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 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40
  41 #include <asm/processor.h>
  42 #include <asm/msr.h>
  43 #include <asm/io.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/desc.h>
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 static cpumask_t cpus_hardware_enabled;
  54
  55 struct kvm_arch_ops *kvm_arch_ops;
  56
  57 static void hardware_disable(void *ignored);
  58
  59 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  60
  61 static struct kvm_stats_debugfs_item {
  62         const char *name;
  63         int offset;
  64         struct dentry *dentry;
  65 } debugfs_entries[] = {
  66         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  67         { "pf_guest", STAT_OFFSET(pf_guest) },
  68         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  69         { "invlpg", STAT_OFFSET(invlpg) },
  70         { "exits", STAT_OFFSET(exits) },
  71         { "io_exits", STAT_OFFSET(io_exits) },
  72         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  73         { "signal_exits", STAT_OFFSET(signal_exits) },
  74         { "irq_window", STAT_OFFSET(irq_window_exits) },
  75         { "halt_exits", STAT_OFFSET(halt_exits) },
  76         { "request_irq", STAT_OFFSET(request_irq_exits) },
  77         { "irq_exits", STAT_OFFSET(irq_exits) },
  78         { "light_exits", STAT_OFFSET(light_exits) },
  79         { "efer_reload", STAT_OFFSET(efer_reload) },
  80         { NULL }
  81 };
  82
  83 static struct dentry *debugfs_dir;
  84
  85 #define MAX_IO_MSRS 256
  86
  87 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
  88 #define LMSW_GUEST_MASK 0x0eULL
  89 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  90 #define CR8_RESEVED_BITS (~0x0fULL)
  91 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  92
  93 #ifdef CONFIG_X86_64
  94 // LDT or TSS descriptor in the GDT. 16 bytes.
  95 struct segment_descriptor_64 {
  96         struct segment_descriptor s;
  97         u32 base_higher;
  98         u32 pad_zero;
  99 };
 100
 101 #endif
 102
 103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 104                            unsigned long arg);
 105
 106 unsigned long segment_base(u16 selector)
 107 {
 108         struct descriptor_table gdt;
 109         struct segment_descriptor *d;
 110         unsigned long table_base;
 111         typedef unsigned long ul;
 112         unsigned long v;
 113
 114         if (selector == 0)
 115                 return 0;
 116
 117         asm ("sgdt %0" : "=m"(gdt));
 118         table_base = gdt.base;
 119
 120         if (selector & 4) {           /* from ldt */
 121                 u16 ldt_selector;
 122
 123                 asm ("sldt %0" : "=g"(ldt_selector));
 124                 table_base = segment_base(ldt_selector);
 125         }
 126         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 127         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 128 #ifdef CONFIG_X86_64
 129         if (d->system == 0
 130             && (d->type == 2 || d->type == 9 || d->type == 11))
 131                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 132 #endif
 133         return v;
 134 }
 135 EXPORT_SYMBOL_GPL(segment_base);
 136
 137 static inline int valid_vcpu(int n)
 138 {
 139         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 140 }
 141
 142 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 143                    void *dest)
 144 {
 145         unsigned char *host_buf = dest;
 146         unsigned long req_size = size;
 147
 148         while (size) {
 149                 hpa_t paddr;
 150                 unsigned now;
 151                 unsigned offset;
 152                 hva_t guest_buf;
 153
 154                 paddr = gva_to_hpa(vcpu, addr);
 155
 156                 if (is_error_hpa(paddr))
 157                         break;
 158
 159                 guest_buf = (hva_t)kmap_atomic(
 160                                         pfn_to_page(paddr >> PAGE_SHIFT),
 161                                         KM_USER0);
 162                 offset = addr & ~PAGE_MASK;
 163                 guest_buf |= offset;
 164                 now = min(size, PAGE_SIZE - offset);
 165                 memcpy(host_buf, (void*)guest_buf, now);
 166                 host_buf += now;
 167                 addr += now;
 168                 size -= now;
 169                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 170         }
 171         return req_size - size;
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_read_guest);
 174
 175 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 176                     void *data)
 177 {
 178         unsigned char *host_buf = data;
 179         unsigned long req_size = size;
 180
 181         while (size) {
 182                 hpa_t paddr;
 183                 unsigned now;
 184                 unsigned offset;
 185                 hva_t guest_buf;
 186                 gfn_t gfn;
 187
 188                 paddr = gva_to_hpa(vcpu, addr);
 189
 190                 if (is_error_hpa(paddr))
 191                         break;
 192
 193                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 194                 mark_page_dirty(vcpu->kvm, gfn);
 195                 guest_buf = (hva_t)kmap_atomic(
 196                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 197                 offset = addr & ~PAGE_MASK;
 198                 guest_buf |= offset;
 199                 now = min(size, PAGE_SIZE - offset);
 200                 memcpy((void*)guest_buf, host_buf, now);
 201                 host_buf += now;
 202                 addr += now;
 203                 size -= now;
 204                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 205         }
 206         return req_size - size;
 207 }
 208 EXPORT_SYMBOL_GPL(kvm_write_guest);
 209
 210 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 211 {
 212         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 213                 return;
 214
 215         vcpu->guest_fpu_loaded = 1;
 216         fx_save(vcpu->host_fx_image);
 217         fx_restore(vcpu->guest_fx_image);
 218 }
 219 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 220
 221 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 222 {
 223         if (!vcpu->guest_fpu_loaded)
 224                 return;
 225
 226         vcpu->guest_fpu_loaded = 0;
 227         fx_save(vcpu->guest_fx_image);
 228         fx_restore(vcpu->host_fx_image);
 229 }
 230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 231
 232 /*
 233  * Switches to specified vcpu, until a matching vcpu_put()
 234  */
 235 static void vcpu_load(struct kvm_vcpu *vcpu)
 236 {
 237         mutex_lock(&vcpu->mutex);
 238         kvm_arch_ops->vcpu_load(vcpu);
 239 }
 240
 241 static void vcpu_put(struct kvm_vcpu *vcpu)
 242 {
 243         kvm_arch_ops->vcpu_put(vcpu);
 244         mutex_unlock(&vcpu->mutex);
 245 }
 246
 247 static void ack_flush(void *_completed)
 248 {
 249         atomic_t *completed = _completed;
 250
 251         atomic_inc(completed);
 252 }
 253
 254 void kvm_flush_remote_tlbs(struct kvm *kvm)
 255 {
 256         int i, cpu, needed;
 257         cpumask_t cpus;
 258         struct kvm_vcpu *vcpu;
 259         atomic_t completed;
 260
 261         atomic_set(&completed, 0);
 262         cpus_clear(cpus);
 263         needed = 0;
 264         for (i = 0; i < kvm->nvcpus; ++i) {
 265                 vcpu = &kvm->vcpus[i];
 266                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 267                         continue;
 268                 cpu = vcpu->cpu;
 269                 if (cpu != -1 && cpu != raw_smp_processor_id())
 270                         if (!cpu_isset(cpu, cpus)) {
 271                                 cpu_set(cpu, cpus);
 272                                 ++needed;
 273                         }
 274         }
 275
 276         /*
 277          * We really want smp_call_function_mask() here.  But that's not
 278          * available, so ipi all cpus in parallel and wait for them
 279          * to complete.
 280          */
 281         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 282                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 283         while (atomic_read(&completed) != needed) {
 284                 cpu_relax();
 285                 barrier();
 286         }
 287 }
 288
 289 static struct kvm *kvm_create_vm(void)
 290 {
 291         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 292         int i;
 293
 294         if (!kvm)
 295                 return ERR_PTR(-ENOMEM);
 296
 297         kvm_io_bus_init(&kvm->pio_bus);
 298         spin_lock_init(&kvm->lock);
 299         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 300         spin_lock(&kvm_lock);
 301         list_add(&kvm->vm_list, &vm_list);
 302         spin_unlock(&kvm_lock);
 303         kvm_io_bus_init(&kvm->mmio_bus);
 304         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 305                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 306
 307                 mutex_init(&vcpu->mutex);
 308                 vcpu->cpu = -1;
 309                 vcpu->kvm = kvm;
 310                 vcpu->mmu.root_hpa = INVALID_PAGE;
 311         }
 312         return kvm;
 313 }
 314
 315 static int kvm_dev_open(struct inode *inode, struct file *filp)
 316 {
 317         return 0;
 318 }
 319
 320 /*
 321  * Free any memory in @free but not in @dont.
 322  */
 323 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 324                                   struct kvm_memory_slot *dont)
 325 {
 326         int i;
 327
 328         if (!dont || free->phys_mem != dont->phys_mem)
 329                 if (free->phys_mem) {
 330                         for (i = 0; i < free->npages; ++i)
 331                                 if (free->phys_mem[i])
 332                                         __free_page(free->phys_mem[i]);
 333                         vfree(free->phys_mem);
 334                 }
 335
 336         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 337                 vfree(free->dirty_bitmap);
 338
 339         free->phys_mem = NULL;
 340         free->npages = 0;
 341         free->dirty_bitmap = NULL;
 342 }
 343
 344 static void kvm_free_physmem(struct kvm *kvm)
 345 {
 346         int i;
 347
 348         for (i = 0; i < kvm->nmemslots; ++i)
 349                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 350 }
 351
 352 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 353 {
 354         int i;
 355
 356         for (i = 0; i < 2; ++i)
 357                 if (vcpu->pio.guest_pages[i]) {
 358                         __free_page(vcpu->pio.guest_pages[i]);
 359                         vcpu->pio.guest_pages[i] = NULL;
 360                 }
 361 }
 362
 363 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 364 {
 365         if (!vcpu->vmcs)
 366                 return;
 367
 368         vcpu_load(vcpu);
 369         kvm_mmu_unload(vcpu);
 370         vcpu_put(vcpu);
 371 }
 372
 373 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 374 {
 375         if (!vcpu->vmcs)
 376                 return;
 377
 378         vcpu_load(vcpu);
 379         kvm_mmu_destroy(vcpu);
 380         vcpu_put(vcpu);
 381         kvm_arch_ops->vcpu_free(vcpu);
 382         free_page((unsigned long)vcpu->run);
 383         vcpu->run = NULL;
 384         free_page((unsigned long)vcpu->pio_data);
 385         vcpu->pio_data = NULL;
 386         free_pio_guest_pages(vcpu);
 387 }
 388
 389 static void kvm_free_vcpus(struct kvm *kvm)
 390 {
 391         unsigned int i;
 392
 393         /*
 394          * Unpin any mmu pages first.
 395          */
 396         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 397                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
 398         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 399                 kvm_free_vcpu(&kvm->vcpus[i]);
 400 }
 401
 402 static int kvm_dev_release(struct inode *inode, struct file *filp)
 403 {
 404         return 0;
 405 }
 406
 407 static void kvm_destroy_vm(struct kvm *kvm)
 408 {
 409         spin_lock(&kvm_lock);
 410         list_del(&kvm->vm_list);
 411         spin_unlock(&kvm_lock);
 412         kvm_io_bus_destroy(&kvm->pio_bus);
 413         kvm_io_bus_destroy(&kvm->mmio_bus);
 414         kvm_free_vcpus(kvm);
 415         kvm_free_physmem(kvm);
 416         kfree(kvm);
 417 }
 418
 419 static int kvm_vm_release(struct inode *inode, struct file *filp)
 420 {
 421         struct kvm *kvm = filp->private_data;
 422
 423         kvm_destroy_vm(kvm);
 424         return 0;
 425 }
 426
 427 static void inject_gp(struct kvm_vcpu *vcpu)
 428 {
 429         kvm_arch_ops->inject_gp(vcpu, 0);
 430 }
 431
 432 /*
 433  * Load the pae pdptrs.  Return true is they are all valid.
 434  */
 435 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 436 {
 437         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 438         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 439         int i;
 440         u64 pdpte;
 441         u64 *pdpt;
 442         int ret;
 443         struct page *page;
 444
 445         spin_lock(&vcpu->kvm->lock);
 446         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 447         /* FIXME: !page - emulate? 0xff? */
 448         pdpt = kmap_atomic(page, KM_USER0);
 449
 450         ret = 1;
 451         for (i = 0; i < 4; ++i) {
 452                 pdpte = pdpt[offset + i];
 453                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 454                         ret = 0;
 455                         goto out;
 456                 }
 457         }
 458
 459         for (i = 0; i < 4; ++i)
 460                 vcpu->pdptrs[i] = pdpt[offset + i];
 461
 462 out:
 463         kunmap_atomic(pdpt, KM_USER0);
 464         spin_unlock(&vcpu->kvm->lock);
 465
 466         return ret;
 467 }
 468
 469 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 470 {
 471         if (cr0 & CR0_RESEVED_BITS) {
 472                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 473                        cr0, vcpu->cr0);
 474                 inject_gp(vcpu);
 475                 return;
 476         }
 477
 478         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 479                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 480                 inject_gp(vcpu);
 481                 return;
 482         }
 483
 484         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 485                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 486                        "and a clear PE flag\n");
 487                 inject_gp(vcpu);
 488                 return;
 489         }
 490
 491         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 492 #ifdef CONFIG_X86_64
 493                 if ((vcpu->shadow_efer & EFER_LME)) {
 494                         int cs_db, cs_l;
 495
 496                         if (!is_pae(vcpu)) {
 497                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 498                                        "in long mode while PAE is disabled\n");
 499                                 inject_gp(vcpu);
 500                                 return;
 501                         }
 502                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 503                         if (cs_l) {
 504                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 505                                        "in long mode while CS.L == 1\n");
 506                                 inject_gp(vcpu);
 507                                 return;
 508
 509                         }
 510                 } else
 511 #endif
 512                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 513                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 514                                "reserved bits\n");
 515                         inject_gp(vcpu);
 516                         return;
 517                 }
 518
 519         }
 520
 521         kvm_arch_ops->set_cr0(vcpu, cr0);
 522         vcpu->cr0 = cr0;
 523
 524         spin_lock(&vcpu->kvm->lock);
 525         kvm_mmu_reset_context(vcpu);
 526         spin_unlock(&vcpu->kvm->lock);
 527         return;
 528 }
 529 EXPORT_SYMBOL_GPL(set_cr0);
 530
 531 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 532 {
 533         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 534 }
 535 EXPORT_SYMBOL_GPL(lmsw);
 536
 537 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 538 {
 539         if (cr4 & CR4_RESEVED_BITS) {
 540                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 541                 inject_gp(vcpu);
 542                 return;
 543         }
 544
 545         if (is_long_mode(vcpu)) {
 546                 if (!(cr4 & CR4_PAE_MASK)) {
 547                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 548                                "in long mode\n");
 549                         inject_gp(vcpu);
 550                         return;
 551                 }
 552         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 553                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 554                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 555                 inject_gp(vcpu);
 556         }
 557
 558         if (cr4 & CR4_VMXE_MASK) {
 559                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 560                 inject_gp(vcpu);
 561                 return;
 562         }
 563         kvm_arch_ops->set_cr4(vcpu, cr4);
 564         spin_lock(&vcpu->kvm->lock);
 565         kvm_mmu_reset_context(vcpu);
 566         spin_unlock(&vcpu->kvm->lock);
 567 }
 568 EXPORT_SYMBOL_GPL(set_cr4);
 569
 570 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 571 {
 572         if (is_long_mode(vcpu)) {
 573                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 574                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 575                         inject_gp(vcpu);
 576                         return;
 577                 }
 578         } else {
 579                 if (cr3 & CR3_RESEVED_BITS) {
 580                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 581                         inject_gp(vcpu);
 582                         return;
 583                 }
 584                 if (is_paging(vcpu) && is_pae(vcpu) &&
 585                     !load_pdptrs(vcpu, cr3)) {
 586                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 587                                "reserved bits\n");
 588                         inject_gp(vcpu);
 589                         return;
 590                 }
 591         }
 592
 593         vcpu->cr3 = cr3;
 594         spin_lock(&vcpu->kvm->lock);
 595         /*
 596          * Does the new cr3 value map to physical memory? (Note, we
 597          * catch an invalid cr3 even in real-mode, because it would
 598          * cause trouble later on when we turn on paging anyway.)
 599          *
 600          * A real CPU would silently accept an invalid cr3 and would
 601          * attempt to use it - with largely undefined (and often hard
 602          * to debug) behavior on the guest side.
 603          */
 604         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 605                 inject_gp(vcpu);
 606         else
 607                 vcpu->mmu.new_cr3(vcpu);
 608         spin_unlock(&vcpu->kvm->lock);
 609 }
 610 EXPORT_SYMBOL_GPL(set_cr3);
 611
 612 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 613 {
 614         if ( cr8 & CR8_RESEVED_BITS) {
 615                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 616                 inject_gp(vcpu);
 617                 return;
 618         }
 619         vcpu->cr8 = cr8;
 620 }
 621 EXPORT_SYMBOL_GPL(set_cr8);
 622
 623 void fx_init(struct kvm_vcpu *vcpu)
 624 {
 625         struct __attribute__ ((__packed__)) fx_image_s {
 626                 u16 control; //fcw
 627                 u16 status; //fsw
 628                 u16 tag; // ftw
 629                 u16 opcode; //fop
 630                 u64 ip; // fpu ip
 631                 u64 operand;// fpu dp
 632                 u32 mxcsr;
 633                 u32 mxcsr_mask;
 634
 635         } *fx_image;
 636
 637         fx_save(vcpu->host_fx_image);
 638         fpu_init();
 639         fx_save(vcpu->guest_fx_image);
 640         fx_restore(vcpu->host_fx_image);
 641
 642         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 643         fx_image->mxcsr = 0x1f80;
 644         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 645                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 646 }
 647 EXPORT_SYMBOL_GPL(fx_init);
 648
 649 /*
 650  * Allocate some memory and give it an address in the guest physical address
 651  * space.
 652  *
 653  * Discontiguous memory is allowed, mostly for framebuffers.
 654  */
 655 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 656                                           struct kvm_memory_region *mem)
 657 {
 658         int r;
 659         gfn_t base_gfn;
 660         unsigned long npages;
 661         unsigned long i;
 662         struct kvm_memory_slot *memslot;
 663         struct kvm_memory_slot old, new;
 664         int memory_config_version;
 665
 666         r = -EINVAL;
 667         /* General sanity checks */
 668         if (mem->memory_size & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 671                 goto out;
 672         if (mem->slot >= KVM_MEMORY_SLOTS)
 673                 goto out;
 674         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 675                 goto out;
 676
 677         memslot = &kvm->memslots[mem->slot];
 678         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 679         npages = mem->memory_size >> PAGE_SHIFT;
 680
 681         if (!npages)
 682                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 683
 684 raced:
 685         spin_lock(&kvm->lock);
 686
 687         memory_config_version = kvm->memory_config_version;
 688         new = old = *memslot;
 689
 690         new.base_gfn = base_gfn;
 691         new.npages = npages;
 692         new.flags = mem->flags;
 693
 694         /* Disallow changing a memory slot's size. */
 695         r = -EINVAL;
 696         if (npages && old.npages && npages != old.npages)
 697                 goto out_unlock;
 698
 699         /* Check for overlaps */
 700         r = -EEXIST;
 701         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 702                 struct kvm_memory_slot *s = &kvm->memslots[i];
 703
 704                 if (s == memslot)
 705                         continue;
 706                 if (!((base_gfn + npages <= s->base_gfn) ||
 707                       (base_gfn >= s->base_gfn + s->npages)))
 708                         goto out_unlock;
 709         }
 710         /*
 711          * Do memory allocations outside lock.  memory_config_version will
 712          * detect any races.
 713          */
 714         spin_unlock(&kvm->lock);
 715
 716         /* Deallocate if slot is being removed */
 717         if (!npages)
 718                 new.phys_mem = NULL;
 719
 720         /* Free page dirty bitmap if unneeded */
 721         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 722                 new.dirty_bitmap = NULL;
 723
 724         r = -ENOMEM;
 725
 726         /* Allocate if a slot is being created */
 727         if (npages && !new.phys_mem) {
 728                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 729
 730                 if (!new.phys_mem)
 731                         goto out_free;
 732
 733                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 734                 for (i = 0; i < npages; ++i) {
 735                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 736                                                      | __GFP_ZERO);
 737                         if (!new.phys_mem[i])
 738                                 goto out_free;
 739                         set_page_private(new.phys_mem[i],0);
 740                 }
 741         }
 742
 743         /* Allocate page dirty bitmap if needed */
 744         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 745                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 746
 747                 new.dirty_bitmap = vmalloc(dirty_bytes);
 748                 if (!new.dirty_bitmap)
 749                         goto out_free;
 750                 memset(new.dirty_bitmap, 0, dirty_bytes);
 751         }
 752
 753         spin_lock(&kvm->lock);
 754
 755         if (memory_config_version != kvm->memory_config_version) {
 756                 spin_unlock(&kvm->lock);
 757                 kvm_free_physmem_slot(&new, &old);
 758                 goto raced;
 759         }
 760
 761         r = -EAGAIN;
 762         if (kvm->busy)
 763                 goto out_unlock;
 764
 765         if (mem->slot >= kvm->nmemslots)
 766                 kvm->nmemslots = mem->slot + 1;
 767
 768         *memslot = new;
 769         ++kvm->memory_config_version;
 770
 771         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 772         kvm_flush_remote_tlbs(kvm);
 773
 774         spin_unlock(&kvm->lock);
 775
 776         kvm_free_physmem_slot(&old, &new);
 777         return 0;
 778
 779 out_unlock:
 780         spin_unlock(&kvm->lock);
 781 out_free:
 782         kvm_free_physmem_slot(&new, &old);
 783 out:
 784         return r;
 785 }
 786
 787 /*
 788  * Get (and clear) the dirty memory log for a memory slot.
 789  */
 790 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 791                                       struct kvm_dirty_log *log)
 792 {
 793         struct kvm_memory_slot *memslot;
 794         int r, i;
 795         int n;
 796         unsigned long any = 0;
 797
 798         spin_lock(&kvm->lock);
 799
 800         /*
 801          * Prevent changes to guest memory configuration even while the lock
 802          * is not taken.
 803          */
 804         ++kvm->busy;
 805         spin_unlock(&kvm->lock);
 806         r = -EINVAL;
 807         if (log->slot >= KVM_MEMORY_SLOTS)
 808                 goto out;
 809
 810         memslot = &kvm->memslots[log->slot];
 811         r = -ENOENT;
 812         if (!memslot->dirty_bitmap)
 813                 goto out;
 814
 815         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 816
 817         for (i = 0; !any && i < n/sizeof(long); ++i)
 818                 any = memslot->dirty_bitmap[i];
 819
 820         r = -EFAULT;
 821         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 822                 goto out;
 823
 824         spin_lock(&kvm->lock);
 825         kvm_mmu_slot_remove_write_access(kvm, log->slot);
 826         kvm_flush_remote_tlbs(kvm);
 827         memset(memslot->dirty_bitmap, 0, n);
 828         spin_unlock(&kvm->lock);
 829
 830         r = 0;
 831
 832 out:
 833         spin_lock(&kvm->lock);
 834         --kvm->busy;
 835         spin_unlock(&kvm->lock);
 836         return r;
 837 }
 838
 839 /*
 840  * Set a new alias region.  Aliases map a portion of physical memory into
 841  * another portion.  This is useful for memory windows, for example the PC
 842  * VGA region.
 843  */
 844 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 845                                          struct kvm_memory_alias *alias)
 846 {
 847         int r, n;
 848         struct kvm_mem_alias *p;
 849
 850         r = -EINVAL;
 851         /* General sanity checks */
 852         if (alias->memory_size & (PAGE_SIZE - 1))
 853                 goto out;
 854         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 855                 goto out;
 856         if (alias->slot >= KVM_ALIAS_SLOTS)
 857                 goto out;
 858         if (alias->guest_phys_addr + alias->memory_size
 859             < alias->guest_phys_addr)
 860                 goto out;
 861         if (alias->target_phys_addr + alias->memory_size
 862             < alias->target_phys_addr)
 863                 goto out;
 864
 865         spin_lock(&kvm->lock);
 866
 867         p = &kvm->aliases[alias->slot];
 868         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 869         p->npages = alias->memory_size >> PAGE_SHIFT;
 870         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 871
 872         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 873                 if (kvm->aliases[n - 1].npages)
 874                         break;
 875         kvm->naliases = n;
 876
 877         kvm_mmu_zap_all(kvm);
 878
 879         spin_unlock(&kvm->lock);
 880
 881         return 0;
 882
 883 out:
 884         return r;
 885 }
 886
 887 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 888 {
 889         int i;
 890         struct kvm_mem_alias *alias;
 891
 892         for (i = 0; i < kvm->naliases; ++i) {
 893                 alias = &kvm->aliases[i];
 894                 if (gfn >= alias->base_gfn
 895                     && gfn < alias->base_gfn + alias->npages)
 896                         return alias->target_gfn + gfn - alias->base_gfn;
 897         }
 898         return gfn;
 899 }
 900
 901 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 902 {
 903         int i;
 904
 905         for (i = 0; i < kvm->nmemslots; ++i) {
 906                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 907
 908                 if (gfn >= memslot->base_gfn
 909                     && gfn < memslot->base_gfn + memslot->npages)
 910                         return memslot;
 911         }
 912         return NULL;
 913 }
 914
 915 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 916 {
 917         gfn = unalias_gfn(kvm, gfn);
 918         return __gfn_to_memslot(kvm, gfn);
 919 }
 920
 921 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 922 {
 923         struct kvm_memory_slot *slot;
 924
 925         gfn = unalias_gfn(kvm, gfn);
 926         slot = __gfn_to_memslot(kvm, gfn);
 927         if (!slot)
 928                 return NULL;
 929         return slot->phys_mem[gfn - slot->base_gfn];
 930 }
 931 EXPORT_SYMBOL_GPL(gfn_to_page);
 932
 933 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 934 {
 935         int i;
 936         struct kvm_memory_slot *memslot;
 937         unsigned long rel_gfn;
 938
 939         for (i = 0; i < kvm->nmemslots; ++i) {
 940                 memslot = &kvm->memslots[i];
 941
 942                 if (gfn >= memslot->base_gfn
 943                     && gfn < memslot->base_gfn + memslot->npages) {
 944
 945                         if (!memslot->dirty_bitmap)
 946                                 return;
 947
 948                         rel_gfn = gfn - memslot->base_gfn;
 949
 950                         /* avoid RMW */
 951                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 952                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 953                         return;
 954                 }
 955         }
 956 }
 957
 958 static int emulator_read_std(unsigned long addr,
 959                              void *val,
 960                              unsigned int bytes,
 961                              struct x86_emulate_ctxt *ctxt)
 962 {
 963         struct kvm_vcpu *vcpu = ctxt->vcpu;
 964         void *data = val;
 965
 966         while (bytes) {
 967                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 968                 unsigned offset = addr & (PAGE_SIZE-1);
 969                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 970                 unsigned long pfn;
 971                 struct page *page;
 972                 void *page_virt;
 973
 974                 if (gpa == UNMAPPED_GVA)
 975                         return X86EMUL_PROPAGATE_FAULT;
 976                 pfn = gpa >> PAGE_SHIFT;
 977                 page = gfn_to_page(vcpu->kvm, pfn);
 978                 if (!page)
 979                         return X86EMUL_UNHANDLEABLE;
 980                 page_virt = kmap_atomic(page, KM_USER0);
 981
 982                 memcpy(data, page_virt + offset, tocopy);
 983
 984                 kunmap_atomic(page_virt, KM_USER0);
 985
 986                 bytes -= tocopy;
 987                 data += tocopy;
 988                 addr += tocopy;
 989         }
 990
 991         return X86EMUL_CONTINUE;
 992 }
 993
 994 static int emulator_write_std(unsigned long addr,
 995                               const void *val,
 996                               unsigned int bytes,
 997                               struct x86_emulate_ctxt *ctxt)
 998 {
 999         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1000                addr, bytes);
1001         return X86EMUL_UNHANDLEABLE;
1002 }
1003
1004 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1005                                                 gpa_t addr)
1006 {
1007         /*
1008          * Note that its important to have this wrapper function because
1009          * in the very near future we will be checking for MMIOs against
1010          * the LAPIC as well as the general MMIO bus
1011          */
1012         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1013 }
1014
1015 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1016                                                gpa_t addr)
1017 {
1018         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1019 }
1020
1021 static int emulator_read_emulated(unsigned long addr,
1022                                   void *val,
1023                                   unsigned int bytes,
1024                                   struct x86_emulate_ctxt *ctxt)
1025 {
1026         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1027         struct kvm_io_device *mmio_dev;
1028         gpa_t                 gpa;
1029
1030         if (vcpu->mmio_read_completed) {
1031                 memcpy(val, vcpu->mmio_data, bytes);
1032                 vcpu->mmio_read_completed = 0;
1033                 return X86EMUL_CONTINUE;
1034         } else if (emulator_read_std(addr, val, bytes, ctxt)
1035                    == X86EMUL_CONTINUE)
1036                 return X86EMUL_CONTINUE;
1037
1038         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1039         if (gpa == UNMAPPED_GVA)
1040                 return X86EMUL_PROPAGATE_FAULT;
1041
1042         /*
1043          * Is this MMIO handled locally?
1044          */
1045         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1046         if (mmio_dev) {
1047                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1048                 return X86EMUL_CONTINUE;
1049         }
1050
1051         vcpu->mmio_needed = 1;
1052         vcpu->mmio_phys_addr = gpa;
1053         vcpu->mmio_size = bytes;
1054         vcpu->mmio_is_write = 0;
1055
1056         return X86EMUL_UNHANDLEABLE;
1057 }
1058
1059 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1060                                const void *val, int bytes)
1061 {
1062         struct page *page;
1063         void *virt;
1064         unsigned offset = offset_in_page(gpa);
1065
1066         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1067                 return 0;
1068         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1069         if (!page)
1070                 return 0;
1071         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1072         virt = kmap_atomic(page, KM_USER0);
1073         if (memcmp(virt + offset_in_page(gpa), val, bytes)) {
1074                 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1075                 memcpy(virt + offset_in_page(gpa), val, bytes);
1076         }
1077         kunmap_atomic(virt, KM_USER0);
1078         return 1;
1079 }
1080
1081 static int emulator_write_emulated(unsigned long addr,
1082                                    const void *val,
1083                                    unsigned int bytes,
1084                                    struct x86_emulate_ctxt *ctxt)
1085 {
1086         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1087         struct kvm_io_device *mmio_dev;
1088         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1089
1090         if (gpa == UNMAPPED_GVA) {
1091                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1092                 return X86EMUL_PROPAGATE_FAULT;
1093         }
1094
1095         if (emulator_write_phys(vcpu, gpa, val, bytes))
1096                 return X86EMUL_CONTINUE;
1097
1098         /*
1099          * Is this MMIO handled locally?
1100          */
1101         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1102         if (mmio_dev) {
1103                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1104                 return X86EMUL_CONTINUE;
1105         }
1106
1107         vcpu->mmio_needed = 1;
1108         vcpu->mmio_phys_addr = gpa;
1109         vcpu->mmio_size = bytes;
1110         vcpu->mmio_is_write = 1;
1111         memcpy(vcpu->mmio_data, val, bytes);
1112
1113         return X86EMUL_CONTINUE;
1114 }
1115
1116 static int emulator_cmpxchg_emulated(unsigned long addr,
1117                                      const void *old,
1118                                      const void *new,
1119                                      unsigned int bytes,
1120                                      struct x86_emulate_ctxt *ctxt)
1121 {
1122         static int reported;
1123
1124         if (!reported) {
1125                 reported = 1;
1126                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1127         }
1128         return emulator_write_emulated(addr, new, bytes, ctxt);
1129 }
1130
1131 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1132 {
1133         return kvm_arch_ops->get_segment_base(vcpu, seg);
1134 }
1135
1136 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1137 {
1138         return X86EMUL_CONTINUE;
1139 }
1140
1141 int emulate_clts(struct kvm_vcpu *vcpu)
1142 {
1143         unsigned long cr0;
1144
1145         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1146         kvm_arch_ops->set_cr0(vcpu, cr0);
1147         return X86EMUL_CONTINUE;
1148 }
1149
1150 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1151 {
1152         struct kvm_vcpu *vcpu = ctxt->vcpu;
1153
1154         switch (dr) {
1155         case 0 ... 3:
1156                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1157                 return X86EMUL_CONTINUE;
1158         default:
1159                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1160                        __FUNCTION__, dr);
1161                 return X86EMUL_UNHANDLEABLE;
1162         }
1163 }
1164
1165 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1166 {
1167         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1168         int exception;
1169
1170         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1171         if (exception) {
1172                 /* FIXME: better handling */
1173                 return X86EMUL_UNHANDLEABLE;
1174         }
1175         return X86EMUL_CONTINUE;
1176 }
1177
1178 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1179 {
1180         static int reported;
1181         u8 opcodes[4];
1182         unsigned long rip = ctxt->vcpu->rip;
1183         unsigned long rip_linear;
1184
1185         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1186
1187         if (reported)
1188                 return;
1189
1190         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1191
1192         printk(KERN_ERR "emulation failed but !mmio_needed?"
1193                " rip %lx %02x %02x %02x %02x\n",
1194                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1195         reported = 1;
1196 }
1197
1198 struct x86_emulate_ops emulate_ops = {
1199         .read_std            = emulator_read_std,
1200         .write_std           = emulator_write_std,
1201         .read_emulated       = emulator_read_emulated,
1202         .write_emulated      = emulator_write_emulated,
1203         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1204 };
1205
1206 int emulate_instruction(struct kvm_vcpu *vcpu,
1207                         struct kvm_run *run,
1208                         unsigned long cr2,
1209                         u16 error_code)
1210 {
1211         struct x86_emulate_ctxt emulate_ctxt;
1212         int r;
1213         int cs_db, cs_l;
1214
1215         vcpu->mmio_fault_cr2 = cr2;
1216         kvm_arch_ops->cache_regs(vcpu);
1217
1218         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1219
1220         emulate_ctxt.vcpu = vcpu;
1221         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1222         emulate_ctxt.cr2 = cr2;
1223         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1224                 ? X86EMUL_MODE_REAL : cs_l
1225                 ? X86EMUL_MODE_PROT64 : cs_db
1226                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1227
1228         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1229                 emulate_ctxt.cs_base = 0;
1230                 emulate_ctxt.ds_base = 0;
1231                 emulate_ctxt.es_base = 0;
1232                 emulate_ctxt.ss_base = 0;
1233         } else {
1234                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1235                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1236                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1237                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1238         }
1239
1240         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1241         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1242
1243         vcpu->mmio_is_write = 0;
1244         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1245
1246         if ((r || vcpu->mmio_is_write) && run) {
1247                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1248                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1249                 run->mmio.len = vcpu->mmio_size;
1250                 run->mmio.is_write = vcpu->mmio_is_write;
1251         }
1252
1253         if (r) {
1254                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1255                         return EMULATE_DONE;
1256                 if (!vcpu->mmio_needed) {
1257                         report_emulation_failure(&emulate_ctxt);
1258                         return EMULATE_FAIL;
1259                 }
1260                 return EMULATE_DO_MMIO;
1261         }
1262
1263         kvm_arch_ops->decache_regs(vcpu);
1264         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1265
1266         if (vcpu->mmio_is_write) {
1267                 vcpu->mmio_needed = 0;
1268                 return EMULATE_DO_MMIO;
1269         }
1270
1271         return EMULATE_DONE;
1272 }
1273 EXPORT_SYMBOL_GPL(emulate_instruction);
1274
1275 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1276 {
1277         if (vcpu->irq_summary)
1278                 return 1;
1279
1280         vcpu->run->exit_reason = KVM_EXIT_HLT;
1281         ++vcpu->stat.halt_exits;
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1285
1286 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1287 {
1288         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1289
1290         kvm_arch_ops->cache_regs(vcpu);
1291         ret = -KVM_EINVAL;
1292 #ifdef CONFIG_X86_64
1293         if (is_long_mode(vcpu)) {
1294                 nr = vcpu->regs[VCPU_REGS_RAX];
1295                 a0 = vcpu->regs[VCPU_REGS_RDI];
1296                 a1 = vcpu->regs[VCPU_REGS_RSI];
1297                 a2 = vcpu->regs[VCPU_REGS_RDX];
1298                 a3 = vcpu->regs[VCPU_REGS_RCX];
1299                 a4 = vcpu->regs[VCPU_REGS_R8];
1300                 a5 = vcpu->regs[VCPU_REGS_R9];
1301         } else
1302 #endif
1303         {
1304                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1305                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1306                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1307                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1308                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1309                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1310                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1311         }
1312         switch (nr) {
1313         default:
1314                 run->hypercall.args[0] = a0;
1315                 run->hypercall.args[1] = a1;
1316                 run->hypercall.args[2] = a2;
1317                 run->hypercall.args[3] = a3;
1318                 run->hypercall.args[4] = a4;
1319                 run->hypercall.args[5] = a5;
1320                 run->hypercall.ret = ret;
1321                 run->hypercall.longmode = is_long_mode(vcpu);
1322                 kvm_arch_ops->decache_regs(vcpu);
1323                 return 0;
1324         }
1325         vcpu->regs[VCPU_REGS_RAX] = ret;
1326         kvm_arch_ops->decache_regs(vcpu);
1327         return 1;
1328 }
1329 EXPORT_SYMBOL_GPL(kvm_hypercall);
1330
1331 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1332 {
1333         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1334 }
1335
1336 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1337 {
1338         struct descriptor_table dt = { limit, base };
1339
1340         kvm_arch_ops->set_gdt(vcpu, &dt);
1341 }
1342
1343 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1344 {
1345         struct descriptor_table dt = { limit, base };
1346
1347         kvm_arch_ops->set_idt(vcpu, &dt);
1348 }
1349
1350 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1351                    unsigned long *rflags)
1352 {
1353         lmsw(vcpu, msw);
1354         *rflags = kvm_arch_ops->get_rflags(vcpu);
1355 }
1356
1357 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1358 {
1359         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1360         switch (cr) {
1361         case 0:
1362                 return vcpu->cr0;
1363         case 2:
1364                 return vcpu->cr2;
1365         case 3:
1366                 return vcpu->cr3;
1367         case 4:
1368                 return vcpu->cr4;
1369         default:
1370                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1371                 return 0;
1372         }
1373 }
1374
1375 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1376                      unsigned long *rflags)
1377 {
1378         switch (cr) {
1379         case 0:
1380                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1381                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1382                 break;
1383         case 2:
1384                 vcpu->cr2 = val;
1385                 break;
1386         case 3:
1387                 set_cr3(vcpu, val);
1388                 break;
1389         case 4:
1390                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1391                 break;
1392         default:
1393                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1394         }
1395 }
1396
1397 /*
1398  * Register the para guest with the host:
1399  */
1400 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1401 {
1402         struct kvm_vcpu_para_state *para_state;
1403         hpa_t para_state_hpa, hypercall_hpa;
1404         struct page *para_state_page;
1405         unsigned char *hypercall;
1406         gpa_t hypercall_gpa;
1407
1408         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1409         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1410
1411         /*
1412          * Needs to be page aligned:
1413          */
1414         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1415                 goto err_gp;
1416
1417         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1418         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1419         if (is_error_hpa(para_state_hpa))
1420                 goto err_gp;
1421
1422         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1423         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1424         para_state = kmap_atomic(para_state_page, KM_USER0);
1425
1426         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1427         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1428
1429         para_state->host_version = KVM_PARA_API_VERSION;
1430         /*
1431          * We cannot support guests that try to register themselves
1432          * with a newer API version than the host supports:
1433          */
1434         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1435                 para_state->ret = -KVM_EINVAL;
1436                 goto err_kunmap_skip;
1437         }
1438
1439         hypercall_gpa = para_state->hypercall_gpa;
1440         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1441         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1442         if (is_error_hpa(hypercall_hpa)) {
1443                 para_state->ret = -KVM_EINVAL;
1444                 goto err_kunmap_skip;
1445         }
1446
1447         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1448         vcpu->para_state_page = para_state_page;
1449         vcpu->para_state_gpa = para_state_gpa;
1450         vcpu->hypercall_gpa = hypercall_gpa;
1451
1452         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1453         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1454                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1455         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1456         kunmap_atomic(hypercall, KM_USER1);
1457
1458         para_state->ret = 0;
1459 err_kunmap_skip:
1460         kunmap_atomic(para_state, KM_USER0);
1461         return 0;
1462 err_gp:
1463         return 1;
1464 }
1465
1466 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1467 {
1468         u64 data;
1469
1470         switch (msr) {
1471         case 0xc0010010: /* SYSCFG */
1472         case 0xc0010015: /* HWCR */
1473         case MSR_IA32_PLATFORM_ID:
1474         case MSR_IA32_P5_MC_ADDR:
1475         case MSR_IA32_P5_MC_TYPE:
1476         case MSR_IA32_MC0_CTL:
1477         case MSR_IA32_MCG_STATUS:
1478         case MSR_IA32_MCG_CAP:
1479         case MSR_IA32_MC0_MISC:
1480         case MSR_IA32_MC0_MISC+4:
1481         case MSR_IA32_MC0_MISC+8:
1482         case MSR_IA32_MC0_MISC+12:
1483         case MSR_IA32_MC0_MISC+16:
1484         case MSR_IA32_UCODE_REV:
1485         case MSR_IA32_PERF_STATUS:
1486         case MSR_IA32_EBL_CR_POWERON:
1487                 /* MTRR registers */
1488         case 0xfe:
1489         case 0x200 ... 0x2ff:
1490                 data = 0;
1491                 break;
1492         case 0xcd: /* fsb frequency */
1493                 data = 3;
1494                 break;
1495         case MSR_IA32_APICBASE:
1496                 data = vcpu->apic_base;
1497                 break;
1498         case MSR_IA32_MISC_ENABLE:
1499                 data = vcpu->ia32_misc_enable_msr;
1500                 break;
1501 #ifdef CONFIG_X86_64
1502         case MSR_EFER:
1503                 data = vcpu->shadow_efer;
1504                 break;
1505 #endif
1506         default:
1507                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1508                 return 1;
1509         }
1510         *pdata = data;
1511         return 0;
1512 }
1513 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1514
1515 /*
1516  * Reads an msr value (of 'msr_index') into 'pdata'.
1517  * Returns 0 on success, non-0 otherwise.
1518  * Assumes vcpu_load() was already called.
1519  */
1520 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1521 {
1522         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1523 }
1524
1525 #ifdef CONFIG_X86_64
1526
1527 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1528 {
1529         if (efer & EFER_RESERVED_BITS) {
1530                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1531                        efer);
1532                 inject_gp(vcpu);
1533                 return;
1534         }
1535
1536         if (is_paging(vcpu)
1537             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1538                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1539                 inject_gp(vcpu);
1540                 return;
1541         }
1542
1543         kvm_arch_ops->set_efer(vcpu, efer);
1544
1545         efer &= ~EFER_LMA;
1546         efer |= vcpu->shadow_efer & EFER_LMA;
1547
1548         vcpu->shadow_efer = efer;
1549 }
1550
1551 #endif
1552
1553 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1554 {
1555         switch (msr) {
1556 #ifdef CONFIG_X86_64
1557         case MSR_EFER:
1558                 set_efer(vcpu, data);
1559                 break;
1560 #endif
1561         case MSR_IA32_MC0_STATUS:
1562                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1563                        __FUNCTION__, data);
1564                 break;
1565         case MSR_IA32_MCG_STATUS:
1566                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1567                         __FUNCTION__, data);
1568                 break;
1569         case MSR_IA32_UCODE_REV:
1570         case MSR_IA32_UCODE_WRITE:
1571         case 0x200 ... 0x2ff: /* MTRRs */
1572                 break;
1573         case MSR_IA32_APICBASE:
1574                 vcpu->apic_base = data;
1575                 break;
1576         case MSR_IA32_MISC_ENABLE:
1577                 vcpu->ia32_misc_enable_msr = data;
1578                 break;
1579         /*
1580          * This is the 'probe whether the host is KVM' logic:
1581          */
1582         case MSR_KVM_API_MAGIC:
1583                 return vcpu_register_para(vcpu, data);
1584
1585         default:
1586                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1587                 return 1;
1588         }
1589         return 0;
1590 }
1591 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1592
1593 /*
1594  * Writes msr value into into the appropriate "register".
1595  * Returns 0 on success, non-0 otherwise.
1596  * Assumes vcpu_load() was already called.
1597  */
1598 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1599 {
1600         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1601 }
1602
1603 void kvm_resched(struct kvm_vcpu *vcpu)
1604 {
1605         if (!need_resched())
1606                 return;
1607         vcpu_put(vcpu);
1608         cond_resched();
1609         vcpu_load(vcpu);
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_resched);
1612
1613 void load_msrs(struct vmx_msr_entry *e, int n)
1614 {
1615         int i;
1616
1617         for (i = 0; i < n; ++i)
1618                 wrmsrl(e[i].index, e[i].data);
1619 }
1620 EXPORT_SYMBOL_GPL(load_msrs);
1621
1622 void save_msrs(struct vmx_msr_entry *e, int n)
1623 {
1624         int i;
1625
1626         for (i = 0; i < n; ++i)
1627                 rdmsrl(e[i].index, e[i].data);
1628 }
1629 EXPORT_SYMBOL_GPL(save_msrs);
1630
1631 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1632 {
1633         int i;
1634         u32 function;
1635         struct kvm_cpuid_entry *e, *best;
1636
1637         kvm_arch_ops->cache_regs(vcpu);
1638         function = vcpu->regs[VCPU_REGS_RAX];
1639         vcpu->regs[VCPU_REGS_RAX] = 0;
1640         vcpu->regs[VCPU_REGS_RBX] = 0;
1641         vcpu->regs[VCPU_REGS_RCX] = 0;
1642         vcpu->regs[VCPU_REGS_RDX] = 0;
1643         best = NULL;
1644         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1645                 e = &vcpu->cpuid_entries[i];
1646                 if (e->function == function) {
1647                         best = e;
1648                         break;
1649                 }
1650                 /*
1651                  * Both basic or both extended?
1652                  */
1653                 if (((e->function ^ function) & 0x80000000) == 0)
1654                         if (!best || e->function > best->function)
1655                                 best = e;
1656         }
1657         if (best) {
1658                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1659                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1660                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1661                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1662         }
1663         kvm_arch_ops->decache_regs(vcpu);
1664         kvm_arch_ops->skip_emulated_instruction(vcpu);
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1667
1668 static int pio_copy_data(struct kvm_vcpu *vcpu)
1669 {
1670         void *p = vcpu->pio_data;
1671         void *q;
1672         unsigned bytes;
1673         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1674
1675         kvm_arch_ops->vcpu_put(vcpu);
1676         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1677                  PAGE_KERNEL);
1678         if (!q) {
1679                 kvm_arch_ops->vcpu_load(vcpu);
1680                 free_pio_guest_pages(vcpu);
1681                 return -ENOMEM;
1682         }
1683         q += vcpu->pio.guest_page_offset;
1684         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1685         if (vcpu->pio.in)
1686                 memcpy(q, p, bytes);
1687         else
1688                 memcpy(p, q, bytes);
1689         q -= vcpu->pio.guest_page_offset;
1690         vunmap(q);
1691         kvm_arch_ops->vcpu_load(vcpu);
1692         free_pio_guest_pages(vcpu);
1693         return 0;
1694 }
1695
1696 static int complete_pio(struct kvm_vcpu *vcpu)
1697 {
1698         struct kvm_pio_request *io = &vcpu->pio;
1699         long delta;
1700         int r;
1701
1702         kvm_arch_ops->cache_regs(vcpu);
1703
1704         if (!io->string) {
1705                 if (io->in)
1706                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1707                                io->size);
1708         } else {
1709                 if (io->in) {
1710                         r = pio_copy_data(vcpu);
1711                         if (r) {
1712                                 kvm_arch_ops->cache_regs(vcpu);
1713                                 return r;
1714                         }
1715                 }
1716
1717                 delta = 1;
1718                 if (io->rep) {
1719                         delta *= io->cur_count;
1720                         /*
1721                          * The size of the register should really depend on
1722                          * current address size.
1723                          */
1724                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1725                 }
1726                 if (io->down)
1727                         delta = -delta;
1728                 delta *= io->size;
1729                 if (io->in)
1730                         vcpu->regs[VCPU_REGS_RDI] += delta;
1731                 else
1732                         vcpu->regs[VCPU_REGS_RSI] += delta;
1733         }
1734
1735         kvm_arch_ops->decache_regs(vcpu);
1736
1737         io->count -= io->cur_count;
1738         io->cur_count = 0;
1739
1740         if (!io->count)
1741                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1742         return 0;
1743 }
1744
1745 void kernel_pio(struct kvm_io_device *pio_dev, struct kvm_vcpu *vcpu)
1746 {
1747         /* TODO: String I/O for in kernel device */
1748
1749         if (vcpu->pio.in)
1750                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1751                                   vcpu->pio.size,
1752                                   vcpu->pio_data);
1753         else
1754                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1755                                    vcpu->pio.size,
1756                                    vcpu->pio_data);
1757 }
1758
1759 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1760                   int size, unsigned long count, int string, int down,
1761                   gva_t address, int rep, unsigned port)
1762 {
1763         unsigned now, in_page;
1764         int i;
1765         int nr_pages = 1;
1766         struct page *page;
1767         struct kvm_io_device *pio_dev;
1768
1769         vcpu->run->exit_reason = KVM_EXIT_IO;
1770         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1771         vcpu->run->io.size = size;
1772         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1773         vcpu->run->io.count = count;
1774         vcpu->run->io.port = port;
1775         vcpu->pio.count = count;
1776         vcpu->pio.cur_count = count;
1777         vcpu->pio.size = size;
1778         vcpu->pio.in = in;
1779         vcpu->pio.port = port;
1780         vcpu->pio.string = string;
1781         vcpu->pio.down = down;
1782         vcpu->pio.guest_page_offset = offset_in_page(address);
1783         vcpu->pio.rep = rep;
1784
1785         pio_dev = vcpu_find_pio_dev(vcpu, port);
1786         if (!string) {
1787                 kvm_arch_ops->cache_regs(vcpu);
1788                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1789                 kvm_arch_ops->decache_regs(vcpu);
1790                 if (pio_dev) {
1791                         kernel_pio(pio_dev, vcpu);
1792                         complete_pio(vcpu);
1793                         return 1;
1794                 }
1795                 return 0;
1796         }
1797         /* TODO: String I/O for in kernel device */
1798         if (pio_dev)
1799                 printk(KERN_ERR "kvm_setup_pio: no string io support\n");
1800
1801         if (!count) {
1802                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1803                 return 1;
1804         }
1805
1806         now = min(count, PAGE_SIZE / size);
1807
1808         if (!down)
1809                 in_page = PAGE_SIZE - offset_in_page(address);
1810         else
1811                 in_page = offset_in_page(address) + size;
1812         now = min(count, (unsigned long)in_page / size);
1813         if (!now) {
1814                 /*
1815                  * String I/O straddles page boundary.  Pin two guest pages
1816                  * so that we satisfy atomicity constraints.  Do just one
1817                  * transaction to avoid complexity.
1818                  */
1819                 nr_pages = 2;
1820                 now = 1;
1821         }
1822         if (down) {
1823                 /*
1824                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1825                  */
1826                 printk(KERN_ERR "kvm: guest string pio down\n");
1827                 inject_gp(vcpu);
1828                 return 1;
1829         }
1830         vcpu->run->io.count = now;
1831         vcpu->pio.cur_count = now;
1832
1833         for (i = 0; i < nr_pages; ++i) {
1834                 spin_lock(&vcpu->kvm->lock);
1835                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1836                 if (page)
1837                         get_page(page);
1838                 vcpu->pio.guest_pages[i] = page;
1839                 spin_unlock(&vcpu->kvm->lock);
1840                 if (!page) {
1841                         inject_gp(vcpu);
1842                         free_pio_guest_pages(vcpu);
1843                         return 1;
1844                 }
1845         }
1846
1847         if (!vcpu->pio.in)
1848                 return pio_copy_data(vcpu);
1849         return 0;
1850 }
1851 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1852
1853 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1854 {
1855         int r;
1856         sigset_t sigsaved;
1857
1858         vcpu_load(vcpu);
1859
1860         if (vcpu->sigset_active)
1861                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1862
1863         /* re-sync apic's tpr */
1864         vcpu->cr8 = kvm_run->cr8;
1865
1866         if (vcpu->pio.cur_count) {
1867                 r = complete_pio(vcpu);
1868                 if (r)
1869                         goto out;
1870         }
1871
1872         if (vcpu->mmio_needed) {
1873                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1874                 vcpu->mmio_read_completed = 1;
1875                 vcpu->mmio_needed = 0;
1876                 r = emulate_instruction(vcpu, kvm_run,
1877                                         vcpu->mmio_fault_cr2, 0);
1878                 if (r == EMULATE_DO_MMIO) {
1879                         /*
1880                          * Read-modify-write.  Back to userspace.
1881                          */
1882                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1883                         r = 0;
1884                         goto out;
1885                 }
1886         }
1887
1888         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1889                 kvm_arch_ops->cache_regs(vcpu);
1890                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1891                 kvm_arch_ops->decache_regs(vcpu);
1892         }
1893
1894         r = kvm_arch_ops->run(vcpu, kvm_run);
1895
1896 out:
1897         if (vcpu->sigset_active)
1898                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1899
1900         vcpu_put(vcpu);
1901         return r;
1902 }
1903
1904 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1905                                    struct kvm_regs *regs)
1906 {
1907         vcpu_load(vcpu);
1908
1909         kvm_arch_ops->cache_regs(vcpu);
1910
1911         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1912         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1913         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1914         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1915         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1916         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1917         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1918         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1919 #ifdef CONFIG_X86_64
1920         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1921         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1922         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1923         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1924         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1925         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1926         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1927         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1928 #endif
1929
1930         regs->rip = vcpu->rip;
1931         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1932
1933         /*
1934          * Don't leak debug flags in case they were set for guest debugging
1935          */
1936         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1937                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1938
1939         vcpu_put(vcpu);
1940
1941         return 0;
1942 }
1943
1944 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1945                                    struct kvm_regs *regs)
1946 {
1947         vcpu_load(vcpu);
1948
1949         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1950         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1951         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1952         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1953         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1954         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1955         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1956         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1957 #ifdef CONFIG_X86_64
1958         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1959         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1960         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1961         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1962         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1963         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1964         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1965         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1966 #endif
1967
1968         vcpu->rip = regs->rip;
1969         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1970
1971         kvm_arch_ops->decache_regs(vcpu);
1972
1973         vcpu_put(vcpu);
1974
1975         return 0;
1976 }
1977
1978 static void get_segment(struct kvm_vcpu *vcpu,
1979                         struct kvm_segment *var, int seg)
1980 {
1981         return kvm_arch_ops->get_segment(vcpu, var, seg);
1982 }
1983
1984 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1985                                     struct kvm_sregs *sregs)
1986 {
1987         struct descriptor_table dt;
1988
1989         vcpu_load(vcpu);
1990
1991         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1992         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1993         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1994         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1995         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1996         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1997
1998         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1999         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2000
2001         kvm_arch_ops->get_idt(vcpu, &dt);
2002         sregs->idt.limit = dt.limit;
2003         sregs->idt.base = dt.base;
2004         kvm_arch_ops->get_gdt(vcpu, &dt);
2005         sregs->gdt.limit = dt.limit;
2006         sregs->gdt.base = dt.base;
2007
2008         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2009         sregs->cr0 = vcpu->cr0;
2010         sregs->cr2 = vcpu->cr2;
2011         sregs->cr3 = vcpu->cr3;
2012         sregs->cr4 = vcpu->cr4;
2013         sregs->cr8 = vcpu->cr8;
2014         sregs->efer = vcpu->shadow_efer;
2015         sregs->apic_base = vcpu->apic_base;
2016
2017         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2018                sizeof sregs->interrupt_bitmap);
2019
2020         vcpu_put(vcpu);
2021
2022         return 0;
2023 }
2024
2025 static void set_segment(struct kvm_vcpu *vcpu,
2026                         struct kvm_segment *var, int seg)
2027 {
2028         return kvm_arch_ops->set_segment(vcpu, var, seg);
2029 }
2030
2031 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2032                                     struct kvm_sregs *sregs)
2033 {
2034         int mmu_reset_needed = 0;
2035         int i;
2036         struct descriptor_table dt;
2037
2038         vcpu_load(vcpu);
2039
2040         dt.limit = sregs->idt.limit;
2041         dt.base = sregs->idt.base;
2042         kvm_arch_ops->set_idt(vcpu, &dt);
2043         dt.limit = sregs->gdt.limit;
2044         dt.base = sregs->gdt.base;
2045         kvm_arch_ops->set_gdt(vcpu, &dt);
2046
2047         vcpu->cr2 = sregs->cr2;
2048         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2049         vcpu->cr3 = sregs->cr3;
2050
2051         vcpu->cr8 = sregs->cr8;
2052
2053         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2054 #ifdef CONFIG_X86_64
2055         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2056 #endif
2057         vcpu->apic_base = sregs->apic_base;
2058
2059         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2060
2061         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2062         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2063
2064         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2065         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2066         if (!is_long_mode(vcpu) && is_pae(vcpu))
2067                 load_pdptrs(vcpu, vcpu->cr3);
2068
2069         if (mmu_reset_needed)
2070                 kvm_mmu_reset_context(vcpu);
2071
2072         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2073                sizeof vcpu->irq_pending);
2074         vcpu->irq_summary = 0;
2075         for (i = 0; i < NR_IRQ_WORDS; ++i)
2076                 if (vcpu->irq_pending[i])
2077                         __set_bit(i, &vcpu->irq_summary);
2078
2079         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2080         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2081         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2082         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2083         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2084         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2085
2086         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2087         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2088
2089         vcpu_put(vcpu);
2090
2091         return 0;
2092 }
2093
2094 /*
2095  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2096  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2097  *
2098  * This list is modified at module load time to reflect the
2099  * capabilities of the host cpu.
2100  */
2101 static u32 msrs_to_save[] = {
2102         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2103         MSR_K6_STAR,
2104 #ifdef CONFIG_X86_64
2105         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2106 #endif
2107         MSR_IA32_TIME_STAMP_COUNTER,
2108 };
2109
2110 static unsigned num_msrs_to_save;
2111
2112 static u32 emulated_msrs[] = {
2113         MSR_IA32_MISC_ENABLE,
2114 };
2115
2116 static __init void kvm_init_msr_list(void)
2117 {
2118         u32 dummy[2];
2119         unsigned i, j;
2120
2121         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2122                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2123                         continue;
2124                 if (j < i)
2125                         msrs_to_save[j] = msrs_to_save[i];
2126                 j++;
2127         }
2128         num_msrs_to_save = j;
2129 }
2130
2131 /*
2132  * Adapt set_msr() to msr_io()'s calling convention
2133  */
2134 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2135 {
2136         return kvm_set_msr(vcpu, index, *data);
2137 }
2138
2139 /*
2140  * Read or write a bunch of msrs. All parameters are kernel addresses.
2141  *
2142  * @return number of msrs set successfully.
2143  */
2144 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2145                     struct kvm_msr_entry *entries,
2146                     int (*do_msr)(struct kvm_vcpu *vcpu,
2147                                   unsigned index, u64 *data))
2148 {
2149         int i;
2150
2151         vcpu_load(vcpu);
2152
2153         for (i = 0; i < msrs->nmsrs; ++i)
2154                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2155                         break;
2156
2157         vcpu_put(vcpu);
2158
2159         return i;
2160 }
2161
2162 /*
2163  * Read or write a bunch of msrs. Parameters are user addresses.
2164  *
2165  * @return number of msrs set successfully.
2166  */
2167 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2168                   int (*do_msr)(struct kvm_vcpu *vcpu,
2169                                 unsigned index, u64 *data),
2170                   int writeback)
2171 {
2172         struct kvm_msrs msrs;
2173         struct kvm_msr_entry *entries;
2174         int r, n;
2175         unsigned size;
2176
2177         r = -EFAULT;
2178         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2179                 goto out;
2180
2181         r = -E2BIG;
2182         if (msrs.nmsrs >= MAX_IO_MSRS)
2183                 goto out;
2184
2185         r = -ENOMEM;
2186         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2187         entries = vmalloc(size);
2188         if (!entries)
2189                 goto out;
2190
2191         r = -EFAULT;
2192         if (copy_from_user(entries, user_msrs->entries, size))
2193                 goto out_free;
2194
2195         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2196         if (r < 0)
2197                 goto out_free;
2198
2199         r = -EFAULT;
2200         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2201                 goto out_free;
2202
2203         r = n;
2204
2205 out_free:
2206         vfree(entries);
2207 out:
2208         return r;
2209 }
2210
2211 /*
2212  * Translate a guest virtual address to a guest physical address.
2213  */
2214 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2215                                     struct kvm_translation *tr)
2216 {
2217         unsigned long vaddr = tr->linear_address;
2218         gpa_t gpa;
2219
2220         vcpu_load(vcpu);
2221         spin_lock(&vcpu->kvm->lock);
2222         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2223         tr->physical_address = gpa;
2224         tr->valid = gpa != UNMAPPED_GVA;
2225         tr->writeable = 1;
2226         tr->usermode = 0;
2227         spin_unlock(&vcpu->kvm->lock);
2228         vcpu_put(vcpu);
2229
2230         return 0;
2231 }
2232
2233 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2234                                     struct kvm_interrupt *irq)
2235 {
2236         if (irq->irq < 0 || irq->irq >= 256)
2237                 return -EINVAL;
2238         vcpu_load(vcpu);
2239
2240         set_bit(irq->irq, vcpu->irq_pending);
2241         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2242
2243         vcpu_put(vcpu);
2244
2245         return 0;
2246 }
2247
2248 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2249                                       struct kvm_debug_guest *dbg)
2250 {
2251         int r;
2252
2253         vcpu_load(vcpu);
2254
2255         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2256
2257         vcpu_put(vcpu);
2258
2259         return r;
2260 }
2261
2262 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2263                                     unsigned long address,
2264                                     int *type)
2265 {
2266         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2267         unsigned long pgoff;
2268         struct page *page;
2269
2270         *type = VM_FAULT_MINOR;
2271         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2272         if (pgoff == 0)
2273                 page = virt_to_page(vcpu->run);
2274         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2275                 page = virt_to_page(vcpu->pio_data);
2276         else
2277                 return NOPAGE_SIGBUS;
2278         get_page(page);
2279         return page;
2280 }
2281
2282 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2283         .nopage = kvm_vcpu_nopage,
2284 };
2285
2286 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2287 {
2288         vma->vm_ops = &kvm_vcpu_vm_ops;
2289         return 0;
2290 }
2291
2292 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2293 {
2294         struct kvm_vcpu *vcpu = filp->private_data;
2295
2296         fput(vcpu->kvm->filp);
2297         return 0;
2298 }
2299
2300 static struct file_operations kvm_vcpu_fops = {
2301         .release        = kvm_vcpu_release,
2302         .unlocked_ioctl = kvm_vcpu_ioctl,
2303         .compat_ioctl   = kvm_vcpu_ioctl,
2304         .mmap           = kvm_vcpu_mmap,
2305 };
2306
2307 /*
2308  * Allocates an inode for the vcpu.
2309  */
2310 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2311 {
2312         int fd, r;
2313         struct inode *inode;
2314         struct file *file;
2315
2316         r = anon_inode_getfd(&fd, &inode, &file,
2317                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2318         if (r)
2319                 return r;
2320         atomic_inc(&vcpu->kvm->filp->f_count);
2321         return fd;
2322 }
2323
2324 /*
2325  * Creates some virtual cpus.  Good luck creating more than one.
2326  */
2327 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2328 {
2329         int r;
2330         struct kvm_vcpu *vcpu;
2331         struct page *page;
2332
2333         r = -EINVAL;
2334         if (!valid_vcpu(n))
2335                 goto out;
2336
2337         vcpu = &kvm->vcpus[n];
2338
2339         mutex_lock(&vcpu->mutex);
2340
2341         if (vcpu->vmcs) {
2342                 mutex_unlock(&vcpu->mutex);
2343                 return -EEXIST;
2344         }
2345
2346         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2347         r = -ENOMEM;
2348         if (!page)
2349                 goto out_unlock;
2350         vcpu->run = page_address(page);
2351
2352         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2353         r = -ENOMEM;
2354         if (!page)
2355                 goto out_free_run;
2356         vcpu->pio_data = page_address(page);
2357
2358         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2359                                            FX_IMAGE_ALIGN);
2360         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2361         vcpu->cr0 = 0x10;
2362
2363         r = kvm_arch_ops->vcpu_create(vcpu);
2364         if (r < 0)
2365                 goto out_free_vcpus;
2366
2367         r = kvm_mmu_create(vcpu);
2368         if (r < 0)
2369                 goto out_free_vcpus;
2370
2371         kvm_arch_ops->vcpu_load(vcpu);
2372         r = kvm_mmu_setup(vcpu);
2373         if (r >= 0)
2374                 r = kvm_arch_ops->vcpu_setup(vcpu);
2375         vcpu_put(vcpu);
2376
2377         if (r < 0)
2378                 goto out_free_vcpus;
2379
2380         r = create_vcpu_fd(vcpu);
2381         if (r < 0)
2382                 goto out_free_vcpus;
2383
2384         spin_lock(&kvm_lock);
2385         if (n >= kvm->nvcpus)
2386                 kvm->nvcpus = n + 1;
2387         spin_unlock(&kvm_lock);
2388
2389         return r;
2390
2391 out_free_vcpus:
2392         kvm_free_vcpu(vcpu);
2393 out_free_run:
2394         free_page((unsigned long)vcpu->run);
2395         vcpu->run = NULL;
2396 out_unlock:
2397         mutex_unlock(&vcpu->mutex);
2398 out:
2399         return r;
2400 }
2401
2402 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2403 {
2404         u64 efer;
2405         int i;
2406         struct kvm_cpuid_entry *e, *entry;
2407
2408         rdmsrl(MSR_EFER, efer);
2409         entry = NULL;
2410         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2411                 e = &vcpu->cpuid_entries[i];
2412                 if (e->function == 0x80000001) {
2413                         entry = e;
2414                         break;
2415                 }
2416         }
2417         if (entry && (entry->edx & EFER_NX) && !(efer & EFER_NX)) {
2418                 entry->edx &= ~(1 << 20);
2419                 printk(KERN_INFO ": guest NX capability removed\n");
2420         }
2421 }
2422
2423 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2424                                     struct kvm_cpuid *cpuid,
2425                                     struct kvm_cpuid_entry __user *entries)
2426 {
2427         int r;
2428
2429         r = -E2BIG;
2430         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2431                 goto out;
2432         r = -EFAULT;
2433         if (copy_from_user(&vcpu->cpuid_entries, entries,
2434                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2435                 goto out;
2436         vcpu->cpuid_nent = cpuid->nent;
2437         cpuid_fix_nx_cap(vcpu);
2438         return 0;
2439
2440 out:
2441         return r;
2442 }
2443
2444 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2445 {
2446         if (sigset) {
2447                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2448                 vcpu->sigset_active = 1;
2449                 vcpu->sigset = *sigset;
2450         } else
2451                 vcpu->sigset_active = 0;
2452         return 0;
2453 }
2454
2455 /*
2456  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2457  * we have asm/x86/processor.h
2458  */
2459 struct fxsave {
2460         u16     cwd;
2461         u16     swd;
2462         u16     twd;
2463         u16     fop;
2464         u64     rip;
2465         u64     rdp;
2466         u32     mxcsr;
2467         u32     mxcsr_mask;
2468         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2469 #ifdef CONFIG_X86_64
2470         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2471 #else
2472         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2473 #endif
2474 };
2475
2476 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2477 {
2478         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2479
2480         vcpu_load(vcpu);
2481
2482         memcpy(fpu->fpr, fxsave->st_space, 128);
2483         fpu->fcw = fxsave->cwd;
2484         fpu->fsw = fxsave->swd;
2485         fpu->ftwx = fxsave->twd;
2486         fpu->last_opcode = fxsave->fop;
2487         fpu->last_ip = fxsave->rip;
2488         fpu->last_dp = fxsave->rdp;
2489         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2490
2491         vcpu_put(vcpu);
2492
2493         return 0;
2494 }
2495
2496 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2497 {
2498         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2499
2500         vcpu_load(vcpu);
2501
2502         memcpy(fxsave->st_space, fpu->fpr, 128);
2503         fxsave->cwd = fpu->fcw;
2504         fxsave->swd = fpu->fsw;
2505         fxsave->twd = fpu->ftwx;
2506         fxsave->fop = fpu->last_opcode;
2507         fxsave->rip = fpu->last_ip;
2508         fxsave->rdp = fpu->last_dp;
2509         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2510
2511         vcpu_put(vcpu);
2512
2513         return 0;
2514 }
2515
2516 static long kvm_vcpu_ioctl(struct file *filp,
2517                            unsigned int ioctl, unsigned long arg)
2518 {
2519         struct kvm_vcpu *vcpu = filp->private_data;
2520         void __user *argp = (void __user *)arg;
2521         int r = -EINVAL;
2522
2523         switch (ioctl) {
2524         case KVM_RUN:
2525                 r = -EINVAL;
2526                 if (arg)
2527                         goto out;
2528                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2529                 break;
2530         case KVM_GET_REGS: {
2531                 struct kvm_regs kvm_regs;
2532
2533                 memset(&kvm_regs, 0, sizeof kvm_regs);
2534                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2535                 if (r)
2536                         goto out;
2537                 r = -EFAULT;
2538                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2539                         goto out;
2540                 r = 0;
2541                 break;
2542         }
2543         case KVM_SET_REGS: {
2544                 struct kvm_regs kvm_regs;
2545
2546                 r = -EFAULT;
2547                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2548                         goto out;
2549                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2550                 if (r)
2551                         goto out;
2552                 r = 0;
2553                 break;
2554         }
2555         case KVM_GET_SREGS: {
2556                 struct kvm_sregs kvm_sregs;
2557
2558                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2559                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2560                 if (r)
2561                         goto out;
2562                 r = -EFAULT;
2563                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2564                         goto out;
2565                 r = 0;
2566                 break;
2567         }
2568         case KVM_SET_SREGS: {
2569                 struct kvm_sregs kvm_sregs;
2570
2571                 r = -EFAULT;
2572                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2573                         goto out;
2574                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2575                 if (r)
2576                         goto out;
2577                 r = 0;
2578                 break;
2579         }
2580         case KVM_TRANSLATE: {
2581                 struct kvm_translation tr;
2582
2583                 r = -EFAULT;
2584                 if (copy_from_user(&tr, argp, sizeof tr))
2585                         goto out;
2586                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2587                 if (r)
2588                         goto out;
2589                 r = -EFAULT;
2590                 if (copy_to_user(argp, &tr, sizeof tr))
2591                         goto out;
2592                 r = 0;
2593                 break;
2594         }
2595         case KVM_INTERRUPT: {
2596                 struct kvm_interrupt irq;
2597
2598                 r = -EFAULT;
2599                 if (copy_from_user(&irq, argp, sizeof irq))
2600                         goto out;
2601                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2602                 if (r)
2603                         goto out;
2604                 r = 0;
2605                 break;
2606         }
2607         case KVM_DEBUG_GUEST: {
2608                 struct kvm_debug_guest dbg;
2609
2610                 r = -EFAULT;
2611                 if (copy_from_user(&dbg, argp, sizeof dbg))
2612                         goto out;
2613                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2614                 if (r)
2615                         goto out;
2616                 r = 0;
2617                 break;
2618         }
2619         case KVM_GET_MSRS:
2620                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2621                 break;
2622         case KVM_SET_MSRS:
2623                 r = msr_io(vcpu, argp, do_set_msr, 0);
2624                 break;
2625         case KVM_SET_CPUID: {
2626                 struct kvm_cpuid __user *cpuid_arg = argp;
2627                 struct kvm_cpuid cpuid;
2628
2629                 r = -EFAULT;
2630                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2631                         goto out;
2632                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2633                 if (r)
2634                         goto out;
2635                 break;
2636         }
2637         case KVM_SET_SIGNAL_MASK: {
2638                 struct kvm_signal_mask __user *sigmask_arg = argp;
2639                 struct kvm_signal_mask kvm_sigmask;
2640                 sigset_t sigset, *p;
2641
2642                 p = NULL;
2643                 if (argp) {
2644                         r = -EFAULT;
2645                         if (copy_from_user(&kvm_sigmask, argp,
2646                                            sizeof kvm_sigmask))
2647                                 goto out;
2648                         r = -EINVAL;
2649                         if (kvm_sigmask.len != sizeof sigset)
2650                                 goto out;
2651                         r = -EFAULT;
2652                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2653                                            sizeof sigset))
2654                                 goto out;
2655                         p = &sigset;
2656                 }
2657                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2658                 break;
2659         }
2660         case KVM_GET_FPU: {
2661                 struct kvm_fpu fpu;
2662
2663                 memset(&fpu, 0, sizeof fpu);
2664                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2665                 if (r)
2666                         goto out;
2667                 r = -EFAULT;
2668                 if (copy_to_user(argp, &fpu, sizeof fpu))
2669                         goto out;
2670                 r = 0;
2671                 break;
2672         }
2673         case KVM_SET_FPU: {
2674                 struct kvm_fpu fpu;
2675
2676                 r = -EFAULT;
2677                 if (copy_from_user(&fpu, argp, sizeof fpu))
2678                         goto out;
2679                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2680                 if (r)
2681                         goto out;
2682                 r = 0;
2683                 break;
2684         }
2685         default:
2686                 ;
2687         }
2688 out:
2689         return r;
2690 }
2691
2692 static long kvm_vm_ioctl(struct file *filp,
2693                            unsigned int ioctl, unsigned long arg)
2694 {
2695         struct kvm *kvm = filp->private_data;
2696         void __user *argp = (void __user *)arg;
2697         int r = -EINVAL;
2698
2699         switch (ioctl) {
2700         case KVM_CREATE_VCPU:
2701                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2702                 if (r < 0)
2703                         goto out;
2704                 break;
2705         case KVM_SET_MEMORY_REGION: {
2706                 struct kvm_memory_region kvm_mem;
2707
2708                 r = -EFAULT;
2709                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2710                         goto out;
2711                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2712                 if (r)
2713                         goto out;
2714                 break;
2715         }
2716         case KVM_GET_DIRTY_LOG: {
2717                 struct kvm_dirty_log log;
2718
2719                 r = -EFAULT;
2720                 if (copy_from_user(&log, argp, sizeof log))
2721                         goto out;
2722                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2723                 if (r)
2724                         goto out;
2725                 break;
2726         }
2727         case KVM_SET_MEMORY_ALIAS: {
2728                 struct kvm_memory_alias alias;
2729
2730                 r = -EFAULT;
2731                 if (copy_from_user(&alias, argp, sizeof alias))
2732                         goto out;
2733                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2734                 if (r)
2735                         goto out;
2736                 break;
2737         }
2738         default:
2739                 ;
2740         }
2741 out:
2742         return r;
2743 }
2744
2745 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2746                                   unsigned long address,
2747                                   int *type)
2748 {
2749         struct kvm *kvm = vma->vm_file->private_data;
2750         unsigned long pgoff;
2751         struct page *page;
2752
2753         *type = VM_FAULT_MINOR;
2754         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2755         page = gfn_to_page(kvm, pgoff);
2756         if (!page)
2757                 return NOPAGE_SIGBUS;
2758         get_page(page);
2759         return page;
2760 }
2761
2762 static struct vm_operations_struct kvm_vm_vm_ops = {
2763         .nopage = kvm_vm_nopage,
2764 };
2765
2766 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2767 {
2768         vma->vm_ops = &kvm_vm_vm_ops;
2769         return 0;
2770 }
2771
2772 static struct file_operations kvm_vm_fops = {
2773         .release        = kvm_vm_release,
2774         .unlocked_ioctl = kvm_vm_ioctl,
2775         .compat_ioctl   = kvm_vm_ioctl,
2776         .mmap           = kvm_vm_mmap,
2777 };
2778
2779 static int kvm_dev_ioctl_create_vm(void)
2780 {
2781         int fd, r;
2782         struct inode *inode;
2783         struct file *file;
2784         struct kvm *kvm;
2785
2786         kvm = kvm_create_vm();
2787         if (IS_ERR(kvm))
2788                 return PTR_ERR(kvm);
2789         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2790         if (r) {
2791                 kvm_destroy_vm(kvm);
2792                 return r;
2793         }
2794
2795         kvm->filp = file;
2796
2797         return fd;
2798 }
2799
2800 static long kvm_dev_ioctl(struct file *filp,
2801                           unsigned int ioctl, unsigned long arg)
2802 {
2803         void __user *argp = (void __user *)arg;
2804         long r = -EINVAL;
2805
2806         switch (ioctl) {
2807         case KVM_GET_API_VERSION:
2808                 r = -EINVAL;
2809                 if (arg)
2810                         goto out;
2811                 r = KVM_API_VERSION;
2812                 break;
2813         case KVM_CREATE_VM:
2814                 r = -EINVAL;
2815                 if (arg)
2816                         goto out;
2817                 r = kvm_dev_ioctl_create_vm();
2818                 break;
2819         case KVM_GET_MSR_INDEX_LIST: {
2820                 struct kvm_msr_list __user *user_msr_list = argp;
2821                 struct kvm_msr_list msr_list;
2822                 unsigned n;
2823
2824                 r = -EFAULT;
2825                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2826                         goto out;
2827                 n = msr_list.nmsrs;
2828                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2829                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2830                         goto out;
2831                 r = -E2BIG;
2832                 if (n < num_msrs_to_save)
2833                         goto out;
2834                 r = -EFAULT;
2835                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2836                                  num_msrs_to_save * sizeof(u32)))
2837                         goto out;
2838                 if (copy_to_user(user_msr_list->indices
2839                                  + num_msrs_to_save * sizeof(u32),
2840                                  &emulated_msrs,
2841                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2842                         goto out;
2843                 r = 0;
2844                 break;
2845         }
2846         case KVM_CHECK_EXTENSION:
2847                 /*
2848                  * No extensions defined at present.
2849                  */
2850                 r = 0;
2851                 break;
2852         case KVM_GET_VCPU_MMAP_SIZE:
2853                 r = -EINVAL;
2854                 if (arg)
2855                         goto out;
2856                 r = 2 * PAGE_SIZE;
2857                 break;
2858         default:
2859                 ;
2860         }
2861 out:
2862         return r;
2863 }
2864
2865 static struct file_operations kvm_chardev_ops = {
2866         .open           = kvm_dev_open,
2867         .release        = kvm_dev_release,
2868         .unlocked_ioctl = kvm_dev_ioctl,
2869         .compat_ioctl   = kvm_dev_ioctl,
2870 };
2871
2872 static struct miscdevice kvm_dev = {
2873         KVM_MINOR,
2874         "kvm",
2875         &kvm_chardev_ops,
2876 };
2877
2878 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2879                        void *v)
2880 {
2881         if (val == SYS_RESTART) {
2882                 /*
2883                  * Some (well, at least mine) BIOSes hang on reboot if
2884                  * in vmx root mode.
2885                  */
2886                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2887                 on_each_cpu(hardware_disable, NULL, 0, 1);
2888         }
2889         return NOTIFY_OK;
2890 }
2891
2892 static struct notifier_block kvm_reboot_notifier = {
2893         .notifier_call = kvm_reboot,
2894         .priority = 0,
2895 };
2896
2897 /*
2898  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2899  * cached on it.
2900  */
2901 static void decache_vcpus_on_cpu(int cpu)
2902 {
2903         struct kvm *vm;
2904         struct kvm_vcpu *vcpu;
2905         int i;
2906
2907         spin_lock(&kvm_lock);
2908         list_for_each_entry(vm, &vm_list, vm_list)
2909                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2910                         vcpu = &vm->vcpus[i];
2911                         /*
2912                          * If the vcpu is locked, then it is running on some
2913                          * other cpu and therefore it is not cached on the
2914                          * cpu in question.
2915                          *
2916                          * If it's not locked, check the last cpu it executed
2917                          * on.
2918                          */
2919                         if (mutex_trylock(&vcpu->mutex)) {
2920                                 if (vcpu->cpu == cpu) {
2921                                         kvm_arch_ops->vcpu_decache(vcpu);
2922                                         vcpu->cpu = -1;
2923                                 }
2924                                 mutex_unlock(&vcpu->mutex);
2925                         }
2926                 }
2927         spin_unlock(&kvm_lock);
2928 }
2929
2930 static void hardware_enable(void *junk)
2931 {
2932         int cpu = raw_smp_processor_id();
2933
2934         if (cpu_isset(cpu, cpus_hardware_enabled))
2935                 return;
2936         cpu_set(cpu, cpus_hardware_enabled);
2937         kvm_arch_ops->hardware_enable(NULL);
2938 }
2939
2940 static void hardware_disable(void *junk)
2941 {
2942         int cpu = raw_smp_processor_id();
2943
2944         if (!cpu_isset(cpu, cpus_hardware_enabled))
2945                 return;
2946         cpu_clear(cpu, cpus_hardware_enabled);
2947         decache_vcpus_on_cpu(cpu);
2948         kvm_arch_ops->hardware_disable(NULL);
2949 }
2950
2951 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2952                            void *v)
2953 {
2954         int cpu = (long)v;
2955
2956         switch (val) {
2957         case CPU_DYING:
2958         case CPU_DYING_FROZEN:
2959         case CPU_UP_CANCELED:
2960         case CPU_UP_CANCELED_FROZEN:
2961                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2962                        cpu);
2963                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2964                 break;
2965         case CPU_ONLINE:
2966         case CPU_ONLINE_FROZEN:
2967                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2968                        cpu);
2969                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2970                 break;
2971         }
2972         return NOTIFY_OK;
2973 }
2974
2975 void kvm_io_bus_init(struct kvm_io_bus *bus)
2976 {
2977         memset(bus, 0, sizeof(*bus));
2978 }
2979
2980 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2981 {
2982         int i;
2983
2984         for (i = 0; i < bus->dev_count; i++) {
2985                 struct kvm_io_device *pos = bus->devs[i];
2986
2987                 kvm_iodevice_destructor(pos);
2988         }
2989 }
2990
2991 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2992 {
2993         int i;
2994
2995         for (i = 0; i < bus->dev_count; i++) {
2996                 struct kvm_io_device *pos = bus->devs[i];
2997
2998                 if (pos->in_range(pos, addr))
2999                         return pos;
3000         }
3001
3002         return NULL;
3003 }
3004
3005 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3006 {
3007         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3008
3009         bus->devs[bus->dev_count++] = dev;
3010 }
3011
3012 static struct notifier_block kvm_cpu_notifier = {
3013         .notifier_call = kvm_cpu_hotplug,
3014         .priority = 20, /* must be > scheduler priority */
3015 };
3016
3017 static u64 stat_get(void *_offset)
3018 {
3019         unsigned offset = (long)_offset;
3020         u64 total = 0;
3021         struct kvm *kvm;
3022         struct kvm_vcpu *vcpu;
3023         int i;
3024
3025         spin_lock(&kvm_lock);
3026         list_for_each_entry(kvm, &vm_list, vm_list)
3027                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3028                         vcpu = &kvm->vcpus[i];
3029                         total += *(u32 *)((void *)vcpu + offset);
3030                 }
3031         spin_unlock(&kvm_lock);
3032         return total;
3033 }
3034
3035 static void stat_set(void *offset, u64 val)
3036 {
3037 }
3038
3039 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3040
3041 static __init void kvm_init_debug(void)
3042 {
3043         struct kvm_stats_debugfs_item *p;
3044
3045         debugfs_dir = debugfs_create_dir("kvm", NULL);
3046         for (p = debugfs_entries; p->name; ++p)
3047                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3048                                                 (void *)(long)p->offset,
3049                                                 &stat_fops);
3050 }
3051
3052 static void kvm_exit_debug(void)
3053 {
3054         struct kvm_stats_debugfs_item *p;
3055
3056         for (p = debugfs_entries; p->name; ++p)
3057                 debugfs_remove(p->dentry);
3058         debugfs_remove(debugfs_dir);
3059 }
3060
3061 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3062 {
3063         hardware_disable(NULL);
3064         return 0;
3065 }
3066
3067 static int kvm_resume(struct sys_device *dev)
3068 {
3069         hardware_enable(NULL);
3070         return 0;
3071 }
3072
3073 static struct sysdev_class kvm_sysdev_class = {
3074         set_kset_name("kvm"),
3075         .suspend = kvm_suspend,
3076         .resume = kvm_resume,
3077 };
3078
3079 static struct sys_device kvm_sysdev = {
3080         .id = 0,
3081         .cls = &kvm_sysdev_class,
3082 };
3083
3084 hpa_t bad_page_address;
3085
3086 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3087 {
3088         int r;
3089
3090         if (kvm_arch_ops) {
3091                 printk(KERN_ERR "kvm: already loaded the other module\n");
3092                 return -EEXIST;
3093         }
3094
3095         if (!ops->cpu_has_kvm_support()) {
3096                 printk(KERN_ERR "kvm: no hardware support\n");
3097                 return -EOPNOTSUPP;
3098         }
3099         if (ops->disabled_by_bios()) {
3100                 printk(KERN_ERR "kvm: disabled by bios\n");
3101                 return -EOPNOTSUPP;
3102         }
3103
3104         kvm_arch_ops = ops;
3105
3106         r = kvm_arch_ops->hardware_setup();
3107         if (r < 0)
3108                 goto out;
3109
3110         on_each_cpu(hardware_enable, NULL, 0, 1);
3111         r = register_cpu_notifier(&kvm_cpu_notifier);
3112         if (r)
3113                 goto out_free_1;
3114         register_reboot_notifier(&kvm_reboot_notifier);
3115
3116         r = sysdev_class_register(&kvm_sysdev_class);
3117         if (r)
3118                 goto out_free_2;
3119
3120         r = sysdev_register(&kvm_sysdev);
3121         if (r)
3122                 goto out_free_3;
3123
3124         kvm_chardev_ops.owner = module;
3125
3126         r = misc_register(&kvm_dev);
3127         if (r) {
3128                 printk (KERN_ERR "kvm: misc device register failed\n");
3129                 goto out_free;
3130         }
3131
3132         return r;
3133
3134 out_free:
3135         sysdev_unregister(&kvm_sysdev);
3136 out_free_3:
3137         sysdev_class_unregister(&kvm_sysdev_class);
3138 out_free_2:
3139         unregister_reboot_notifier(&kvm_reboot_notifier);
3140         unregister_cpu_notifier(&kvm_cpu_notifier);
3141 out_free_1:
3142         on_each_cpu(hardware_disable, NULL, 0, 1);
3143         kvm_arch_ops->hardware_unsetup();
3144 out:
3145         kvm_arch_ops = NULL;
3146         return r;
3147 }
3148
3149 void kvm_exit_arch(void)
3150 {
3151         misc_deregister(&kvm_dev);
3152         sysdev_unregister(&kvm_sysdev);
3153         sysdev_class_unregister(&kvm_sysdev_class);
3154         unregister_reboot_notifier(&kvm_reboot_notifier);
3155         unregister_cpu_notifier(&kvm_cpu_notifier);
3156         on_each_cpu(hardware_disable, NULL, 0, 1);
3157         kvm_arch_ops->hardware_unsetup();
3158         kvm_arch_ops = NULL;
3159 }
3160
3161 static __init int kvm_init(void)
3162 {
3163         static struct page *bad_page;
3164         int r;
3165
3166         r = kvm_mmu_module_init();
3167         if (r)
3168                 goto out4;
3169
3170         kvm_init_debug();
3171
3172         kvm_init_msr_list();
3173
3174         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3175                 r = -ENOMEM;
3176                 goto out;
3177         }
3178
3179         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3180         memset(__va(bad_page_address), 0, PAGE_SIZE);
3181
3182         return 0;
3183
3184 out:
3185         kvm_exit_debug();
3186         kvm_mmu_module_exit();
3187 out4:
3188         return r;
3189 }
3190
3191 static __exit void kvm_exit(void)
3192 {
3193         kvm_exit_debug();
3194         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3195         kvm_mmu_module_exit();
3196 }
3197
3198 module_init(kvm_init)
3199 module_exit(kvm_exit)
3200
3201 EXPORT_SYMBOL_GPL(kvm_init_arch);
3202 EXPORT_SYMBOL_GPL(kvm_exit_arch);