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                 return X86EMUL_PROPAGATE_FAULT;
1076
1077         if (emulator_write_phys(vcpu, gpa, val, bytes))
1078                 return X86EMUL_CONTINUE;
1079
1080         vcpu->mmio_needed = 1;
1081         vcpu->mmio_phys_addr = gpa;
1082         vcpu->mmio_size = bytes;
1083         vcpu->mmio_is_write = 1;
1084         memcpy(vcpu->mmio_data, &val, bytes);
1085
1086         return X86EMUL_CONTINUE;
1087 }
1088
1089 static int emulator_cmpxchg_emulated(unsigned long addr,
1090                                      unsigned long old,
1091                                      unsigned long new,
1092                                      unsigned int bytes,
1093                                      struct x86_emulate_ctxt *ctxt)
1094 {
1095         static int reported;
1096
1097         if (!reported) {
1098                 reported = 1;
1099                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1100         }
1101         return emulator_write_emulated(addr, new, bytes, ctxt);
1102 }
1103
1104 #ifdef CONFIG_X86_32
1105
1106 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1107                                        unsigned long old_lo,
1108                                        unsigned long old_hi,
1109                                        unsigned long new_lo,
1110                                        unsigned long new_hi,
1111                                        struct x86_emulate_ctxt *ctxt)
1112 {
1113         static int reported;
1114         int r;
1115
1116         if (!reported) {
1117                 reported = 1;
1118                 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1119         }
1120         r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1121         if (r != X86EMUL_CONTINUE)
1122                 return r;
1123         return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1124 }
1125
1126 #endif
1127
1128 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1129 {
1130         return kvm_arch_ops->get_segment_base(vcpu, seg);
1131 }
1132
1133 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1134 {
1135         return X86EMUL_CONTINUE;
1136 }
1137
1138 int emulate_clts(struct kvm_vcpu *vcpu)
1139 {
1140         unsigned long cr0;
1141
1142         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1143         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1144         kvm_arch_ops->set_cr0(vcpu, cr0);
1145         return X86EMUL_CONTINUE;
1146 }
1147
1148 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1149 {
1150         struct kvm_vcpu *vcpu = ctxt->vcpu;
1151
1152         switch (dr) {
1153         case 0 ... 3:
1154                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1155                 return X86EMUL_CONTINUE;
1156         default:
1157                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1158                        __FUNCTION__, dr);
1159                 return X86EMUL_UNHANDLEABLE;
1160         }
1161 }
1162
1163 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1164 {
1165         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1166         int exception;
1167
1168         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1169         if (exception) {
1170                 /* FIXME: better handling */
1171                 return X86EMUL_UNHANDLEABLE;
1172         }
1173         return X86EMUL_CONTINUE;
1174 }
1175
1176 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1177 {
1178         static int reported;
1179         u8 opcodes[4];
1180         unsigned long rip = ctxt->vcpu->rip;
1181         unsigned long rip_linear;
1182
1183         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1184
1185         if (reported)
1186                 return;
1187
1188         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1189
1190         printk(KERN_ERR "emulation failed but !mmio_needed?"
1191                " rip %lx %02x %02x %02x %02x\n",
1192                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1193         reported = 1;
1194 }
1195
1196 struct x86_emulate_ops emulate_ops = {
1197         .read_std            = emulator_read_std,
1198         .write_std           = emulator_write_std,
1199         .read_emulated       = emulator_read_emulated,
1200         .write_emulated      = emulator_write_emulated,
1201         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1202 #ifdef CONFIG_X86_32
1203         .cmpxchg8b_emulated  = emulator_cmpxchg8b_emulated,
1204 #endif
1205 };
1206
1207 int emulate_instruction(struct kvm_vcpu *vcpu,
1208                         struct kvm_run *run,
1209                         unsigned long cr2,
1210                         u16 error_code)
1211 {
1212         struct x86_emulate_ctxt emulate_ctxt;
1213         int r;
1214         int cs_db, cs_l;
1215
1216         kvm_arch_ops->cache_regs(vcpu);
1217
1218         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1219
1220         emulate_ctxt.vcpu = vcpu;
1221         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1222         emulate_ctxt.cr2 = cr2;
1223         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1224                 ? X86EMUL_MODE_REAL : cs_l
1225                 ? X86EMUL_MODE_PROT64 : cs_db
1226                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1227
1228         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1229                 emulate_ctxt.cs_base = 0;
1230                 emulate_ctxt.ds_base = 0;
1231                 emulate_ctxt.es_base = 0;
1232                 emulate_ctxt.ss_base = 0;
1233         } else {
1234                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1235                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1236                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1237                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1238         }
1239
1240         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1241         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1242
1243         vcpu->mmio_is_write = 0;
1244         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1245
1246         if ((r || vcpu->mmio_is_write) && run) {
1247                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1248                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1249                 run->mmio.len = vcpu->mmio_size;
1250                 run->mmio.is_write = vcpu->mmio_is_write;
1251         }
1252
1253         if (r) {
1254                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1255                         return EMULATE_DONE;
1256                 if (!vcpu->mmio_needed) {
1257                         report_emulation_failure(&emulate_ctxt);
1258                         return EMULATE_FAIL;
1259                 }
1260                 return EMULATE_DO_MMIO;
1261         }
1262
1263         kvm_arch_ops->decache_regs(vcpu);
1264         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1265
1266         if (vcpu->mmio_is_write)
1267                 return EMULATE_DO_MMIO;
1268
1269         return EMULATE_DONE;
1270 }
1271 EXPORT_SYMBOL_GPL(emulate_instruction);
1272
1273 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1274 {
1275         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1276
1277         kvm_arch_ops->cache_regs(vcpu);
1278         ret = -KVM_EINVAL;
1279 #ifdef CONFIG_X86_64
1280         if (is_long_mode(vcpu)) {
1281                 nr = vcpu->regs[VCPU_REGS_RAX];
1282                 a0 = vcpu->regs[VCPU_REGS_RDI];
1283                 a1 = vcpu->regs[VCPU_REGS_RSI];
1284                 a2 = vcpu->regs[VCPU_REGS_RDX];
1285                 a3 = vcpu->regs[VCPU_REGS_RCX];
1286                 a4 = vcpu->regs[VCPU_REGS_R8];
1287                 a5 = vcpu->regs[VCPU_REGS_R9];
1288         } else
1289 #endif
1290         {
1291                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1292                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1293                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1294                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1295                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1296                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1297                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1298         }
1299         switch (nr) {
1300         default:
1301                 run->hypercall.args[0] = a0;
1302                 run->hypercall.args[1] = a1;
1303                 run->hypercall.args[2] = a2;
1304                 run->hypercall.args[3] = a3;
1305                 run->hypercall.args[4] = a4;
1306                 run->hypercall.args[5] = a5;
1307                 run->hypercall.ret = ret;
1308                 run->hypercall.longmode = is_long_mode(vcpu);
1309                 kvm_arch_ops->decache_regs(vcpu);
1310                 return 0;
1311         }
1312         vcpu->regs[VCPU_REGS_RAX] = ret;
1313         kvm_arch_ops->decache_regs(vcpu);
1314         return 1;
1315 }
1316 EXPORT_SYMBOL_GPL(kvm_hypercall);
1317
1318 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1319 {
1320         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1321 }
1322
1323 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1324 {
1325         struct descriptor_table dt = { limit, base };
1326
1327         kvm_arch_ops->set_gdt(vcpu, &dt);
1328 }
1329
1330 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1331 {
1332         struct descriptor_table dt = { limit, base };
1333
1334         kvm_arch_ops->set_idt(vcpu, &dt);
1335 }
1336
1337 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1338                    unsigned long *rflags)
1339 {
1340         lmsw(vcpu, msw);
1341         *rflags = kvm_arch_ops->get_rflags(vcpu);
1342 }
1343
1344 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1345 {
1346         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1347         switch (cr) {
1348         case 0:
1349                 return vcpu->cr0;
1350         case 2:
1351                 return vcpu->cr2;
1352         case 3:
1353                 return vcpu->cr3;
1354         case 4:
1355                 return vcpu->cr4;
1356         default:
1357                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1358                 return 0;
1359         }
1360 }
1361
1362 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1363                      unsigned long *rflags)
1364 {
1365         switch (cr) {
1366         case 0:
1367                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1368                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1369                 break;
1370         case 2:
1371                 vcpu->cr2 = val;
1372                 break;
1373         case 3:
1374                 set_cr3(vcpu, val);
1375                 break;
1376         case 4:
1377                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1378                 break;
1379         default:
1380                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1381         }
1382 }
1383
1384 /*
1385  * Register the para guest with the host:
1386  */
1387 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1388 {
1389         struct kvm_vcpu_para_state *para_state;
1390         hpa_t para_state_hpa, hypercall_hpa;
1391         struct page *para_state_page;
1392         unsigned char *hypercall;
1393         gpa_t hypercall_gpa;
1394
1395         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1396         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1397
1398         /*
1399          * Needs to be page aligned:
1400          */
1401         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1402                 goto err_gp;
1403
1404         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1405         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1406         if (is_error_hpa(para_state_hpa))
1407                 goto err_gp;
1408
1409         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1410         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1411         para_state = kmap_atomic(para_state_page, KM_USER0);
1412
1413         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1414         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1415
1416         para_state->host_version = KVM_PARA_API_VERSION;
1417         /*
1418          * We cannot support guests that try to register themselves
1419          * with a newer API version than the host supports:
1420          */
1421         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1422                 para_state->ret = -KVM_EINVAL;
1423                 goto err_kunmap_skip;
1424         }
1425
1426         hypercall_gpa = para_state->hypercall_gpa;
1427         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1428         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1429         if (is_error_hpa(hypercall_hpa)) {
1430                 para_state->ret = -KVM_EINVAL;
1431                 goto err_kunmap_skip;
1432         }
1433
1434         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1435         vcpu->para_state_page = para_state_page;
1436         vcpu->para_state_gpa = para_state_gpa;
1437         vcpu->hypercall_gpa = hypercall_gpa;
1438
1439         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1440         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1441                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1442         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1443         kunmap_atomic(hypercall, KM_USER1);
1444
1445         para_state->ret = 0;
1446 err_kunmap_skip:
1447         kunmap_atomic(para_state, KM_USER0);
1448         return 0;
1449 err_gp:
1450         return 1;
1451 }
1452
1453 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1454 {
1455         u64 data;
1456
1457         switch (msr) {
1458         case 0xc0010010: /* SYSCFG */
1459         case 0xc0010015: /* HWCR */
1460         case MSR_IA32_PLATFORM_ID:
1461         case MSR_IA32_P5_MC_ADDR:
1462         case MSR_IA32_P5_MC_TYPE:
1463         case MSR_IA32_MC0_CTL:
1464         case MSR_IA32_MCG_STATUS:
1465         case MSR_IA32_MCG_CAP:
1466         case MSR_IA32_MC0_MISC:
1467         case MSR_IA32_MC0_MISC+4:
1468         case MSR_IA32_MC0_MISC+8:
1469         case MSR_IA32_MC0_MISC+12:
1470         case MSR_IA32_MC0_MISC+16:
1471         case MSR_IA32_UCODE_REV:
1472         case MSR_IA32_PERF_STATUS:
1473                 /* MTRR registers */
1474         case 0xfe:
1475         case 0x200 ... 0x2ff:
1476                 data = 0;
1477                 break;
1478         case 0xcd: /* fsb frequency */
1479                 data = 3;
1480                 break;
1481         case MSR_IA32_APICBASE:
1482                 data = vcpu->apic_base;
1483                 break;
1484         case MSR_IA32_MISC_ENABLE:
1485                 data = vcpu->ia32_misc_enable_msr;
1486                 break;
1487 #ifdef CONFIG_X86_64
1488         case MSR_EFER:
1489                 data = vcpu->shadow_efer;
1490                 break;
1491 #endif
1492         default:
1493                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1494                 return 1;
1495         }
1496         *pdata = data;
1497         return 0;
1498 }
1499 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1500
1501 /*
1502  * Reads an msr value (of 'msr_index') into 'pdata'.
1503  * Returns 0 on success, non-0 otherwise.
1504  * Assumes vcpu_load() was already called.
1505  */
1506 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1507 {
1508         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1509 }
1510
1511 #ifdef CONFIG_X86_64
1512
1513 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1514 {
1515         if (efer & EFER_RESERVED_BITS) {
1516                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1517                        efer);
1518                 inject_gp(vcpu);
1519                 return;
1520         }
1521
1522         if (is_paging(vcpu)
1523             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1524                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1525                 inject_gp(vcpu);
1526                 return;
1527         }
1528
1529         kvm_arch_ops->set_efer(vcpu, efer);
1530
1531         efer &= ~EFER_LMA;
1532         efer |= vcpu->shadow_efer & EFER_LMA;
1533
1534         vcpu->shadow_efer = efer;
1535 }
1536
1537 #endif
1538
1539 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1540 {
1541         switch (msr) {
1542 #ifdef CONFIG_X86_64
1543         case MSR_EFER:
1544                 set_efer(vcpu, data);
1545                 break;
1546 #endif
1547         case MSR_IA32_MC0_STATUS:
1548                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1549                        __FUNCTION__, data);
1550                 break;
1551         case MSR_IA32_MCG_STATUS:
1552                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1553                         __FUNCTION__, data);
1554                 break;
1555         case MSR_IA32_UCODE_REV:
1556         case MSR_IA32_UCODE_WRITE:
1557         case 0x200 ... 0x2ff: /* MTRRs */
1558                 break;
1559         case MSR_IA32_APICBASE:
1560                 vcpu->apic_base = data;
1561                 break;
1562         case MSR_IA32_MISC_ENABLE:
1563                 vcpu->ia32_misc_enable_msr = data;
1564                 break;
1565         /*
1566          * This is the 'probe whether the host is KVM' logic:
1567          */
1568         case MSR_KVM_API_MAGIC:
1569                 return vcpu_register_para(vcpu, data);
1570
1571         default:
1572                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1573                 return 1;
1574         }
1575         return 0;
1576 }
1577 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1578
1579 /*
1580  * Writes msr value into into the appropriate "register".
1581  * Returns 0 on success, non-0 otherwise.
1582  * Assumes vcpu_load() was already called.
1583  */
1584 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1585 {
1586         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1587 }
1588
1589 void kvm_resched(struct kvm_vcpu *vcpu)
1590 {
1591         vcpu_put(vcpu);
1592         cond_resched();
1593         vcpu_load(vcpu);
1594 }
1595 EXPORT_SYMBOL_GPL(kvm_resched);
1596
1597 void load_msrs(struct vmx_msr_entry *e, int n)
1598 {
1599         int i;
1600
1601         for (i = 0; i < n; ++i)
1602                 wrmsrl(e[i].index, e[i].data);
1603 }
1604 EXPORT_SYMBOL_GPL(load_msrs);
1605
1606 void save_msrs(struct vmx_msr_entry *e, int n)
1607 {
1608         int i;
1609
1610         for (i = 0; i < n; ++i)
1611                 rdmsrl(e[i].index, e[i].data);
1612 }
1613 EXPORT_SYMBOL_GPL(save_msrs);
1614
1615 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1616 {
1617         int i;
1618         u32 function;
1619         struct kvm_cpuid_entry *e, *best;
1620
1621         kvm_arch_ops->cache_regs(vcpu);
1622         function = vcpu->regs[VCPU_REGS_RAX];
1623         vcpu->regs[VCPU_REGS_RAX] = 0;
1624         vcpu->regs[VCPU_REGS_RBX] = 0;
1625         vcpu->regs[VCPU_REGS_RCX] = 0;
1626         vcpu->regs[VCPU_REGS_RDX] = 0;
1627         best = NULL;
1628         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1629                 e = &vcpu->cpuid_entries[i];
1630                 if (e->function == function) {
1631                         best = e;
1632                         break;
1633                 }
1634                 /*
1635                  * Both basic or both extended?
1636                  */
1637                 if (((e->function ^ function) & 0x80000000) == 0)
1638                         if (!best || e->function > best->function)
1639                                 best = e;
1640         }
1641         if (best) {
1642                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1643                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1644                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1645                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1646         }
1647         kvm_arch_ops->decache_regs(vcpu);
1648         kvm_arch_ops->skip_emulated_instruction(vcpu);
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1651
1652 static int pio_copy_data(struct kvm_vcpu *vcpu)
1653 {
1654         void *p = vcpu->pio_data;
1655         void *q;
1656         unsigned bytes;
1657         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1658
1659         kvm_arch_ops->vcpu_put(vcpu);
1660         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1661                  PAGE_KERNEL);
1662         if (!q) {
1663                 kvm_arch_ops->vcpu_load(vcpu);
1664                 free_pio_guest_pages(vcpu);
1665                 return -ENOMEM;
1666         }
1667         q += vcpu->pio.guest_page_offset;
1668         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1669         if (vcpu->pio.in)
1670                 memcpy(q, p, bytes);
1671         else
1672                 memcpy(p, q, bytes);
1673         q -= vcpu->pio.guest_page_offset;
1674         vunmap(q);
1675         kvm_arch_ops->vcpu_load(vcpu);
1676         free_pio_guest_pages(vcpu);
1677         return 0;
1678 }
1679
1680 static int complete_pio(struct kvm_vcpu *vcpu)
1681 {
1682         struct kvm_pio_request *io = &vcpu->pio;
1683         long delta;
1684         int r;
1685
1686         kvm_arch_ops->cache_regs(vcpu);
1687
1688         if (!io->string) {
1689                 if (io->in)
1690                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1691                                io->size);
1692         } else {
1693                 if (io->in) {
1694                         r = pio_copy_data(vcpu);
1695                         if (r) {
1696                                 kvm_arch_ops->cache_regs(vcpu);
1697                                 return r;
1698                         }
1699                 }
1700
1701                 delta = 1;
1702                 if (io->rep) {
1703                         delta *= io->cur_count;
1704                         /*
1705                          * The size of the register should really depend on
1706                          * current address size.
1707                          */
1708                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1709                 }
1710                 if (io->down)
1711                         delta = -delta;
1712                 delta *= io->size;
1713                 if (io->in)
1714                         vcpu->regs[VCPU_REGS_RDI] += delta;
1715                 else
1716                         vcpu->regs[VCPU_REGS_RSI] += delta;
1717         }
1718
1719         vcpu->run->io_completed = 0;
1720
1721         kvm_arch_ops->decache_regs(vcpu);
1722
1723         io->count -= io->cur_count;
1724         io->cur_count = 0;
1725
1726         if (!io->count)
1727                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1728         return 0;
1729 }
1730
1731 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1732                   int size, unsigned long count, int string, int down,
1733                   gva_t address, int rep, unsigned port)
1734 {
1735         unsigned now, in_page;
1736         int i;
1737         int nr_pages = 1;
1738         struct page *page;
1739
1740         vcpu->run->exit_reason = KVM_EXIT_IO;
1741         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1742         vcpu->run->io.size = size;
1743         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1744         vcpu->run->io.count = count;
1745         vcpu->run->io.port = port;
1746         vcpu->pio.count = count;
1747         vcpu->pio.cur_count = count;
1748         vcpu->pio.size = size;
1749         vcpu->pio.in = in;
1750         vcpu->pio.string = string;
1751         vcpu->pio.down = down;
1752         vcpu->pio.guest_page_offset = offset_in_page(address);
1753         vcpu->pio.rep = rep;
1754
1755         if (!string) {
1756                 kvm_arch_ops->cache_regs(vcpu);
1757                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1758                 kvm_arch_ops->decache_regs(vcpu);
1759                 return 0;
1760         }
1761
1762         if (!count) {
1763                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1764                 return 1;
1765         }
1766
1767         now = min(count, PAGE_SIZE / size);
1768
1769         if (!down)
1770                 in_page = PAGE_SIZE - offset_in_page(address);
1771         else
1772                 in_page = offset_in_page(address) + size;
1773         now = min(count, (unsigned long)in_page / size);
1774         if (!now) {
1775                 /*
1776                  * String I/O straddles page boundary.  Pin two guest pages
1777                  * so that we satisfy atomicity constraints.  Do just one
1778                  * transaction to avoid complexity.
1779                  */
1780                 nr_pages = 2;
1781                 now = 1;
1782         }
1783         if (down) {
1784                 /*
1785                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1786                  */
1787                 printk(KERN_ERR "kvm: guest string pio down\n");
1788                 inject_gp(vcpu);
1789                 return 1;
1790         }
1791         vcpu->run->io.count = now;
1792         vcpu->pio.cur_count = now;
1793
1794         for (i = 0; i < nr_pages; ++i) {
1795                 spin_lock(&vcpu->kvm->lock);
1796                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1797                 if (page)
1798                         get_page(page);
1799                 vcpu->pio.guest_pages[i] = page;
1800                 spin_unlock(&vcpu->kvm->lock);
1801                 if (!page) {
1802                         inject_gp(vcpu);
1803                         free_pio_guest_pages(vcpu);
1804                         return 1;
1805                 }
1806         }
1807
1808         if (!vcpu->pio.in)
1809                 return pio_copy_data(vcpu);
1810         return 0;
1811 }
1812 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1813
1814 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1815 {
1816         int r;
1817         sigset_t sigsaved;
1818
1819         vcpu_load(vcpu);
1820
1821         if (vcpu->sigset_active)
1822                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1823
1824         /* re-sync apic's tpr */
1825         vcpu->cr8 = kvm_run->cr8;
1826
1827         if (kvm_run->io_completed) {
1828                 if (vcpu->pio.cur_count) {
1829                         r = complete_pio(vcpu);
1830                         if (r)
1831                                 goto out;
1832                 } else {
1833                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1834                         vcpu->mmio_read_completed = 1;
1835                 }
1836         }
1837
1838         vcpu->mmio_needed = 0;
1839
1840         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1841                 kvm_arch_ops->cache_regs(vcpu);
1842                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1843                 kvm_arch_ops->decache_regs(vcpu);
1844         }
1845
1846         r = kvm_arch_ops->run(vcpu, kvm_run);
1847
1848 out:
1849         if (vcpu->sigset_active)
1850                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1851
1852         vcpu_put(vcpu);
1853         return r;
1854 }
1855
1856 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1857                                    struct kvm_regs *regs)
1858 {
1859         vcpu_load(vcpu);
1860
1861         kvm_arch_ops->cache_regs(vcpu);
1862
1863         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1864         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1865         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1866         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1867         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1868         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1869         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1870         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1871 #ifdef CONFIG_X86_64
1872         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1873         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1874         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1875         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1876         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1877         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1878         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1879         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1880 #endif
1881
1882         regs->rip = vcpu->rip;
1883         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1884
1885         /*
1886          * Don't leak debug flags in case they were set for guest debugging
1887          */
1888         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1889                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1890
1891         vcpu_put(vcpu);
1892
1893         return 0;
1894 }
1895
1896 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1897                                    struct kvm_regs *regs)
1898 {
1899         vcpu_load(vcpu);
1900
1901         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1902         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1903         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1904         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1905         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1906         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1907         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1908         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1909 #ifdef CONFIG_X86_64
1910         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1911         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1912         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1913         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1914         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1915         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1916         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1917         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1918 #endif
1919
1920         vcpu->rip = regs->rip;
1921         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1922
1923         kvm_arch_ops->decache_regs(vcpu);
1924
1925         vcpu_put(vcpu);
1926
1927         return 0;
1928 }
1929
1930 static void get_segment(struct kvm_vcpu *vcpu,
1931                         struct kvm_segment *var, int seg)
1932 {
1933         return kvm_arch_ops->get_segment(vcpu, var, seg);
1934 }
1935
1936 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1937                                     struct kvm_sregs *sregs)
1938 {
1939         struct descriptor_table dt;
1940
1941         vcpu_load(vcpu);
1942
1943         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1944         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1945         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1946         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1947         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1948         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1949
1950         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1951         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1952
1953         kvm_arch_ops->get_idt(vcpu, &dt);
1954         sregs->idt.limit = dt.limit;
1955         sregs->idt.base = dt.base;
1956         kvm_arch_ops->get_gdt(vcpu, &dt);
1957         sregs->gdt.limit = dt.limit;
1958         sregs->gdt.base = dt.base;
1959
1960         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1961         sregs->cr0 = vcpu->cr0;
1962         sregs->cr2 = vcpu->cr2;
1963         sregs->cr3 = vcpu->cr3;
1964         sregs->cr4 = vcpu->cr4;
1965         sregs->cr8 = vcpu->cr8;
1966         sregs->efer = vcpu->shadow_efer;
1967         sregs->apic_base = vcpu->apic_base;
1968
1969         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1970                sizeof sregs->interrupt_bitmap);
1971
1972         vcpu_put(vcpu);
1973
1974         return 0;
1975 }
1976
1977 static void set_segment(struct kvm_vcpu *vcpu,
1978                         struct kvm_segment *var, int seg)
1979 {
1980         return kvm_arch_ops->set_segment(vcpu, var, seg);
1981 }
1982
1983 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1984                                     struct kvm_sregs *sregs)
1985 {
1986         int mmu_reset_needed = 0;
1987         int i;
1988         struct descriptor_table dt;
1989
1990         vcpu_load(vcpu);
1991
1992         dt.limit = sregs->idt.limit;
1993         dt.base = sregs->idt.base;
1994         kvm_arch_ops->set_idt(vcpu, &dt);
1995         dt.limit = sregs->gdt.limit;
1996         dt.base = sregs->gdt.base;
1997         kvm_arch_ops->set_gdt(vcpu, &dt);
1998
1999         vcpu->cr2 = sregs->cr2;
2000         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2001         vcpu->cr3 = sregs->cr3;
2002
2003         vcpu->cr8 = sregs->cr8;
2004
2005         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2006 #ifdef CONFIG_X86_64
2007         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2008 #endif
2009         vcpu->apic_base = sregs->apic_base;
2010
2011         kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
2012
2013         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2014         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2015
2016         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2017         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2018         if (!is_long_mode(vcpu) && is_pae(vcpu))
2019                 load_pdptrs(vcpu, vcpu->cr3);
2020
2021         if (mmu_reset_needed)
2022                 kvm_mmu_reset_context(vcpu);
2023
2024         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2025                sizeof vcpu->irq_pending);
2026         vcpu->irq_summary = 0;
2027         for (i = 0; i < NR_IRQ_WORDS; ++i)
2028                 if (vcpu->irq_pending[i])
2029                         __set_bit(i, &vcpu->irq_summary);
2030
2031         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2032         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2033         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2034         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2035         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2036         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2037
2038         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2039         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2040
2041         vcpu_put(vcpu);
2042
2043         return 0;
2044 }
2045
2046 /*
2047  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2048  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2049  *
2050  * This list is modified at module load time to reflect the
2051  * capabilities of the host cpu.
2052  */
2053 static u32 msrs_to_save[] = {
2054         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2055         MSR_K6_STAR,
2056 #ifdef CONFIG_X86_64
2057         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2058 #endif
2059         MSR_IA32_TIME_STAMP_COUNTER,
2060 };
2061
2062 static unsigned num_msrs_to_save;
2063
2064 static u32 emulated_msrs[] = {
2065         MSR_IA32_MISC_ENABLE,
2066 };
2067
2068 static __init void kvm_init_msr_list(void)
2069 {
2070         u32 dummy[2];
2071         unsigned i, j;
2072
2073         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2074                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2075                         continue;
2076                 if (j < i)
2077                         msrs_to_save[j] = msrs_to_save[i];
2078                 j++;
2079         }
2080         num_msrs_to_save = j;
2081 }
2082
2083 /*
2084  * Adapt set_msr() to msr_io()'s calling convention
2085  */
2086 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2087 {
2088         return set_msr(vcpu, index, *data);
2089 }
2090
2091 /*
2092  * Read or write a bunch of msrs. All parameters are kernel addresses.
2093  *
2094  * @return number of msrs set successfully.
2095  */
2096 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2097                     struct kvm_msr_entry *entries,
2098                     int (*do_msr)(struct kvm_vcpu *vcpu,
2099                                   unsigned index, u64 *data))
2100 {
2101         int i;
2102
2103         vcpu_load(vcpu);
2104
2105         for (i = 0; i < msrs->nmsrs; ++i)
2106                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2107                         break;
2108
2109         vcpu_put(vcpu);
2110
2111         return i;
2112 }
2113
2114 /*
2115  * Read or write a bunch of msrs. Parameters are user addresses.
2116  *
2117  * @return number of msrs set successfully.
2118  */
2119 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2120                   int (*do_msr)(struct kvm_vcpu *vcpu,
2121                                 unsigned index, u64 *data),
2122                   int writeback)
2123 {
2124         struct kvm_msrs msrs;
2125         struct kvm_msr_entry *entries;
2126         int r, n;
2127         unsigned size;
2128
2129         r = -EFAULT;
2130         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2131                 goto out;
2132
2133         r = -E2BIG;
2134         if (msrs.nmsrs >= MAX_IO_MSRS)
2135                 goto out;
2136
2137         r = -ENOMEM;
2138         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2139         entries = vmalloc(size);
2140         if (!entries)
2141                 goto out;
2142
2143         r = -EFAULT;
2144         if (copy_from_user(entries, user_msrs->entries, size))
2145                 goto out_free;
2146
2147         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2148         if (r < 0)
2149                 goto out_free;
2150
2151         r = -EFAULT;
2152         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2153                 goto out_free;
2154
2155         r = n;
2156
2157 out_free:
2158         vfree(entries);
2159 out:
2160         return r;
2161 }
2162
2163 /*
2164  * Translate a guest virtual address to a guest physical address.
2165  */
2166 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2167                                     struct kvm_translation *tr)
2168 {
2169         unsigned long vaddr = tr->linear_address;
2170         gpa_t gpa;
2171
2172         vcpu_load(vcpu);
2173         spin_lock(&vcpu->kvm->lock);
2174         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2175         tr->physical_address = gpa;
2176         tr->valid = gpa != UNMAPPED_GVA;
2177         tr->writeable = 1;
2178         tr->usermode = 0;
2179         spin_unlock(&vcpu->kvm->lock);
2180         vcpu_put(vcpu);
2181
2182         return 0;
2183 }
2184
2185 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2186                                     struct kvm_interrupt *irq)
2187 {
2188         if (irq->irq < 0 || irq->irq >= 256)
2189                 return -EINVAL;
2190         vcpu_load(vcpu);
2191
2192         set_bit(irq->irq, vcpu->irq_pending);
2193         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2194
2195         vcpu_put(vcpu);
2196
2197         return 0;
2198 }
2199
2200 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2201                                       struct kvm_debug_guest *dbg)
2202 {
2203         int r;
2204
2205         vcpu_load(vcpu);
2206
2207         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2208
2209         vcpu_put(vcpu);
2210
2211         return r;
2212 }
2213
2214 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2215                                     unsigned long address,
2216                                     int *type)
2217 {
2218         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2219         unsigned long pgoff;
2220         struct page *page;
2221
2222         *type = VM_FAULT_MINOR;
2223         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2224         if (pgoff == 0)
2225                 page = virt_to_page(vcpu->run);
2226         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2227                 page = virt_to_page(vcpu->pio_data);
2228         else
2229                 return NOPAGE_SIGBUS;
2230         get_page(page);
2231         return page;
2232 }
2233
2234 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2235         .nopage = kvm_vcpu_nopage,
2236 };
2237
2238 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2239 {
2240         vma->vm_ops = &kvm_vcpu_vm_ops;
2241         return 0;
2242 }
2243
2244 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2245 {
2246         struct kvm_vcpu *vcpu = filp->private_data;
2247
2248         fput(vcpu->kvm->filp);
2249         return 0;
2250 }
2251
2252 static struct file_operations kvm_vcpu_fops = {
2253         .release        = kvm_vcpu_release,
2254         .unlocked_ioctl = kvm_vcpu_ioctl,
2255         .compat_ioctl   = kvm_vcpu_ioctl,
2256         .mmap           = kvm_vcpu_mmap,
2257 };
2258
2259 /*
2260  * Allocates an inode for the vcpu.
2261  */
2262 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2263 {
2264         int fd, r;
2265         struct inode *inode;
2266         struct file *file;
2267
2268         atomic_inc(&vcpu->kvm->filp->f_count);
2269         inode = kvmfs_inode(&kvm_vcpu_fops);
2270         if (IS_ERR(inode)) {
2271                 r = PTR_ERR(inode);
2272                 goto out1;
2273         }
2274
2275         file = kvmfs_file(inode, vcpu);
2276         if (IS_ERR(file)) {
2277                 r = PTR_ERR(file);
2278                 goto out2;
2279         }
2280
2281         r = get_unused_fd();
2282         if (r < 0)
2283                 goto out3;
2284         fd = r;
2285         fd_install(fd, file);
2286
2287         return fd;
2288
2289 out3:
2290         fput(file);
2291 out2:
2292         iput(inode);
2293 out1:
2294         fput(vcpu->kvm->filp);
2295         return r;
2296 }
2297
2298 /*
2299  * Creates some virtual cpus.  Good luck creating more than one.
2300  */
2301 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2302 {
2303         int r;
2304         struct kvm_vcpu *vcpu;
2305         struct page *page;
2306
2307         r = -EINVAL;
2308         if (!valid_vcpu(n))
2309                 goto out;
2310
2311         vcpu = &kvm->vcpus[n];
2312
2313         mutex_lock(&vcpu->mutex);
2314
2315         if (vcpu->vmcs) {
2316                 mutex_unlock(&vcpu->mutex);
2317                 return -EEXIST;
2318         }
2319
2320         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2321         r = -ENOMEM;
2322         if (!page)
2323                 goto out_unlock;
2324         vcpu->run = page_address(page);
2325
2326         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2327         r = -ENOMEM;
2328         if (!page)
2329                 goto out_free_run;
2330         vcpu->pio_data = page_address(page);
2331
2332         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2333                                            FX_IMAGE_ALIGN);
2334         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2335
2336         r = kvm_arch_ops->vcpu_create(vcpu);
2337         if (r < 0)
2338                 goto out_free_vcpus;
2339
2340         r = kvm_mmu_create(vcpu);
2341         if (r < 0)
2342                 goto out_free_vcpus;
2343
2344         kvm_arch_ops->vcpu_load(vcpu);
2345         r = kvm_mmu_setup(vcpu);
2346         if (r >= 0)
2347                 r = kvm_arch_ops->vcpu_setup(vcpu);
2348         vcpu_put(vcpu);
2349
2350         if (r < 0)
2351                 goto out_free_vcpus;
2352
2353         r = create_vcpu_fd(vcpu);
2354         if (r < 0)
2355                 goto out_free_vcpus;
2356
2357         return r;
2358
2359 out_free_vcpus:
2360         kvm_free_vcpu(vcpu);
2361 out_free_run:
2362         free_page((unsigned long)vcpu->run);
2363         vcpu->run = NULL;
2364 out_unlock:
2365         mutex_unlock(&vcpu->mutex);
2366 out:
2367         return r;
2368 }
2369
2370 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2371                                     struct kvm_cpuid *cpuid,
2372                                     struct kvm_cpuid_entry __user *entries)
2373 {
2374         int r;
2375
2376         r = -E2BIG;
2377         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2378                 goto out;
2379         r = -EFAULT;
2380         if (copy_from_user(&vcpu->cpuid_entries, entries,
2381                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2382                 goto out;
2383         vcpu->cpuid_nent = cpuid->nent;
2384         return 0;
2385
2386 out:
2387         return r;
2388 }
2389
2390 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2391 {
2392         if (sigset) {
2393                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2394                 vcpu->sigset_active = 1;
2395                 vcpu->sigset = *sigset;
2396         } else
2397                 vcpu->sigset_active = 0;
2398         return 0;
2399 }
2400
2401 static long kvm_vcpu_ioctl(struct file *filp,
2402                            unsigned int ioctl, unsigned long arg)
2403 {
2404         struct kvm_vcpu *vcpu = filp->private_data;
2405         void __user *argp = (void __user *)arg;
2406         int r = -EINVAL;
2407
2408         switch (ioctl) {
2409         case KVM_RUN:
2410                 r = -EINVAL;
2411                 if (arg)
2412                         goto out;
2413                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2414                 break;
2415         case KVM_GET_REGS: {
2416                 struct kvm_regs kvm_regs;
2417
2418                 memset(&kvm_regs, 0, sizeof kvm_regs);
2419                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2420                 if (r)
2421                         goto out;
2422                 r = -EFAULT;
2423                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2424                         goto out;
2425                 r = 0;
2426                 break;
2427         }
2428         case KVM_SET_REGS: {
2429                 struct kvm_regs kvm_regs;
2430
2431                 r = -EFAULT;
2432                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2433                         goto out;
2434                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2435                 if (r)
2436                         goto out;
2437                 r = 0;
2438                 break;
2439         }
2440         case KVM_GET_SREGS: {
2441                 struct kvm_sregs kvm_sregs;
2442
2443                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2444                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2445                 if (r)
2446                         goto out;
2447                 r = -EFAULT;
2448                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2449                         goto out;
2450                 r = 0;
2451                 break;
2452         }
2453         case KVM_SET_SREGS: {
2454                 struct kvm_sregs kvm_sregs;
2455
2456                 r = -EFAULT;
2457                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2458                         goto out;
2459                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2460                 if (r)
2461                         goto out;
2462                 r = 0;
2463                 break;
2464         }
2465         case KVM_TRANSLATE: {
2466                 struct kvm_translation tr;
2467
2468                 r = -EFAULT;
2469                 if (copy_from_user(&tr, argp, sizeof tr))
2470                         goto out;
2471                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2472                 if (r)
2473                         goto out;
2474                 r = -EFAULT;
2475                 if (copy_to_user(argp, &tr, sizeof tr))
2476                         goto out;
2477                 r = 0;
2478                 break;
2479         }
2480         case KVM_INTERRUPT: {
2481                 struct kvm_interrupt irq;
2482
2483                 r = -EFAULT;
2484                 if (copy_from_user(&irq, argp, sizeof irq))
2485                         goto out;
2486                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2487                 if (r)
2488                         goto out;
2489                 r = 0;
2490                 break;
2491         }
2492         case KVM_DEBUG_GUEST: {
2493                 struct kvm_debug_guest dbg;
2494
2495                 r = -EFAULT;
2496                 if (copy_from_user(&dbg, argp, sizeof dbg))
2497                         goto out;
2498                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2499                 if (r)
2500                         goto out;
2501                 r = 0;
2502                 break;
2503         }
2504         case KVM_GET_MSRS:
2505                 r = msr_io(vcpu, argp, get_msr, 1);
2506                 break;
2507         case KVM_SET_MSRS:
2508                 r = msr_io(vcpu, argp, do_set_msr, 0);
2509                 break;
2510         case KVM_SET_CPUID: {
2511                 struct kvm_cpuid __user *cpuid_arg = argp;
2512                 struct kvm_cpuid cpuid;
2513
2514                 r = -EFAULT;
2515                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2516                         goto out;
2517                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2518                 if (r)
2519                         goto out;
2520                 break;
2521         }
2522         case KVM_SET_SIGNAL_MASK: {
2523                 struct kvm_signal_mask __user *sigmask_arg = argp;
2524                 struct kvm_signal_mask kvm_sigmask;
2525                 sigset_t sigset, *p;
2526
2527                 p = NULL;
2528                 if (argp) {
2529                         r = -EFAULT;
2530                         if (copy_from_user(&kvm_sigmask, argp,
2531                                            sizeof kvm_sigmask))
2532                                 goto out;
2533                         r = -EINVAL;
2534                         if (kvm_sigmask.len != sizeof sigset)
2535                                 goto out;
2536                         r = -EFAULT;
2537                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2538                                            sizeof sigset))
2539                                 goto out;
2540                         p = &sigset;
2541                 }
2542                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2543                 break;
2544         }
2545         default:
2546                 ;
2547         }
2548 out:
2549         return r;
2550 }
2551
2552 static long kvm_vm_ioctl(struct file *filp,
2553                            unsigned int ioctl, unsigned long arg)
2554 {
2555         struct kvm *kvm = filp->private_data;
2556         void __user *argp = (void __user *)arg;
2557         int r = -EINVAL;
2558
2559         switch (ioctl) {
2560         case KVM_CREATE_VCPU:
2561                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2562                 if (r < 0)
2563                         goto out;
2564                 break;
2565         case KVM_SET_MEMORY_REGION: {
2566                 struct kvm_memory_region kvm_mem;
2567
2568                 r = -EFAULT;
2569                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2570                         goto out;
2571                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2572                 if (r)
2573                         goto out;
2574                 break;
2575         }
2576         case KVM_GET_DIRTY_LOG: {
2577                 struct kvm_dirty_log log;
2578
2579                 r = -EFAULT;
2580                 if (copy_from_user(&log, argp, sizeof log))
2581                         goto out;
2582                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2583                 if (r)
2584                         goto out;
2585                 break;
2586         }
2587         case KVM_SET_MEMORY_ALIAS: {
2588                 struct kvm_memory_alias alias;
2589
2590                 r = -EFAULT;
2591                 if (copy_from_user(&alias, argp, sizeof alias))
2592                         goto out;
2593                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2594                 if (r)
2595                         goto out;
2596                 break;
2597         }
2598         default:
2599                 ;
2600         }
2601 out:
2602         return r;
2603 }
2604
2605 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2606                                   unsigned long address,
2607                                   int *type)
2608 {
2609         struct kvm *kvm = vma->vm_file->private_data;
2610         unsigned long pgoff;
2611         struct page *page;
2612
2613         *type = VM_FAULT_MINOR;
2614         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2615         page = gfn_to_page(kvm, pgoff);
2616         if (!page)
2617                 return NOPAGE_SIGBUS;
2618         get_page(page);
2619         return page;
2620 }
2621
2622 static struct vm_operations_struct kvm_vm_vm_ops = {
2623         .nopage = kvm_vm_nopage,
2624 };
2625
2626 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2627 {
2628         vma->vm_ops = &kvm_vm_vm_ops;
2629         return 0;
2630 }
2631
2632 static struct file_operations kvm_vm_fops = {
2633         .release        = kvm_vm_release,
2634         .unlocked_ioctl = kvm_vm_ioctl,
2635         .compat_ioctl   = kvm_vm_ioctl,
2636         .mmap           = kvm_vm_mmap,
2637 };
2638
2639 static int kvm_dev_ioctl_create_vm(void)
2640 {
2641         int fd, r;
2642         struct inode *inode;
2643         struct file *file;
2644         struct kvm *kvm;
2645
2646         inode = kvmfs_inode(&kvm_vm_fops);
2647         if (IS_ERR(inode)) {
2648                 r = PTR_ERR(inode);
2649                 goto out1;
2650         }
2651
2652         kvm = kvm_create_vm();
2653         if (IS_ERR(kvm)) {
2654                 r = PTR_ERR(kvm);
2655                 goto out2;
2656         }
2657
2658         file = kvmfs_file(inode, kvm);
2659         if (IS_ERR(file)) {
2660                 r = PTR_ERR(file);
2661                 goto out3;
2662         }
2663         kvm->filp = file;
2664
2665         r = get_unused_fd();
2666         if (r < 0)
2667                 goto out4;
2668         fd = r;
2669         fd_install(fd, file);
2670
2671         return fd;
2672
2673 out4:
2674         fput(file);
2675 out3:
2676         kvm_destroy_vm(kvm);
2677 out2:
2678         iput(inode);
2679 out1:
2680         return r;
2681 }
2682
2683 static long kvm_dev_ioctl(struct file *filp,
2684                           unsigned int ioctl, unsigned long arg)
2685 {
2686         void __user *argp = (void __user *)arg;
2687         long r = -EINVAL;
2688
2689         switch (ioctl) {
2690         case KVM_GET_API_VERSION:
2691                 r = -EINVAL;
2692                 if (arg)
2693                         goto out;
2694                 r = KVM_API_VERSION;
2695                 break;
2696         case KVM_CREATE_VM:
2697                 r = -EINVAL;
2698                 if (arg)
2699                         goto out;
2700                 r = kvm_dev_ioctl_create_vm();
2701                 break;
2702         case KVM_GET_MSR_INDEX_LIST: {
2703                 struct kvm_msr_list __user *user_msr_list = argp;
2704                 struct kvm_msr_list msr_list;
2705                 unsigned n;
2706
2707                 r = -EFAULT;
2708                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2709                         goto out;
2710                 n = msr_list.nmsrs;
2711                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2712                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2713                         goto out;
2714                 r = -E2BIG;
2715                 if (n < num_msrs_to_save)
2716                         goto out;
2717                 r = -EFAULT;
2718                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2719                                  num_msrs_to_save * sizeof(u32)))
2720                         goto out;
2721                 if (copy_to_user(user_msr_list->indices
2722                                  + num_msrs_to_save * sizeof(u32),
2723                                  &emulated_msrs,
2724                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2725                         goto out;
2726                 r = 0;
2727                 break;
2728         }
2729         case KVM_CHECK_EXTENSION:
2730                 /*
2731                  * No extensions defined at present.
2732                  */
2733                 r = 0;
2734                 break;
2735         case KVM_GET_VCPU_MMAP_SIZE:
2736                 r = -EINVAL;
2737                 if (arg)
2738                         goto out;
2739                 r = 2 * PAGE_SIZE;
2740                 break;
2741         default:
2742                 ;
2743         }
2744 out:
2745         return r;
2746 }
2747
2748 static struct file_operations kvm_chardev_ops = {
2749         .open           = kvm_dev_open,
2750         .release        = kvm_dev_release,
2751         .unlocked_ioctl = kvm_dev_ioctl,
2752         .compat_ioctl   = kvm_dev_ioctl,
2753 };
2754
2755 static struct miscdevice kvm_dev = {
2756         KVM_MINOR,
2757         "kvm",
2758         &kvm_chardev_ops,
2759 };
2760
2761 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2762                        void *v)
2763 {
2764         if (val == SYS_RESTART) {
2765                 /*
2766                  * Some (well, at least mine) BIOSes hang on reboot if
2767                  * in vmx root mode.
2768                  */
2769                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2770                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2771         }
2772         return NOTIFY_OK;
2773 }
2774
2775 static struct notifier_block kvm_reboot_notifier = {
2776         .notifier_call = kvm_reboot,
2777         .priority = 0,
2778 };
2779
2780 /*
2781  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2782  * cached on it.
2783  */
2784 static void decache_vcpus_on_cpu(int cpu)
2785 {
2786         struct kvm *vm;
2787         struct kvm_vcpu *vcpu;
2788         int i;
2789
2790         spin_lock(&kvm_lock);
2791         list_for_each_entry(vm, &vm_list, vm_list)
2792                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2793                         vcpu = &vm->vcpus[i];
2794                         /*
2795                          * If the vcpu is locked, then it is running on some
2796                          * other cpu and therefore it is not cached on the
2797                          * cpu in question.
2798                          *
2799                          * If it's not locked, check the last cpu it executed
2800                          * on.
2801                          */
2802                         if (mutex_trylock(&vcpu->mutex)) {
2803                                 if (vcpu->cpu == cpu) {
2804                                         kvm_arch_ops->vcpu_decache(vcpu);
2805                                         vcpu->cpu = -1;
2806                                 }
2807                                 mutex_unlock(&vcpu->mutex);
2808                         }
2809                 }
2810         spin_unlock(&kvm_lock);
2811 }
2812
2813 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2814                            void *v)
2815 {
2816         int cpu = (long)v;
2817
2818         switch (val) {
2819         case CPU_DOWN_PREPARE:
2820         case CPU_UP_CANCELED:
2821                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2822                        cpu);
2823                 decache_vcpus_on_cpu(cpu);
2824                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2825                                          NULL, 0, 1);
2826                 break;
2827         case CPU_ONLINE:
2828                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2829                        cpu);
2830                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2831                                          NULL, 0, 1);
2832                 break;
2833         }
2834         return NOTIFY_OK;
2835 }
2836
2837 static struct notifier_block kvm_cpu_notifier = {
2838         .notifier_call = kvm_cpu_hotplug,
2839         .priority = 20, /* must be > scheduler priority */
2840 };
2841
2842 static __init void kvm_init_debug(void)
2843 {
2844         struct kvm_stats_debugfs_item *p;
2845
2846         debugfs_dir = debugfs_create_dir("kvm", NULL);
2847         for (p = debugfs_entries; p->name; ++p)
2848                 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2849                                                p->data);
2850 }
2851
2852 static void kvm_exit_debug(void)
2853 {
2854         struct kvm_stats_debugfs_item *p;
2855
2856         for (p = debugfs_entries; p->name; ++p)
2857                 debugfs_remove(p->dentry);
2858         debugfs_remove(debugfs_dir);
2859 }
2860
2861 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2862 {
2863         decache_vcpus_on_cpu(raw_smp_processor_id());
2864         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2865         return 0;
2866 }
2867
2868 static int kvm_resume(struct sys_device *dev)
2869 {
2870         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2871         return 0;
2872 }
2873
2874 static struct sysdev_class kvm_sysdev_class = {
2875         set_kset_name("kvm"),
2876         .suspend = kvm_suspend,
2877         .resume = kvm_resume,
2878 };
2879
2880 static struct sys_device kvm_sysdev = {
2881         .id = 0,
2882         .cls = &kvm_sysdev_class,
2883 };
2884
2885 hpa_t bad_page_address;
2886
2887 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2888                         const char *dev_name, void *data, struct vfsmount *mnt)
2889 {
2890         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2891 }
2892
2893 static struct file_system_type kvm_fs_type = {
2894         .name           = "kvmfs",
2895         .get_sb         = kvmfs_get_sb,
2896         .kill_sb        = kill_anon_super,
2897 };
2898
2899 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2900 {
2901         int r;
2902
2903         if (kvm_arch_ops) {
2904                 printk(KERN_ERR "kvm: already loaded the other module\n");
2905                 return -EEXIST;
2906         }
2907
2908         if (!ops->cpu_has_kvm_support()) {
2909                 printk(KERN_ERR "kvm: no hardware support\n");
2910                 return -EOPNOTSUPP;
2911         }
2912         if (ops->disabled_by_bios()) {
2913                 printk(KERN_ERR "kvm: disabled by bios\n");
2914                 return -EOPNOTSUPP;
2915         }
2916
2917         kvm_arch_ops = ops;
2918
2919         r = kvm_arch_ops->hardware_setup();
2920         if (r < 0)
2921                 goto out;
2922
2923         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2924         r = register_cpu_notifier(&kvm_cpu_notifier);
2925         if (r)
2926                 goto out_free_1;
2927         register_reboot_notifier(&kvm_reboot_notifier);
2928
2929         r = sysdev_class_register(&kvm_sysdev_class);
2930         if (r)
2931                 goto out_free_2;
2932
2933         r = sysdev_register(&kvm_sysdev);
2934         if (r)
2935                 goto out_free_3;
2936
2937         kvm_chardev_ops.owner = module;
2938
2939         r = misc_register(&kvm_dev);
2940         if (r) {
2941                 printk (KERN_ERR "kvm: misc device register failed\n");
2942                 goto out_free;
2943         }
2944
2945         return r;
2946
2947 out_free:
2948         sysdev_unregister(&kvm_sysdev);
2949 out_free_3:
2950         sysdev_class_unregister(&kvm_sysdev_class);
2951 out_free_2:
2952         unregister_reboot_notifier(&kvm_reboot_notifier);
2953         unregister_cpu_notifier(&kvm_cpu_notifier);
2954 out_free_1:
2955         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2956         kvm_arch_ops->hardware_unsetup();
2957 out:
2958         kvm_arch_ops = NULL;
2959         return r;
2960 }
2961
2962 void kvm_exit_arch(void)
2963 {
2964         misc_deregister(&kvm_dev);
2965         sysdev_unregister(&kvm_sysdev);
2966         sysdev_class_unregister(&kvm_sysdev_class);
2967         unregister_reboot_notifier(&kvm_reboot_notifier);
2968         unregister_cpu_notifier(&kvm_cpu_notifier);
2969         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2970         kvm_arch_ops->hardware_unsetup();
2971         kvm_arch_ops = NULL;
2972 }
2973
2974 static __init int kvm_init(void)
2975 {
2976         static struct page *bad_page;
2977         int r;
2978
2979         r = register_filesystem(&kvm_fs_type);
2980         if (r)
2981                 goto out3;
2982
2983         kvmfs_mnt = kern_mount(&kvm_fs_type);
2984         r = PTR_ERR(kvmfs_mnt);
2985         if (IS_ERR(kvmfs_mnt))
2986                 goto out2;
2987         kvm_init_debug();
2988
2989         kvm_init_msr_list();
2990
2991         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2992                 r = -ENOMEM;
2993                 goto out;
2994         }
2995
2996         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2997         memset(__va(bad_page_address), 0, PAGE_SIZE);
2998
2999         return 0;
3000
3001 out:
3002         kvm_exit_debug();
3003         mntput(kvmfs_mnt);
3004 out2:
3005         unregister_filesystem(&kvm_fs_type);
3006 out3:
3007         return r;
3008 }
3009
3010 static __exit void kvm_exit(void)
3011 {
3012         kvm_exit_debug();
3013         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3014         mntput(kvmfs_mnt);
3015         unregister_filesystem(&kvm_fs_type);
3016 }
3017
3018 module_init(kvm_init)
3019 module_exit(kvm_exit)
3020
3021 EXPORT_SYMBOL_GPL(kvm_init_arch);
3022 EXPORT_SYMBOL_GPL(kvm_exit_arch);