offset = (long)_offset;
2982         u64 total = 0;
2983         struct kvm *kvm;
2984         struct kvm_vcpu *vcpu;
2985         int i;
2986
2987         spin_lock(&kvm_lock);
2988         list_for_each_entry(kvm, &vm_list, vm_list)
2989                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2990                         vcpu = &kvm->vcpus[i];
2991                         total += *(u32 *)((void *)vcpu + offset);
2992                 }
2993         spin_unlock(&kvm_lock);
2994         return total;
2995 }
2996
2997 static void stat_set(void *offset, u64 val)
2998 {
2999 }
3000
3001 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3002
3003 static __init void kvm_init_debug(void)
3004 {
3005         struct kvm_stats_debugfs_item *p;
3006
3007         debugfs_dir = debugfs_create_dir("kvm", NULL);
3008         for (p = debugfs_entries; p->name; ++p)
3009                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3010                                                 (void *)(long)p->offset,
3011                                                 &stat_fops);
3012 }
3013
3014 static void kvm_exit_debug(void)
3015 {
3016         struct kvm_stats_debugfs_item *p;
3017
3018         for (p = debugfs_entries; p->name; ++p)
3019                 debugfs_remove(p->dentry);
3020         debugfs_remove(debugfs_dir);
3021 }
3022
3023 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3024 {
3025         decache_vcpus_on_cpu(raw_smp_processor_id());
3026         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3027         return 0;
3028 }
3029
3030 static int kvm_resume(struct sys_device *dev)
3031 {
3032         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3033         return 0;
3034 }
3035
3036 static struct sysdev_class kvm_sysdev_class = {
3037         set_kset_name("kvm"),
3038         .suspend = kvm_suspend,
3039         .resume = kvm_resume,
3040 };
3041
3042 static struct sys_device kvm_sysdev = {
3043         .id = 0,
3044         .cls = &kvm_sysdev_class,
3045 };
3046
3047 hpa_t bad_page_address;
3048
3049 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
3050                         const char *dev_name, void *data, struct vfsmount *mnt)
3051 {
3052         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
3053 }
3054
3055 static struct file_system_type kvm_fs_type = {
3056         .name           = "kvmfs",
3057         .get_sb         = kvmfs_get_sb,
3058         .kill_sb        = kill_anon_super,
3059 };
3060
3061 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3062 {
3063         int r;
3064
3065         if (kvm_arch_ops) {
3066                 printk(KERN_ERR "kvm: already loaded the other module\n");
3067                 return -EEXIST;
3068         }
3069
3070         if (!ops->cpu_has_kvm_support()) {
3071                 printk(KERN_ERR "kvm: no hardware support\n");
3072                 return -EOPNOTSUPP;
3073         }
3074         if (ops->disabled_by_bios()) {
3075                 printk(KERN_ERR "kvm: disabled by bios\n");
3076                 return -EOPNOTSUPP;
3077         }
3078
3079         kvm_arch_ops = ops;
3080
3081         r = kvm_arch_ops->hardware_setup();
3082         if (r < 0)
3083                 goto out;
3084
3085         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3086         r = register_cpu_notifier(&kvm_cpu_notifier);
3087         if (r)
3088                 goto out_free_1;
3089         register_reboot_notifier(&kvm_reboot_notifier);
3090
3091         r = sysdev_class_register(&kvm_sysdev_class);
3092         if (r)
3093                 goto out_free_2;
3094
3095         r = sysdev_register(&kvm_sysdev);
3096         if (r)
3097                 goto out_free_3;
3098
3099         kvm_chardev_ops.owner = module;
3100
3101         r = misc_register(&kvm_dev);
3102         if (r) {
3103                 printk (KERN_ERR "kvm: misc device register failed\n");
3104                 goto out_free;
3105         }
3106
3107         return r;
3108
3109 out_free:
3110         sysdev_unregister(&kvm_sysdev);
3111 out_free_3:
3112         sysdev_class_unregister(&kvm_sysdev_class);
3113 out_free_2:
3114         unregister_reboot_notifier(&kvm_reboot_notifier);
3115         unregister_cpu_notifier(&kvm_cpu_notifier);
3116 out_free_1:
3117         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3118         kvm_arch_ops->hardware_unsetup();
3119 out:
3120         kvm_arch_ops = NULL;
3121         return r;
3122 }
3123
3124 void kvm_exit_arch(void)
3125 {
3126         misc_deregister(&kvm_dev);
3127         sysdev_unregister(&kvm_sysdev);
3128         sysdev_class_unregister(&kvm_sysdev_class);
3129         unregister_reboot_notifier(&kvm_reboot_notifier);
3130         unregister_cpu_notifier(&kvm_cpu_notifier);
3131         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3132         kvm_arch_ops->hardware_unsetup();
3133         kvm_arch_ops = NULL;
3134 }
3135
3136 static __init int kvm_init(void)
3137 {
3138         static struct page *bad_page;
3139         int r;
3140
3141         r = kvm_mmu_module_init();
3142         if (r)
3143                 goto out4;
3144
3145         r = register_filesystem(&kvm_fs_type);
3146         if (r)
3147                 goto out3;
3148
3149         kvmfs_mnt = kern_mount(&kvm_fs_type);
3150         r = PTR_ERR(kvmfs_mnt);
3151         if (IS_ERR(kvmfs_mnt))
3152                 goto out2;
3153         kvm_init_debug();
3154
3155         kvm_init_msr_list();
3156
3157         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3158                 r = -ENOMEM;
3159                 goto out;
3160         }
3161
3162         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3163         memset(__va(bad_page_address), 0, PAGE_SIZE);
3164
3165         return 0;
3166
3167 out:
3168         kvm_exit_debug();
3169         mntput(kvmfs_mnt);
3170 out2:
3171         unregister_filesystem(&kvm_fs_type);
3172 out3:
3173         kvm_mmu_module_exit();
3174 out4:
3175         return r;
3176 }
3177
3178 static __exit void kvm_exit(void)
3179 {
3180         kvm_exit_debug();
3181         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3182         mntput(kvmfs_mnt);
3183         unregister_filesystem(&kvm_fs_type);
3184         kvm_mmu_module_exit();
3185 }
3186
3187 module_init(kvm_init)
3188 module_exit(kvm_exit)
3189
3190 EXPORT_SYMBOL_GPL(kvm_init_arch);
3191 EXPORT_SYMBOL_GPL(kvm_exit_arch);