drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19
  20 #include <linux/kvm.h>
  21 #include <linux/module.h>
  22 #include <linux/errno.h>
  23 #include <linux/magic.h>
  24 #include <asm/processor.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <asm/msr.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <asm/uaccess.h>
  32 #include <linux/reboot.h>
  33 #include <asm/io.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/highmem.h>
  36 #include <linux/file.h>
  37 #include <asm/desc.h>
  38 #include <linux/sysdev.h>
  39 #include <linux/cpu.h>
  40 #include <linux/file.h>
  41 #include <linux/fs.h>
  42 #include <linux/mount.h>
  43 #include <linux/sched.h>
  44
  45 #include "x86_emulate.h"
  46 #include "segment_descriptor.h"
  47
  48 MODULE_AUTHOR("Qumranet");
  49 MODULE_LICENSE("GPL");
  50
  51 static DEFINE_SPINLOCK(kvm_lock);
  52 static LIST_HEAD(vm_list);
  53
  54 struct kvm_arch_ops *kvm_arch_ops;
  55
  56 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  57
  58 static struct kvm_stats_debugfs_item {
  59         const char *name;
  60         int offset;
  61         struct dentry *dentry;
  62 } debugfs_entries[] = {
  63         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  64         { "pf_guest", STAT_OFFSET(pf_guest) },
  65         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  66         { "invlpg", STAT_OFFSET(invlpg) },
  67         { "exits", STAT_OFFSET(exits) },
  68         { "io_exits", STAT_OFFSET(io_exits) },
  69         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  70         { "signal_exits", STAT_OFFSET(signal_exits) },
  71         { "irq_window", STAT_OFFSET(irq_window_exits) },
  72         { "halt_exits", STAT_OFFSET(halt_exits) },
  73         { "request_irq", STAT_OFFSET(request_irq_exits) },
  74         { "irq_exits", STAT_OFFSET(irq_exits) },
  75         { NULL }
  76 };
  77
  78 static struct dentry *debugfs_dir;
  79
  80 struct vfsmount *kvmfs_mnt;
  81
  82 #define MAX_IO_MSRS 256
  83
  84 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  85 #define LMSW_GUEST_MASK 0x0eULL
  86 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  87 #define CR8_RESEVED_BITS (~0x0fULL)
  88 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  89
  90 #ifdef CONFIG_X86_64
  91 // LDT or TSS descriptor in the GDT. 16 bytes.
  92 struct segment_descriptor_64 {
  93         struct segment_descriptor s;
  94         u32 base_higher;
  95         u32 pad_zero;
  96 };
  97
  98 #endif
  99
 100 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 101                            unsigned long arg);
 102
 103 static struct inode *kvmfs_inode(struct file_operations *fops)
 104 {
 105         int error = -ENOMEM;
 106         struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
 107
 108         if (!inode)
 109                 goto eexit_1;
 110
 111         inode->i_fop = fops;
 112
 113         /*
 114          * Mark the inode dirty from the very beginning,
 115          * that way it will never be moved to the dirty
 116          * list because mark_inode_dirty() will think
 117          * that it already _is_ on the dirty list.
 118          */
 119         inode->i_state = I_DIRTY;
 120         inode->i_mode = S_IRUSR | S_IWUSR;
 121         inode->i_uid = current->fsuid;
 122         inode->i_gid = current->fsgid;
 123         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 124         return inode;
 125
 126 eexit_1:
 127         return ERR_PTR(error);
 128 }
 129
 130 static struct file *kvmfs_file(struct inode *inode, void *private_data)
 131 {
 132         struct file *file = get_empty_filp();
 133
 134         if (!file)
 135                 return ERR_PTR(-ENFILE);
 136
 137         file->f_path.mnt = mntget(kvmfs_mnt);
 138         file->f_path.dentry = d_alloc_anon(inode);
 139         if (!file->f_path.dentry)
 140                 return ERR_PTR(-ENOMEM);
 141         file->f_mapping = inode->i_mapping;
 142
 143         file->f_pos = 0;
 144         file->f_flags = O_RDWR;
 145         file->f_op = inode->i_fop;
 146         file->f_mode = FMODE_READ | FMODE_WRITE;
 147         file->f_version = 0;
 148         file->private_data = private_data;
 149         return file;
 150 }
 151
 152 unsigned long segment_base(u16 selector)
 153 {
 154         struct descriptor_table gdt;
 155         struct segment_descriptor *d;
 156         unsigned long table_base;
 157         typedef unsigned long ul;
 158         unsigned long v;
 159
 160         if (selector == 0)
 161                 return 0;
 162
 163         asm ("sgdt %0" : "=m"(gdt));
 164         table_base = gdt.base;
 165
 166         if (selector & 4) {           /* from ldt */
 167                 u16 ldt_selector;
 168
 169                 asm ("sldt %0" : "=g"(ldt_selector));
 170                 table_base = segment_base(ldt_selector);
 171         }
 172         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 173         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 174 #ifdef CONFIG_X86_64
 175         if (d->system == 0
 176             && (d->type == 2 || d->type == 9 || d->type == 11))
 177                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 178 #endif
 179         return v;
 180 }
 181 EXPORT_SYMBOL_GPL(segment_base);
 182
 183 static inline int valid_vcpu(int n)
 184 {
 185         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 186 }
 187
 188 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 189                    void *dest)
 190 {
 191         unsigned char *host_buf = dest;
 192         unsigned long req_size = size;
 193
 194         while (size) {
 195                 hpa_t paddr;
 196                 unsigned now;
 197                 unsigned offset;
 198                 hva_t guest_buf;
 199
 200                 paddr = gva_to_hpa(vcpu, addr);
 201
 202                 if (is_error_hpa(paddr))
 203                         break;
 204
 205                 guest_buf = (hva_t)kmap_atomic(
 206                                         pfn_to_page(paddr >> PAGE_SHIFT),
 207                                         KM_USER0);
 208                 offset = addr & ~PAGE_MASK;
 209                 guest_buf |= offset;
 210                 now = min(size, PAGE_SIZE - offset);
 211                 memcpy(host_buf, (void*)guest_buf, now);
 212                 host_buf += now;
 213                 addr += now;
 214                 size -= now;
 215                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 216         }
 217         return req_size - size;
 218 }
 219 EXPORT_SYMBOL_GPL(kvm_read_guest);
 220
 221 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 222                     void *data)
 223 {
 224         unsigned char *host_buf = data;
 225         unsigned long req_size = size;
 226
 227         while (size) {
 228                 hpa_t paddr;
 229                 unsigned now;
 230                 unsigned offset;
 231                 hva_t guest_buf;
 232                 gfn_t gfn;
 233
 234                 paddr = gva_to_hpa(vcpu, addr);
 235
 236                 if (is_error_hpa(paddr))
 237                         break;
 238
 239                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 240                 mark_page_dirty(vcpu->kvm, gfn);
 241                 guest_buf = (hva_t)kmap_atomic(
 242                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 243                 offset = addr & ~PAGE_MASK;
 244                 guest_buf |= offset;
 245                 now = min(size, PAGE_SIZE - offset);
 246                 memcpy((void*)guest_buf, host_buf, now);
 247                 host_buf += now;
 248                 addr += now;
 249                 size -= now;
 250                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 251         }
 252         return req_size - size;
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_write_guest);
 255
 256 /*
 257  * Switches to specified vcpu, until a matching vcpu_put()
 258  */
 259 static void vcpu_load(struct kvm_vcpu *vcpu)
 260 {
 261         mutex_lock(&vcpu->mutex);
 262         kvm_arch_ops->vcpu_load(vcpu);
 263 }
 264
 265 /*
 266  * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
 267  * if the slot is not populated.
 268  */
 269 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
 270 {
 271         struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
 272
 273         mutex_lock(&vcpu->mutex);
 274         if (!vcpu->vmcs) {
 275                 mutex_unlock(&vcpu->mutex);
 276                 return NULL;
 277         }
 278         kvm_arch_ops->vcpu_load(vcpu);
 279         return vcpu;
 280 }
 281
 282 static void vcpu_put(struct kvm_vcpu *vcpu)
 283 {
 284         kvm_arch_ops->vcpu_put(vcpu);
 285         mutex_unlock(&vcpu->mutex);
 286 }
 287
 288 static struct kvm *kvm_create_vm(void)
 289 {
 290         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 291         int i;
 292
 293         if (!kvm)
 294                 return ERR_PTR(-ENOMEM);
 295
 296         spin_lock_init(&kvm->lock);
 297         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 298         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 299                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 300
 301                 mutex_init(&vcpu->mutex);
 302                 vcpu->cpu = -1;
 303                 vcpu->kvm = kvm;
 304                 vcpu->mmu.root_hpa = INVALID_PAGE;
 305                 INIT_LIST_HEAD(&vcpu->free_pages);
 306                 spin_lock(&kvm_lock);
 307                 list_add(&kvm->vm_list, &vm_list);
 308                 spin_unlock(&kvm_lock);
 309         }
 310         return kvm;
 311 }
 312
 313 static int kvm_dev_open(struct inode *inode, struct file *filp)
 314 {
 315         return 0;
 316 }
 317
 318 /*
 319  * Free any memory in @free but not in @dont.
 320  */
 321 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 322                                   struct kvm_memory_slot *dont)
 323 {
 324         int i;
 325
 326         if (!dont || free->phys_mem != dont->phys_mem)
 327                 if (free->phys_mem) {
 328                         for (i = 0; i < free->npages; ++i)
 329                                 if (free->phys_mem[i])
 330                                         __free_page(free->phys_mem[i]);
 331                         vfree(free->phys_mem);
 332                 }
 333
 334         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 335                 vfree(free->dirty_bitmap);
 336
 337         free->phys_mem = NULL;
 338         free->npages = 0;
 339         free->dirty_bitmap = NULL;
 340 }
 341
 342 static void kvm_free_physmem(struct kvm *kvm)
 343 {
 344         int i;
 345
 346         for (i = 0; i < kvm->nmemslots; ++i)
 347                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 348 }
 349
 350 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 351 {
 352         int i;
 353
 354         for (i = 0; i < 2; ++i)
 355                 if (vcpu->pio.guest_pages[i]) {
 356                         __free_page(vcpu->pio.guest_pages[i]);
 357                         vcpu->pio.guest_pages[i] = NULL;
 358                 }
 359 }
 360
 361 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 362 {
 363         if (!vcpu->vmcs)
 364                 return;
 365
 366         vcpu_load(vcpu);
 367         kvm_mmu_destroy(vcpu);
 368         vcpu_put(vcpu);
 369         kvm_arch_ops->vcpu_free(vcpu);
 370         free_page((unsigned long)vcpu->run);
 371         vcpu->run = NULL;
 372         free_page((unsigned long)vcpu->pio_data);
 373         vcpu->pio_data = NULL;
 374         free_pio_guest_pages(vcpu);
 375 }
 376
 377 static void kvm_free_vcpus(struct kvm *kvm)
 378 {
 379         unsigned int i;
 380
 381         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 382                 kvm_free_vcpu(&kvm->vcpus[i]);
 383 }
 384
 385 static int kvm_dev_release(struct inode *inode, struct file *filp)
 386 {
 387         return 0;
 388 }
 389
 390 static void kvm_destroy_vm(struct kvm *kvm)
 391 {
 392         spin_lock(&kvm_lock);
 393         list_del(&kvm->vm_list);
 394         spin_unlock(&kvm_lock);
 395         kvm_free_vcpus(kvm);
 396         kvm_free_physmem(kvm);
 397         kfree(kvm);
 398 }
 399
 400 static int kvm_vm_release(struct inode *inode, struct file *filp)
 401 {
 402         struct kvm *kvm = filp->private_data;
 403
 404         kvm_destroy_vm(kvm);
 405         return 0;
 406 }
 407
 408 static void inject_gp(struct kvm_vcpu *vcpu)
 409 {
 410         kvm_arch_ops->inject_gp(vcpu, 0);
 411 }
 412
 413 /*
 414  * Load the pae pdptrs.  Return true is they are all valid.
 415  */
 416 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 417 {
 418         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 419         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 420         int i;
 421         u64 pdpte;
 422         u64 *pdpt;
 423         int ret;
 424         struct page *page;
 425
 426         spin_lock(&vcpu->kvm->lock);
 427         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 428         /* FIXME: !page - emulate? 0xff? */
 429         pdpt = kmap_atomic(page, KM_USER0);
 430
 431         ret = 1;
 432         for (i = 0; i < 4; ++i) {
 433                 pdpte = pdpt[offset + i];
 434                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 435                         ret = 0;
 436                         goto out;
 437                 }
 438         }
 439
 440         for (i = 0; i < 4; ++i)
 441                 vcpu->pdptrs[i] = pdpt[offset + i];
 442
 443 out:
 444         kunmap_atomic(pdpt, KM_USER0);
 445         spin_unlock(&vcpu->kvm->lock);
 446
 447         return ret;
 448 }
 449
 450 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 451 {
 452         if (cr0 & CR0_RESEVED_BITS) {
 453                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 454                        cr0, vcpu->cr0);
 455                 inject_gp(vcpu);
 456                 return;
 457         }
 458
 459         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 460                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 461                 inject_gp(vcpu);
 462                 return;
 463         }
 464
 465         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 466                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 467                        "and a clear PE flag\n");
 468                 inject_gp(vcpu);
 469                 return;
 470         }
 471
 472         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 473 #ifdef CONFIG_X86_64
 474                 if ((vcpu->shadow_efer & EFER_LME)) {
 475                         int cs_db, cs_l;
 476
 477                         if (!is_pae(vcpu)) {
 478                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 479                                        "in long mode while PAE is disabled\n");
 480                                 inject_gp(vcpu);
 481                                 return;
 482                         }
 483                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 484                         if (cs_l) {
 485                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 486                                        "in long mode while CS.L == 1\n");
 487                                 inject_gp(vcpu);
 488                                 return;
 489
 490                         }
 491                 } else
 492 #endif
 493                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 494                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 495                                "reserved bits\n");
 496                         inject_gp(vcpu);
 497                         return;
 498                 }
 499
 500         }
 501
 502         kvm_arch_ops->set_cr0(vcpu, cr0);
 503         vcpu->cr0 = cr0;
 504
 505         spin_lock(&vcpu->kvm->lock);
 506         kvm_mmu_reset_context(vcpu);
 507         spin_unlock(&vcpu->kvm->lock);
 508         return;
 509 }
 510 EXPORT_SYMBOL_GPL(set_cr0);
 511
 512 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 513 {
 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                              void *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                               const void *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                                   void *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                                const void *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                                    const void *val,
1068                                    unsigned int bytes,
1069                                    struct x86_emulate_ctxt *ctxt)
1070 {
1071         struct kvm_vcpu *vcpu = ctxt->vcpu;
1072         gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1073
1074         if (gpa == UNMAPPED_GVA) {
1075                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1076                 return X86EMUL_PROPAGATE_FAULT;
1077         }
1078
1079         if (emulator_write_phys(vcpu, gpa, val, bytes))
1080                 return X86EMUL_CONTINUE;
1081
1082         vcpu->mmio_needed = 1;
1083         vcpu->mmio_phys_addr = gpa;
1084         vcpu->mmio_size = bytes;
1085         vcpu->mmio_is_write = 1;
1086         memcpy(vcpu->mmio_data, val, bytes);
1087
1088         return X86EMUL_CONTINUE;
1089 }
1090
1091 static int emulator_cmpxchg_emulated(unsigned long addr,
1092                                      const void *old,
1093                                      const void *new,
1094                                      unsigned int bytes,
1095                                      struct x86_emulate_ctxt *ctxt)
1096 {
1097         static int reported;
1098
1099         if (!reported) {
1100                 reported = 1;
1101                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1102         }
1103         return emulator_write_emulated(addr, new, bytes, ctxt);
1104 }
1105
1106 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1107 {
1108         return kvm_arch_ops->get_segment_base(vcpu, seg);
1109 }
1110
1111 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1112 {
1113         return X86EMUL_CONTINUE;
1114 }
1115
1116 int emulate_clts(struct kvm_vcpu *vcpu)
1117 {
1118         unsigned long cr0;
1119
1120         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1121         kvm_arch_ops->set_cr0(vcpu, cr0);
1122         return X86EMUL_CONTINUE;
1123 }
1124
1125 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1126 {
1127         struct kvm_vcpu *vcpu = ctxt->vcpu;
1128
1129         switch (dr) {
1130         case 0 ... 3:
1131                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1132                 return X86EMUL_CONTINUE;
1133         default:
1134                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1135                        __FUNCTION__, dr);
1136                 return X86EMUL_UNHANDLEABLE;
1137         }
1138 }
1139
1140 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1141 {
1142         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1143         int exception;
1144
1145         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1146         if (exception) {
1147                 /* FIXME: better handling */
1148                 return X86EMUL_UNHANDLEABLE;
1149         }
1150         return X86EMUL_CONTINUE;
1151 }
1152
1153 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1154 {
1155         static int reported;
1156         u8 opcodes[4];
1157         unsigned long rip = ctxt->vcpu->rip;
1158         unsigned long rip_linear;
1159
1160         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1161
1162         if (reported)
1163                 return;
1164
1165         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1166
1167         printk(KERN_ERR "emulation failed but !mmio_needed?"
1168                " rip %lx %02x %02x %02x %02x\n",
1169                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1170         reported = 1;
1171 }
1172
1173 struct x86_emulate_ops emulate_ops = {
1174         .read_std            = emulator_read_std,
1175         .write_std           = emulator_write_std,
1176         .read_emulated       = emulator_read_emulated,
1177         .write_emulated      = emulator_write_emulated,
1178         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1179 };
1180
1181 int emulate_instruction(struct kvm_vcpu *vcpu,
1182                         struct kvm_run *run,
1183                         unsigned long cr2,
1184                         u16 error_code)
1185 {
1186         struct x86_emulate_ctxt emulate_ctxt;
1187         int r;
1188         int cs_db, cs_l;
1189
1190         vcpu->mmio_fault_cr2 = cr2;
1191         kvm_arch_ops->cache_regs(vcpu);
1192
1193         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1194
1195         emulate_ctxt.vcpu = vcpu;
1196         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1197         emulate_ctxt.cr2 = cr2;
1198         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1199                 ? X86EMUL_MODE_REAL : cs_l
1200                 ? X86EMUL_MODE_PROT64 : cs_db
1201                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1202
1203         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1204                 emulate_ctxt.cs_base = 0;
1205                 emulate_ctxt.ds_base = 0;
1206                 emulate_ctxt.es_base = 0;
1207                 emulate_ctxt.ss_base = 0;
1208         } else {
1209                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1210                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1211                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1212                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1213         }
1214
1215         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1216         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1217
1218         vcpu->mmio_is_write = 0;
1219         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1220
1221         if ((r || vcpu->mmio_is_write) && run) {
1222                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1223                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1224                 run->mmio.len = vcpu->mmio_size;
1225                 run->mmio.is_write = vcpu->mmio_is_write;
1226         }
1227
1228         if (r) {
1229                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1230                         return EMULATE_DONE;
1231                 if (!vcpu->mmio_needed) {
1232                         report_emulation_failure(&emulate_ctxt);
1233                         return EMULATE_FAIL;
1234                 }
1235                 return EMULATE_DO_MMIO;
1236         }
1237
1238         kvm_arch_ops->decache_regs(vcpu);
1239         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1240
1241         if (vcpu->mmio_is_write) {
1242                 vcpu->mmio_needed = 0;
1243                 return EMULATE_DO_MMIO;
1244         }
1245
1246         return EMULATE_DONE;
1247 }
1248 EXPORT_SYMBOL_GPL(emulate_instruction);
1249
1250 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1251 {
1252         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1253
1254         kvm_arch_ops->cache_regs(vcpu);
1255         ret = -KVM_EINVAL;
1256 #ifdef CONFIG_X86_64
1257         if (is_long_mode(vcpu)) {
1258                 nr = vcpu->regs[VCPU_REGS_RAX];
1259                 a0 = vcpu->regs[VCPU_REGS_RDI];
1260                 a1 = vcpu->regs[VCPU_REGS_RSI];
1261                 a2 = vcpu->regs[VCPU_REGS_RDX];
1262                 a3 = vcpu->regs[VCPU_REGS_RCX];
1263                 a4 = vcpu->regs[VCPU_REGS_R8];
1264                 a5 = vcpu->regs[VCPU_REGS_R9];
1265         } else
1266 #endif
1267         {
1268                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1269                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1270                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1271                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1272                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1273                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1274                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1275         }
1276         switch (nr) {
1277         default:
1278                 run->hypercall.args[0] = a0;
1279                 run->hypercall.args[1] = a1;
1280                 run->hypercall.args[2] = a2;
1281                 run->hypercall.args[3] = a3;
1282                 run->hypercall.args[4] = a4;
1283                 run->hypercall.args[5] = a5;
1284                 run->hypercall.ret = ret;
1285                 run->hypercall.longmode = is_long_mode(vcpu);
1286                 kvm_arch_ops->decache_regs(vcpu);
1287                 return 0;
1288         }
1289         vcpu->regs[VCPU_REGS_RAX] = ret;
1290         kvm_arch_ops->decache_regs(vcpu);
1291         return 1;
1292 }
1293 EXPORT_SYMBOL_GPL(kvm_hypercall);
1294
1295 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1296 {
1297         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1298 }
1299
1300 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1301 {
1302         struct descriptor_table dt = { limit, base };
1303
1304         kvm_arch_ops->set_gdt(vcpu, &dt);
1305 }
1306
1307 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1308 {
1309         struct descriptor_table dt = { limit, base };
1310
1311         kvm_arch_ops->set_idt(vcpu, &dt);
1312 }
1313
1314 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1315                    unsigned long *rflags)
1316 {
1317         lmsw(vcpu, msw);
1318         *rflags = kvm_arch_ops->get_rflags(vcpu);
1319 }
1320
1321 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1322 {
1323         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1324         switch (cr) {
1325         case 0:
1326                 return vcpu->cr0;
1327         case 2:
1328                 return vcpu->cr2;
1329         case 3:
1330                 return vcpu->cr3;
1331         case 4:
1332                 return vcpu->cr4;
1333         default:
1334                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1335                 return 0;
1336         }
1337 }
1338
1339 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1340                      unsigned long *rflags)
1341 {
1342         switch (cr) {
1343         case 0:
1344                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1345                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1346                 break;
1347         case 2:
1348                 vcpu->cr2 = val;
1349                 break;
1350         case 3:
1351                 set_cr3(vcpu, val);
1352                 break;
1353         case 4:
1354                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1355                 break;
1356         default:
1357                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1358         }
1359 }
1360
1361 /*
1362  * Register the para guest with the host:
1363  */
1364 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1365 {
1366         struct kvm_vcpu_para_state *para_state;
1367         hpa_t para_state_hpa, hypercall_hpa;
1368         struct page *para_state_page;
1369         unsigned char *hypercall;
1370         gpa_t hypercall_gpa;
1371
1372         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1373         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1374
1375         /*
1376          * Needs to be page aligned:
1377          */
1378         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1379                 goto err_gp;
1380
1381         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1382         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1383         if (is_error_hpa(para_state_hpa))
1384                 goto err_gp;
1385
1386         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1387         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1388         para_state = kmap_atomic(para_state_page, KM_USER0);
1389
1390         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1391         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1392
1393         para_state->host_version = KVM_PARA_API_VERSION;
1394         /*
1395          * We cannot support guests that try to register themselves
1396          * with a newer API version than the host supports:
1397          */
1398         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1399                 para_state->ret = -KVM_EINVAL;
1400                 goto err_kunmap_skip;
1401         }
1402
1403         hypercall_gpa = para_state->hypercall_gpa;
1404         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1405         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1406         if (is_error_hpa(hypercall_hpa)) {
1407                 para_state->ret = -KVM_EINVAL;
1408                 goto err_kunmap_skip;
1409         }
1410
1411         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1412         vcpu->para_state_page = para_state_page;
1413         vcpu->para_state_gpa = para_state_gpa;
1414         vcpu->hypercall_gpa = hypercall_gpa;
1415
1416         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1417         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1418                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1419         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1420         kunmap_atomic(hypercall, KM_USER1);
1421
1422         para_state->ret = 0;
1423 err_kunmap_skip:
1424         kunmap_atomic(para_state, KM_USER0);
1425         return 0;
1426 err_gp:
1427         return 1;
1428 }
1429
1430 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1431 {
1432         u64 data;
1433
1434         switch (msr) {
1435         case 0xc0010010: /* SYSCFG */
1436         case 0xc0010015: /* HWCR */
1437         case MSR_IA32_PLATFORM_ID:
1438         case MSR_IA32_P5_MC_ADDR:
1439         case MSR_IA32_P5_MC_TYPE:
1440         case MSR_IA32_MC0_CTL:
1441         case MSR_IA32_MCG_STATUS:
1442         case MSR_IA32_MCG_CAP:
1443         case MSR_IA32_MC0_MISC:
1444         case MSR_IA32_MC0_MISC+4:
1445         case MSR_IA32_MC0_MISC+8:
1446         case MSR_IA32_MC0_MISC+12:
1447         case MSR_IA32_MC0_MISC+16:
1448         case MSR_IA32_UCODE_REV:
1449         case MSR_IA32_PERF_STATUS:
1450                 /* MTRR registers */
1451         case 0xfe:
1452         case 0x200 ... 0x2ff:
1453                 data = 0;
1454                 break;
1455         case 0xcd: /* fsb frequency */
1456                 data = 3;
1457                 break;
1458         case MSR_IA32_APICBASE:
1459                 data = vcpu->apic_base;
1460                 break;
1461         case MSR_IA32_MISC_ENABLE:
1462                 data = vcpu->ia32_misc_enable_msr;
1463                 break;
1464 #ifdef CONFIG_X86_64
1465         case MSR_EFER:
1466                 data = vcpu->shadow_efer;
1467                 break;
1468 #endif
1469         default:
1470                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1471                 return 1;
1472         }
1473         *pdata = data;
1474         return 0;
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1477
1478 /*
1479  * Reads an msr value (of 'msr_index') into 'pdata'.
1480  * Returns 0 on success, non-0 otherwise.
1481  * Assumes vcpu_load() was already called.
1482  */
1483 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1484 {
1485         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1486 }
1487
1488 #ifdef CONFIG_X86_64
1489
1490 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1491 {
1492         if (efer & EFER_RESERVED_BITS) {
1493                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1494                        efer);
1495                 inject_gp(vcpu);
1496                 return;
1497         }
1498
1499         if (is_paging(vcpu)
1500             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1501                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1502                 inject_gp(vcpu);
1503                 return;
1504         }
1505
1506         kvm_arch_ops->set_efer(vcpu, efer);
1507
1508         efer &= ~EFER_LMA;
1509         efer |= vcpu->shadow_efer & EFER_LMA;
1510
1511         vcpu->shadow_efer = efer;
1512 }
1513
1514 #endif
1515
1516 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1517 {
1518         switch (msr) {
1519 #ifdef CONFIG_X86_64
1520         case MSR_EFER:
1521                 set_efer(vcpu, data);
1522                 break;
1523 #endif
1524         case MSR_IA32_MC0_STATUS:
1525                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1526                        __FUNCTION__, data);
1527                 break;
1528         case MSR_IA32_MCG_STATUS:
1529                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1530                         __FUNCTION__, data);
1531                 break;
1532         case MSR_IA32_UCODE_REV:
1533         case MSR_IA32_UCODE_WRITE:
1534         case 0x200 ... 0x2ff: /* MTRRs */
1535                 break;
1536         case MSR_IA32_APICBASE:
1537                 vcpu->apic_base = data;
1538                 break;
1539         case MSR_IA32_MISC_ENABLE:
1540                 vcpu->ia32_misc_enable_msr = data;
1541                 break;
1542         /*
1543          * This is the 'probe whether the host is KVM' logic:
1544          */
1545         case MSR_KVM_API_MAGIC:
1546                 return vcpu_register_para(vcpu, data);
1547
1548         default:
1549                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1550                 return 1;
1551         }
1552         return 0;
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1555
1556 /*
1557  * Writes msr value into into the appropriate "register".
1558  * Returns 0 on success, non-0 otherwise.
1559  * Assumes vcpu_load() was already called.
1560  */
1561 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1562 {
1563         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1564 }
1565
1566 void kvm_resched(struct kvm_vcpu *vcpu)
1567 {
1568         if (!need_resched())
1569                 return;
1570         vcpu_put(vcpu);
1571         cond_resched();
1572         vcpu_load(vcpu);
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_resched);
1575
1576 void load_msrs(struct vmx_msr_entry *e, int n)
1577 {
1578         int i;
1579
1580         for (i = 0; i < n; ++i)
1581                 wrmsrl(e[i].index, e[i].data);
1582 }
1583 EXPORT_SYMBOL_GPL(load_msrs);
1584
1585 void save_msrs(struct vmx_msr_entry *e, int n)
1586 {
1587         int i;
1588
1589         for (i = 0; i < n; ++i)
1590                 rdmsrl(e[i].index, e[i].data);
1591 }
1592 EXPORT_SYMBOL_GPL(save_msrs);
1593
1594 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1595 {
1596         int i;
1597         u32 function;
1598         struct kvm_cpuid_entry *e, *best;
1599
1600         kvm_arch_ops->cache_regs(vcpu);
1601         function = vcpu->regs[VCPU_REGS_RAX];
1602         vcpu->regs[VCPU_REGS_RAX] = 0;
1603         vcpu->regs[VCPU_REGS_RBX] = 0;
1604         vcpu->regs[VCPU_REGS_RCX] = 0;
1605         vcpu->regs[VCPU_REGS_RDX] = 0;
1606         best = NULL;
1607         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1608                 e = &vcpu->cpuid_entries[i];
1609                 if (e->function == function) {
1610                         best = e;
1611                         break;
1612                 }
1613                 /*
1614                  * Both basic or both extended?
1615                  */
1616                 if (((e->function ^ function) & 0x80000000) == 0)
1617                         if (!best || e->function > best->function)
1618                                 best = e;
1619         }
1620         if (best) {
1621                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1622                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1623                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1624                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1625         }
1626         kvm_arch_ops->decache_regs(vcpu);
1627         kvm_arch_ops->skip_emulated_instruction(vcpu);
1628 }
1629 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1630
1631 static int pio_copy_data(struct kvm_vcpu *vcpu)
1632 {
1633         void *p = vcpu->pio_data;
1634         void *q;
1635         unsigned bytes;
1636         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1637
1638         kvm_arch_ops->vcpu_put(vcpu);
1639         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1640                  PAGE_KERNEL);
1641         if (!q) {
1642                 kvm_arch_ops->vcpu_load(vcpu);
1643                 free_pio_guest_pages(vcpu);
1644                 return -ENOMEM;
1645         }
1646         q += vcpu->pio.guest_page_offset;
1647         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1648         if (vcpu->pio.in)
1649                 memcpy(q, p, bytes);
1650         else
1651                 memcpy(p, q, bytes);
1652         q -= vcpu->pio.guest_page_offset;
1653         vunmap(q);
1654         kvm_arch_ops->vcpu_load(vcpu);
1655         free_pio_guest_pages(vcpu);
1656         return 0;
1657 }
1658
1659 static int complete_pio(struct kvm_vcpu *vcpu)
1660 {
1661         struct kvm_pio_request *io = &vcpu->pio;
1662         long delta;
1663         int r;
1664
1665         kvm_arch_ops->cache_regs(vcpu);
1666
1667         if (!io->string) {
1668                 if (io->in)
1669                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1670                                io->size);
1671         } else {
1672                 if (io->in) {
1673                         r = pio_copy_data(vcpu);
1674                         if (r) {
1675                                 kvm_arch_ops->cache_regs(vcpu);
1676                                 return r;
1677                         }
1678                 }
1679
1680                 delta = 1;
1681                 if (io->rep) {
1682                         delta *= io->cur_count;
1683                         /*
1684                          * The size of the register should really depend on
1685                          * current address size.
1686                          */
1687                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1688                 }
1689                 if (io->down)
1690                         delta = -delta;
1691                 delta *= io->size;
1692                 if (io->in)
1693                         vcpu->regs[VCPU_REGS_RDI] += delta;
1694                 else
1695                         vcpu->regs[VCPU_REGS_RSI] += delta;
1696         }
1697
1698         kvm_arch_ops->decache_regs(vcpu);
1699
1700         io->count -= io->cur_count;
1701         io->cur_count = 0;
1702
1703         if (!io->count)
1704                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1705         return 0;
1706 }
1707
1708 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1709                   int size, unsigned long count, int string, int down,
1710                   gva_t address, int rep, unsigned port)
1711 {
1712         unsigned now, in_page;
1713         int i;
1714         int nr_pages = 1;
1715         struct page *page;
1716
1717         vcpu->run->exit_reason = KVM_EXIT_IO;
1718         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1719         vcpu->run->io.size = size;
1720         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1721         vcpu->run->io.count = count;
1722         vcpu->run->io.port = port;
1723         vcpu->pio.count = count;
1724         vcpu->pio.cur_count = count;
1725         vcpu->pio.size = size;
1726         vcpu->pio.in = in;
1727         vcpu->pio.string = string;
1728         vcpu->pio.down = down;
1729         vcpu->pio.guest_page_offset = offset_in_page(address);
1730         vcpu->pio.rep = rep;
1731
1732         if (!string) {
1733                 kvm_arch_ops->cache_regs(vcpu);
1734                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1735                 kvm_arch_ops->decache_regs(vcpu);
1736                 return 0;
1737         }
1738
1739         if (!count) {
1740                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1741                 return 1;
1742         }
1743
1744         now = min(count, PAGE_SIZE / size);
1745
1746         if (!down)
1747                 in_page = PAGE_SIZE - offset_in_page(address);
1748         else
1749                 in_page = offset_in_page(address) + size;
1750         now = min(count, (unsigned long)in_page / size);
1751         if (!now) {
1752                 /*
1753                  * String I/O straddles page boundary.  Pin two guest pages
1754                  * so that we satisfy atomicity constraints.  Do just one
1755                  * transaction to avoid complexity.
1756                  */
1757                 nr_pages = 2;
1758                 now = 1;
1759         }
1760         if (down) {
1761                 /*
1762                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1763                  */
1764                 printk(KERN_ERR "kvm: guest string pio down\n");
1765                 inject_gp(vcpu);
1766                 return 1;
1767         }
1768         vcpu->run->io.count = now;
1769         vcpu->pio.cur_count = now;
1770
1771         for (i = 0; i < nr_pages; ++i) {
1772                 spin_lock(&vcpu->kvm->lock);
1773                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1774                 if (page)
1775                         get_page(page);
1776                 vcpu->pio.guest_pages[i] = page;
1777                 spin_unlock(&vcpu->kvm->lock);
1778                 if (!page) {
1779                         inject_gp(vcpu);
1780                         free_pio_guest_pages(vcpu);
1781                         return 1;
1782                 }
1783         }
1784
1785         if (!vcpu->pio.in)
1786                 return pio_copy_data(vcpu);
1787         return 0;
1788 }
1789 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1790
1791 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1792 {
1793         int r;
1794         sigset_t sigsaved;
1795
1796         vcpu_load(vcpu);
1797
1798         if (vcpu->sigset_active)
1799                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1800
1801         /* re-sync apic's tpr */
1802         vcpu->cr8 = kvm_run->cr8;
1803
1804         if (vcpu->pio.cur_count) {
1805                 r = complete_pio(vcpu);
1806                 if (r)
1807                         goto out;
1808         }
1809
1810         if (vcpu->mmio_needed) {
1811                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1812                 vcpu->mmio_read_completed = 1;
1813                 vcpu->mmio_needed = 0;
1814                 r = emulate_instruction(vcpu, kvm_run,
1815                                         vcpu->mmio_fault_cr2, 0);
1816                 if (r == EMULATE_DO_MMIO) {
1817                         /*
1818                          * Read-modify-write.  Back to userspace.
1819                          */
1820                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1821                         r = 0;
1822                         goto out;
1823                 }
1824         }
1825
1826         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1827                 kvm_arch_ops->cache_regs(vcpu);
1828                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1829                 kvm_arch_ops->decache_regs(vcpu);
1830         }
1831
1832         r = kvm_arch_ops->run(vcpu, kvm_run);
1833
1834 out:
1835         if (vcpu->sigset_active)
1836                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1837
1838         vcpu_put(vcpu);
1839         return r;
1840 }
1841
1842 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1843                                    struct kvm_regs *regs)
1844 {
1845         vcpu_load(vcpu);
1846
1847         kvm_arch_ops->cache_regs(vcpu);
1848
1849         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1850         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1851         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1852         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1853         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1854         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1855         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1856         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1857 #ifdef CONFIG_X86_64
1858         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1859         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1860         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1861         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1862         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1863         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1864         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1865         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1866 #endif
1867
1868         regs->rip = vcpu->rip;
1869         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1870
1871         /*
1872          * Don't leak debug flags in case they were set for guest debugging
1873          */
1874         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1875                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1876
1877         vcpu_put(vcpu);
1878
1879         return 0;
1880 }
1881
1882 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1883                                    struct kvm_regs *regs)
1884 {
1885         vcpu_load(vcpu);
1886
1887         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1888         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1889         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1890         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1891         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1892         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1893         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1894         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1895 #ifdef CONFIG_X86_64
1896         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1897         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1898         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1899         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1900         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1901         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1902         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1903         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1904 #endif
1905
1906         vcpu->rip = regs->rip;
1907         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1908
1909         kvm_arch_ops->decache_regs(vcpu);
1910
1911         vcpu_put(vcpu);
1912
1913         return 0;
1914 }
1915
1916 static void get_segment(struct kvm_vcpu *vcpu,
1917                         struct kvm_segment *var, int seg)
1918 {
1919         return kvm_arch_ops->get_segment(vcpu, var, seg);
1920 }
1921
1922 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1923                                     struct kvm_sregs *sregs)
1924 {
1925         struct descriptor_table dt;
1926
1927         vcpu_load(vcpu);
1928
1929         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1930         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1931         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1932         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1933         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1934         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1935
1936         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1937         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1938
1939         kvm_arch_ops->get_idt(vcpu, &dt);
1940         sregs->idt.limit = dt.limit;
1941         sregs->idt.base = dt.base;
1942         kvm_arch_ops->get_gdt(vcpu, &dt);
1943         sregs->gdt.limit = dt.limit;
1944         sregs->gdt.base = dt.base;
1945
1946         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1947         sregs->cr0 = vcpu->cr0;
1948         sregs->cr2 = vcpu->cr2;
1949         sregs->cr3 = vcpu->cr3;
1950         sregs->cr4 = vcpu->cr4;
1951         sregs->cr8 = vcpu->cr8;
1952         sregs->efer = vcpu->shadow_efer;
1953         sregs->apic_base = vcpu->apic_base;
1954
1955         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1956                sizeof sregs->interrupt_bitmap);
1957
1958         vcpu_put(vcpu);
1959
1960         return 0;
1961 }
1962
1963 static void set_segment(struct kvm_vcpu *vcpu,
1964                         struct kvm_segment *var, int seg)
1965 {
1966         return kvm_arch_ops->set_segment(vcpu, var, seg);
1967 }
1968
1969 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1970                                     struct kvm_sregs *sregs)
1971 {
1972         int mmu_reset_needed = 0;
1973         int i;
1974         struct descriptor_table dt;
1975
1976         vcpu_load(vcpu);
1977
1978         dt.limit = sregs->idt.limit;
1979         dt.base = sregs->idt.base;
1980         kvm_arch_ops->set_idt(vcpu, &dt);
1981         dt.limit = sregs->gdt.limit;
1982         dt.base = sregs->gdt.base;
1983         kvm_arch_ops->set_gdt(vcpu, &dt);
1984
1985         vcpu->cr2 = sregs->cr2;
1986         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1987         vcpu->cr3 = sregs->cr3;
1988
1989         vcpu->cr8 = sregs->cr8;
1990
1991         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1992 #ifdef CONFIG_X86_64
1993         kvm_arch_ops->set_efer(vcpu, sregs->efer);
1994 #endif
1995         vcpu->apic_base = sregs->apic_base;
1996
1997         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1998
1999         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2000         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2001
2002         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2003         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2004         if (!is_long_mode(vcpu) && is_pae(vcpu))
2005                 load_pdptrs(vcpu, vcpu->cr3);
2006
2007         if (mmu_reset_needed)
2008                 kvm_mmu_reset_context(vcpu);
2009
2010         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2011                sizeof vcpu->irq_pending);
2012         vcpu->irq_summary = 0;
2013         for (i = 0; i < NR_IRQ_WORDS; ++i)
2014                 if (vcpu->irq_pending[i])
2015                         __set_bit(i, &vcpu->irq_summary);
2016
2017         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2018         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2019         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2020         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2021         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2022         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2023
2024         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2025         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2026
2027         vcpu_put(vcpu);
2028
2029         return 0;
2030 }
2031
2032 /*
2033  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2034  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2035  *
2036  * This list is modified at module load time to reflect the
2037  * capabilities of the host cpu.
2038  */
2039 static u32 msrs_to_save[] = {
2040         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2041         MSR_K6_STAR,
2042 #ifdef CONFIG_X86_64
2043         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2044 #endif
2045         MSR_IA32_TIME_STAMP_COUNTER,
2046 };
2047
2048 static unsigned num_msrs_to_save;
2049
2050 static u32 emulated_msrs[] = {
2051         MSR_IA32_MISC_ENABLE,
2052 };
2053
2054 static __init void kvm_init_msr_list(void)
2055 {
2056         u32 dummy[2];
2057         unsigned i, j;
2058
2059         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2060                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2061                         continue;
2062                 if (j < i)
2063                         msrs_to_save[j] = msrs_to_save[i];
2064                 j++;
2065         }
2066         num_msrs_to_save = j;
2067 }
2068
2069 /*
2070  * Adapt set_msr() to msr_io()'s calling convention
2071  */
2072 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2073 {
2074         return set_msr(vcpu, index, *data);
2075 }
2076
2077 /*
2078  * Read or write a bunch of msrs. All parameters are kernel addresses.
2079  *
2080  * @return number of msrs set successfully.
2081  */
2082 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2083                     struct kvm_msr_entry *entries,
2084                     int (*do_msr)(struct kvm_vcpu *vcpu,
2085                                   unsigned index, u64 *data))
2086 {
2087         int i;
2088
2089         vcpu_load(vcpu);
2090
2091         for (i = 0; i < msrs->nmsrs; ++i)
2092                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2093                         break;
2094
2095         vcpu_put(vcpu);
2096
2097         return i;
2098 }
2099
2100 /*
2101  * Read or write a bunch of msrs. Parameters are user addresses.
2102  *
2103  * @return number of msrs set successfully.
2104  */
2105 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2106                   int (*do_msr)(struct kvm_vcpu *vcpu,
2107                                 unsigned index, u64 *data),
2108                   int writeback)
2109 {
2110         struct kvm_msrs msrs;
2111         struct kvm_msr_entry *entries;
2112         int r, n;
2113         unsigned size;
2114
2115         r = -EFAULT;
2116         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2117                 goto out;
2118
2119         r = -E2BIG;
2120         if (msrs.nmsrs >= MAX_IO_MSRS)
2121                 goto out;
2122
2123         r = -ENOMEM;
2124         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2125         entries = vmalloc(size);
2126         if (!entries)
2127                 goto out;
2128
2129         r = -EFAULT;
2130         if (copy_from_user(entries, user_msrs->entries, size))
2131                 goto out_free;
2132
2133         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2134         if (r < 0)
2135                 goto out_free;
2136
2137         r = -EFAULT;
2138         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2139                 goto out_free;
2140
2141         r = n;
2142
2143 out_free:
2144         vfree(entries);
2145 out:
2146         return r;
2147 }
2148
2149 /*
2150  * Translate a guest virtual address to a guest physical address.
2151  */
2152 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2153                                     struct kvm_translation *tr)
2154 {
2155         unsigned long vaddr = tr->linear_address;
2156         gpa_t gpa;
2157
2158         vcpu_load(vcpu);
2159         spin_lock(&vcpu->kvm->lock);
2160         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2161         tr->physical_address = gpa;
2162         tr->valid = gpa != UNMAPPED_GVA;
2163         tr->writeable = 1;
2164         tr->usermode = 0;
2165         spin_unlock(&vcpu->kvm->lock);
2166         vcpu_put(vcpu);
2167
2168         return 0;
2169 }
2170
2171 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2172                                     struct kvm_interrupt *irq)
2173 {
2174         if (irq->irq < 0 || irq->irq >= 256)
2175                 return -EINVAL;
2176         vcpu_load(vcpu);
2177
2178         set_bit(irq->irq, vcpu->irq_pending);
2179         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2180
2181         vcpu_put(vcpu);
2182
2183         return 0;
2184 }
2185
2186 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2187                                       struct kvm_debug_guest *dbg)
2188 {
2189         int r;
2190
2191         vcpu_load(vcpu);
2192
2193         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2194
2195         vcpu_put(vcpu);
2196
2197         return r;
2198 }
2199
2200 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2201                                     unsigned long address,
2202                                     int *type)
2203 {
2204         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2205         unsigned long pgoff;
2206         struct page *page;
2207
2208         *type = VM_FAULT_MINOR;
2209         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2210         if (pgoff == 0)
2211                 page = virt_to_page(vcpu->run);
2212         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2213                 page = virt_to_page(vcpu->pio_data);
2214         else
2215                 return NOPAGE_SIGBUS;
2216         get_page(page);
2217         return page;
2218 }
2219
2220 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2221         .nopage = kvm_vcpu_nopage,
2222 };
2223
2224 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2225 {
2226         vma->vm_ops = &kvm_vcpu_vm_ops;
2227         return 0;
2228 }
2229
2230 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2231 {
2232         struct kvm_vcpu *vcpu = filp->private_data;
2233
2234         fput(vcpu->kvm->filp);
2235         return 0;
2236 }
2237
2238 static struct file_operations kvm_vcpu_fops = {
2239         .release        = kvm_vcpu_release,
2240         .unlocked_ioctl = kvm_vcpu_ioctl,
2241         .compat_ioctl   = kvm_vcpu_ioctl,
2242         .mmap           = kvm_vcpu_mmap,
2243 };
2244
2245 /*
2246  * Allocates an inode for the vcpu.
2247  */
2248 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2249 {
2250         int fd, r;
2251         struct inode *inode;
2252         struct file *file;
2253
2254         atomic_inc(&vcpu->kvm->filp->f_count);
2255         inode = kvmfs_inode(&kvm_vcpu_fops);
2256         if (IS_ERR(inode)) {
2257                 r = PTR_ERR(inode);
2258                 goto out1;
2259         }
2260
2261         file = kvmfs_file(inode, vcpu);
2262         if (IS_ERR(file)) {
2263                 r = PTR_ERR(file);
2264                 goto out2;
2265         }
2266
2267         r = get_unused_fd();
2268         if (r < 0)
2269                 goto out3;
2270         fd = r;
2271         fd_install(fd, file);
2272
2273         return fd;
2274
2275 out3:
2276         fput(file);
2277 out2:
2278         iput(inode);
2279 out1:
2280         fput(vcpu->kvm->filp);
2281         return r;
2282 }
2283
2284 /*
2285  * Creates some virtual cpus.  Good luck creating more than one.
2286  */
2287 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2288 {
2289         int r;
2290         struct kvm_vcpu *vcpu;
2291         struct page *page;
2292
2293         r = -EINVAL;
2294         if (!valid_vcpu(n))
2295                 goto out;
2296
2297         vcpu = &kvm->vcpus[n];
2298
2299         mutex_lock(&vcpu->mutex);
2300
2301         if (vcpu->vmcs) {
2302                 mutex_unlock(&vcpu->mutex);
2303                 return -EEXIST;
2304         }
2305
2306         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2307         r = -ENOMEM;
2308         if (!page)
2309                 goto out_unlock;
2310         vcpu->run = page_address(page);
2311
2312         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2313         r = -ENOMEM;
2314         if (!page)
2315                 goto out_free_run;
2316         vcpu->pio_data = page_address(page);
2317
2318         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2319                                            FX_IMAGE_ALIGN);
2320         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2321         vcpu->cr0 = 0x10;
2322
2323         r = kvm_arch_ops->vcpu_create(vcpu);
2324         if (r < 0)
2325                 goto out_free_vcpus;
2326
2327         r = kvm_mmu_create(vcpu);
2328         if (r < 0)
2329                 goto out_free_vcpus;
2330
2331         kvm_arch_ops->vcpu_load(vcpu);
2332         r = kvm_mmu_setup(vcpu);
2333         if (r >= 0)
2334                 r = kvm_arch_ops->vcpu_setup(vcpu);
2335         vcpu_put(vcpu);
2336
2337         if (r < 0)
2338                 goto out_free_vcpus;
2339
2340         r = create_vcpu_fd(vcpu);
2341         if (r < 0)
2342                 goto out_free_vcpus;
2343
2344         return r;
2345
2346 out_free_vcpus:
2347         kvm_free_vcpu(vcpu);
2348 out_free_run:
2349         free_page((unsigned long)vcpu->run);
2350         vcpu->run = NULL;
2351 out_unlock:
2352         mutex_unlock(&vcpu->mutex);
2353 out:
2354         return r;
2355 }
2356
2357 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2358                                     struct kvm_cpuid *cpuid,
2359                                     struct kvm_cpuid_entry __user *entries)
2360 {
2361         int r;
2362
2363         r = -E2BIG;
2364         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2365                 goto out;
2366         r = -EFAULT;
2367         if (copy_from_user(&vcpu->cpuid_entries, entries,
2368                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2369                 goto out;
2370         vcpu->cpuid_nent = cpuid->nent;
2371         return 0;
2372
2373 out:
2374         return r;
2375 }
2376
2377 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2378 {
2379         if (sigset) {
2380                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2381                 vcpu->sigset_active = 1;
2382                 vcpu->sigset = *sigset;
2383         } else
2384                 vcpu->sigset_active = 0;
2385         return 0;
2386 }
2387
2388 /*
2389  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2390  * we have asm/x86/processor.h
2391  */
2392 struct fxsave {
2393         u16     cwd;
2394         u16     swd;
2395         u16     twd;
2396         u16     fop;
2397         u64     rip;
2398         u64     rdp;
2399         u32     mxcsr;
2400         u32     mxcsr_mask;
2401         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2402 #ifdef CONFIG_X86_64
2403         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2404 #else
2405         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2406 #endif
2407 };
2408
2409 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2410 {
2411         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2412
2413         vcpu_load(vcpu);
2414
2415         memcpy(fpu->fpr, fxsave->st_space, 128);
2416         fpu->fcw = fxsave->cwd;
2417         fpu->fsw = fxsave->swd;
2418         fpu->ftwx = fxsave->twd;
2419         fpu->last_opcode = fxsave->fop;
2420         fpu->last_ip = fxsave->rip;
2421         fpu->last_dp = fxsave->rdp;
2422         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2423
2424         vcpu_put(vcpu);
2425
2426         return 0;
2427 }
2428
2429 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2430 {
2431         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2432
2433         vcpu_load(vcpu);
2434
2435         memcpy(fxsave->st_space, fpu->fpr, 128);
2436         fxsave->cwd = fpu->fcw;
2437         fxsave->swd = fpu->fsw;
2438         fxsave->twd = fpu->ftwx;
2439         fxsave->fop = fpu->last_opcode;
2440         fxsave->rip = fpu->last_ip;
2441         fxsave->rdp = fpu->last_dp;
2442         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2443
2444         vcpu_put(vcpu);
2445
2446         return 0;
2447 }
2448
2449 static long kvm_vcpu_ioctl(struct file *filp,
2450                            unsigned int ioctl, unsigned long arg)
2451 {
2452         struct kvm_vcpu *vcpu = filp->private_data;
2453         void __user *argp = (void __user *)arg;
2454         int r = -EINVAL;
2455
2456         switch (ioctl) {
2457         case KVM_RUN:
2458                 r = -EINVAL;
2459                 if (arg)
2460                         goto out;
2461                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2462                 break;
2463         case KVM_GET_REGS: {
2464                 struct kvm_regs kvm_regs;
2465
2466                 memset(&kvm_regs, 0, sizeof kvm_regs);
2467                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2468                 if (r)
2469                         goto out;
2470                 r = -EFAULT;
2471                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2472                         goto out;
2473                 r = 0;
2474                 break;
2475         }
2476         case KVM_SET_REGS: {
2477                 struct kvm_regs kvm_regs;
2478
2479                 r = -EFAULT;
2480                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2481                         goto out;
2482                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2483                 if (r)
2484                         goto out;
2485                 r = 0;
2486                 break;
2487         }
2488         case KVM_GET_SREGS: {
2489                 struct kvm_sregs kvm_sregs;
2490
2491                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2492                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2493                 if (r)
2494                         goto out;
2495                 r = -EFAULT;
2496                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2497                         goto out;
2498                 r = 0;
2499                 break;
2500         }
2501         case KVM_SET_SREGS: {
2502                 struct kvm_sregs kvm_sregs;
2503
2504                 r = -EFAULT;
2505                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2506                         goto out;
2507                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2508                 if (r)
2509                         goto out;
2510                 r = 0;
2511                 break;
2512         }
2513         case KVM_TRANSLATE: {
2514                 struct kvm_translation tr;
2515
2516                 r = -EFAULT;
2517                 if (copy_from_user(&tr, argp, sizeof tr))
2518                         goto out;
2519                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2520                 if (r)
2521                         goto out;
2522                 r = -EFAULT;
2523                 if (copy_to_user(argp, &tr, sizeof tr))
2524                         goto out;
2525                 r = 0;
2526                 break;
2527         }
2528         case KVM_INTERRUPT: {
2529                 struct kvm_interrupt irq;
2530
2531                 r = -EFAULT;
2532                 if (copy_from_user(&irq, argp, sizeof irq))
2533                         goto out;
2534                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2535                 if (r)
2536                         goto out;
2537                 r = 0;
2538                 break;
2539         }
2540         case KVM_DEBUG_GUEST: {
2541                 struct kvm_debug_guest dbg;
2542
2543                 r = -EFAULT;
2544                 if (copy_from_user(&dbg, argp, sizeof dbg))
2545                         goto out;
2546                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2547                 if (r)
2548                         goto out;
2549                 r = 0;
2550                 break;
2551         }
2552         case KVM_GET_MSRS:
2553                 r = msr_io(vcpu, argp, get_msr, 1);
2554                 break;
2555         case KVM_SET_MSRS:
2556                 r = msr_io(vcpu, argp, do_set_msr, 0);
2557                 break;
2558         case KVM_SET_CPUID: {
2559                 struct kvm_cpuid __user *cpuid_arg = argp;
2560                 struct kvm_cpuid cpuid;
2561
2562                 r = -EFAULT;
2563                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2564                         goto out;
2565                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2566                 if (r)
2567                         goto out;
2568                 break;
2569         }
2570         case KVM_SET_SIGNAL_MASK: {
2571                 struct kvm_signal_mask __user *sigmask_arg = argp;
2572                 struct kvm_signal_mask kvm_sigmask;
2573                 sigset_t sigset, *p;
2574
2575                 p = NULL;
2576                 if (argp) {
2577                         r = -EFAULT;
2578                         if (copy_from_user(&kvm_sigmask, argp,
2579                                            sizeof kvm_sigmask))
2580                                 goto out;
2581                         r = -EINVAL;
2582                         if (kvm_sigmask.len != sizeof sigset)
2583                                 goto out;
2584                         r = -EFAULT;
2585                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2586                                            sizeof sigset))
2587                                 goto out;
2588                         p = &sigset;
2589                 }
2590                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2591                 break;
2592         }
2593         case KVM_GET_FPU: {
2594                 struct kvm_fpu fpu;
2595
2596                 memset(&fpu, 0, sizeof fpu);
2597                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2598                 if (r)
2599                         goto out;
2600                 r = -EFAULT;
2601                 if (copy_to_user(argp, &fpu, sizeof fpu))
2602                         goto out;
2603                 r = 0;
2604                 break;
2605         }
2606         case KVM_SET_FPU: {
2607                 struct kvm_fpu fpu;
2608
2609                 r = -EFAULT;
2610                 if (copy_from_user(&fpu, argp, sizeof fpu))
2611                         goto out;
2612                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2613                 if (r)
2614                         goto out;
2615                 r = 0;
2616                 break;
2617         }
2618         default:
2619                 ;
2620         }
2621 out:
2622         return r;
2623 }
2624
2625 static long kvm_vm_ioctl(struct file *filp,
2626                            unsigned int ioctl, unsigned long arg)
2627 {
2628         struct kvm *kvm = filp->private_data;
2629         void __user *argp = (void __user *)arg;
2630         int r = -EINVAL;
2631
2632         switch (ioctl) {
2633         case KVM_CREATE_VCPU:
2634                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2635                 if (r < 0)
2636                         goto out;
2637                 break;
2638         case KVM_SET_MEMORY_REGION: {
2639                 struct kvm_memory_region kvm_mem;
2640
2641                 r = -EFAULT;
2642                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2643                         goto out;
2644                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2645                 if (r)
2646                         goto out;
2647                 break;
2648         }
2649         case KVM_GET_DIRTY_LOG: {
2650                 struct kvm_dirty_log log;
2651
2652                 r = -EFAULT;
2653                 if (copy_from_user(&log, argp, sizeof log))
2654                         goto out;
2655                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2656                 if (r)
2657                         goto out;
2658                 break;
2659         }
2660         case KVM_SET_MEMORY_ALIAS: {
2661                 struct kvm_memory_alias alias;
2662
2663                 r = -EFAULT;
2664                 if (copy_from_user(&alias, argp, sizeof alias))
2665                         goto out;
2666                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2667                 if (r)
2668                         goto out;
2669                 break;
2670         }
2671         default:
2672                 ;
2673         }
2674 out:
2675         return r;
2676 }
2677
2678 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2679                                   unsigned long address,
2680                                   int *type)
2681 {
2682         struct kvm *kvm = vma->vm_file->private_data;
2683         unsigned long pgoff;
2684         struct page *page;
2685
2686         *type = VM_FAULT_MINOR;
2687         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2688         page = gfn_to_page(kvm, pgoff);
2689         if (!page)
2690                 return NOPAGE_SIGBUS;
2691         get_page(page);
2692         return page;
2693 }
2694
2695 static struct vm_operations_struct kvm_vm_vm_ops = {
2696         .nopage = kvm_vm_nopage,
2697 };
2698
2699 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2700 {
2701         vma->vm_ops = &kvm_vm_vm_ops;
2702         return 0;
2703 }
2704
2705 static struct file_operations kvm_vm_fops = {
2706         .release        = kvm_vm_release,
2707         .unlocked_ioctl = kvm_vm_ioctl,
2708         .compat_ioctl   = kvm_vm_ioctl,
2709         .mmap           = kvm_vm_mmap,
2710 };
2711
2712 static int kvm_dev_ioctl_create_vm(void)
2713 {
2714         int fd, r;
2715         struct inode *inode;
2716         struct file *file;
2717         struct kvm *kvm;
2718
2719         inode = kvmfs_inode(&kvm_vm_fops);
2720         if (IS_ERR(inode)) {
2721                 r = PTR_ERR(inode);
2722                 goto out1;
2723         }
2724
2725         kvm = kvm_create_vm();
2726         if (IS_ERR(kvm)) {
2727                 r = PTR_ERR(kvm);
2728                 goto out2;
2729         }
2730
2731         file = kvmfs_file(inode, kvm);
2732         if (IS_ERR(file)) {
2733                 r = PTR_ERR(file);
2734                 goto out3;
2735         }
2736         kvm->filp = file;
2737
2738         r = get_unused_fd();
2739         if (r < 0)
2740                 goto out4;
2741         fd = r;
2742         fd_install(fd, file);
2743
2744         return fd;
2745
2746 out4:
2747         fput(file);
2748 out3:
2749         kvm_destroy_vm(kvm);
2750 out2:
2751         iput(inode);
2752 out1:
2753         return r;
2754 }
2755
2756 static long kvm_dev_ioctl(struct file *filp,
2757                           unsigned int ioctl, unsigned long arg)
2758 {
2759         void __user *argp = (void __user *)arg;
2760         long r = -EINVAL;
2761
2762         switch (ioctl) {
2763         case KVM_GET_API_VERSION:
2764                 r = -EINVAL;
2765                 if (arg)
2766                         goto out;
2767                 r = KVM_API_VERSION;
2768                 break;
2769         case KVM_CREATE_VM:
2770                 r = -EINVAL;
2771                 if (arg)
2772                         goto out;
2773                 r = kvm_dev_ioctl_create_vm();
2774                 break;
2775         case KVM_GET_MSR_INDEX_LIST: {
2776                 struct kvm_msr_list __user *user_msr_list = argp;
2777                 struct kvm_msr_list msr_list;
2778                 unsigned n;
2779
2780                 r = -EFAULT;
2781                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2782                         goto out;
2783                 n = msr_list.nmsrs;
2784                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2785                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2786                         goto out;
2787                 r = -E2BIG;
2788                 if (n < num_msrs_to_save)
2789                         goto out;
2790                 r = -EFAULT;
2791                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2792                                  num_msrs_to_save * sizeof(u32)))
2793                         goto out;
2794                 if (copy_to_user(user_msr_list->indices
2795                                  + num_msrs_to_save * sizeof(u32),
2796                                  &emulated_msrs,
2797                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2798                         goto out;
2799                 r = 0;
2800                 break;
2801         }
2802         case KVM_CHECK_EXTENSION:
2803                 /*
2804                  * No extensions defined at present.
2805                  */
2806                 r = 0;
2807                 break;
2808         case KVM_GET_VCPU_MMAP_SIZE:
2809                 r = -EINVAL;
2810                 if (arg)
2811                         goto out;
2812                 r = 2 * PAGE_SIZE;
2813                 break;
2814         default:
2815                 ;
2816         }
2817 out:
2818         return r;
2819 }
2820
2821 static struct file_operations kvm_chardev_ops = {
2822         .open           = kvm_dev_open,
2823         .release        = kvm_dev_release,
2824         .unlocked_ioctl = kvm_dev_ioctl,
2825         .compat_ioctl   = kvm_dev_ioctl,
2826 };
2827
2828 static struct miscdevice kvm_dev = {
2829         KVM_MINOR,
2830         "kvm",
2831         &kvm_chardev_ops,
2832 };
2833
2834 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2835                        void *v)
2836 {
2837         if (val == SYS_RESTART) {
2838                 /*
2839                  * Some (well, at least mine) BIOSes hang on reboot if
2840                  * in vmx root mode.
2841                  */
2842                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2843                 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2844         }
2845         return NOTIFY_OK;
2846 }
2847
2848 static struct notifier_block kvm_reboot_notifier = {
2849         .notifier_call = kvm_reboot,
2850         .priority = 0,
2851 };
2852
2853 /*
2854  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2855  * cached on it.
2856  */
2857 static void decache_vcpus_on_cpu(int cpu)
2858 {
2859         struct kvm *vm;
2860         struct kvm_vcpu *vcpu;
2861         int i;
2862
2863         spin_lock(&kvm_lock);
2864         list_for_each_entry(vm, &vm_list, vm_list)
2865                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2866                         vcpu = &vm->vcpus[i];
2867                         /*
2868                          * If the vcpu is locked, then it is running on some
2869                          * other cpu and therefore it is not cached on the
2870                          * cpu in question.
2871                          *
2872                          * If it's not locked, check the last cpu it executed
2873                          * on.
2874                          */
2875                         if (mutex_trylock(&vcpu->mutex)) {
2876                                 if (vcpu->cpu == cpu) {
2877                                         kvm_arch_ops->vcpu_decache(vcpu);
2878                                         vcpu->cpu = -1;
2879                                 }
2880                                 mutex_unlock(&vcpu->mutex);
2881                         }
2882                 }
2883         spin_unlock(&kvm_lock);
2884 }
2885
2886 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2887                            void *v)
2888 {
2889         int cpu = (long)v;
2890
2891         switch (val) {
2892         case CPU_DOWN_PREPARE:
2893         case CPU_DOWN_PREPARE_FROZEN:
2894         case CPU_UP_CANCELED:
2895         case CPU_UP_CANCELED_FROZEN:
2896                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2897                        cpu);
2898                 decache_vcpus_on_cpu(cpu);
2899                 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2900                                          NULL, 0, 1);
2901                 break;
2902         case CPU_ONLINE:
2903         case CPU_ONLINE_FROZEN:
2904                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2905                        cpu);
2906                 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2907                                          NULL, 0, 1);
2908                 break;
2909         }
2910         return NOTIFY_OK;
2911 }
2912
2913 static struct notifier_block kvm_cpu_notifier = {
2914         .notifier_call = kvm_cpu_hotplug,
2915         .priority = 20, /* must be > scheduler priority */
2916 };
2917
2918 static u64 stat_get(void *_offset)
2919 {
2920         unsigned offset = (long)_offset;
2921         u64 total = 0;
2922         struct kvm *kvm;
2923         struct kvm_vcpu *vcpu;
2924         int i;
2925
2926         spin_lock(&kvm_lock);
2927         list_for_each_entry(kvm, &vm_list, vm_list)
2928                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2929                         vcpu = &kvm->vcpus[i];
2930                         total += *(u32 *)((void *)vcpu + offset);
2931                 }
2932         spin_unlock(&kvm_lock);
2933         return total;
2934 }
2935
2936 static void stat_set(void *offset, u64 val)
2937 {
2938 }
2939
2940 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
2941
2942 static __init void kvm_init_debug(void)
2943 {
2944         struct kvm_stats_debugfs_item *p;
2945
2946         debugfs_dir = debugfs_create_dir("kvm", NULL);
2947         for (p = debugfs_entries; p->name; ++p)
2948                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2949                                                 (void *)(long)p->offset,
2950                                                 &stat_fops);
2951 }
2952
2953 static void kvm_exit_debug(void)
2954 {
2955         struct kvm_stats_debugfs_item *p;
2956
2957         for (p = debugfs_entries; p->name; ++p)
2958                 debugfs_remove(p->dentry);
2959         debugfs_remove(debugfs_dir);
2960 }
2961
2962 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2963 {
2964         decache_vcpus_on_cpu(raw_smp_processor_id());
2965         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2966         return 0;
2967 }
2968
2969 static int kvm_resume(struct sys_device *dev)
2970 {
2971         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2972         return 0;
2973 }
2974
2975 static struct sysdev_class kvm_sysdev_class = {
2976         set_kset_name("kvm"),
2977         .suspend = kvm_suspend,
2978         .resume = kvm_resume,
2979 };
2980
2981 static struct sys_device kvm_sysdev = {
2982         .id = 0,
2983         .cls = &kvm_sysdev_class,
2984 };
2985
2986 hpa_t bad_page_address;
2987
2988 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2989                         const char *dev_name, void *data, struct vfsmount *mnt)
2990 {
2991         return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2992 }
2993
2994 static struct file_system_type kvm_fs_type = {
2995         .name           = "kvmfs",
2996         .get_sb         = kvmfs_get_sb,
2997         .kill_sb        = kill_anon_super,
2998 };
2999
3000 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3001 {
3002         int r;
3003
3004         if (kvm_arch_ops) {
3005                 printk(KERN_ERR "kvm: already loaded the other module\n");
3006                 return -EEXIST;
3007         }
3008
3009         if (!ops->cpu_has_kvm_support()) {
3010                 printk(KERN_ERR "kvm: no hardware support\n");
3011                 return -EOPNOTSUPP;
3012         }
3013         if (ops->disabled_by_bios()) {
3014                 printk(KERN_ERR "kvm: disabled by bios\n");
3015                 return -EOPNOTSUPP;
3016         }
3017
3018         kvm_arch_ops = ops;
3019
3020         r = kvm_arch_ops->hardware_setup();
3021         if (r < 0)
3022                 goto out;
3023
3024         on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3025         r = register_cpu_notifier(&kvm_cpu_notifier);
3026         if (r)
3027                 goto out_free_1;
3028         register_reboot_notifier(&kvm_reboot_notifier);
3029
3030         r = sysdev_class_register(&kvm_sysdev_class);
3031         if (r)
3032                 goto out_free_2;
3033
3034         r = sysdev_register(&kvm_sysdev);
3035         if (r)
3036                 goto out_free_3;
3037
3038         kvm_chardev_ops.owner = module;
3039
3040         r = misc_register(&kvm_dev);
3041         if (r) {
3042                 printk (KERN_ERR "kvm: misc device register failed\n");
3043                 goto out_free;
3044         }
3045
3046         return r;
3047
3048 out_free:
3049         sysdev_unregister(&kvm_sysdev);
3050 out_free_3:
3051         sysdev_class_unregister(&kvm_sysdev_class);
3052 out_free_2:
3053         unregister_reboot_notifier(&kvm_reboot_notifier);
3054         unregister_cpu_notifier(&kvm_cpu_notifier);
3055 out_free_1:
3056         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3057         kvm_arch_ops->hardware_unsetup();
3058 out:
3059         kvm_arch_ops = NULL;
3060         return r;
3061 }
3062
3063 void kvm_exit_arch(void)
3064 {
3065         misc_deregister(&kvm_dev);
3066         sysdev_unregister(&kvm_sysdev);
3067         sysdev_class_unregister(&kvm_sysdev_class);
3068         unregister_reboot_notifier(&kvm_reboot_notifier);
3069         unregister_cpu_notifier(&kvm_cpu_notifier);
3070         on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3071         kvm_arch_ops->hardware_unsetup();
3072         kvm_arch_ops = NULL;
3073 }
3074
3075 static __init int kvm_init(void)
3076 {
3077         static struct page *bad_page;
3078         int r;
3079
3080         r = kvm_mmu_module_init();
3081         if (r)
3082                 goto out4;
3083
3084         r = register_filesystem(&kvm_fs_type);
3085         if (r)
3086                 goto out3;
3087
3088         kvmfs_mnt = kern_mount(&kvm_fs_type);
3089         r = PTR_ERR(kvmfs_mnt);
3090         if (IS_ERR(kvmfs_mnt))
3091                 goto out2;
3092         kvm_init_debug();
3093
3094         kvm_init_msr_list();
3095
3096         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3097                 r = -ENOMEM;
3098                 goto out;
3099         }
3100
3101         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3102         memset(__va(bad_page_address), 0, PAGE_SIZE);
3103
3104         return 0;
3105
3106 out:
3107         kvm_exit_debug();
3108         mntput(kvmfs_mnt);
3109 out2:
3110         unregister_filesystem(&kvm_fs_type);
3111 out3:
3112         kvm_mmu_module_exit();
3113 out4:
3114         return r;
3115 }
3116
3117 static __exit void kvm_exit(void)
3118 {
3119         kvm_exit_debug();
3120         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3121         mntput(kvmfs_mnt);
3122         unregister_filesystem(&kvm_fs_type);
3123         kvm_mmu_module_exit();
3124 }
3125
3126 module_init(kvm_init)
3127 module_exit(kvm_exit)
3128
3129 EXPORT_SYMBOL_GPL(kvm_init_arch);
3130 EXPORT_SYMBOL_GPL(kvm_exit_arch);