git.oblomov.eu Git - linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <linux/magic.h>
  24 #include <asm/processor.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <asm/msr.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/reboot.h>
  33 #include <asm/io.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/highmem.h>
  36 #include <linux/file.h>
  37 #include <asm/desc.h>
  38 #include <linux/sysdev.h>
  39 #include <linux/cpu.h>
  40 #include <linux/file.h>
  41 #include <linux/fs.h>
  42 #include <linux/mount.h>
  43
  44 #include "x86_emulate.h"
  45 #include "segment_descriptor.h"
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 struct kvm_arch_ops *kvm_arch_ops;
  54 struct kvm_stat kvm_stat;
  55 EXPORT_SYMBOL_GPL(kvm_stat);
  56
  57 static struct kvm_stats_debugfs_item {
  58         const char *name;
  59         u32 *data;
  60         struct dentry *dentry;
  61 } debugfs_entries[] = {
  62         { "pf_fixed", &kvm_stat.pf_fixed },
  63         { "pf_guest", &kvm_stat.pf_guest },
  64         { "tlb_flush", &kvm_stat.tlb_flush },
  65         { "invlpg", &kvm_stat.invlpg },
  66         { "exits", &kvm_stat.exits },
  67         { "io_exits", &kvm_stat.io_exits },
  68         { "mmio_exits", &kvm_stat.mmio_exits },
  69         { "signal_exits", &kvm_stat.signal_exits },
  70         { "irq_window", &kvm_stat.irq_window_exits },
  71         { "halt_exits", &kvm_stat.halt_exits },
  72         { "request_irq", &kvm_stat.request_irq_exits },
  73         { "irq_exits", &kvm_stat.irq_exits },
  74         { NULL, NULL }
  75 };
  76
  77 static struct dentry *debugfs_dir;
  78
  79 struct vfsmount *kvmfs_mnt;
  80
  81 #define MAX_IO_MSRS 256
  82
  83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  84 #define LMSW_GUEST_MASK 0x0eULL
  85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  86 #define CR8_RESEVED_BITS (~0x0fULL)
  87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  88
  89 #ifdef CONFIG_X86_64
  90 // LDT or TSS descriptor in the GDT. 16 bytes.
  91 struct segment_descriptor_64 {
  92         struct segment_descriptor s;
  93         u32 base_higher;
  94         u32 pad_zero;
  95 };
  96
  97 #endif
  98
  99 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 100                            unsigned long arg);
 101
 102 static struct inode *kvmfs_inode(struct file_operations *fops)
 103 {
 104         int error = -ENOMEM;
 105         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 106
 107         if (!inode)
 108                 goto eexit_1;
 109
 110         inode->i_fop = fops;
 111
 112         /*
 113          * Mark the inode dirty from the very beginning,
 114          * that way it will never be moved to the dirty
 115          * list because mark_inode_dirty() will think
 116          * that it already _is_ on the dirty list.
 117          */
 118         inode->i_state = I_DIRTY;
 119         inode->i_mode = S_IRUSR | S_IWUSR;
 120         inode->i_uid = current->fsuid;
 121         inode->i_gid = current->fsgid;
 122         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 123         return inode;
 124
 125 eexit_1:
 126         return ERR_PTR(error);
 127 }
 128
 129 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 130 {
 131         struct file *file = get_empty_filp();
 132
 133         if (!file)
 134                 return ERR_PTR(-ENFILE);
 135
 136         file->f_path.mnt = mntget(kvmfs_mnt);
 137         file->f_path.dentry = d_alloc_anon(inode);
 138         if (!file->f_path.dentry)
 139                 return ERR_PTR(-ENOMEM);
 140         file->f_mapping = inode->i_mapping;
 141
 142         file->f_pos = 0;
 143         file->f_flags = O_RDWR;
 144         file->f_op = inode->i_fop;
 145         file->f_mode = FMODE_READ | FMODE_WRITE;
 146         file->f_version = 0;
 147         file->private_data = private_data;
 148         return file;
 149 }
 150
 151 unsigned long segment_base(u16 selector)
 152 {
 153         struct descriptor_table gdt;
 154         struct segment_descriptor *d;
 155         unsigned long table_base;
 156         typedef unsigned long ul;
 157         unsigned long v;
 158
 159         if (selector == 0)
 160                 return 0;
 161
 162         asm ("sgdt %0" : "=m"(gdt));
 163         table_base = gdt.base;
 164
 165         if (selector & 4) {           /* from ldt */
 166                 u16 ldt_selector;
 167
 168                 asm ("sldt %0" : "=g"(ldt_selector));
 169                 table_base = segment_base(ldt_selector);
 170         }
 171         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 172         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 173 #ifdef CONFIG_X86_64
 174         if (d->system == 0
 175             && (d->type == 2 || d->type == 9 || d->type == 11))
 176                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 177 #endif
 178         return v;
 179 }
 180 EXPORT_SYMBOL_GPL(segment_base);
 181
 182 static inline int valid_vcpu(int n)
 183 {
 184         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 185 }
 186
 187 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 188                    void *dest)
 189 {
 190         unsigned char *host_buf = dest;
 191         unsigned long req_size = size;
 192
 193         while (size) {
 194                 hpa_t paddr;
 195                 unsigned now;
 196                 unsigned offset;
 197                 hva_t guest_buf;
 198
 199                 paddr = gva_to_hpa(vcpu, addr);
 200
 201                 if (is_error_hpa(paddr))
 202                         break;
 203
 204                 guest_buf = (hva_t)kmap_atomic(
 205                                         pfn_to_page(paddr >> PAGE_SHIFT),
 206                                         KM_USER0);
 207                 offset = addr & ~PAGE_MASK;
 208                 guest_buf |= offset;
 209                 now = min(size, PAGE_SIZE - offset);
 210                 memcpy(host_buf, (void*)guest_buf, now);
 211                 host_buf += now;
 212                 addr += now;
 213                 size -= now;
 214                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 215         }
 216         return req_size - size;
 217 }
 218 EXPORT_SYMBOL_GPL(kvm_read_guest);
 219
 220 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 221                     void *data)
 222 {
 223         unsigned char *host_buf = data;
 224         unsigned long req_size = size;
 225
 226         while (size) {
 227                 hpa_t paddr;
 228                 unsigned now;
 229                 unsigned offset;
 230                 hva_t guest_buf;
 231                 gfn_t gfn;
 232
 233                 paddr = gva_to_hpa(vcpu, addr);
 234
 235                 if (is_error_hpa(paddr))
 236                         break;
 237
 238                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 239                 mark_page_dirty(vcpu->kvm, gfn);
 240                 guest_buf = (hva_t)kmap_atomic(
 241                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 242                 offset = addr & ~PAGE_MASK;
 243                 guest_buf |= offset;
 244                 now = min(size, PAGE_SIZE - offset);
 245                 memcpy((void*)guest_buf, host_buf, now);
 246                 host_buf += now;
 247                 addr += now;
 248                 size -= now;
 249                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 250         }
 251         return req_size - size;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_write_guest);
 254
 255 /*
 256  * Switches to specified vcpu, until a matching vcpu_put()
 257  */
 258 static void vcpu_load(struct kvm_vcpu *vcpu)
 259 {
 260         mutex_lock(&vcpu->mutex);
 261         kvm_arch_ops->vcpu_load(vcpu);
 262 }
 263
 264 /*
 265  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 266  * if the slot is not populated.
 267  */
 268 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 269 {
 270         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 271
 272         mutex_lock(&vcpu->mutex);
 273         if (!vcpu->vmcs) {
 274                 mutex_unlock(&vcpu->mutex);
 275                 return NULL;
 276         }
 277         kvm_arch_ops->vcpu_load(vcpu);
 278         return vcpu;
 279 }
 280
 281 static void vcpu_put(struct kvm_vcpu *vcpu)
 282 {
 283         kvm_arch_ops->vcpu_put(vcpu);
 284         mutex_unlock(&vcpu->mutex);
 285 }
 286
 287 static struct kvm *kvm_create_vm(void)
 288 {
 289         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 290         int i;
 291
 292         if (!kvm)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         spin_lock_init(&kvm->lock);
 296         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 297         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 298                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 299
 300                 mutex_init(&vcpu->mutex);
 301                 vcpu->cpu = -1;
 302                 vcpu->kvm = kvm;
 303                 vcpu->mmu.root_hpa = INVALID_PAGE;
 304                 INIT_LIST_HEAD(&vcpu->free_pages);
 305                 spin_lock(&kvm_lock);
 306                 list_add(&kvm->vm_list, &vm_list);
 307                 spin_unlock(&kvm_lock);
 308         }
 309         return kvm;
 310 }
 311
 312 static int kvm_dev_open(struct inode *inode, struct file *filp)
 313 {
 314         return 0;
 315 }
 316
 317 /*
 318  * Free any memory in @free but not in @dont.
 319  */
 320 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 321                                   struct kvm_memory_slot *dont)
 322 {
 323         int i;
 324
 325         if (!dont || free->phys_mem != dont->phys_mem)
 326                 if (free->phys_mem) {
 327                         for (i = 0; i < free->npages; ++i)
 328                                 if (free->phys_mem[i])
 329                                         __free_page(free->phys_mem[i]);
 330                         vfree(free->phys_mem);
 331                 }
 332
 333         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 334                 vfree(free->dirty_bitmap);
 335
 336         free->phys_mem = NULL;
 337         free->npages = 0;
 338         free->dirty_bitmap = NULL;
 339 }
 340
 341 static void kvm_free_physmem(struct kvm *kvm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < kvm->nmemslots; ++i)
 346                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 347 }
 348
 349 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 350 {
 351         int i;
 352
 353         for (i = 0; i < 2; ++i)
 354                 if (vcpu->pio.guest_pages[i]) {
 355                         __free_page(vcpu->pio.guest_pages[i]);
 356                         vcpu->pio.guest_pages[i] = NULL;
 357                 }
 358 }
 359
 360 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 361 {
 362         if (!vcpu->vmcs)
 363                 return;
 364
 365         vcpu_load(vcpu);
 366         kvm_mmu_destroy(vcpu);
 367         vcpu_put(vcpu);
 368         kvm_arch_ops->vcpu_free(vcpu);
 369         free_page((unsigned long)vcpu->run);
 370         vcpu->run = NULL;
 371         free_page((unsigned long)vcpu->pio_data);
 372         vcpu->pio_data = NULL;
 373         free_pio_guest_pages(vcpu);
 374 }
 375
 376 static void kvm_free_vcpus(struct kvm *kvm)
 377 {
 378         unsigned int i;
 379
 380         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 381                 kvm_free_vcpu(&kvm->vcpus[i]);
 382 }
 383
 384 static int kvm_dev_release(struct inode *inode, struct file *filp)
 385 {
 386         return 0;
 387 }
 388
 389 static void kvm_destroy_vm(struct kvm *kvm)
 390 {
 391         spin_lock(&kvm_lock);
 392         list_del(&kvm->vm_list);
 393         spin_unlock(&kvm_lock);
 394         kvm_free_vcpus(kvm);
 395         kvm_free_physmem(kvm);
 396         kfree(kvm);
 397 }
 398
 399 static int kvm_vm_release(struct inode *inode, struct file *filp)
 400 {
 401         struct kvm *kvm = filp->private_data;
 402
 403         kvm_destroy_vm(kvm);
 404         return 0;
 405 }
 406
 407 static void inject_gp(struct kvm_vcpu *vcpu)
 408 {
 409         kvm_arch_ops->inject_gp(vcpu, 0);
 410 }
 411
 412 /*
 413  * Load the pae pdptrs.  Return true is they are all valid.
 414  */
 415 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 416 {
 417         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 418         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 419         int i;
 420         u64 pdpte;
 421         u64 *pdpt;
 422         int ret;
 423         struct page *page;
 424
 425         spin_lock(&vcpu->kvm->lock);
 426         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 427         /* FIXME: !page - emulate? 0xff? */
 428         pdpt = kmap_atomic(page, KM_USER0);
 429
 430         ret = 1;
 431         for (i = 0; i < 4; ++i) {
 432                 pdpte = pdpt[offset + i];
 433                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 434                         ret = 0;
 435                         goto out;
 436                 }
 437         }
 438
 439         for (i = 0; i < 4; ++i)
 440                 vcpu->pdptrs[i] = pdpt[offset + i];
 441
 442 out:
 443         kunmap_atomic(pdpt, KM_USER0);
 444         spin_unlock(&vcpu->kvm->lock);
 445
 446         return ret;
 447 }
 448
 449 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 450 {
 451         if (cr0 & CR0_RESEVED_BITS) {
 452                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 453                        cr0, vcpu->cr0);
 454                 inject_gp(vcpu);
 455                 return;
 456         }
 457
 458         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 459                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 460                 inject_gp(vcpu);
 461                 return;
 462         }
 463
 464         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 465                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 466                        "and a clear PE flag\n");
 467                 inject_gp(vcpu);
 468                 return;
 469         }
 470
 471         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 472 #ifdef CONFIG_X86_64
 473                 if ((vcpu->shadow_efer & EFER_LME)) {
 474                         int cs_db, cs_l;
 475
 476                         if (!is_pae(vcpu)) {
 477                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 478                                        "in long mode while PAE is disabled\n");
 479                                 inject_gp(vcpu);
 480                                 return;
 481                         }
 482                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 483                         if (cs_l) {
 484                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 485                                        "in long mode while CS.L == 1\n");
 486                                 inject_gp(vcpu);
 487                                 return;
 488
 489                         }
 490                 } else
 491 #endif
 492                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 493                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 494                                "reserved bits\n");
 495                         inject_gp(vcpu);
 496                         return;
 497                 }
 498
 499         }
 500
 501         kvm_arch_ops->set_cr0(vcpu, cr0);
 502         vcpu->cr0 = cr0;
 503
 504         spin_lock(&vcpu->kvm->lock);
 505         kvm_mmu_reset_context(vcpu);
 506         spin_unlock(&vcpu->kvm->lock);
 507         return;
 508 }
 509 EXPORT_SYMBOL_GPL(set_cr0);
 510
 511 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 512 {
 513         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
 514         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 515 }
 516 EXPORT_SYMBOL_GPL(lmsw);
 517
 518 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 519 {
 520         if (cr4 & CR4_RESEVED_BITS) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 522                 inject_gp(vcpu);
 523                 return;
 524         }
 525
 526         if (is_long_mode(vcpu)) {
 527                 if (!(cr4 & CR4_PAE_MASK)) {
 528                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 529                                "in long mode\n");
 530                         inject_gp(vcpu);
 531                         return;
 532                 }
 533         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 534                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 535                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 536                 inject_gp(vcpu);
 537         }
 538
 539         if (cr4 & CR4_VMXE_MASK) {
 540                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 541                 inject_gp(vcpu);
 542                 return;
 543         }
 544         kvm_arch_ops->set_cr4(vcpu, cr4);
 545         spin_lock(&vcpu->kvm->lock);
 546         kvm_mmu_reset_context(vcpu);
 547         spin_unlock(&vcpu->kvm->lock);
 548 }
 549 EXPORT_SYMBOL_GPL(set_cr4);
 550
 551 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 552 {
 553         if (is_long_mode(vcpu)) {
 554                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 555                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 556                         inject_gp(vcpu);
 557                         return;
 558                 }
 559         } else {
 560                 if (cr3 & CR3_RESEVED_BITS) {
 561                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 562                         inject_gp(vcpu);
 563                         return;
 564                 }
 565                 if (is_paging(vcpu) && is_pae(vcpu) &&
 566                     !load_pdptrs(vcpu, cr3)) {
 567                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 568                                "reserved bits\n");
 569                         inject_gp(vcpu);
 570                         return;
 571                 }
 572         }
 573
 574         vcpu->cr3 = cr3;
 575         spin_lock(&vcpu->kvm->lock);
 576         /*
 577          * Does the new cr3 value map to physical memory? (Note, we
 578          * catch an invalid cr3 even in real-mode, because it would
 579          * cause trouble later on when we turn on paging anyway.)
 580          *
 581          * A real CPU would silently accept an invalid cr3 and would
 582          * attempt to use it - with largely undefined (and often hard
 583          * to debug) behavior on the guest side.
 584          */
 585         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 586                 inject_gp(vcpu);
 587         else
 588                 vcpu->mmu.new_cr3(vcpu);
 589         spin_unlock(&vcpu->kvm->lock);
 590 }
 591 EXPORT_SYMBOL_GPL(set_cr3);
 592
 593 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 594 {
 595         if ( cr8 & CR8_RESEVED_BITS) {
 596                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 597                 inject_gp(vcpu);
 598                 return;
 599         }
 600         vcpu->cr8 = cr8;
 601 }
 602 EXPORT_SYMBOL_GPL(set_cr8);
 603
 604 void fx_init(struct kvm_vcpu *vcpu)
 605 {
 606         struct __attribute__ ((__packed__)) fx_image_s {
 607                 u16 control; //fcw
 608                 u16 status; //fsw
 609                 u16 tag; // ftw
 610                 u16 opcode; //fop
 611                 u64 ip; // fpu ip
 612                 u64 operand;// fpu dp
 613                 u32 mxcsr;
 614                 u32 mxcsr_mask;
 615
 616         } *fx_image;
 617
 618         fx_save(vcpu->host_fx_image);
 619         fpu_init();
 620         fx_save(vcpu->guest_fx_image);
 621         fx_restore(vcpu->host_fx_image);
 622
 623         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 624         fx_image->mxcsr = 0x1f80;
 625         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 626                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 627 }
 628 EXPORT_SYMBOL_GPL(fx_init);
 629
 630 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
 631 {
 632         spin_lock(&vcpu->kvm->lock);
 633         kvm_mmu_slot_remove_write_access(vcpu, slot);
 634         spin_unlock(&vcpu->kvm->lock);
 635 }
 636
 637 /*
 638  * Allocate some memory and give it an address in the guest physical address
 639  * space.
 640  *
 641  * Discontiguous memory is allowed, mostly for framebuffers.
 642  */
 643 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 644                                           struct kvm_memory_region *mem)
 645 {
 646         int r;
 647         gfn_t base_gfn;
 648         unsigned long npages;
 649         unsigned long i;
 650         struct kvm_memory_slot *memslot;
 651         struct kvm_memory_slot old, new;
 652         int memory_config_version;
 653
 654         r = -EINVAL;
 655         /* General sanity checks */
 656         if (mem->memory_size & (PAGE_SIZE - 1))
 657                 goto out;
 658         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 659                 goto out;
 660         if (mem->slot >= KVM_MEMORY_SLOTS)
 661                 goto out;
 662         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 663                 goto out;
 664
 665         memslot = &kvm->memslots[mem->slot];
 666         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 667         npages = mem->memory_size >> PAGE_SHIFT;
 668
 669         if (!npages)
 670                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 671
 672 raced:
 673         spin_lock(&kvm->lock);
 674
 675         memory_config_version = kvm->memory_config_version;
 676         new = old = *memslot;
 677
 678         new.base_gfn = base_gfn;
 679         new.npages = npages;
 680         new.flags = mem->flags;
 681
 682         /* Disallow changing a memory slot's size. */
 683         r = -EINVAL;
 684         if (npages && old.npages && npages != old.npages)
 685                 goto out_unlock;
 686
 687         /* Check for overlaps */
 688         r = -EEXIST;
 689         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 690                 struct kvm_memory_slot *s = &kvm->memslots[i];
 691
 692                 if (s == memslot)
 693                         continue;
 694                 if (!((base_gfn + npages <= s->base_gfn) ||
 695                       (base_gfn >= s->base_gfn + s->npages)))
 696                         goto out_unlock;
 697         }
 698         /*
 699          * Do memory allocations outside lock.  memory_config_version will
 700          * detect any races.
 701          */
 702         spin_unlock(&kvm->lock);
 703
 704         /* Deallocate if slot is being removed */
 705         if (!npages)
 706                 new.phys_mem = NULL;
 707
 708         /* Free page dirty bitmap if unneeded */
 709         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 710                 new.dirty_bitmap = NULL;
 711
 712         r = -ENOMEM;
 713
 714         /* Allocate if a slot is being created */
 715         if (npages && !new.phys_mem) {
 716                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 717
 718                 if (!new.phys_mem)
 719                         goto out_free;
 720
 721                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 722                 for (i = 0; i < npages; ++i) {
 723                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 724                                                      | __GFP_ZERO);
 725                         if (!new.phys_mem[i])
 726                                 goto out_free;
 727                         set_page_private(new.phys_mem[i],0);
 728                 }
 729         }
 730
 731         /* Allocate page dirty bitmap if needed */
 732         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 733                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 734
 735                 new.dirty_bitmap = vmalloc(dirty_bytes);
 736                 if (!new.dirty_bitmap)
 737                         goto out_free;
 738                 memset(new.dirty_bitmap, 0, dirty_bytes);
 739         }
 740
 741         spin_lock(&kvm->lock);
 742
 743         if (memory_config_version != kvm->memory_config_version) {
 744                 spin_unlock(&kvm->lock);
 745                 kvm_free_physmem_slot(&new, &old);
 746                 goto raced;
 747         }
 748
 749         r = -EAGAIN;
 750         if (kvm->busy)
 751                 goto out_unlock;
 752
 753         if (mem->slot >= kvm->nmemslots)
 754                 kvm->nmemslots = mem->slot + 1;
 755
 756         *memslot = new;
 757         ++kvm->memory_config_version;
 758
 759         spin_unlock(&kvm->lock);
 760
 761         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 762                 struct kvm_vcpu *vcpu;
 763
 764                 vcpu = vcpu_load_slot(kvm, i);
 765                 if (!vcpu)
 766                         continue;
 767                 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
 768                         do_remove_write_access(vcpu, mem->slot);
 769                 kvm_mmu_reset_context(vcpu);
 770                 vcpu_put(vcpu);
 771         }
 772
 773         kvm_free_physmem_slot(&old, &new);
 774         return 0;
 775
 776 out_unlock:
 777         spin_unlock(&kvm->lock);
 778 out_free:
 779         kvm_free_physmem_slot(&new, &old);
 780 out:
 781         return r;
 782 }
 783
 784 /*
 785  * Get (and clear) the dirty memory log for a memory slot.
 786  */
 787 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 788                                       struct kvm_dirty_log *log)
 789 {
 790         struct kvm_memory_slot *memslot;
 791         int r, i;
 792         int n;
 793         int cleared;
 794         unsigned long any = 0;
 795
 796         spin_lock(&kvm->lock);
 797
 798         /*
 799          * Prevent changes to guest memory configuration even while the lock
 800          * is not taken.
 801          */
 802         ++kvm->busy;
 803         spin_unlock(&kvm->lock);
 804         r = -EINVAL;
 805         if (log->slot >= KVM_MEMORY_SLOTS)
 806                 goto out;
 807
 808         memslot = &kvm->memslots[log->slot];
 809         r = -ENOENT;
 810         if (!memslot->dirty_bitmap)
 811                 goto out;
 812
 813         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 814
 815         for (i = 0; !any && i < n/sizeof(long); ++i)
 816                 any = memslot->dirty_bitmap[i];
 817
 818         r = -EFAULT;
 819         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 820                 goto out;
 821
 822         if (any) {
 823                 cleared = 0;
 824                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 825                         struct kvm_vcpu *vcpu;
 826
 827                         vcpu = vcpu_load_slot(kvm, i);
 828                         if (!vcpu)
 829                                 continue;
 830                         if (!cleared) {
 831                                 do_remove_write_access(vcpu, log->slot);
 832                                 memset(memslot->dirty_bitmap, 0, n);
 833                                 cleared = 1;
 834                         }
 835                         kvm_arch_ops->tlb_flush(vcpu);
 836                         vcpu_put(vcpu);
 837                 }
 838         }
 839
 840         r = 0;
 841
 842 out:
 843         spin_lock(&kvm->lock);
 844         --kvm->busy;
 845         spin_unlock(&kvm->lock);
 846         return r;
 847 }
 848
 849 /*
 850  * Set a new alias region.  Aliases map a portion of physical memory into
 851  * another portion.  This is useful for memory windows, for example the PC
 852  * VGA region.
 853  */
 854 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 855                                          struct kvm_memory_alias *alias)
 856 {
 857         int r, n;
 858         struct kvm_mem_alias *p;
 859
 860         r = -EINVAL;
 861         /* General sanity checks */
 862         if (alias->memory_size & (PAGE_SIZE - 1))
 863                 goto out;
 864         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 865                 goto out;
 866         if (alias->slot >= KVM_ALIAS_SLOTS)
 867                 goto out;
 868         if (alias->guest_phys_addr + alias->memory_size
 869             < alias->guest_phys_addr)
 870                 goto out;
 871         if (alias->target_phys_addr + alias->memory_size
 872             < alias->target_phys_addr)
 873                 goto out;
 874
 875         spin_lock(&kvm->lock);
 876
 877         p = &kvm->aliases[alias->slot];
 878         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 879         p->npages = alias->memory_size >> PAGE_SHIFT;
 880         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 881
 882         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 883                 if (kvm->aliases[n - 1].npages)
 884                         break;
 885         kvm->naliases = n;
 886
 887         spin_unlock(&kvm->lock);
 888
 889         vcpu_load(&kvm->vcpus[0]);
 890         spin_lock(&kvm->lock);
 891         kvm_mmu_zap_all(&kvm->vcpus[0]);
 892         spin_unlock(&kvm->lock);
 893         vcpu_put(&kvm->vcpus[0]);
 894
 895         return 0;
 896
 897 out:
 898         return r;
 899 }
 900
 901 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 902 {
 903         int i;
 904         struct kvm_mem_alias *alias;
 905
 906         for (i = 0; i < kvm->naliases; ++i) {
 907                 alias = &kvm->aliases[i];
 908                 if (gfn >= alias->base_gfn
 909                     && gfn < alias->base_gfn + alias->npages)
 910                         return alias->target_gfn + gfn - alias->base_gfn;
 911         }
 912         return gfn;
 913 }
 914
 915 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 916 {
 917         int i;
 918
 919         for (i = 0; i < kvm->nmemslots; ++i) {
 920                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 921
 922                 if (gfn >= memslot->base_gfn
 923                     && gfn < memslot->base_gfn + memslot->npages)
 924                         return memslot;
 925         }
 926         return NULL;
 927 }
 928
 929 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 930 {
 931         gfn = unalias_gfn(kvm, gfn);
 932         return __gfn_to_memslot(kvm, gfn);
 933 }
 934
 935 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 936 {
 937         struct kvm_memory_slot *slot;
 938
 939         gfn = unalias_gfn(kvm, gfn);
 940         slot = __gfn_to_memslot(kvm, gfn);
 941         if (!slot)
 942                 return NULL;
 943         return slot->phys_mem[gfn - slot->base_gfn];
 944 }
 945 EXPORT_SYMBOL_GPL(gfn_to_page);
 946
 947 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 948 {
 949         int i;
 950         struct kvm_memory_slot *memslot = NULL;
 951         unsigned long rel_gfn;
 952
 953         for (i = 0; i < kvm->nmemslots; ++i) {
 954                 memslot = &kvm->memslots[i];
 955
 956                 if (gfn >= memslot->base_gfn
 957                     && gfn < memslot->base_gfn + memslot->npages) {
 958
 959                         if (!memslot || !memslot->dirty_bitmap)
 960                                 return;
 961
 962                         rel_gfn = gfn - memslot->base_gfn;
 963
 964                         /* avoid RMW */
 965                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 966                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 967                         return;
 968                 }
 969         }
 970 }
 971
 972 static int emulator_read_std(unsigned long addr,
 973                              unsigned long *val,
 974                              unsigned int bytes,
 975                              struct x86_emulate_ctxt *ctxt)
 976 {
 977         struct kvm_vcpu *vcpu = ctxt->vcpu;
 978         void *data = val;
 979
 980         while (bytes) {
 981                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 982                 unsigned offset = addr & (PAGE_SIZE-1);
 983                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 984                 unsigned long pfn;
 985                 struct page *page;
 986                 void *page_virt;
 987
 988                 if (gpa == UNMAPPED_GVA)
 989                         return X86EMUL_PROPAGATE_FAULT;
 990                 pfn = gpa >> PAGE_SHIFT;
 991                 page = gfn_to_page(vcpu->kvm, pfn);
 992                 if (!page)
 993                         return X86EMUL_UNHANDLEABLE;
 994                 page_virt = kmap_atomic(page, KM_USER0);
 995
 996                 memcpy(data, page_virt + offset, tocopy);
 997
 998                 kunmap_atomic(page_virt, KM_USER0);
 999
1000                 bytes -= tocopy;
1001                 data += tocopy;
1002                 addr += tocopy;
1003         }
1004
1005         return X86EMUL_CONTINUE;
1006 }
1007
1008 static int emulator_write_std(unsigned long addr,
1009                               unsigned long val,
1010                               unsigned int bytes,
1011                               struct x86_emulate_ctxt *ctxt)
1012 {
1013         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1014                addr, bytes);
1015         return X86EMUL_UNHANDLEABLE;
1016 }
1017
1018 static int emulator_read_emulated(unsigned long addr,
1019                                   unsigned long *val,
1020                                   unsigned int bytes,
1021                                   struct x86_emulate_ctxt *ctxt)
1022 {
1023         struct kvm_vcpu *vcpu = ctxt->vcpu;
1024
1025         if (vcpu->mmio_read_completed) {
1026                 memcpy(val, vcpu->mmio_data, bytes);
1027                 vcpu->mmio_read_completed = 0;
1028                 return X86EMUL_CONTINUE;
1029         } else if (emulator_read_std(addr, val, bytes, ctxt)
1030                    == X86EMUL_CONTINUE)
1031                 return X86EMUL_CONTINUE;
1032         else {
1033                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1034
1035                 if (gpa == UNMAPPED_GVA)
1036                         return X86EMUL_PROPAGATE_FAULT;
1037                 vcpu->mmio_needed = 1;
1038                 vcpu->mmio_phys_addr = gpa;
1039                 vcpu->mmio_size = bytes;
1040                 vcpu->mmio_is_write = 0;
1041
1042                 return X86EMUL_UNHANDLEABLE;
1043         }
1044 }
1045
1046 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1047                                unsigned long val, int bytes)
1048 {
1049         struct page *page;
1050         void *virt;
1051
1052         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1053                 return 0;
1054         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1055         if (!page)
1056                 return 0;
1057         kvm_mmu_pre_write(vcpu, gpa, bytes);
1058         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1059         virt = kmap_atomic(page, KM_USER0);
1060         memcpy(virt + offset_in_page(gpa), &val, bytes);
1061         kunmap_atomic(virt, KM_USER0);
1062         kvm_mmu_post_write(vcpu, gpa, bytes);
1063         return 1;
1064 }
1065
1066 static int emulator_write_emulated(unsigned long addr,
1067                                    unsigned long val,
1068                                    unsigned int bytes,
1069                                    struct x86_emulate_ctxt *ctxt)
1070 {
1071         struct kvm_vcpu *vcpu = ctxt->vcpu;
1072         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1073
1074         if (gpa == UNMAPPED_GVA) {
1075                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1076                 return X86EMUL_PROPAGATE_FAULT;
1077         }
1078
1079         if (emulator_write_phys(vcpu, gpa, val, bytes))
1080                 return X86EMUL_CONTINUE;
1081
1082         vcpu->mmio_needed = 1;
1083         vcpu->mmio_phys_addr = gpa;
1084         vcpu->mmio_size = bytes;
1085         vcpu->mmio_is_write = 1;
1086         memcpy(vcpu->mmio_data, &val, bytes);
1087
1088         return X86EMUL_CONTINUE;
1089 }
1090
1091 static int emulator_cmpxchg_emulated(unsigned long addr,
1092                                      unsigned long old,
1093                                      unsigned long new,
1094                                      unsigned int bytes,
1095                                      struct x86_emulate_ctxt *ctxt)
1096 {
1097         static int reported;
1098
1099         if (!reported) {
1100                 reported = 1;
1101                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1102         }
1103         return emulator_write_emulated(addr, new, bytes, ctxt);
1104 }
1105
1106 #ifdef CONFIG_X86_32
1107
1108 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1109                                        unsigned long old_lo,
1110                                        unsigned long old_hi,
1111                                        unsigned long new_lo,
1112                                        unsigned long new_hi,
1113                                        struct x86_emulate_ctxt *ctxt)
1114 {
1115         static int reported;
1116         int r;
1117
1118         if (!reported) {
1119                 reported = 1;
1120                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1121         }
1122         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1123         if (r != X86EMUL_CONTINUE)
1124                 return r;
1125         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1126 }
1127
1128 #endif
1129
1130 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1131 {
1132         return kvm_arch_ops->get_segment_base(vcpu, seg);
1133 }
1134
1135 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1136 {
1137         return X86EMUL_CONTINUE;
1138 }
1139
1140 int emulate_clts(struct kvm_vcpu *vcpu)
1141 {
1142         unsigned long cr0;
1143
1144         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1145         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1146         kvm_arch_ops->set_cr0(vcpu, cr0);
1147         return X86EMUL_CONTINUE;
1148 }
1149
1150 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1151 {
1152         struct kvm_vcpu *vcpu = ctxt->vcpu;
1153
1154         switch (dr) {
1155         case 0 ... 3:
1156                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1157                 return X86EMUL_CONTINUE;
1158         default:
1159                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1160                        __FUNCTION__, dr);
1161                 return X86EMUL_UNHANDLEABLE;
1162         }
1163 }
1164
1165 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1166 {
1167         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1168         int exception;
1169
1170         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1171         if (exception) {
1172                 /* FIXME: better handling */
1173                 return X86EMUL_UNHANDLEABLE;
1174         }
1175         return X86EMUL_CONTINUE;
1176 }
1177
1178 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1179 {
1180         static int reported;
1181         u8 opcodes[4];
1182         unsigned long rip = ctxt->vcpu->rip;
1183         unsigned long rip_linear;
1184
1185         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1186
1187         if (reported)
1188                 return;
1189
1190         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1191
1192         printk(KERN_ERR "emulation failed but !mmio_needed?"
1193                " rip %lx %02x %02x %02x %02x\n",
1194                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1195         reported = 1;
1196 }
1197
1198 struct x86_emulate_ops emulate_ops = {
1199         .read_std            = emulator_read_std,
1200         .write_std           = emulator_write_std,
1201         .read_emulated       = emulator_read_emulated,
1202         .write_emulated      = emulator_write_emulated,
1203         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1204 #ifdef CONFIG_X86_32
1205         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1206 #endif
1207 };
1208
1209 int emulate_instruction(struct kvm_vcpu *vcpu,
1210                         struct kvm_run *run,
1211                         unsigned long cr2,
1212                         u16 error_code)
1213 {
1214         struct x86_emulate_ctxt emulate_ctxt;
1215         int r;
1216         int cs_db, cs_l;
1217
1218         kvm_arch_ops->cache_regs(vcpu);
1219
1220         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1221
1222         emulate_ctxt.vcpu = vcpu;
1223         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1224         emulate_ctxt.cr2 = cr2;
1225         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1226                 ? X86EMUL_MODE_REAL : cs_l
1227                 ? X86EMUL_MODE_PROT64 : cs_db
1228                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1229
1230         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1231                 emulate_ctxt.cs_base = 0;
1232                 emulate_ctxt.ds_base = 0;
1233                 emulate_ctxt.es_base = 0;
1234                 emulate_ctxt.ss_base = 0;
1235         } else {
1236                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1237                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1238                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1239                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1240         }
1241
1242         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1243         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1244
1245         vcpu->mmio_is_write = 0;
1246         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1247
1248         if ((r || vcpu->mmio_is_write) && run) {
1249                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1250                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1251                 run->mmio.len = vcpu->mmio_size;
1252                 run->mmio.is_write = vcpu->mmio_is_write;
1253         }
1254
1255         if (r) {
1256                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1257                         return EMULATE_DONE;
1258                 if (!vcpu->mmio_needed) {
1259                         report_emulation_failure(&emulate_ctxt);
1260                         return EMULATE_FAIL;
1261                 }
1262                 return EMULATE_DO_MMIO;
1263         }
1264
1265         kvm_arch_ops->decache_regs(vcpu);
1266         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1267
1268         if (vcpu->mmio_is_write)
1269                 return EMULATE_DO_MMIO;
1270
1271         return EMULATE_DONE;
1272 }
1273 EXPORT_SYMBOL_GPL(emulate_instruction);
1274
1275 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1276 {
1277         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1278
1279         kvm_arch_ops->cache_regs(vcpu);
1280         ret = -KVM_EINVAL;
1281 #ifdef CONFIG_X86_64
1282         if (is_long_mode(vcpu)) {
1283                 nr = vcpu->regs[VCPU_REGS_RAX];
1284                 a0 = vcpu->regs[VCPU_REGS_RDI];
1285                 a1 = vcpu->regs[VCPU_REGS_RSI];
1286                 a2 = vcpu->regs[VCPU_REGS_RDX];
1287                 a3 = vcpu->regs[VCPU_REGS_RCX];
1288                 a4 = vcpu->regs[VCPU_REGS_R8];
1289                 a5 = vcpu->regs[VCPU_REGS_R9];
1290         } else
1291 #endif
1292         {
1293                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1294                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1295                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1296                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1297                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1298                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1299                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1300         }
1301         switch (nr) {
1302         default:
1303                 run->hypercall.args[0] = a0;
1304                 run->hypercall.args[1] = a1;
1305                 run->hypercall.args[2] = a2;
1306                 run->hypercall.args[3] = a3;
1307                 run->hypercall.args[4] = a4;
1308                 run->hypercall.args[5] = a5;
1309                 run->hypercall.ret = ret;
1310                 run->hypercall.longmode = is_long_mode(vcpu);
1311                 kvm_arch_ops->decache_regs(vcpu);
1312                 return 0;
1313         }
1314         vcpu->regs[VCPU_REGS_RAX] = ret;
1315         kvm_arch_ops->decache_regs(vcpu);
1316         return 1;
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_hypercall);
1319
1320 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1321 {
1322         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1323 }
1324
1325 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1326 {
1327         struct descriptor_table dt = { limit, base };
1328
1329         kvm_arch_ops->set_gdt(vcpu, &dt);
1330 }
1331
1332 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1333 {
1334         struct descriptor_table dt = { limit, base };
1335
1336         kvm_arch_ops->set_idt(vcpu, &dt);
1337 }
1338
1339 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1340                    unsigned long *rflags)
1341 {
1342         lmsw(vcpu, msw);
1343         *rflags = kvm_arch_ops->get_rflags(vcpu);
1344 }
1345
1346 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1347 {
1348         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1349         switch (cr) {
1350         case 0:
1351                 return vcpu->cr0;
1352         case 2:
1353                 return vcpu->cr2;
1354         case 3:
1355                 return vcpu->cr3;
1356         case 4:
1357                 return vcpu->cr4;
1358         default:
1359                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1360                 return 0;
1361         }
1362 }
1363
1364 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1365                      unsigned long *rflags)
1366 {
1367         switch (cr) {
1368         case 0:
1369                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1370                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1371                 break;
1372         case 2:
1373                 vcpu->cr2 = val;
1374                 break;
1375         case 3:
1376                 set_cr3(vcpu, val);
1377                 break;
1378         case 4:
1379                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1380                 break;
1381         default:
1382                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1383         }
1384 }
1385
1386 /*
1387  * Register the para guest with the host:
1388  */
1389 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1390 {
1391         struct kvm_vcpu_para_state *para_state;
1392         hpa_t para_state_hpa, hypercall_hpa;
1393         struct page *para_state_page;
1394         unsigned char *hypercall;
1395         gpa_t hypercall_gpa;
1396
1397         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1398         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1399
1400         /*
1401          * Needs to be page aligned:
1402          */
1403         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1404                 goto err_gp;
1405
1406         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1407         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1408         if (is_error_hpa(para_state_hpa))
1409                 goto err_gp;
1410
1411         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1412         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1413         para_state = kmap_atomic(para_state_page, KM_USER0);
1414
1415         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1416         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1417
1418         para_state->host_version = KVM_PARA_API_VERSION;
1419         /*
1420          * We cannot support guests that try to register themselves
1421          * with a newer API version than the host supports:
1422          */
1423         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1424                 para_state->ret = -KVM_EINVAL;
1425                 goto err_kunmap_skip;
1426         }
1427
1428         hypercall_gpa = para_state->hypercall_gpa;
1429         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1430         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1431         if (is_error_hpa(hypercall_hpa)) {
1432                 para_state->ret = -KVM_EINVAL;
1433                 goto err_kunmap_skip;
1434         }
1435
1436         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1437         vcpu->para_state_page = para_state_page;
1438         vcpu->para_state_gpa = para_state_gpa;
1439         vcpu->hypercall_gpa = hypercall_gpa;
1440
1441         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1442         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1443                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1444         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1445         kunmap_atomic(hypercall, KM_USER1);
1446
1447         para_state->ret = 0;
1448 err_kunmap_skip:
1449         kunmap_atomic(para_state, KM_USER0);
1450         return 0;
1451 err_gp:
1452         return 1;
1453 }
1454
1455 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1456 {
1457         u64 data;
1458
1459         switch (msr) {
1460         case 0xc0010010: /* SYSCFG */
1461         case 0xc0010015: /* HWCR */
1462         case MSR_IA32_PLATFORM_ID:
1463         case MSR_IA32_P5_MC_ADDR:
1464         case MSR_IA32_P5_MC_TYPE:
1465         case MSR_IA32_MC0_CTL:
1466         case MSR_IA32_MCG_STATUS:
1467         case MSR_IA32_MCG_CAP:
1468         case MSR_IA32_MC0_MISC:
1469         case MSR_IA32_MC0_MISC+4:
1470         case MSR_IA32_MC0_MISC+8:
1471         case MSR_IA32_MC0_MISC+12:
1472         case MSR_IA32_MC0_MISC+16:
1473         case MSR_IA32_UCODE_REV:
1474         case MSR_IA32_PERF_STATUS:
1475                 /* MTRR registers */
1476         case 0xfe:
1477         case 0x200 ... 0x2ff:
1478                 data = 0;
1479                 break;
1480         case 0xcd: /* fsb frequency */
1481                 data = 3;
1482                 break;
1483         case MSR_IA32_APICBASE:
1484                 data = vcpu->apic_base;
1485                 break;
1486         case MSR_IA32_MISC_ENABLE:
1487                 data = vcpu->ia32_misc_enable_msr;
1488                 break;
1489 #ifdef CONFIG_X86_64
1490         case MSR_EFER:
1491                 data = vcpu->shadow_efer;
1492                 break;
1493 #endif
1494         default:
1495                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1496                 return 1;
1497         }
1498         *pdata = data;
1499         return 0;
1500 }
1501 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1502
1503 /*
1504  * Reads an msr value (of 'msr_index') into 'pdata'.
1505  * Returns 0 on success, non-0 otherwise.
1506  * Assumes vcpu_load() was already called.
1507  */
1508 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1509 {
1510         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1511 }
1512
1513 #ifdef CONFIG_X86_64
1514
1515 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1516 {
1517         if (efer & EFER_RESERVED_BITS) {
1518                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1519                        efer);
1520                 inject_gp(vcpu);
1521                 return;
1522         }
1523
1524         if (is_paging(vcpu)
1525             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1526                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1527                 inject_gp(vcpu);
1528                 return;
1529         }
1530
1531         kvm_arch_ops->set_efer(vcpu, efer);
1532
1533         efer &= ~EFER_LMA;
1534         efer |= vcpu->shadow_efer & EFER_LMA;
1535
1536         vcpu->shadow_efer = efer;
1537 }
1538
1539 #endif
1540
1541 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1542 {
1543         switch (msr) {
1544 #ifdef CONFIG_X86_64
1545         case MSR_EFER:
1546                 set_efer(vcpu, data);
1547                 break;
1548 #endif
1549         case MSR_IA32_MC0_STATUS:
1550                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1551                        __FUNCTION__, data);
1552                 break;
1553         case MSR_IA32_MCG_STATUS:
1554                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1555                         __FUNCTION__, data);
1556                 break;
1557         case MSR_IA32_UCODE_REV:
1558         case MSR_IA32_UCODE_WRITE:
1559         case 0x200 ... 0x2ff: /* MTRRs */
1560                 break;
1561         case MSR_IA32_APICBASE:
1562                 vcpu->apic_base = data;
1563                 break;
1564         case MSR_IA32_MISC_ENABLE:
1565                 vcpu->ia32_misc_enable_msr = data;
1566                 break;
1567         /*
1568          * This is the 'probe whether the host is KVM' logic:
1569          */
1570         case MSR_KVM_API_MAGIC:
1571                 return vcpu_register_para(vcpu, data);
1572
1573         default:
1574                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1575                 return 1;
1576         }
1577         return 0;
1578 }
1579 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1580
1581 /*
1582  * Writes msr value into into the appropriate "register".
1583  * Returns 0 on success, non-0 otherwise.
1584  * Assumes vcpu_load() was already called.
1585  */
1586 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1587 {
1588         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1589 }
1590
1591 void kvm_resched(struct kvm_vcpu *vcpu)
1592 {
1593         vcpu_put(vcpu);
1594         cond_resched();
1595         vcpu_load(vcpu);
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_resched);
1598
1599 void load_msrs(struct vmx_msr_entry *e, int n)
1600 {
1601         int i;
1602
1603         for (i = 0; i < n; ++i)
1604                 wrmsrl(e[i].index, e[i].data);
1605 }
1606 EXPORT_SYMBOL_GPL(load_msrs);
1607
1608 void save_msrs(struct vmx_msr_entry *e, int n)
1609 {
1610         int i;
1611
1612         for (i = 0; i < n; ++i)
1613                 rdmsrl(e[i].index, e[i].data);
1614 }
1615 EXPORT_SYMBOL_GPL(save_msrs);
1616
1617 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1618 {
1619         int i;
1620         u32 function;
1621         struct kvm_cpuid_entry *e, *best;
1622
1623         kvm_arch_ops->cache_regs(vcpu);
1624         function = vcpu->regs[VCPU_REGS_RAX];
1625         vcpu->regs[VCPU_REGS_RAX] = 0;
1626         vcpu->regs[VCPU_REGS_RBX] = 0;
1627         vcpu->regs[VCPU_REGS_RCX] = 0;
1628         vcpu->regs[VCPU_REGS_RDX] = 0;
1629         best = NULL;
1630         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1631                 e = &vcpu->cpuid_entries[i];
1632                 if (e->function == function) {
1633                         best = e;
1634                         break;
1635                 }
1636                 /*
1637                  * Both basic or both extended?
1638                  */
1639                 if (((e->function ^ function) & 0x80000000) == 0)
1640                         if (!best || e->function > best->function)
1641                                 best = e;
1642         }
1643         if (best) {
1644                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1645                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1646                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1647                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1648         }
1649         kvm_arch_ops->decache_regs(vcpu);
1650         kvm_arch_ops->skip_emulated_instruction(vcpu);
1651 }
1652 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1653
1654 static int pio_copy_data(struct kvm_vcpu *vcpu)
1655 {
1656         void *p = vcpu->pio_data;
1657         void *q;
1658         unsigned bytes;
1659         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1660
1661         kvm_arch_ops->vcpu_put(vcpu);
1662         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1663                  PAGE_KERNEL);
1664         if (!q) {
1665                 kvm_arch_ops->vcpu_load(vcpu);
1666                 free_pio_guest_pages(vcpu);
1667                 return -ENOMEM;
1668         }
1669         q += vcpu->pio.guest_page_offset;
1670         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1671         if (vcpu->pio.in)
1672                 memcpy(q, p, bytes);
1673         else
1674                 memcpy(p, q, bytes);
1675         q -= vcpu->pio.guest_page_offset;
1676         vunmap(q);
1677         kvm_arch_ops->vcpu_load(vcpu);
1678         free_pio_guest_pages(vcpu);
1679         return 0;
1680 }
1681
1682 static int complete_pio(struct kvm_vcpu *vcpu)
1683 {
1684         struct kvm_pio_request *io = &vcpu->pio;
1685         long delta;
1686         int r;
1687
1688         kvm_arch_ops->cache_regs(vcpu);
1689
1690         if (!io->string) {
1691                 if (io->in)
1692                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1693                                io->size);
1694         } else {
1695                 if (io->in) {
1696                         r = pio_copy_data(vcpu);
1697                         if (r) {
1698                                 kvm_arch_ops->cache_regs(vcpu);
1699                                 return r;
1700                         }
1701                 }
1702
1703                 delta = 1;
1704                 if (io->rep) {
1705                         delta *= io->cur_count;
1706                         /*
1707                          * The size of the register should really depend on
1708                          * current address size.
1709                          */
1710                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1711                 }
1712                 if (io->down)
1713                         delta = -delta;
1714                 delta *= io->size;
1715                 if (io->in)
1716                         vcpu->regs[VCPU_REGS_RDI] += delta;
1717                 else
1718                         vcpu->regs[VCPU_REGS_RSI] += delta;
1719         }
1720
1721         vcpu->run->io_completed = 0;
1722
1723         kvm_arch_ops->decache_regs(vcpu);
1724
1725         io->count -= io->cur_count;
1726         io->cur_count = 0;
1727
1728         if (!io->count)
1729                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1730         return 0;
1731 }
1732
1733 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1734                   int size, unsigned long count, int string, int down,
1735                   gva_t address, int rep, unsigned port)
1736 {
1737         unsigned now, in_page;
1738         int i;
1739         int nr_pages = 1;
1740         struct page *page;
1741
1742         vcpu->run->exit_reason = KVM_EXIT_IO;
1743         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1744         vcpu->run->io.size = size;
1745         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1746         vcpu->run->io.count = count;
1747         vcpu->run->io.port = port;
1748         vcpu->pio.count = count;
1749         vcpu->pio.cur_count = count;
1750         vcpu->pio.size = size;
1751         vcpu->pio.in = in;
1752         vcpu->pio.string = string;
1753         vcpu->pio.down = down;
1754         vcpu->pio.guest_page_offset = offset_in_page(address);
1755         vcpu->pio.rep = rep;
1756
1757         if (!string) {
1758                 kvm_arch_ops->cache_regs(vcpu);
1759                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1760                 kvm_arch_ops->decache_regs(vcpu);
1761                 return 0;
1762         }
1763
1764         if (!count) {
1765                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1766                 return 1;
1767         }
1768
1769         now = min(count, PAGE_SIZE / size);
1770
1771         if (!down)
1772                 in_page = PAGE_SIZE - offset_in_page(address);
1773         else
1774                 in_page = offset_in_page(address) + size;
1775         now = min(count, (unsigned long)in_page / size);
1776         if (!now) {
1777                 /*
1778                  * String I/O straddles page boundary.  Pin two guest pages
1779                  * so that we satisfy atomicity constraints.  Do just one
1780                  * transaction to avoid complexity.
1781                  */
1782                 nr_pages = 2;
1783                 now = 1;
1784         }
1785         if (down) {
1786                 /*
1787                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1788                  */
1789                 printk(KERN_ERR "kvm: guest string pio down\n");
1790                 inject_gp(vcpu);
1791                 return 1;
1792         }
1793         vcpu->run->io.count = now;
1794         vcpu->pio.cur_count = now;
1795
1796         for (i = 0; i < nr_pages; ++i) {
1797                 spin_lock(&vcpu->kvm->lock);
1798                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1799                 if (page)
1800                         get_page(page);
1801                 vcpu->pio.guest_pages[i] = page;
1802                 spin_unlock(&vcpu->kvm->lock);
1803                 if (!page) {
1804                         inject_gp(vcpu);
1805                         free_pio_guest_pages(vcpu);
1806                         return 1;
1807                 }
1808         }
1809
1810         if (!vcpu->pio.in)
1811                 return pio_copy_data(vcpu);
1812         return 0;
1813 }
1814 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1815
1816 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1817 {
1818         int r;
1819         sigset_t sigsaved;
1820
1821         vcpu_load(vcpu);
1822
1823         if (vcpu->sigset_active)
1824                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1825
1826         /* re-sync apic's tpr */
1827         vcpu->cr8 = kvm_run->cr8;
1828
1829         if (kvm_run->io_completed) {
1830                 if (vcpu->pio.cur_count) {
1831                         r = complete_pio(vcpu);
1832                         if (r)
1833                                 goto out;
1834                 } else {
1835                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1836                         vcpu->mmio_read_completed = 1;
1837                 }
1838         }
1839
1840         vcpu->mmio_needed = 0;
1841
1842         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1843                 kvm_arch_ops->cache_regs(vcpu);
1844                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1845                 kvm_arch_ops->decache_regs(vcpu);
1846         }
1847
1848         r = kvm_arch_ops->run(vcpu, kvm_run);
1849
1850 out:
1851         if (vcpu->sigset_active)
1852                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1853
1854         vcpu_put(vcpu);
1855         return r;
1856 }
1857
1858 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1859                                    struct kvm_regs *regs)
1860 {
1861         vcpu_load(vcpu);
1862
1863         kvm_arch_ops->cache_regs(vcpu);
1864
1865         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1866         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1867         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1868         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1869         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1870         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1871         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1872         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1873 #ifdef CONFIG_X86_64
1874         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1875         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1876         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1877         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1878         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1879         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1880         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1881         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1882 #endif
1883
1884         regs->rip = vcpu->rip;
1885         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1886
1887         /*
1888          * Don't leak debug flags in case they were set for guest debugging
1889          */
1890         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1891                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1892
1893         vcpu_put(vcpu);
1894
1895         return 0;
1896 }
1897
1898 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1899                                    struct kvm_regs *regs)
1900 {
1901         vcpu_load(vcpu);
1902
1903         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1904         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1905         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1906         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1907         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1908         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1909         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1910         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1911 #ifdef CONFIG_X86_64
1912         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1913         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1914         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1915         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1916         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1917         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1918         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1919         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1920 #endif
1921
1922         vcpu->rip = regs->rip;
1923         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1924
1925         kvm_arch_ops->decache_regs(vcpu);
1926
1927         vcpu_put(vcpu);
1928
1929         return 0;
1930 }
1931
1932 static void get_segment(struct kvm_vcpu *vcpu,
1933                         struct kvm_segment *var, int seg)
1934 {
1935         return kvm_arch_ops->get_segment(vcpu, var, seg);
1936 }
1937
1938 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1939                                     struct kvm_sregs *sregs)
1940 {
1941         struct descriptor_table dt;
1942
1943         vcpu_load(vcpu);
1944
1945         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1946         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1947         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1948         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1949         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1950         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1951
1952         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1953         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1954
1955         kvm_arch_ops->get_idt(vcpu, &dt);
1956         sregs->idt.limit = dt.limit;
1957         sregs->idt.base = dt.base;
1958         kvm_arch_ops->get_gdt(vcpu, &dt);
1959         sregs->gdt.limit = dt.limit;
1960         sregs->gdt.base = dt.base;
1961
1962         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1963         sregs->cr0 = vcpu->cr0;
1964         sregs->cr2 = vcpu->cr2;
1965         sregs->cr3 = vcpu->cr3;
1966         sregs->cr4 = vcpu->cr4;
1967         sregs->cr8 = vcpu->cr8;
1968         sregs->efer = vcpu->shadow_efer;
1969         sregs->apic_base = vcpu->apic_base;
1970
1971         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1972                sizeof sregs->interrupt_bitmap);
1973
1974         vcpu_put(vcpu);
1975
1976         return 0;
1977 }
1978
1979 static void set_segment(struct kvm_vcpu *vcpu,
1980                         struct kvm_segment *var, int seg)
1981 {
1982         return kvm_arch_ops->set_segment(vcpu, var, seg);
1983 }
1984
1985 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1986                                     struct kvm_sregs *sregs)
1987 {
1988         int mmu_reset_needed = 0;
1989         int i;
1990         struct descriptor_table dt;
1991
1992         vcpu_load(vcpu);
1993
1994         dt.limit = sregs->idt.limit;
1995         dt.base = sregs->idt.base;
1996         kvm_arch_ops->set_idt(vcpu, &dt);
1997         dt.limit = sregs->gdt.limit;
1998         dt.base = sregs->gdt.base;
1999         kvm_arch_ops->set_gdt(vcpu, &dt);
2000
2001         vcpu->cr2 = sregs->cr2;
2002         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2003         vcpu->cr3 = sregs->cr3;
2004
2005         vcpu->cr8 = sregs->cr8;
2006
2007         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2008 #ifdef CONFIG_X86_64
2009         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2010 #endif
2011         vcpu->apic_base = sregs->apic_base;
2012
2013         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
2014
2015         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2016         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2017
2018         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2019         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2020         if (!is_long_mode(vcpu) && is_pae(vcpu))
2021                 load_pdptrs(vcpu, vcpu->cr3);
2022
2023         if (mmu_reset_needed)
2024                 kvm_mmu_reset_context(vcpu);
2025
2026         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2027                sizeof vcpu->irq_pending);
2028         vcpu->irq_summary = 0;
2029         for (i = 0; i < NR_IRQ_WORDS; ++i)
2030                 if (vcpu->irq_pending[i])
2031                         __set_bit(i, &vcpu->irq_summary);
2032
2033         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2034         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2035         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2036         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2037         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2038         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2039
2040         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2041         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2042
2043         vcpu_put(vcpu);
2044
2045         return 0;
2046 }
2047
2048 /*
2049  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2050  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2051  *
2052  * This list is modified at module load time to reflect the
2053  * capabilities of the host cpu.
2054  */
2055 static u32 msrs_to_save[] = {
2056         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2057         MSR_K6_STAR,
2058 #ifdef CONFIG_X86_64
2059         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2060 #endif
2061         MSR_IA32_TIME_STAMP_COUNTER,
2062 };
2063
2064 static unsigned num_msrs_to_save;
2065
2066 static u32 emulated_msrs[] = {
2067         MSR_IA32_MISC_ENABLE,
2068 };
2069
2070 static __init void kvm_init_msr_list(void)
2071 {
2072         u32 dummy[2];
2073         unsigned i, j;
2074
2075         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2076                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2077                         continue;
2078                 if (j < i)
2079                         msrs_to_save[j] = msrs_to_save[i];
2080                 j++;
2081         }
2082         num_msrs_to_save = j;
2083 }
2084
2085 /*
2086  * Adapt set_msr() to msr_io()'s calling convention
2087  */
2088 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2089 {
2090         return set_msr(vcpu, index, *data);
2091 }
2092
2093 /*
2094  * Read or write a bunch of msrs. All parameters are kernel addresses.
2095  *
2096  * @return number of msrs set successfully.
2097  */
2098 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2099                     struct kvm_msr_entry *entries,
2100                     int (*do_msr)(struct kvm_vcpu *vcpu,
2101                                   unsigned index, u64 *data))
2102 {
2103         int i;
2104
2105         vcpu_load(vcpu);
2106
2107         for (i = 0; i < msrs->nmsrs; ++i)
2108                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2109                         break;
2110
2111         vcpu_put(vcpu);
2112
2113         return i;
2114 }
2115
2116 /*
2117  * Read or write a bunch of msrs. Parameters are user addresses.
2118  *
2119  * @return number of msrs set successfully.
2120  */
2121 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2122                   int (*do_msr)(struct kvm_vcpu *vcpu,
2123                                 unsigned index, u64 *data),
2124                   int writeback)
2125 {
2126         struct kvm_msrs msrs;
2127         struct kvm_msr_entry *entries;
2128         int r, n;
2129         unsigned size;
2130
2131         r = -EFAULT;
2132         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2133                 goto out;
2134
2135         r = -E2BIG;
2136         if (msrs.nmsrs >= MAX_IO_MSRS)
2137                 goto out;
2138
2139         r = -ENOMEM;
2140         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2141         entries = vmalloc(size);
2142         if (!entries)
2143                 goto out;
2144
2145         r = -EFAULT;
2146         if (copy_from_user(entries, user_msrs->entries, size))
2147                 goto out_free;
2148
2149         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2150         if (r < 0)
2151                 goto out_free;
2152
2153         r = -EFAULT;
2154         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2155                 goto out_free;
2156
2157         r = n;
2158
2159 out_free:
2160         vfree(entries);
2161 out:
2162         return r;
2163 }
2164
2165 /*
2166  * Translate a guest virtual address to a guest physical address.
2167  */
2168 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2169                                     struct kvm_translation *tr)
2170 {
2171         unsigned long vaddr = tr->linear_address;
2172         gpa_t gpa;
2173
2174         vcpu_load(vcpu);
2175         spin_lock(&vcpu->kvm->lock);
2176         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2177         tr->physical_address = gpa;
2178         tr->valid = gpa != UNMAPPED_GVA;
2179         tr->writeable = 1;
2180         tr->usermode = 0;
2181         spin_unlock(&vcpu->kvm->lock);
2182         vcpu_put(vcpu);
2183
2184         return 0;
2185 }
2186
2187 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2188                                     struct kvm_interrupt *irq)
2189 {
2190         if (irq->irq < 0 || irq->irq >= 256)
2191                 return -EINVAL;
2192         vcpu_load(vcpu);
2193
2194         set_bit(irq->irq, vcpu->irq_pending);
2195         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2196
2197         vcpu_put(vcpu);
2198
2199         return 0;
2200 }
2201
2202 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2203                                       struct kvm_debug_guest *dbg)
2204 {
2205         int r;
2206
2207         vcpu_load(vcpu);
2208
2209         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2210
2211         vcpu_put(vcpu);
2212
2213         return r;
2214 }
2215
2216 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2217                                     unsigned long address,
2218                                     int *type)
2219 {
2220         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2221         unsigned long pgoff;
2222         struct page *page;
2223
2224         *type = VM_FAULT_MINOR;
2225         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2226         if (pgoff == 0)
2227                 page = virt_to_page(vcpu->run);
2228         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2229                 page = virt_to_page(vcpu->pio_data);
2230         else
2231                 return NOPAGE_SIGBUS;
2232         get_page(page);
2233         return page;
2234 }
2235
2236 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2237         .nopage = kvm_vcpu_nopage,
2238 };
2239
2240 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2241 {
2242         vma->vm_ops = &kvm_vcpu_vm_ops;
2243         return 0;
2244 }
2245
2246 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2247 {
2248         struct kvm_vcpu *vcpu = filp->private_data;
2249
2250         fput(vcpu->kvm->filp);
2251         return 0;
2252 }
2253
2254 static struct file_operations kvm_vcpu_fops = {
2255         .release        = kvm_vcpu_release,
2256         .unlocked_ioctl = kvm_vcpu_ioctl,
2257         .compat_ioctl   = kvm_vcpu_ioctl,
2258         .mmap           = kvm_vcpu_mmap,
2259 };
2260
2261 /*
2262  * Allocates an inode for the vcpu.
2263  */
2264 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2265 {
2266         int fd, r;
2267         struct inode *inode;
2268         struct file *file;
2269
2270         atomic_inc(&vcpu->kvm->filp->f_count);
2271         inode = kvmfs_inode(&kvm_vcpu_fops);
2272         if (IS_ERR(inode)) {
2273                 r = PTR_ERR(inode);
2274                 goto out1;
2275         }
2276
2277         file = kvmfs_file(inode, vcpu);
2278         if (IS_ERR(file)) {
2279                 r = PTR_ERR(file);
2280                 goto out2;
2281         }
2282
2283         r = get_unused_fd();
2284         if (r < 0)
2285                 goto out3;
2286         fd = r;
2287         fd_install(fd, file);
2288
2289         return fd;
2290
2291 out3:
2292         fput(file);
2293 out2:
2294         iput(inode);
2295 out1:
2296         fput(vcpu->kvm->filp);
2297         return r;
2298 }
2299
2300 /*
2301  * Creates some virtual cpus.  Good luck creating more than one.
2302  */
2303 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2304 {
2305         int r;
2306         struct kvm_vcpu *vcpu;
2307         struct page *page;
2308
2309         r = -EINVAL;
2310         if (!valid_vcpu(n))
2311                 goto out;
2312
2313         vcpu = &kvm->vcpus[n];
2314
2315         mutex_lock(&vcpu->mutex);
2316
2317         if (vcpu->vmcs) {
2318                 mutex_unlock(&vcpu->mutex);
2319                 return -EEXIST;
2320         }
2321
2322         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2323         r = -ENOMEM;
2324         if (!page)
2325                 goto out_unlock;
2326         vcpu->run = page_address(page);
2327
2328         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2329         r = -ENOMEM;
2330         if (!page)
2331                 goto out_free_run;
2332         vcpu->pio_data = page_address(page);
2333
2334         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2335                                            FX_IMAGE_ALIGN);
2336         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2337         vcpu->cr0 = 0x10;
2338
2339         r = kvm_arch_ops->vcpu_create(vcpu);
2340         if (r < 0)
2341                 goto out_free_vcpus;
2342
2343         r = kvm_mmu_create(vcpu);
2344         if (r < 0)
2345                 goto out_free_vcpus;
2346
2347         kvm_arch_ops->vcpu_load(vcpu);
2348         r = kvm_mmu_setup(vcpu);
2349         if (r >= 0)
2350                 r = kvm_arch_ops->vcpu_setup(vcpu);
2351         vcpu_put(vcpu);
2352
2353         if (r < 0)
2354                 goto out_free_vcpus;
2355
2356         r = create_vcpu_fd(vcpu);
2357         if (r < 0)
2358                 goto out_free_vcpus;
2359
2360         return r;
2361
2362 out_free_vcpus:
2363         kvm_free_vcpu(vcpu);
2364 out_free_run:
2365         free_page((unsigned long)vcpu->run);
2366         vcpu->run = NULL;
2367 out_unlock:
2368         mutex_unlock(&vcpu->mutex);
2369 out:
2370         return r;
2371 }
2372
2373 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2374                                     struct kvm_cpuid *cpuid,
2375                                     struct kvm_cpuid_entry __user *entries)
2376 {
2377         int r;
2378
2379         r = -E2BIG;
2380         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2381                 goto out;
2382         r = -EFAULT;
2383         if (copy_from_user(&vcpu->cpuid_entries, entries,
2384                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2385                 goto out;
2386         vcpu->cpuid_nent = cpuid->nent;
2387         return 0;
2388
2389 out:
2390         return r;
2391 }
2392
2393 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2394 {
2395         if (sigset) {
2396                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2397                 vcpu->sigset_active = 1;
2398                 vcpu->sigset = *sigset;
2399         } else
2400                 vcpu->sigset_active = 0;
2401         return 0;
2402 }
2403
2404 /*
2405  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2406  * we have asm/x86/processor.h
2407  */
2408 struct fxsave {
2409         u16     cwd;
2410         u16     swd;
2411         u16     twd;
2412         u16     fop;
2413         u64     rip;
2414         u64     rdp;
2415         u32     mxcsr;
2416         u32     mxcsr_mask;
2417         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2418 #ifdef CONFIG_X86_64
2419         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2420 #else
2421         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2422 #endif
2423 };
2424
2425 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2426 {
2427         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2428
2429         vcpu_load(vcpu);
2430
2431         memcpy(fpu->fpr, fxsave->st_space, 128);
2432         fpu->fcw = fxsave->cwd;
2433         fpu->fsw = fxsave->swd;
2434         fpu->ftwx = fxsave->twd;
2435         fpu->last_opcode = fxsave->fop;
2436         fpu->last_ip = fxsave->rip;
2437         fpu->last_dp = fxsave->rdp;
2438         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2439
2440         vcpu_put(vcpu);
2441
2442         return 0;
2443 }
2444
2445 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2446 {
2447         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2448
2449         vcpu_load(vcpu);
2450
2451         memcpy(fxsave->st_space, fpu->fpr, 128);
2452         fxsave->cwd = fpu->fcw;
2453         fxsave->swd = fpu->fsw;
2454         fxsave->twd = fpu->ftwx;
2455         fxsave->fop = fpu->last_opcode;
2456         fxsave->rip = fpu->last_ip;
2457         fxsave->rdp = fpu->last_dp;
2458         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2459
2460         vcpu_put(vcpu);
2461
2462         return 0;
2463 }
2464
2465 static long kvm_vcpu_ioctl(struct file *filp,
2466                            unsigned int ioctl, unsigned long arg)
2467 {
2468         struct kvm_vcpu *vcpu = filp->private_data;
2469         void __user *argp = (void __user *)arg;
2470         int r = -EINVAL;
2471
2472         switch (ioctl) {
2473         case KVM_RUN:
2474                 r = -EINVAL;
2475                 if (arg)
2476                         goto out;
2477                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2478                 break;
2479         case KVM_GET_REGS: {
2480                 struct kvm_regs kvm_regs;
2481
2482                 memset(&kvm_regs, 0, sizeof kvm_regs);
2483                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2484                 if (r)
2485                         goto out;
2486                 r = -EFAULT;
2487                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2488                         goto out;
2489                 r = 0;
2490                 break;
2491         }
2492         case KVM_SET_REGS: {
2493                 struct kvm_regs kvm_regs;
2494
2495                 r = -EFAULT;
2496                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2497                         goto out;
2498                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2499                 if (r)
2500                         goto out;
2501                 r = 0;
2502                 break;
2503         }
2504         case KVM_GET_SREGS: {
2505                 struct kvm_sregs kvm_sregs;
2506
2507                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2508                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2509                 if (r)
2510                         goto out;
2511                 r = -EFAULT;
2512                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2513                         goto out;
2514                 r = 0;
2515                 break;
2516         }
2517         case KVM_SET_SREGS: {
2518                 struct kvm_sregs kvm_sregs;
2519
2520                 r = -EFAULT;
2521                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2522                         goto out;
2523                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2524                 if (r)
2525                         goto out;
2526                 r = 0;
2527                 break;
2528         }
2529         case KVM_TRANSLATE: {
2530                 struct kvm_translation tr;
2531
2532                 r = -EFAULT;
2533                 if (copy_from_user(&tr, argp, sizeof tr))
2534                         goto out;
2535                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2536                 if (r)
2537                         goto out;
2538                 r = -EFAULT;
2539                 if (copy_to_user(argp, &tr, sizeof tr))
2540                         goto out;
2541                 r = 0;
2542                 break;
2543         }
2544         case KVM_INTERRUPT: {
2545                 struct kvm_interrupt irq;
2546
2547                 r = -EFAULT;
2548                 if (copy_from_user(&irq, argp, sizeof irq))
2549                         goto out;
2550                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2551                 if (r)
2552                         goto out;
2553                 r = 0;
2554                 break;
2555         }
2556         case KVM_DEBUG_GUEST: {
2557                 struct kvm_debug_guest dbg;
2558
2559                 r = -EFAULT;
2560                 if (copy_from_user(&dbg, argp, sizeof dbg))
2561                         goto out;
2562                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2563                 if (r)
2564                         goto out;
2565                 r = 0;
2566                 break;
2567         }
2568         case KVM_GET_MSRS:
2569                 r = msr_io(vcpu, argp, get_msr, 1);
2570                 break;
2571         case KVM_SET_MSRS:
2572                 r = msr_io(vcpu, argp, do_set_msr, 0);
2573                 break;
2574         case KVM_SET_CPUID: {
2575                 struct kvm_cpuid __user *cpuid_arg = argp;
2576                 struct kvm_cpuid cpuid;
2577
2578                 r = -EFAULT;
2579                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2580                         goto out;
2581                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2582                 if (r)
2583                         goto out;
2584                 break;
2585         }
2586         case KVM_SET_SIGNAL_MASK: {
2587                 struct kvm_signal_mask __user *sigmask_arg = argp;
2588                 struct kvm_signal_mask kvm_sigmask;
2589                 sigset_t sigset, *p;
2590
2591                 p = NULL;
2592                 if (argp) {
2593                         r = -EFAULT;
2594                         if (copy_from_user(&kvm_sigmask, argp,
2595                                            sizeof kvm_sigmask))
2596                                 goto out;
2597                         r = -EINVAL;
2598                         if (kvm_sigmask.len != sizeof sigset)
2599                                 goto out;
2600                         r = -EFAULT;
2601                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2602                                            sizeof sigset))
2603                                 goto out;
2604                         p = &sigset;
2605                 }
2606                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2607                 break;
2608         }
2609         case KVM_GET_FPU: {
2610                 struct kvm_fpu fpu;
2611
2612                 memset(&fpu, 0, sizeof fpu);
2613                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2614                 if (r)
2615                         goto out;
2616                 r = -EFAULT;
2617                 if (copy_to_user(argp, &fpu, sizeof fpu))
2618                         goto out;
2619                 r = 0;
2620                 break;
2621         }
2622         case KVM_SET_FPU: {
2623                 struct kvm_fpu fpu;
2624
2625                 r = -EFAULT;
2626                 if (copy_from_user(&fpu, argp, sizeof fpu))
2627                         goto out;
2628                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2629                 if (r)
2630                         goto out;
2631                 r = 0;
2632                 break;
2633         }
2634         default:
2635                 ;
2636         }
2637 out:
2638         return r;
2639 }
2640
2641 static long kvm_vm_ioctl(struct file *filp,
2642                            unsigned int ioctl, unsigned long arg)
2643 {
2644         struct kvm *kvm = filp->private_data;
2645         void __user *argp = (void __user *)arg;
2646         int r = -EINVAL;
2647
2648         switch (ioctl) {
2649         case KVM_CREATE_VCPU:
2650                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2651                 if (r < 0)
2652                         goto out;
2653                 break;
2654         case KVM_SET_MEMORY_REGION: {
2655                 struct kvm_memory_region kvm_mem;
2656
2657                 r = -EFAULT;
2658                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2659                         goto out;
2660                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2661                 if (r)
2662                         goto out;
2663                 break;
2664         }
2665         case KVM_GET_DIRTY_LOG: {
2666                 struct kvm_dirty_log log;
2667
2668                 r = -EFAULT;
2669                 if (copy_from_user(&log, argp, sizeof log))
2670                         goto out;
2671                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2672                 if (r)
2673                         goto out;
2674                 break;
2675         }
2676         case KVM_SET_MEMORY_ALIAS: {
2677                 struct kvm_memory_alias alias;
2678
2679                 r = -EFAULT;
2680                 if (copy_from_user(&alias, argp, sizeof alias))
2681                         goto out;
2682                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2683                 if (r)
2684                         goto out;
2685                 break;
2686         }
2687         default:
2688                 ;
2689         }
2690 out:
2691         return r;
2692 }
2693
2694 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2695                                   unsigned long address,
2696                                   int *type)
2697 {
2698         struct kvm *kvm = vma->vm_file->private_data;
2699         unsigned long pgoff;
2700         struct page *page;
2701
2702         *type = VM_FAULT_MINOR;
2703         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2704         page = gfn_to_page(kvm, pgoff);
2705         if (!page)
2706                 return NOPAGE_SIGBUS;
2707         get_page(page);
2708         return page;
2709 }
2710
2711 static struct vm_operations_struct kvm_vm_vm_ops = {
2712         .nopage = kvm_vm_nopage,
2713 };
2714
2715 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2716 {
2717         vma->vm_ops = &kvm_vm_vm_ops;
2718         return 0;
2719 }
2720
2721 static struct file_operations kvm_vm_fops = {
2722         .release        = kvm_vm_release,
2723         .unlocked_ioctl = kvm_vm_ioctl,
2724         .compat_ioctl   = kvm_vm_ioctl,
2725         .mmap           = kvm_vm_mmap,
2726 };
2727
2728 static int kvm_dev_ioctl_create_vm(void)
2729 {
2730         int fd, r;
2731         struct inode *inode;
2732         struct file *file;
2733         struct kvm *kvm;
2734
2735         inode = kvmfs_inode(&kvm_vm_fops);
2736         if (IS_ERR(inode)) {
2737                 r = PTR_ERR(inode);
2738                 goto out1;
2739         }
2740
2741         kvm = kvm_create_vm();
2742         if (IS_ERR(kvm)) {
2743                 r = PTR_ERR(kvm);
2744                 goto out2;
2745         }
2746
2747         file = kvmfs_file(inode, kvm);
2748         if (IS_ERR(file)) {
2749                 r = PTR_ERR(file);
2750                 goto out3;
2751         }
2752         kvm->filp = file;
2753
2754         r = get_unused_fd();
2755         if (r < 0)
2756                 goto out4;
2757         fd = r;
2758         fd_install(fd, file);
2759
2760         return fd;
2761
2762 out4:
2763         fput(file);
2764 out3:
2765         kvm_destroy_vm(kvm);
2766 out2:
2767         iput(inode);
2768 out1:
2769         return r;
2770 }
2771
2772 static long kvm_dev_ioctl(struct file *filp,
2773                           unsigned int ioctl, unsigned long arg)
2774 {
2775         void __user *argp = (void __user *)arg;
2776         long r = -EINVAL;
2777
2778         switch (ioctl) {
2779         case KVM_GET_API_VERSION:
2780                 r = -EINVAL;
2781                 if (arg)
2782                         goto out;
2783                 r = KVM_API_VERSION;
2784                 break;
2785         case KVM_CREATE_VM:
2786                 r = -EINVAL;
2787                 if (arg)
2788                         goto out;
2789                 r = kvm_dev_ioctl_create_vm();
2790                 break;
2791         case KVM_GET_MSR_INDEX_LIST: {
2792                 struct kvm_msr_list __user *user_msr_list = argp;
2793                 struct kvm_msr_list msr_list;
2794                 unsigned n;
2795
2796                 r = -EFAULT;
2797                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2798                         goto out;
2799                 n = msr_list.nmsrs;
2800                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2801                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2802                         goto out;
2803                 r = -E2BIG;
2804                 if (n < num_msrs_to_save)
2805                         goto out;
2806                 r = -EFAULT;
2807                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2808                                  num_msrs_to_save * sizeof(u32)))
2809                         goto out;
2810                 if (copy_to_user(user_msr_list->indices
2811                                  + num_msrs_to_save * sizeof(u32),
2812                                  &emulated_msrs,
2813                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2814                         goto out;
2815                 r = 0;
2816                 break;
2817         }
2818         case KVM_CHECK_EXTENSION:
2819                 /*
2820                  * No extensions defined at present.
2821                  */
2822                 r = 0;
2823                 break;
2824         case KVM_GET_VCPU_MMAP_SIZE:
2825                 r = -EINVAL;
2826                 if (arg)
2827                         goto out;
2828                 r = 2 * PAGE_SIZE;
2829                 break;
2830         default:
2831                 ;
2832         }
2833 out:
2834         return r;
2835 }
2836
2837 static struct file_operations kvm_chardev_ops = {
2838         .open           = kvm_dev_open,
2839         .release        = kvm_dev_release,
2840         .unlocked_ioctl = kvm_dev_ioctl,
2841         .compat_ioctl   = kvm_dev_ioctl,
2842 };
2843
2844 static struct miscdevice kvm_dev = {
2845         KVM_MINOR,
2846         "kvm",
2847         &kvm_chardev_ops,
2848 };
2849
2850 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2851                        void *v)
2852 {
2853         if (val == SYS_RESTART) {
2854                 /*
2855                  * Some (well, at least mine) BIOSes hang on reboot if
2856                  * in vmx root mode.
2857                  */
2858                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2859                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2860         }
2861         return NOTIFY_OK;
2862 }
2863
2864 static struct notifier_block kvm_reboot_notifier = {
2865         .notifier_call = kvm_reboot,
2866         .priority = 0,
2867 };
2868
2869 /*
2870  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2871  * cached on it.
2872  */
2873 static void decache_vcpus_on_cpu(int cpu)
2874 {
2875         struct kvm *vm;
2876         struct kvm_vcpu *vcpu;
2877         int i;
2878
2879         spin_lock(&kvm_lock);
2880         list_for_each_entry(vm, &vm_list, vm_list)
2881                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2882                         vcpu = &vm->vcpus[i];
2883                         /*
2884                          * If the vcpu is locked, then it is running on some
2885                          * other cpu and therefore it is not cached on the
2886                          * cpu in question.
2887                          *
2888                          * If it's not locked, check the last cpu it executed
2889                          * on.
2890                          */
2891                         if (mutex_trylock(&vcpu->mutex)) {
2892                                 if (vcpu->cpu == cpu) {
2893                                         kvm_arch_ops->vcpu_decache(vcpu);
2894                                         vcpu->cpu = -1;
2895                                 }
2896                                 mutex_unlock(&vcpu->mutex);
2897                         }
2898                 }
2899         spin_unlock(&kvm_lock);
2900 }
2901
2902 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2903                            void *v)
2904 {
2905         int cpu = (long)v;
2906
2907         switch (val) {
2908         case CPU_DOWN_PREPARE:
2909         case CPU_UP_CANCELED:
2910                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2911                        cpu);
2912                 decache_vcpus_on_cpu(cpu);
2913                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2914                                          NULL, 0, 1);
2915                 break;
2916         case CPU_ONLINE:
2917                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2918                        cpu);
2919                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2920                                          NULL, 0, 1);
2921                 break;
2922         }
2923         return NOTIFY_OK;
2924 }
2925
2926 static struct notifier_block kvm_cpu_notifier = {
2927         .notifier_call = kvm_cpu_hotplug,
2928         .priority = 20, /* must be > scheduler priority */
2929 };
2930
2931 static __init void kvm_init_debug(void)
2932 {
2933         struct kvm_stats_debugfs_item *p;
2934
2935         debugfs_dir = debugfs_create_dir("kvm", NULL);
2936         for (p = debugfs_entries; p->name; ++p)
2937                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2938                                                p->data);
2939 }
2940
2941 static void kvm_exit_debug(void)
2942 {
2943         struct kvm_stats_debugfs_item *p;
2944
2945         for (p = debugfs_entries; p->name; ++p)
2946                 debugfs_remove(p->dentry);
2947         debugfs_remove(debugfs_dir);
2948 }
2949
2950 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2951 {
2952         decache_vcpus_on_cpu(raw_smp_processor_id());
2953         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2954         return 0;
2955 }
2956
2957 static int kvm_resume(struct sys_device *dev)
2958 {
2959         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2960         return 0;
2961 }
2962
2963 static struct sysdev_class kvm_sysdev_class = {
2964         set_kset_name("kvm"),
2965         .suspend = kvm_suspend,
2966         .resume = kvm_resume,
2967 };
2968
2969 static struct sys_device kvm_sysdev = {
2970         .id = 0,
2971         .cls = &kvm_sysdev_class,
2972 };
2973
2974 hpa_t bad_page_address;
2975
2976 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2977                         const char *dev_name, void *data, struct vfsmount *mnt)
2978 {
2979         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2980 }
2981
2982 static struct file_system_type kvm_fs_type = {
2983         .name           = "kvmfs",
2984         .get_sb         = kvmfs_get_sb,
2985         .kill_sb        = kill_anon_super,
2986 };
2987
2988 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2989 {
2990         int r;
2991
2992         if (kvm_arch_ops) {
2993                 printk(KERN_ERR "kvm: already loaded the other module\n");
2994                 return -EEXIST;
2995         }
2996
2997         if (!ops->cpu_has_kvm_support()) {
2998                 printk(KERN_ERR "kvm: no hardware support\n");
2999                 return -EOPNOTSUPP;
3000         }
3001         if (ops->disabled_by_bios()) {
3002                 printk(KERN_ERR "kvm: disabled by bios\n");
3003                 return -EOPNOTSUPP;
3004         }
3005
3006         kvm_arch_ops = ops;
3007
3008         r = kvm_arch_ops->hardware_setup();
3009         if (r < 0)
3010                 goto out;
3011
3012         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3013         r = register_cpu_notifier(&kvm_cpu_notifier);
3014         if (r)
3015                 goto out_free_1;
3016         register_reboot_notifier(&kvm_reboot_notifier);
3017
3018         r = sysdev_class_register(&kvm_sysdev_class);
3019         if (r)
3020                 goto out_free_2;
3021
3022         r = sysdev_register(&kvm_sysdev);
3023         if (r)
3024                 goto out_free_3;
3025
3026         kvm_chardev_ops.owner = module;
3027
3028         r = misc_register(&kvm_dev);
3029         if (r) {
3030                 printk (KERN_ERR "kvm: misc device register failed\n");
3031                 goto out_free;
3032         }
3033
3034         return r;
3035
3036 out_free:
3037         sysdev_unregister(&kvm_sysdev);
3038 out_free_3:
3039         sysdev_class_unregister(&kvm_sysdev_class);
3040 out_free_2:
3041         unregister_reboot_notifier(&kvm_reboot_notifier);
3042         unregister_cpu_notifier(&kvm_cpu_notifier);
3043 out_free_1:
3044         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3045         kvm_arch_ops->hardware_unsetup();
3046 out:
3047         kvm_arch_ops = NULL;
3048         return r;
3049 }
3050
3051 void kvm_exit_arch(void)
3052 {
3053         misc_deregister(&kvm_dev);
3054         sysdev_unregister(&kvm_sysdev);
3055         sysdev_class_unregister(&kvm_sysdev_class);
3056         unregister_reboot_notifier(&kvm_reboot_notifier);
3057         unregister_cpu_notifier(&kvm_cpu_notifier);
3058         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3059         kvm_arch_ops->hardware_unsetup();
3060         kvm_arch_ops = NULL;
3061 }
3062
3063 static __init int kvm_init(void)
3064 {
3065         static struct page *bad_page;
3066         int r;
3067
3068         r = kvm_mmu_module_init();
3069         if (r)
3070                 goto out4;
3071
3072         r = register_filesystem(&kvm_fs_type);
3073         if (r)
3074                 goto out3;
3075
3076         kvmfs_mnt = kern_mount(&kvm_fs_type);
3077         r = PTR_ERR(kvmfs_mnt);
3078         if (IS_ERR(kvmfs_mnt))
3079                 goto out2;
3080         kvm_init_debug();
3081
3082         kvm_init_msr_list();
3083
3084         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3085                 r = -ENOMEM;
3086                 goto out;
3087         }
3088
3089         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3090         memset(__va(bad_page_address), 0, PAGE_SIZE);
3091
3092         return 0;
3093
3094 out:
3095         kvm_exit_debug();
3096         mntput(kvmfs_mnt);
3097 out2:
3098         unregister_filesystem(&kvm_fs_type);
3099 out3:
3100         kvm_mmu_module_exit();
3101 out4:
3102         return r;
3103 }
3104
3105 static __exit void kvm_exit(void)
3106 {
3107         kvm_exit_debug();
3108         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3109         mntput(kvmfs_mnt);
3110         unregister_filesystem(&kvm_fs_type);
3111         kvm_mmu_module_exit();
3112 }
3113
3114 module_init(kvm_init)
3115 module_exit(kvm_exit)
3116
3117 EXPORT_SYMBOL_GPL(kvm_init_arch);
3118 EXPORT_SYMBOL_GPL(kvm_exit_arch);