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 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41
  42 #include <asm/processor.h>
  43 #include <asm/msr.h>
  44 #include <asm/io.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/desc.h>
  47
  48 MODULE_AUTHOR("Qumranet");
  49 MODULE_LICENSE("GPL");
  50
  51 static DEFINE_SPINLOCK(kvm_lock);
  52 static LIST_HEAD(vm_list);
  53
  54 static cpumask_t cpus_hardware_enabled;
  55
  56 struct kvm_arch_ops *kvm_arch_ops;
  57 struct kmem_cache *kvm_vcpu_cache;
  58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  59
  60 static __read_mostly struct preempt_ops kvm_preempt_ops;
  61
  62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  63
  64 static struct kvm_stats_debugfs_item {
  65         const char *name;
  66         int offset;
  67         struct dentry *dentry;
  68 } debugfs_entries[] = {
  69         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  70         { "pf_guest", STAT_OFFSET(pf_guest) },
  71         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  72         { "invlpg", STAT_OFFSET(invlpg) },
  73         { "exits", STAT_OFFSET(exits) },
  74         { "io_exits", STAT_OFFSET(io_exits) },
  75         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  76         { "signal_exits", STAT_OFFSET(signal_exits) },
  77         { "irq_window", STAT_OFFSET(irq_window_exits) },
  78         { "halt_exits", STAT_OFFSET(halt_exits) },
  79         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  80         { "request_irq", STAT_OFFSET(request_irq_exits) },
  81         { "irq_exits", STAT_OFFSET(irq_exits) },
  82         { "light_exits", STAT_OFFSET(light_exits) },
  83         { "efer_reload", STAT_OFFSET(efer_reload) },
  84         { NULL }
  85 };
  86
  87 static struct dentry *debugfs_dir;
  88
  89 #define MAX_IO_MSRS 256
  90
  91 #define CR0_RESERVED_BITS                                               \
  92         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  93                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  94                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  95 #define CR4_RESERVED_BITS                                               \
  96         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  97                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  98                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  99                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 100
 101 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 102 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 103
 104 #ifdef CONFIG_X86_64
 105 // LDT or TSS descriptor in the GDT. 16 bytes.
 106 struct segment_descriptor_64 {
 107         struct segment_descriptor s;
 108         u32 base_higher;
 109         u32 pad_zero;
 110 };
 111
 112 #endif
 113
 114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 115                            unsigned long arg);
 116
 117 unsigned long segment_base(u16 selector)
 118 {
 119         struct descriptor_table gdt;
 120         struct segment_descriptor *d;
 121         unsigned long table_base;
 122         typedef unsigned long ul;
 123         unsigned long v;
 124
 125         if (selector == 0)
 126                 return 0;
 127
 128         asm ("sgdt %0" : "=m"(gdt));
 129         table_base = gdt.base;
 130
 131         if (selector & 4) {           /* from ldt */
 132                 u16 ldt_selector;
 133
 134                 asm ("sldt %0" : "=g"(ldt_selector));
 135                 table_base = segment_base(ldt_selector);
 136         }
 137         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 138         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 139 #ifdef CONFIG_X86_64
 140         if (d->system == 0
 141             && (d->type == 2 || d->type == 9 || d->type == 11))
 142                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 143 #endif
 144         return v;
 145 }
 146 EXPORT_SYMBOL_GPL(segment_base);
 147
 148 static inline int valid_vcpu(int n)
 149 {
 150         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 151 }
 152
 153 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 154 {
 155         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 156                 return;
 157
 158         vcpu->guest_fpu_loaded = 1;
 159         fx_save(&vcpu->host_fx_image);
 160         fx_restore(&vcpu->guest_fx_image);
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 163
 164 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 165 {
 166         if (!vcpu->guest_fpu_loaded)
 167                 return;
 168
 169         vcpu->guest_fpu_loaded = 0;
 170         fx_save(&vcpu->guest_fx_image);
 171         fx_restore(&vcpu->host_fx_image);
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 174
 175 /*
 176  * Switches to specified vcpu, until a matching vcpu_put()
 177  */
 178 static void vcpu_load(struct kvm_vcpu *vcpu)
 179 {
 180         int cpu;
 181
 182         mutex_lock(&vcpu->mutex);
 183         cpu = get_cpu();
 184         preempt_notifier_register(&vcpu->preempt_notifier);
 185         kvm_arch_ops->vcpu_load(vcpu, cpu);
 186         put_cpu();
 187 }
 188
 189 static void vcpu_put(struct kvm_vcpu *vcpu)
 190 {
 191         preempt_disable();
 192         kvm_arch_ops->vcpu_put(vcpu);
 193         preempt_notifier_unregister(&vcpu->preempt_notifier);
 194         preempt_enable();
 195         mutex_unlock(&vcpu->mutex);
 196 }
 197
 198 static void ack_flush(void *_completed)
 199 {
 200         atomic_t *completed = _completed;
 201
 202         atomic_inc(completed);
 203 }
 204
 205 void kvm_flush_remote_tlbs(struct kvm *kvm)
 206 {
 207         int i, cpu, needed;
 208         cpumask_t cpus;
 209         struct kvm_vcpu *vcpu;
 210         atomic_t completed;
 211
 212         atomic_set(&completed, 0);
 213         cpus_clear(cpus);
 214         needed = 0;
 215         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 216                 vcpu = kvm->vcpus[i];
 217                 if (!vcpu)
 218                         continue;
 219                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 220                         continue;
 221                 cpu = vcpu->cpu;
 222                 if (cpu != -1 && cpu != raw_smp_processor_id())
 223                         if (!cpu_isset(cpu, cpus)) {
 224                                 cpu_set(cpu, cpus);
 225                                 ++needed;
 226                         }
 227         }
 228
 229         /*
 230          * We really want smp_call_function_mask() here.  But that's not
 231          * available, so ipi all cpus in parallel and wait for them
 232          * to complete.
 233          */
 234         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 235                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 236         while (atomic_read(&completed) != needed) {
 237                 cpu_relax();
 238                 barrier();
 239         }
 240 }
 241
 242 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 243 {
 244         struct page *page;
 245         int r;
 246
 247         mutex_init(&vcpu->mutex);
 248         vcpu->cpu = -1;
 249         vcpu->mmu.root_hpa = INVALID_PAGE;
 250         vcpu->kvm = kvm;
 251         vcpu->vcpu_id = id;
 252         init_waitqueue_head(&vcpu->wq);
 253
 254         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 255         if (!page) {
 256                 r = -ENOMEM;
 257                 goto fail;
 258         }
 259         vcpu->run = page_address(page);
 260
 261         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 262         if (!page) {
 263                 r = -ENOMEM;
 264                 goto fail_free_run;
 265         }
 266         vcpu->pio_data = page_address(page);
 267
 268         r = kvm_mmu_create(vcpu);
 269         if (r < 0)
 270                 goto fail_free_pio_data;
 271
 272         return 0;
 273
 274 fail_free_pio_data:
 275         free_page((unsigned long)vcpu->pio_data);
 276 fail_free_run:
 277         free_page((unsigned long)vcpu->run);
 278 fail:
 279         return -ENOMEM;
 280 }
 281 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 282
 283 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 284 {
 285         kvm_mmu_destroy(vcpu);
 286         if (vcpu->apic)
 287                 hrtimer_cancel(&vcpu->apic->timer.dev);
 288         kvm_free_apic(vcpu->apic);
 289         free_page((unsigned long)vcpu->pio_data);
 290         free_page((unsigned long)vcpu->run);
 291 }
 292 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 293
 294 static struct kvm *kvm_create_vm(void)
 295 {
 296         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 297
 298         if (!kvm)
 299                 return ERR_PTR(-ENOMEM);
 300
 301         kvm_io_bus_init(&kvm->pio_bus);
 302         mutex_init(&kvm->lock);
 303         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 304         kvm_io_bus_init(&kvm->mmio_bus);
 305         spin_lock(&kvm_lock);
 306         list_add(&kvm->vm_list, &vm_list);
 307         spin_unlock(&kvm_lock);
 308         return kvm;
 309 }
 310
 311 /*
 312  * Free any memory in @free but not in @dont.
 313  */
 314 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 315                                   struct kvm_memory_slot *dont)
 316 {
 317         int i;
 318
 319         if (!dont || free->phys_mem != dont->phys_mem)
 320                 if (free->phys_mem) {
 321                         for (i = 0; i < free->npages; ++i)
 322                                 if (free->phys_mem[i])
 323                                         __free_page(free->phys_mem[i]);
 324                         vfree(free->phys_mem);
 325                 }
 326
 327         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 328                 vfree(free->dirty_bitmap);
 329
 330         free->phys_mem = NULL;
 331         free->npages = 0;
 332         free->dirty_bitmap = NULL;
 333 }
 334
 335 static void kvm_free_physmem(struct kvm *kvm)
 336 {
 337         int i;
 338
 339         for (i = 0; i < kvm->nmemslots; ++i)
 340                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 341 }
 342
 343 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 344 {
 345         int i;
 346
 347         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 348                 if (vcpu->pio.guest_pages[i]) {
 349                         __free_page(vcpu->pio.guest_pages[i]);
 350                         vcpu->pio.guest_pages[i] = NULL;
 351                 }
 352 }
 353
 354 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 355 {
 356         vcpu_load(vcpu);
 357         kvm_mmu_unload(vcpu);
 358         vcpu_put(vcpu);
 359 }
 360
 361 static void kvm_free_vcpus(struct kvm *kvm)
 362 {
 363         unsigned int i;
 364
 365         /*
 366          * Unpin any mmu pages first.
 367          */
 368         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 369                 if (kvm->vcpus[i])
 370                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 371         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 372                 if (kvm->vcpus[i]) {
 373                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 374                         kvm->vcpus[i] = NULL;
 375                 }
 376         }
 377
 378 }
 379
 380 static void kvm_destroy_vm(struct kvm *kvm)
 381 {
 382         spin_lock(&kvm_lock);
 383         list_del(&kvm->vm_list);
 384         spin_unlock(&kvm_lock);
 385         kvm_io_bus_destroy(&kvm->pio_bus);
 386         kvm_io_bus_destroy(&kvm->mmio_bus);
 387         kfree(kvm->vpic);
 388         kfree(kvm->vioapic);
 389         kvm_free_vcpus(kvm);
 390         kvm_free_physmem(kvm);
 391         kfree(kvm);
 392 }
 393
 394 static int kvm_vm_release(struct inode *inode, struct file *filp)
 395 {
 396         struct kvm *kvm = filp->private_data;
 397
 398         kvm_destroy_vm(kvm);
 399         return 0;
 400 }
 401
 402 static void inject_gp(struct kvm_vcpu *vcpu)
 403 {
 404         kvm_arch_ops->inject_gp(vcpu, 0);
 405 }
 406
 407 /*
 408  * Load the pae pdptrs.  Return true is they are all valid.
 409  */
 410 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 411 {
 412         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 413         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 414         int i;
 415         u64 *pdpt;
 416         int ret;
 417         struct page *page;
 418         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 419
 420         mutex_lock(&vcpu->kvm->lock);
 421         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 422         if (!page) {
 423                 ret = 0;
 424                 goto out;
 425         }
 426
 427         pdpt = kmap_atomic(page, KM_USER0);
 428         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 429         kunmap_atomic(pdpt, KM_USER0);
 430
 431         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 432                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 433                         ret = 0;
 434                         goto out;
 435                 }
 436         }
 437         ret = 1;
 438
 439         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 440 out:
 441         mutex_unlock(&vcpu->kvm->lock);
 442
 443         return ret;
 444 }
 445
 446 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 447 {
 448         if (cr0 & CR0_RESERVED_BITS) {
 449                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 450                        cr0, vcpu->cr0);
 451                 inject_gp(vcpu);
 452                 return;
 453         }
 454
 455         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 456                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 457                 inject_gp(vcpu);
 458                 return;
 459         }
 460
 461         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 462                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 463                        "and a clear PE flag\n");
 464                 inject_gp(vcpu);
 465                 return;
 466         }
 467
 468         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 469 #ifdef CONFIG_X86_64
 470                 if ((vcpu->shadow_efer & EFER_LME)) {
 471                         int cs_db, cs_l;
 472
 473                         if (!is_pae(vcpu)) {
 474                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 475                                        "in long mode while PAE is disabled\n");
 476                                 inject_gp(vcpu);
 477                                 return;
 478                         }
 479                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 480                         if (cs_l) {
 481                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 482                                        "in long mode while CS.L == 1\n");
 483                                 inject_gp(vcpu);
 484                                 return;
 485
 486                         }
 487                 } else
 488 #endif
 489                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 490                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 491                                "reserved bits\n");
 492                         inject_gp(vcpu);
 493                         return;
 494                 }
 495
 496         }
 497
 498         kvm_arch_ops->set_cr0(vcpu, cr0);
 499         vcpu->cr0 = cr0;
 500
 501         mutex_lock(&vcpu->kvm->lock);
 502         kvm_mmu_reset_context(vcpu);
 503         mutex_unlock(&vcpu->kvm->lock);
 504         return;
 505 }
 506 EXPORT_SYMBOL_GPL(set_cr0);
 507
 508 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 509 {
 510         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 511 }
 512 EXPORT_SYMBOL_GPL(lmsw);
 513
 514 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 515 {
 516         if (cr4 & CR4_RESERVED_BITS) {
 517                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 518                 inject_gp(vcpu);
 519                 return;
 520         }
 521
 522         if (is_long_mode(vcpu)) {
 523                 if (!(cr4 & X86_CR4_PAE)) {
 524                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 525                                "in long mode\n");
 526                         inject_gp(vcpu);
 527                         return;
 528                 }
 529         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 530                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 531                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 532                 inject_gp(vcpu);
 533                 return;
 534         }
 535
 536         if (cr4 & X86_CR4_VMXE) {
 537                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 538                 inject_gp(vcpu);
 539                 return;
 540         }
 541         kvm_arch_ops->set_cr4(vcpu, cr4);
 542         mutex_lock(&vcpu->kvm->lock);
 543         kvm_mmu_reset_context(vcpu);
 544         mutex_unlock(&vcpu->kvm->lock);
 545 }
 546 EXPORT_SYMBOL_GPL(set_cr4);
 547
 548 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 549 {
 550         if (is_long_mode(vcpu)) {
 551                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 552                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 553                         inject_gp(vcpu);
 554                         return;
 555                 }
 556         } else {
 557                 if (is_pae(vcpu)) {
 558                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 559                                 printk(KERN_DEBUG
 560                                        "set_cr3: #GP, reserved bits\n");
 561                                 inject_gp(vcpu);
 562                                 return;
 563                         }
 564                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 565                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 566                                        "reserved bits\n");
 567                                 inject_gp(vcpu);
 568                                 return;
 569                         }
 570                 } else {
 571                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 572                                 printk(KERN_DEBUG
 573                                        "set_cr3: #GP, reserved bits\n");
 574                                 inject_gp(vcpu);
 575                                 return;
 576                         }
 577                 }
 578         }
 579
 580         mutex_lock(&vcpu->kvm->lock);
 581         /*
 582          * Does the new cr3 value map to physical memory? (Note, we
 583          * catch an invalid cr3 even in real-mode, because it would
 584          * cause trouble later on when we turn on paging anyway.)
 585          *
 586          * A real CPU would silently accept an invalid cr3 and would
 587          * attempt to use it - with largely undefined (and often hard
 588          * to debug) behavior on the guest side.
 589          */
 590         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 591                 inject_gp(vcpu);
 592         else {
 593                 vcpu->cr3 = cr3;
 594                 vcpu->mmu.new_cr3(vcpu);
 595         }
 596         mutex_unlock(&vcpu->kvm->lock);
 597 }
 598 EXPORT_SYMBOL_GPL(set_cr3);
 599
 600 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 601 {
 602         if (cr8 & CR8_RESERVED_BITS) {
 603                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 604                 inject_gp(vcpu);
 605                 return;
 606         }
 607         if (irqchip_in_kernel(vcpu->kvm))
 608                 kvm_lapic_set_tpr(vcpu, cr8);
 609         else
 610                 vcpu->cr8 = cr8;
 611 }
 612 EXPORT_SYMBOL_GPL(set_cr8);
 613
 614 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 615 {
 616         if (irqchip_in_kernel(vcpu->kvm))
 617                 return kvm_lapic_get_cr8(vcpu);
 618         else
 619                 return vcpu->cr8;
 620 }
 621 EXPORT_SYMBOL_GPL(get_cr8);
 622
 623 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 624 {
 625         if (irqchip_in_kernel(vcpu->kvm))
 626                 return vcpu->apic_base;
 627         else
 628                 return vcpu->apic_base;
 629 }
 630 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 631
 632 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 633 {
 634         /* TODO: reserve bits check */
 635         if (irqchip_in_kernel(vcpu->kvm))
 636                 kvm_lapic_set_base(vcpu, data);
 637         else
 638                 vcpu->apic_base = data;
 639 }
 640 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 641
 642 void fx_init(struct kvm_vcpu *vcpu)
 643 {
 644         unsigned after_mxcsr_mask;
 645
 646         /* Initialize guest FPU by resetting ours and saving into guest's */
 647         preempt_disable();
 648         fx_save(&vcpu->host_fx_image);
 649         fpu_init();
 650         fx_save(&vcpu->guest_fx_image);
 651         fx_restore(&vcpu->host_fx_image);
 652         preempt_enable();
 653
 654         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 655         vcpu->guest_fx_image.mxcsr = 0x1f80;
 656         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 657                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 658 }
 659 EXPORT_SYMBOL_GPL(fx_init);
 660
 661 /*
 662  * Allocate some memory and give it an address in the guest physical address
 663  * space.
 664  *
 665  * Discontiguous memory is allowed, mostly for framebuffers.
 666  */
 667 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 668                                           struct kvm_memory_region *mem)
 669 {
 670         int r;
 671         gfn_t base_gfn;
 672         unsigned long npages;
 673         unsigned long i;
 674         struct kvm_memory_slot *memslot;
 675         struct kvm_memory_slot old, new;
 676         int memory_config_version;
 677
 678         r = -EINVAL;
 679         /* General sanity checks */
 680         if (mem->memory_size & (PAGE_SIZE - 1))
 681                 goto out;
 682         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 683                 goto out;
 684         if (mem->slot >= KVM_MEMORY_SLOTS)
 685                 goto out;
 686         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 687                 goto out;
 688
 689         memslot = &kvm->memslots[mem->slot];
 690         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 691         npages = mem->memory_size >> PAGE_SHIFT;
 692
 693         if (!npages)
 694                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 695
 696 raced:
 697         mutex_lock(&kvm->lock);
 698
 699         memory_config_version = kvm->memory_config_version;
 700         new = old = *memslot;
 701
 702         new.base_gfn = base_gfn;
 703         new.npages = npages;
 704         new.flags = mem->flags;
 705
 706         /* Disallow changing a memory slot's size. */
 707         r = -EINVAL;
 708         if (npages && old.npages && npages != old.npages)
 709                 goto out_unlock;
 710
 711         /* Check for overlaps */
 712         r = -EEXIST;
 713         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 714                 struct kvm_memory_slot *s = &kvm->memslots[i];
 715
 716                 if (s == memslot)
 717                         continue;
 718                 if (!((base_gfn + npages <= s->base_gfn) ||
 719                       (base_gfn >= s->base_gfn + s->npages)))
 720                         goto out_unlock;
 721         }
 722         /*
 723          * Do memory allocations outside lock.  memory_config_version will
 724          * detect any races.
 725          */
 726         mutex_unlock(&kvm->lock);
 727
 728         /* Deallocate if slot is being removed */
 729         if (!npages)
 730                 new.phys_mem = NULL;
 731
 732         /* Free page dirty bitmap if unneeded */
 733         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 734                 new.dirty_bitmap = NULL;
 735
 736         r = -ENOMEM;
 737
 738         /* Allocate if a slot is being created */
 739         if (npages && !new.phys_mem) {
 740                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 741
 742                 if (!new.phys_mem)
 743                         goto out_free;
 744
 745                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 746                 for (i = 0; i < npages; ++i) {
 747                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 748                                                      | __GFP_ZERO);
 749                         if (!new.phys_mem[i])
 750                                 goto out_free;
 751                         set_page_private(new.phys_mem[i],0);
 752                 }
 753         }
 754
 755         /* Allocate page dirty bitmap if needed */
 756         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 757                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 758
 759                 new.dirty_bitmap = vmalloc(dirty_bytes);
 760                 if (!new.dirty_bitmap)
 761                         goto out_free;
 762                 memset(new.dirty_bitmap, 0, dirty_bytes);
 763         }
 764
 765         mutex_lock(&kvm->lock);
 766
 767         if (memory_config_version != kvm->memory_config_version) {
 768                 mutex_unlock(&kvm->lock);
 769                 kvm_free_physmem_slot(&new, &old);
 770                 goto raced;
 771         }
 772
 773         r = -EAGAIN;
 774         if (kvm->busy)
 775                 goto out_unlock;
 776
 777         if (mem->slot >= kvm->nmemslots)
 778                 kvm->nmemslots = mem->slot + 1;
 779
 780         *memslot = new;
 781         ++kvm->memory_config_version;
 782
 783         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 784         kvm_flush_remote_tlbs(kvm);
 785
 786         mutex_unlock(&kvm->lock);
 787
 788         kvm_free_physmem_slot(&old, &new);
 789         return 0;
 790
 791 out_unlock:
 792         mutex_unlock(&kvm->lock);
 793 out_free:
 794         kvm_free_physmem_slot(&new, &old);
 795 out:
 796         return r;
 797 }
 798
 799 /*
 800  * Get (and clear) the dirty memory log for a memory slot.
 801  */
 802 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 803                                       struct kvm_dirty_log *log)
 804 {
 805         struct kvm_memory_slot *memslot;
 806         int r, i;
 807         int n;
 808         unsigned long any = 0;
 809
 810         mutex_lock(&kvm->lock);
 811
 812         /*
 813          * Prevent changes to guest memory configuration even while the lock
 814          * is not taken.
 815          */
 816         ++kvm->busy;
 817         mutex_unlock(&kvm->lock);
 818         r = -EINVAL;
 819         if (log->slot >= KVM_MEMORY_SLOTS)
 820                 goto out;
 821
 822         memslot = &kvm->memslots[log->slot];
 823         r = -ENOENT;
 824         if (!memslot->dirty_bitmap)
 825                 goto out;
 826
 827         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 828
 829         for (i = 0; !any && i < n/sizeof(long); ++i)
 830                 any = memslot->dirty_bitmap[i];
 831
 832         r = -EFAULT;
 833         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 834                 goto out;
 835
 836         /* If nothing is dirty, don't bother messing with page tables. */
 837         if (any) {
 838                 mutex_lock(&kvm->lock);
 839                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 840                 kvm_flush_remote_tlbs(kvm);
 841                 memset(memslot->dirty_bitmap, 0, n);
 842                 mutex_unlock(&kvm->lock);
 843         }
 844
 845         r = 0;
 846
 847 out:
 848         mutex_lock(&kvm->lock);
 849         --kvm->busy;
 850         mutex_unlock(&kvm->lock);
 851         return r;
 852 }
 853
 854 /*
 855  * Set a new alias region.  Aliases map a portion of physical memory into
 856  * another portion.  This is useful for memory windows, for example the PC
 857  * VGA region.
 858  */
 859 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 860                                          struct kvm_memory_alias *alias)
 861 {
 862         int r, n;
 863         struct kvm_mem_alias *p;
 864
 865         r = -EINVAL;
 866         /* General sanity checks */
 867         if (alias->memory_size & (PAGE_SIZE - 1))
 868                 goto out;
 869         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 870                 goto out;
 871         if (alias->slot >= KVM_ALIAS_SLOTS)
 872                 goto out;
 873         if (alias->guest_phys_addr + alias->memory_size
 874             < alias->guest_phys_addr)
 875                 goto out;
 876         if (alias->target_phys_addr + alias->memory_size
 877             < alias->target_phys_addr)
 878                 goto out;
 879
 880         mutex_lock(&kvm->lock);
 881
 882         p = &kvm->aliases[alias->slot];
 883         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 884         p->npages = alias->memory_size >> PAGE_SHIFT;
 885         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 886
 887         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 888                 if (kvm->aliases[n - 1].npages)
 889                         break;
 890         kvm->naliases = n;
 891
 892         kvm_mmu_zap_all(kvm);
 893
 894         mutex_unlock(&kvm->lock);
 895
 896         return 0;
 897
 898 out:
 899         return r;
 900 }
 901
 902 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 903 {
 904         int r;
 905
 906         r = 0;
 907         switch (chip->chip_id) {
 908         case KVM_IRQCHIP_PIC_MASTER:
 909                 memcpy (&chip->chip.pic,
 910                         &pic_irqchip(kvm)->pics[0],
 911                         sizeof(struct kvm_pic_state));
 912                 break;
 913         case KVM_IRQCHIP_PIC_SLAVE:
 914                 memcpy (&chip->chip.pic,
 915                         &pic_irqchip(kvm)->pics[1],
 916                         sizeof(struct kvm_pic_state));
 917                 break;
 918         case KVM_IRQCHIP_IOAPIC:
 919                 memcpy (&chip->chip.ioapic,
 920                         ioapic_irqchip(kvm),
 921                         sizeof(struct kvm_ioapic_state));
 922                 break;
 923         default:
 924                 r = -EINVAL;
 925                 break;
 926         }
 927         return r;
 928 }
 929
 930 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 931 {
 932         int r;
 933
 934         r = 0;
 935         switch (chip->chip_id) {
 936         case KVM_IRQCHIP_PIC_MASTER:
 937                 memcpy (&pic_irqchip(kvm)->pics[0],
 938                         &chip->chip.pic,
 939                         sizeof(struct kvm_pic_state));
 940                 break;
 941         case KVM_IRQCHIP_PIC_SLAVE:
 942                 memcpy (&pic_irqchip(kvm)->pics[1],
 943                         &chip->chip.pic,
 944                         sizeof(struct kvm_pic_state));
 945                 break;
 946         case KVM_IRQCHIP_IOAPIC:
 947                 memcpy (ioapic_irqchip(kvm),
 948                         &chip->chip.ioapic,
 949                         sizeof(struct kvm_ioapic_state));
 950                 break;
 951         default:
 952                 r = -EINVAL;
 953                 break;
 954         }
 955         kvm_pic_update_irq(pic_irqchip(kvm));
 956         return r;
 957 }
 958
 959 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 960 {
 961         int i;
 962         struct kvm_mem_alias *alias;
 963
 964         for (i = 0; i < kvm->naliases; ++i) {
 965                 alias = &kvm->aliases[i];
 966                 if (gfn >= alias->base_gfn
 967                     && gfn < alias->base_gfn + alias->npages)
 968                         return alias->target_gfn + gfn - alias->base_gfn;
 969         }
 970         return gfn;
 971 }
 972
 973 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 974 {
 975         int i;
 976
 977         for (i = 0; i < kvm->nmemslots; ++i) {
 978                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 979
 980                 if (gfn >= memslot->base_gfn
 981                     && gfn < memslot->base_gfn + memslot->npages)
 982                         return memslot;
 983         }
 984         return NULL;
 985 }
 986
 987 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 988 {
 989         gfn = unalias_gfn(kvm, gfn);
 990         return __gfn_to_memslot(kvm, gfn);
 991 }
 992
 993 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 994 {
 995         struct kvm_memory_slot *slot;
 996
 997         gfn = unalias_gfn(kvm, gfn);
 998         slot = __gfn_to_memslot(kvm, gfn);
 999         if (!slot)
1000                 return NULL;
1001         return slot->phys_mem[gfn - slot->base_gfn];
1002 }
1003 EXPORT_SYMBOL_GPL(gfn_to_page);
1004
1005 /* WARNING: Does not work on aliased pages. */
1006 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1007 {
1008         struct kvm_memory_slot *memslot;
1009
1010         memslot = __gfn_to_memslot(kvm, gfn);
1011         if (memslot && memslot->dirty_bitmap) {
1012                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1013
1014                 /* avoid RMW */
1015                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1016                         set_bit(rel_gfn, memslot->dirty_bitmap);
1017         }
1018 }
1019
1020 int emulator_read_std(unsigned long addr,
1021                              void *val,
1022                              unsigned int bytes,
1023                              struct kvm_vcpu *vcpu)
1024 {
1025         void *data = val;
1026
1027         while (bytes) {
1028                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1029                 unsigned offset = addr & (PAGE_SIZE-1);
1030                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1031                 unsigned long pfn;
1032                 struct page *page;
1033                 void *page_virt;
1034
1035                 if (gpa == UNMAPPED_GVA)
1036                         return X86EMUL_PROPAGATE_FAULT;
1037                 pfn = gpa >> PAGE_SHIFT;
1038                 page = gfn_to_page(vcpu->kvm, pfn);
1039                 if (!page)
1040                         return X86EMUL_UNHANDLEABLE;
1041                 page_virt = kmap_atomic(page, KM_USER0);
1042
1043                 memcpy(data, page_virt + offset, tocopy);
1044
1045                 kunmap_atomic(page_virt, KM_USER0);
1046
1047                 bytes -= tocopy;
1048                 data += tocopy;
1049                 addr += tocopy;
1050         }
1051
1052         return X86EMUL_CONTINUE;
1053 }
1054 EXPORT_SYMBOL_GPL(emulator_read_std);
1055
1056 static int emulator_write_std(unsigned long addr,
1057                               const void *val,
1058                               unsigned int bytes,
1059                               struct kvm_vcpu *vcpu)
1060 {
1061         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1062         return X86EMUL_UNHANDLEABLE;
1063 }
1064
1065 /*
1066  * Only apic need an MMIO device hook, so shortcut now..
1067  */
1068 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1069                                                 gpa_t addr)
1070 {
1071         struct kvm_io_device *dev;
1072
1073         if (vcpu->apic) {
1074                 dev = &vcpu->apic->dev;
1075                 if (dev->in_range(dev, addr))
1076                         return dev;
1077         }
1078         return NULL;
1079 }
1080
1081 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1082                                                 gpa_t addr)
1083 {
1084         struct kvm_io_device *dev;
1085
1086         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1087         if (dev == NULL)
1088                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1089         return dev;
1090 }
1091
1092 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1093                                                gpa_t addr)
1094 {
1095         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1096 }
1097
1098 static int emulator_read_emulated(unsigned long addr,
1099                                   void *val,
1100                                   unsigned int bytes,
1101                                   struct kvm_vcpu *vcpu)
1102 {
1103         struct kvm_io_device *mmio_dev;
1104         gpa_t                 gpa;
1105
1106         if (vcpu->mmio_read_completed) {
1107                 memcpy(val, vcpu->mmio_data, bytes);
1108                 vcpu->mmio_read_completed = 0;
1109                 return X86EMUL_CONTINUE;
1110         } else if (emulator_read_std(addr, val, bytes, vcpu)
1111                    == X86EMUL_CONTINUE)
1112                 return X86EMUL_CONTINUE;
1113
1114         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1115         if (gpa == UNMAPPED_GVA)
1116                 return X86EMUL_PROPAGATE_FAULT;
1117
1118         /*
1119          * Is this MMIO handled locally?
1120          */
1121         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1122         if (mmio_dev) {
1123                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1124                 return X86EMUL_CONTINUE;
1125         }
1126
1127         vcpu->mmio_needed = 1;
1128         vcpu->mmio_phys_addr = gpa;
1129         vcpu->mmio_size = bytes;
1130         vcpu->mmio_is_write = 0;
1131
1132         return X86EMUL_UNHANDLEABLE;
1133 }
1134
1135 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1136                                const void *val, int bytes)
1137 {
1138         struct page *page;
1139         void *virt;
1140
1141         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1142                 return 0;
1143         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1144         if (!page)
1145                 return 0;
1146         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1147         virt = kmap_atomic(page, KM_USER0);
1148         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1149         memcpy(virt + offset_in_page(gpa), val, bytes);
1150         kunmap_atomic(virt, KM_USER0);
1151         return 1;
1152 }
1153
1154 static int emulator_write_emulated_onepage(unsigned long addr,
1155                                            const void *val,
1156                                            unsigned int bytes,
1157                                            struct kvm_vcpu *vcpu)
1158 {
1159         struct kvm_io_device *mmio_dev;
1160         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1161
1162         if (gpa == UNMAPPED_GVA) {
1163                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1164                 return X86EMUL_PROPAGATE_FAULT;
1165         }
1166
1167         if (emulator_write_phys(vcpu, gpa, val, bytes))
1168                 return X86EMUL_CONTINUE;
1169
1170         /*
1171          * Is this MMIO handled locally?
1172          */
1173         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1174         if (mmio_dev) {
1175                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1176                 return X86EMUL_CONTINUE;
1177         }
1178
1179         vcpu->mmio_needed = 1;
1180         vcpu->mmio_phys_addr = gpa;
1181         vcpu->mmio_size = bytes;
1182         vcpu->mmio_is_write = 1;
1183         memcpy(vcpu->mmio_data, val, bytes);
1184
1185         return X86EMUL_CONTINUE;
1186 }
1187
1188 int emulator_write_emulated(unsigned long addr,
1189                                    const void *val,
1190                                    unsigned int bytes,
1191                                    struct kvm_vcpu *vcpu)
1192 {
1193         /* Crossing a page boundary? */
1194         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1195                 int rc, now;
1196
1197                 now = -addr & ~PAGE_MASK;
1198                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1199                 if (rc != X86EMUL_CONTINUE)
1200                         return rc;
1201                 addr += now;
1202                 val += now;
1203                 bytes -= now;
1204         }
1205         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1206 }
1207 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1208
1209 static int emulator_cmpxchg_emulated(unsigned long addr,
1210                                      const void *old,
1211                                      const void *new,
1212                                      unsigned int bytes,
1213                                      struct kvm_vcpu *vcpu)
1214 {
1215         static int reported;
1216
1217         if (!reported) {
1218                 reported = 1;
1219                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1220         }
1221         return emulator_write_emulated(addr, new, bytes, vcpu);
1222 }
1223
1224 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1225 {
1226         return kvm_arch_ops->get_segment_base(vcpu, seg);
1227 }
1228
1229 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1230 {
1231         return X86EMUL_CONTINUE;
1232 }
1233
1234 int emulate_clts(struct kvm_vcpu *vcpu)
1235 {
1236         unsigned long cr0;
1237
1238         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1239         kvm_arch_ops->set_cr0(vcpu, cr0);
1240         return X86EMUL_CONTINUE;
1241 }
1242
1243 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1244 {
1245         struct kvm_vcpu *vcpu = ctxt->vcpu;
1246
1247         switch (dr) {
1248         case 0 ... 3:
1249                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1250                 return X86EMUL_CONTINUE;
1251         default:
1252                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1253                 return X86EMUL_UNHANDLEABLE;
1254         }
1255 }
1256
1257 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1258 {
1259         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1260         int exception;
1261
1262         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1263         if (exception) {
1264                 /* FIXME: better handling */
1265                 return X86EMUL_UNHANDLEABLE;
1266         }
1267         return X86EMUL_CONTINUE;
1268 }
1269
1270 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1271 {
1272         static int reported;
1273         u8 opcodes[4];
1274         unsigned long rip = ctxt->vcpu->rip;
1275         unsigned long rip_linear;
1276
1277         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1278
1279         if (reported)
1280                 return;
1281
1282         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1283
1284         printk(KERN_ERR "emulation failed but !mmio_needed?"
1285                " rip %lx %02x %02x %02x %02x\n",
1286                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1287         reported = 1;
1288 }
1289
1290 struct x86_emulate_ops emulate_ops = {
1291         .read_std            = emulator_read_std,
1292         .write_std           = emulator_write_std,
1293         .read_emulated       = emulator_read_emulated,
1294         .write_emulated      = emulator_write_emulated,
1295         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1296 };
1297
1298 int emulate_instruction(struct kvm_vcpu *vcpu,
1299                         struct kvm_run *run,
1300                         unsigned long cr2,
1301                         u16 error_code)
1302 {
1303         struct x86_emulate_ctxt emulate_ctxt;
1304         int r;
1305         int cs_db, cs_l;
1306
1307         vcpu->mmio_fault_cr2 = cr2;
1308         kvm_arch_ops->cache_regs(vcpu);
1309
1310         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1311
1312         emulate_ctxt.vcpu = vcpu;
1313         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1314         emulate_ctxt.cr2 = cr2;
1315         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1316                 ? X86EMUL_MODE_REAL : cs_l
1317                 ? X86EMUL_MODE_PROT64 : cs_db
1318                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1319
1320         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1321                 emulate_ctxt.cs_base = 0;
1322                 emulate_ctxt.ds_base = 0;
1323                 emulate_ctxt.es_base = 0;
1324                 emulate_ctxt.ss_base = 0;
1325         } else {
1326                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1327                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1328                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1329                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1330         }
1331
1332         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1333         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1334
1335         vcpu->mmio_is_write = 0;
1336         vcpu->pio.string = 0;
1337         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1338         if (vcpu->pio.string)
1339                 return EMULATE_DO_MMIO;
1340
1341         if ((r || vcpu->mmio_is_write) && run) {
1342                 run->exit_reason = KVM_EXIT_MMIO;
1343                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1344                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1345                 run->mmio.len = vcpu->mmio_size;
1346                 run->mmio.is_write = vcpu->mmio_is_write;
1347         }
1348
1349         if (r) {
1350                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1351                         return EMULATE_DONE;
1352                 if (!vcpu->mmio_needed) {
1353                         report_emulation_failure(&emulate_ctxt);
1354                         return EMULATE_FAIL;
1355                 }
1356                 return EMULATE_DO_MMIO;
1357         }
1358
1359         kvm_arch_ops->decache_regs(vcpu);
1360         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1361
1362         if (vcpu->mmio_is_write) {
1363                 vcpu->mmio_needed = 0;
1364                 return EMULATE_DO_MMIO;
1365         }
1366
1367         return EMULATE_DONE;
1368 }
1369 EXPORT_SYMBOL_GPL(emulate_instruction);
1370
1371 /*
1372  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1373  */
1374 static void kvm_vcpu_kernel_halt(struct kvm_vcpu *vcpu)
1375 {
1376         DECLARE_WAITQUEUE(wait, current);
1377
1378         add_wait_queue(&vcpu->wq, &wait);
1379
1380         /*
1381          * We will block until either an interrupt or a signal wakes us up
1382          */
1383         while(!(irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu))
1384               && !vcpu->irq_summary
1385               && !signal_pending(current)) {
1386                 set_current_state(TASK_INTERRUPTIBLE);
1387                 vcpu_put(vcpu);
1388                 schedule();
1389                 vcpu_load(vcpu);
1390         }
1391
1392         remove_wait_queue(&vcpu->wq, &wait);
1393         set_current_state(TASK_RUNNING);
1394 }
1395
1396 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1397 {
1398         ++vcpu->stat.halt_exits;
1399         if (irqchip_in_kernel(vcpu->kvm)) {
1400                 kvm_vcpu_kernel_halt(vcpu);
1401                 return 1;
1402         } else {
1403                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1404                 return 0;
1405         }
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1408
1409 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1410 {
1411         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1412
1413         kvm_arch_ops->cache_regs(vcpu);
1414         ret = -KVM_EINVAL;
1415 #ifdef CONFIG_X86_64
1416         if (is_long_mode(vcpu)) {
1417                 nr = vcpu->regs[VCPU_REGS_RAX];
1418                 a0 = vcpu->regs[VCPU_REGS_RDI];
1419                 a1 = vcpu->regs[VCPU_REGS_RSI];
1420                 a2 = vcpu->regs[VCPU_REGS_RDX];
1421                 a3 = vcpu->regs[VCPU_REGS_RCX];
1422                 a4 = vcpu->regs[VCPU_REGS_R8];
1423                 a5 = vcpu->regs[VCPU_REGS_R9];
1424         } else
1425 #endif
1426         {
1427                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1428                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1429                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1430                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1431                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1432                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1433                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1434         }
1435         switch (nr) {
1436         default:
1437                 run->hypercall.nr = nr;
1438                 run->hypercall.args[0] = a0;
1439                 run->hypercall.args[1] = a1;
1440                 run->hypercall.args[2] = a2;
1441                 run->hypercall.args[3] = a3;
1442                 run->hypercall.args[4] = a4;
1443                 run->hypercall.args[5] = a5;
1444                 run->hypercall.ret = ret;
1445                 run->hypercall.longmode = is_long_mode(vcpu);
1446                 kvm_arch_ops->decache_regs(vcpu);
1447                 return 0;
1448         }
1449         vcpu->regs[VCPU_REGS_RAX] = ret;
1450         kvm_arch_ops->decache_regs(vcpu);
1451         return 1;
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_hypercall);
1454
1455 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1456 {
1457         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1458 }
1459
1460 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1461 {
1462         struct descriptor_table dt = { limit, base };
1463
1464         kvm_arch_ops->set_gdt(vcpu, &dt);
1465 }
1466
1467 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1468 {
1469         struct descriptor_table dt = { limit, base };
1470
1471         kvm_arch_ops->set_idt(vcpu, &dt);
1472 }
1473
1474 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1475                    unsigned long *rflags)
1476 {
1477         lmsw(vcpu, msw);
1478         *rflags = kvm_arch_ops->get_rflags(vcpu);
1479 }
1480
1481 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1482 {
1483         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1484         switch (cr) {
1485         case 0:
1486                 return vcpu->cr0;
1487         case 2:
1488                 return vcpu->cr2;
1489         case 3:
1490                 return vcpu->cr3;
1491         case 4:
1492                 return vcpu->cr4;
1493         default:
1494                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1495                 return 0;
1496         }
1497 }
1498
1499 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1500                      unsigned long *rflags)
1501 {
1502         switch (cr) {
1503         case 0:
1504                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1505                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1506                 break;
1507         case 2:
1508                 vcpu->cr2 = val;
1509                 break;
1510         case 3:
1511                 set_cr3(vcpu, val);
1512                 break;
1513         case 4:
1514                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1515                 break;
1516         default:
1517                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1518         }
1519 }
1520
1521 /*
1522  * Register the para guest with the host:
1523  */
1524 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1525 {
1526         struct kvm_vcpu_para_state *para_state;
1527         hpa_t para_state_hpa, hypercall_hpa;
1528         struct page *para_state_page;
1529         unsigned char *hypercall;
1530         gpa_t hypercall_gpa;
1531
1532         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1533         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1534
1535         /*
1536          * Needs to be page aligned:
1537          */
1538         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1539                 goto err_gp;
1540
1541         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1542         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1543         if (is_error_hpa(para_state_hpa))
1544                 goto err_gp;
1545
1546         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1547         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1548         para_state = kmap(para_state_page);
1549
1550         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1551         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1552
1553         para_state->host_version = KVM_PARA_API_VERSION;
1554         /*
1555          * We cannot support guests that try to register themselves
1556          * with a newer API version than the host supports:
1557          */
1558         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1559                 para_state->ret = -KVM_EINVAL;
1560                 goto err_kunmap_skip;
1561         }
1562
1563         hypercall_gpa = para_state->hypercall_gpa;
1564         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1565         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1566         if (is_error_hpa(hypercall_hpa)) {
1567                 para_state->ret = -KVM_EINVAL;
1568                 goto err_kunmap_skip;
1569         }
1570
1571         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1572         vcpu->para_state_page = para_state_page;
1573         vcpu->para_state_gpa = para_state_gpa;
1574         vcpu->hypercall_gpa = hypercall_gpa;
1575
1576         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1577         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1578                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1579         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1580         kunmap_atomic(hypercall, KM_USER1);
1581
1582         para_state->ret = 0;
1583 err_kunmap_skip:
1584         kunmap(para_state_page);
1585         return 0;
1586 err_gp:
1587         return 1;
1588 }
1589
1590 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1591 {
1592         u64 data;
1593
1594         switch (msr) {
1595         case 0xc0010010: /* SYSCFG */
1596         case 0xc0010015: /* HWCR */
1597         case MSR_IA32_PLATFORM_ID:
1598         case MSR_IA32_P5_MC_ADDR:
1599         case MSR_IA32_P5_MC_TYPE:
1600         case MSR_IA32_MC0_CTL:
1601         case MSR_IA32_MCG_STATUS:
1602         case MSR_IA32_MCG_CAP:
1603         case MSR_IA32_MC0_MISC:
1604         case MSR_IA32_MC0_MISC+4:
1605         case MSR_IA32_MC0_MISC+8:
1606         case MSR_IA32_MC0_MISC+12:
1607         case MSR_IA32_MC0_MISC+16:
1608         case MSR_IA32_UCODE_REV:
1609         case MSR_IA32_PERF_STATUS:
1610         case MSR_IA32_EBL_CR_POWERON:
1611                 /* MTRR registers */
1612         case 0xfe:
1613         case 0x200 ... 0x2ff:
1614                 data = 0;
1615                 break;
1616         case 0xcd: /* fsb frequency */
1617                 data = 3;
1618                 break;
1619         case MSR_IA32_APICBASE:
1620                 data = kvm_get_apic_base(vcpu);
1621                 break;
1622         case MSR_IA32_MISC_ENABLE:
1623                 data = vcpu->ia32_misc_enable_msr;
1624                 break;
1625 #ifdef CONFIG_X86_64
1626         case MSR_EFER:
1627                 data = vcpu->shadow_efer;
1628                 break;
1629 #endif
1630         default:
1631                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1632                 return 1;
1633         }
1634         *pdata = data;
1635         return 0;
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1638
1639 /*
1640  * Reads an msr value (of 'msr_index') into 'pdata'.
1641  * Returns 0 on success, non-0 otherwise.
1642  * Assumes vcpu_load() was already called.
1643  */
1644 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1645 {
1646         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1647 }
1648
1649 #ifdef CONFIG_X86_64
1650
1651 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1652 {
1653         if (efer & EFER_RESERVED_BITS) {
1654                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1655                        efer);
1656                 inject_gp(vcpu);
1657                 return;
1658         }
1659
1660         if (is_paging(vcpu)
1661             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1662                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1663                 inject_gp(vcpu);
1664                 return;
1665         }
1666
1667         kvm_arch_ops->set_efer(vcpu, efer);
1668
1669         efer &= ~EFER_LMA;
1670         efer |= vcpu->shadow_efer & EFER_LMA;
1671
1672         vcpu->shadow_efer = efer;
1673 }
1674
1675 #endif
1676
1677 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1678 {
1679         switch (msr) {
1680 #ifdef CONFIG_X86_64
1681         case MSR_EFER:
1682                 set_efer(vcpu, data);
1683                 break;
1684 #endif
1685         case MSR_IA32_MC0_STATUS:
1686                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1687                        __FUNCTION__, data);
1688                 break;
1689         case MSR_IA32_MCG_STATUS:
1690                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1691                         __FUNCTION__, data);
1692                 break;
1693         case MSR_IA32_UCODE_REV:
1694         case MSR_IA32_UCODE_WRITE:
1695         case 0x200 ... 0x2ff: /* MTRRs */
1696                 break;
1697         case MSR_IA32_APICBASE:
1698                 kvm_set_apic_base(vcpu, data);
1699                 break;
1700         case MSR_IA32_MISC_ENABLE:
1701                 vcpu->ia32_misc_enable_msr = data;
1702                 break;
1703         /*
1704          * This is the 'probe whether the host is KVM' logic:
1705          */
1706         case MSR_KVM_API_MAGIC:
1707                 return vcpu_register_para(vcpu, data);
1708
1709         default:
1710                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1711                 return 1;
1712         }
1713         return 0;
1714 }
1715 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1716
1717 /*
1718  * Writes msr value into into the appropriate "register".
1719  * Returns 0 on success, non-0 otherwise.
1720  * Assumes vcpu_load() was already called.
1721  */
1722 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1723 {
1724         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1725 }
1726
1727 void kvm_resched(struct kvm_vcpu *vcpu)
1728 {
1729         if (!need_resched())
1730                 return;
1731         cond_resched();
1732 }
1733 EXPORT_SYMBOL_GPL(kvm_resched);
1734
1735 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1736 {
1737         int i;
1738         u32 function;
1739         struct kvm_cpuid_entry *e, *best;
1740
1741         kvm_arch_ops->cache_regs(vcpu);
1742         function = vcpu->regs[VCPU_REGS_RAX];
1743         vcpu->regs[VCPU_REGS_RAX] = 0;
1744         vcpu->regs[VCPU_REGS_RBX] = 0;
1745         vcpu->regs[VCPU_REGS_RCX] = 0;
1746         vcpu->regs[VCPU_REGS_RDX] = 0;
1747         best = NULL;
1748         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1749                 e = &vcpu->cpuid_entries[i];
1750                 if (e->function == function) {
1751                         best = e;
1752                         break;
1753                 }
1754                 /*
1755                  * Both basic or both extended?
1756                  */
1757                 if (((e->function ^ function) & 0x80000000) == 0)
1758                         if (!best || e->function > best->function)
1759                                 best = e;
1760         }
1761         if (best) {
1762                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1763                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1764                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1765                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1766         }
1767         kvm_arch_ops->decache_regs(vcpu);
1768         kvm_arch_ops->skip_emulated_instruction(vcpu);
1769 }
1770 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1771
1772 static int pio_copy_data(struct kvm_vcpu *vcpu)
1773 {
1774         void *p = vcpu->pio_data;
1775         void *q;
1776         unsigned bytes;
1777         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1778
1779         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1780                  PAGE_KERNEL);
1781         if (!q) {
1782                 free_pio_guest_pages(vcpu);
1783                 return -ENOMEM;
1784         }
1785         q += vcpu->pio.guest_page_offset;
1786         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1787         if (vcpu->pio.in)
1788                 memcpy(q, p, bytes);
1789         else
1790                 memcpy(p, q, bytes);
1791         q -= vcpu->pio.guest_page_offset;
1792         vunmap(q);
1793         free_pio_guest_pages(vcpu);
1794         return 0;
1795 }
1796
1797 static int complete_pio(struct kvm_vcpu *vcpu)
1798 {
1799         struct kvm_pio_request *io = &vcpu->pio;
1800         long delta;
1801         int r;
1802
1803         kvm_arch_ops->cache_regs(vcpu);
1804
1805         if (!io->string) {
1806                 if (io->in)
1807                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1808                                io->size);
1809         } else {
1810                 if (io->in) {
1811                         r = pio_copy_data(vcpu);
1812                         if (r) {
1813                                 kvm_arch_ops->cache_regs(vcpu);
1814                                 return r;
1815                         }
1816                 }
1817
1818                 delta = 1;
1819                 if (io->rep) {
1820                         delta *= io->cur_count;
1821                         /*
1822                          * The size of the register should really depend on
1823                          * current address size.
1824                          */
1825                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1826                 }
1827                 if (io->down)
1828                         delta = -delta;
1829                 delta *= io->size;
1830                 if (io->in)
1831                         vcpu->regs[VCPU_REGS_RDI] += delta;
1832                 else
1833                         vcpu->regs[VCPU_REGS_RSI] += delta;
1834         }
1835
1836         kvm_arch_ops->decache_regs(vcpu);
1837
1838         io->count -= io->cur_count;
1839         io->cur_count = 0;
1840
1841         if (!io->count)
1842                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1843         return 0;
1844 }
1845
1846 static void kernel_pio(struct kvm_io_device *pio_dev,
1847                        struct kvm_vcpu *vcpu,
1848                        void *pd)
1849 {
1850         /* TODO: String I/O for in kernel device */
1851
1852         mutex_lock(&vcpu->kvm->lock);
1853         if (vcpu->pio.in)
1854                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1855                                   vcpu->pio.size,
1856                                   pd);
1857         else
1858                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1859                                    vcpu->pio.size,
1860                                    pd);
1861         mutex_unlock(&vcpu->kvm->lock);
1862 }
1863
1864 static void pio_string_write(struct kvm_io_device *pio_dev,
1865                              struct kvm_vcpu *vcpu)
1866 {
1867         struct kvm_pio_request *io = &vcpu->pio;
1868         void *pd = vcpu->pio_data;
1869         int i;
1870
1871         mutex_lock(&vcpu->kvm->lock);
1872         for (i = 0; i < io->cur_count; i++) {
1873                 kvm_iodevice_write(pio_dev, io->port,
1874                                    io->size,
1875                                    pd);
1876                 pd += io->size;
1877         }
1878         mutex_unlock(&vcpu->kvm->lock);
1879 }
1880
1881 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1882                   int size, unsigned port)
1883 {
1884         struct kvm_io_device *pio_dev;
1885
1886         vcpu->run->exit_reason = KVM_EXIT_IO;
1887         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1888         vcpu->run->io.size = vcpu->pio.size = size;
1889         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1890         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1891         vcpu->run->io.port = vcpu->pio.port = port;
1892         vcpu->pio.in = in;
1893         vcpu->pio.string = 0;
1894         vcpu->pio.down = 0;
1895         vcpu->pio.guest_page_offset = 0;
1896         vcpu->pio.rep = 0;
1897
1898         kvm_arch_ops->cache_regs(vcpu);
1899         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1900         kvm_arch_ops->decache_regs(vcpu);
1901
1902         pio_dev = vcpu_find_pio_dev(vcpu, port);
1903         if (pio_dev) {
1904                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1905                 complete_pio(vcpu);
1906                 return 1;
1907         }
1908         return 0;
1909 }
1910 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1911
1912 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1913                   int size, unsigned long count, int down,
1914                   gva_t address, int rep, unsigned port)
1915 {
1916         unsigned now, in_page;
1917         int i, ret = 0;
1918         int nr_pages = 1;
1919         struct page *page;
1920         struct kvm_io_device *pio_dev;
1921
1922         vcpu->run->exit_reason = KVM_EXIT_IO;
1923         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1924         vcpu->run->io.size = vcpu->pio.size = size;
1925         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1926         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1927         vcpu->run->io.port = vcpu->pio.port = port;
1928         vcpu->pio.in = in;
1929         vcpu->pio.string = 1;
1930         vcpu->pio.down = down;
1931         vcpu->pio.guest_page_offset = offset_in_page(address);
1932         vcpu->pio.rep = rep;
1933
1934         if (!count) {
1935                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1936                 return 1;
1937         }
1938
1939         if (!down)
1940                 in_page = PAGE_SIZE - offset_in_page(address);
1941         else
1942                 in_page = offset_in_page(address) + size;
1943         now = min(count, (unsigned long)in_page / size);
1944         if (!now) {
1945                 /*
1946                  * String I/O straddles page boundary.  Pin two guest pages
1947                  * so that we satisfy atomicity constraints.  Do just one
1948                  * transaction to avoid complexity.
1949                  */
1950                 nr_pages = 2;
1951                 now = 1;
1952         }
1953         if (down) {
1954                 /*
1955                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1956                  */
1957                 pr_unimpl(vcpu, "guest string pio down\n");
1958                 inject_gp(vcpu);
1959                 return 1;
1960         }
1961         vcpu->run->io.count = now;
1962         vcpu->pio.cur_count = now;
1963
1964         for (i = 0; i < nr_pages; ++i) {
1965                 mutex_lock(&vcpu->kvm->lock);
1966                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1967                 if (page)
1968                         get_page(page);
1969                 vcpu->pio.guest_pages[i] = page;
1970                 mutex_unlock(&vcpu->kvm->lock);
1971                 if (!page) {
1972                         inject_gp(vcpu);
1973                         free_pio_guest_pages(vcpu);
1974                         return 1;
1975                 }
1976         }
1977
1978         pio_dev = vcpu_find_pio_dev(vcpu, port);
1979         if (!vcpu->pio.in) {
1980                 /* string PIO write */
1981                 ret = pio_copy_data(vcpu);
1982                 if (ret >= 0 && pio_dev) {
1983                         pio_string_write(pio_dev, vcpu);
1984                         complete_pio(vcpu);
1985                         if (vcpu->pio.count == 0)
1986                                 ret = 1;
1987                 }
1988         } else if (pio_dev)
1989                 pr_unimpl(vcpu, "no string pio read support yet, "
1990                        "port %x size %d count %ld\n",
1991                         port, size, count);
1992
1993         return ret;
1994 }
1995 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1996
1997 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1998 {
1999         int r;
2000         sigset_t sigsaved;
2001
2002         vcpu_load(vcpu);
2003
2004         if (vcpu->sigset_active)
2005                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2006
2007         /* re-sync apic's tpr */
2008         set_cr8(vcpu, kvm_run->cr8);
2009
2010         if (vcpu->pio.cur_count) {
2011                 r = complete_pio(vcpu);
2012                 if (r)
2013                         goto out;
2014         }
2015
2016         if (vcpu->mmio_needed) {
2017                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2018                 vcpu->mmio_read_completed = 1;
2019                 vcpu->mmio_needed = 0;
2020                 r = emulate_instruction(vcpu, kvm_run,
2021                                         vcpu->mmio_fault_cr2, 0);
2022                 if (r == EMULATE_DO_MMIO) {
2023                         /*
2024                          * Read-modify-write.  Back to userspace.
2025                          */
2026                         r = 0;
2027                         goto out;
2028                 }
2029         }
2030
2031         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2032                 kvm_arch_ops->cache_regs(vcpu);
2033                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2034                 kvm_arch_ops->decache_regs(vcpu);
2035         }
2036
2037         r = kvm_arch_ops->run(vcpu, kvm_run);
2038
2039 out:
2040         if (vcpu->sigset_active)
2041                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2042
2043         vcpu_put(vcpu);
2044         return r;
2045 }
2046
2047 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2048                                    struct kvm_regs *regs)
2049 {
2050         vcpu_load(vcpu);
2051
2052         kvm_arch_ops->cache_regs(vcpu);
2053
2054         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2055         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2056         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2057         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2058         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2059         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2060         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2061         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2062 #ifdef CONFIG_X86_64
2063         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2064         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2065         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2066         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2067         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2068         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2069         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2070         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2071 #endif
2072
2073         regs->rip = vcpu->rip;
2074         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2075
2076         /*
2077          * Don't leak debug flags in case they were set for guest debugging
2078          */
2079         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2080                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2081
2082         vcpu_put(vcpu);
2083
2084         return 0;
2085 }
2086
2087 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2088                                    struct kvm_regs *regs)
2089 {
2090         vcpu_load(vcpu);
2091
2092         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2093         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2094         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2095         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2096         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2097         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2098         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2099         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2100 #ifdef CONFIG_X86_64
2101         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2102         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2103         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2104         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2105         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2106         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2107         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2108         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2109 #endif
2110
2111         vcpu->rip = regs->rip;
2112         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2113
2114         kvm_arch_ops->decache_regs(vcpu);
2115
2116         vcpu_put(vcpu);
2117
2118         return 0;
2119 }
2120
2121 static void get_segment(struct kvm_vcpu *vcpu,
2122                         struct kvm_segment *var, int seg)
2123 {
2124         return kvm_arch_ops->get_segment(vcpu, var, seg);
2125 }
2126
2127 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2128                                     struct kvm_sregs *sregs)
2129 {
2130         struct descriptor_table dt;
2131         int pending_vec;
2132
2133         vcpu_load(vcpu);
2134
2135         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2136         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2137         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2138         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2139         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2140         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2141
2142         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2143         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2144
2145         kvm_arch_ops->get_idt(vcpu, &dt);
2146         sregs->idt.limit = dt.limit;
2147         sregs->idt.base = dt.base;
2148         kvm_arch_ops->get_gdt(vcpu, &dt);
2149         sregs->gdt.limit = dt.limit;
2150         sregs->gdt.base = dt.base;
2151
2152         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2153         sregs->cr0 = vcpu->cr0;
2154         sregs->cr2 = vcpu->cr2;
2155         sregs->cr3 = vcpu->cr3;
2156         sregs->cr4 = vcpu->cr4;
2157         sregs->cr8 = get_cr8(vcpu);
2158         sregs->efer = vcpu->shadow_efer;
2159         sregs->apic_base = kvm_get_apic_base(vcpu);
2160
2161         if (irqchip_in_kernel(vcpu->kvm)) {
2162                 memset(sregs->interrupt_bitmap, 0,
2163                        sizeof sregs->interrupt_bitmap);
2164                 pending_vec = kvm_arch_ops->get_irq(vcpu);
2165                 if (pending_vec >= 0)
2166                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2167         } else
2168                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2169                        sizeof sregs->interrupt_bitmap);
2170
2171         vcpu_put(vcpu);
2172
2173         return 0;
2174 }
2175
2176 static void set_segment(struct kvm_vcpu *vcpu,
2177                         struct kvm_segment *var, int seg)
2178 {
2179         return kvm_arch_ops->set_segment(vcpu, var, seg);
2180 }
2181
2182 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2183                                     struct kvm_sregs *sregs)
2184 {
2185         int mmu_reset_needed = 0;
2186         int i, pending_vec, max_bits;
2187         struct descriptor_table dt;
2188
2189         vcpu_load(vcpu);
2190
2191         dt.limit = sregs->idt.limit;
2192         dt.base = sregs->idt.base;
2193         kvm_arch_ops->set_idt(vcpu, &dt);
2194         dt.limit = sregs->gdt.limit;
2195         dt.base = sregs->gdt.base;
2196         kvm_arch_ops->set_gdt(vcpu, &dt);
2197
2198         vcpu->cr2 = sregs->cr2;
2199         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2200         vcpu->cr3 = sregs->cr3;
2201
2202         set_cr8(vcpu, sregs->cr8);
2203
2204         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2205 #ifdef CONFIG_X86_64
2206         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2207 #endif
2208         kvm_set_apic_base(vcpu, sregs->apic_base);
2209
2210         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2211
2212         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2213         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2214
2215         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2216         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2217         if (!is_long_mode(vcpu) && is_pae(vcpu))
2218                 load_pdptrs(vcpu, vcpu->cr3);
2219
2220         if (mmu_reset_needed)
2221                 kvm_mmu_reset_context(vcpu);
2222
2223         if (!irqchip_in_kernel(vcpu->kvm)) {
2224                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2225                        sizeof vcpu->irq_pending);
2226                 vcpu->irq_summary = 0;
2227                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2228                         if (vcpu->irq_pending[i])
2229                                 __set_bit(i, &vcpu->irq_summary);
2230         } else {
2231                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2232                 pending_vec = find_first_bit(
2233                         (const unsigned long *)sregs->interrupt_bitmap,
2234                         max_bits);
2235                 /* Only pending external irq is handled here */
2236                 if (pending_vec < max_bits) {
2237                         kvm_arch_ops->set_irq(vcpu, pending_vec);
2238                         printk("Set back pending irq %d\n", pending_vec);
2239                 }
2240         }
2241
2242         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2243         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2244         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2245         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2246         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2247         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2248
2249         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2250         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2251
2252         vcpu_put(vcpu);
2253
2254         return 0;
2255 }
2256
2257 /*
2258  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2259  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2260  *
2261  * This list is modified at module load time to reflect the
2262  * capabilities of the host cpu.
2263  */
2264 static u32 msrs_to_save[] = {
2265         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2266         MSR_K6_STAR,
2267 #ifdef CONFIG_X86_64
2268         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2269 #endif
2270         MSR_IA32_TIME_STAMP_COUNTER,
2271 };
2272
2273 static unsigned num_msrs_to_save;
2274
2275 static u32 emulated_msrs[] = {
2276         MSR_IA32_MISC_ENABLE,
2277 };
2278
2279 static __init void kvm_init_msr_list(void)
2280 {
2281         u32 dummy[2];
2282         unsigned i, j;
2283
2284         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2285                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2286                         continue;
2287                 if (j < i)
2288                         msrs_to_save[j] = msrs_to_save[i];
2289                 j++;
2290         }
2291         num_msrs_to_save = j;
2292 }
2293
2294 /*
2295  * Adapt set_msr() to msr_io()'s calling convention
2296  */
2297 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2298 {
2299         return kvm_set_msr(vcpu, index, *data);
2300 }
2301
2302 /*
2303  * Read or write a bunch of msrs. All parameters are kernel addresses.
2304  *
2305  * @return number of msrs set successfully.
2306  */
2307 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2308                     struct kvm_msr_entry *entries,
2309                     int (*do_msr)(struct kvm_vcpu *vcpu,
2310                                   unsigned index, u64 *data))
2311 {
2312         int i;
2313
2314         vcpu_load(vcpu);
2315
2316         for (i = 0; i < msrs->nmsrs; ++i)
2317                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2318                         break;
2319
2320         vcpu_put(vcpu);
2321
2322         return i;
2323 }
2324
2325 /*
2326  * Read or write a bunch of msrs. Parameters are user addresses.
2327  *
2328  * @return number of msrs set successfully.
2329  */
2330 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2331                   int (*do_msr)(struct kvm_vcpu *vcpu,
2332                                 unsigned index, u64 *data),
2333                   int writeback)
2334 {
2335         struct kvm_msrs msrs;
2336         struct kvm_msr_entry *entries;
2337         int r, n;
2338         unsigned size;
2339
2340         r = -EFAULT;
2341         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2342                 goto out;
2343
2344         r = -E2BIG;
2345         if (msrs.nmsrs >= MAX_IO_MSRS)
2346                 goto out;
2347
2348         r = -ENOMEM;
2349         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2350         entries = vmalloc(size);
2351         if (!entries)
2352                 goto out;
2353
2354         r = -EFAULT;
2355         if (copy_from_user(entries, user_msrs->entries, size))
2356                 goto out_free;
2357
2358         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2359         if (r < 0)
2360                 goto out_free;
2361
2362         r = -EFAULT;
2363         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2364                 goto out_free;
2365
2366         r = n;
2367
2368 out_free:
2369         vfree(entries);
2370 out:
2371         return r;
2372 }
2373
2374 /*
2375  * Translate a guest virtual address to a guest physical address.
2376  */
2377 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2378                                     struct kvm_translation *tr)
2379 {
2380         unsigned long vaddr = tr->linear_address;
2381         gpa_t gpa;
2382
2383         vcpu_load(vcpu);
2384         mutex_lock(&vcpu->kvm->lock);
2385         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2386         tr->physical_address = gpa;
2387         tr->valid = gpa != UNMAPPED_GVA;
2388         tr->writeable = 1;
2389         tr->usermode = 0;
2390         mutex_unlock(&vcpu->kvm->lock);
2391         vcpu_put(vcpu);
2392
2393         return 0;
2394 }
2395
2396 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2397                                     struct kvm_interrupt *irq)
2398 {
2399         if (irq->irq < 0 || irq->irq >= 256)
2400                 return -EINVAL;
2401         if (irqchip_in_kernel(vcpu->kvm))
2402                 return -ENXIO;
2403         vcpu_load(vcpu);
2404
2405         set_bit(irq->irq, vcpu->irq_pending);
2406         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2407
2408         vcpu_put(vcpu);
2409
2410         return 0;
2411 }
2412
2413 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2414                                       struct kvm_debug_guest *dbg)
2415 {
2416         int r;
2417
2418         vcpu_load(vcpu);
2419
2420         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2421
2422         vcpu_put(vcpu);
2423
2424         return r;
2425 }
2426
2427 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2428                                     unsigned long address,
2429                                     int *type)
2430 {
2431         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2432         unsigned long pgoff;
2433         struct page *page;
2434
2435         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2436         if (pgoff == 0)
2437                 page = virt_to_page(vcpu->run);
2438         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2439                 page = virt_to_page(vcpu->pio_data);
2440         else
2441                 return NOPAGE_SIGBUS;
2442         get_page(page);
2443         if (type != NULL)
2444                 *type = VM_FAULT_MINOR;
2445
2446         return page;
2447 }
2448
2449 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2450         .nopage = kvm_vcpu_nopage,
2451 };
2452
2453 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2454 {
2455         vma->vm_ops = &kvm_vcpu_vm_ops;
2456         return 0;
2457 }
2458
2459 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2460 {
2461         struct kvm_vcpu *vcpu = filp->private_data;
2462
2463         fput(vcpu->kvm->filp);
2464         return 0;
2465 }
2466
2467 static struct file_operations kvm_vcpu_fops = {
2468         .release        = kvm_vcpu_release,
2469         .unlocked_ioctl = kvm_vcpu_ioctl,
2470         .compat_ioctl   = kvm_vcpu_ioctl,
2471         .mmap           = kvm_vcpu_mmap,
2472 };
2473
2474 /*
2475  * Allocates an inode for the vcpu.
2476  */
2477 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2478 {
2479         int fd, r;
2480         struct inode *inode;
2481         struct file *file;
2482
2483         r = anon_inode_getfd(&fd, &inode, &file,
2484                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2485         if (r)
2486                 return r;
2487         atomic_inc(&vcpu->kvm->filp->f_count);
2488         return fd;
2489 }
2490
2491 /*
2492  * Creates some virtual cpus.  Good luck creating more than one.
2493  */
2494 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2495 {
2496         int r;
2497         struct kvm_vcpu *vcpu;
2498
2499         if (!valid_vcpu(n))
2500                 return -EINVAL;
2501
2502         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2503         if (IS_ERR(vcpu))
2504                 return PTR_ERR(vcpu);
2505
2506         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2507
2508         /* We do fxsave: this must be aligned. */
2509         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2510
2511         vcpu_load(vcpu);
2512         r = kvm_mmu_setup(vcpu);
2513         vcpu_put(vcpu);
2514         if (r < 0)
2515                 goto free_vcpu;
2516
2517         mutex_lock(&kvm->lock);
2518         if (kvm->vcpus[n]) {
2519                 r = -EEXIST;
2520                 mutex_unlock(&kvm->lock);
2521                 goto mmu_unload;
2522         }
2523         kvm->vcpus[n] = vcpu;
2524         mutex_unlock(&kvm->lock);
2525
2526         /* Now it's all set up, let userspace reach it */
2527         r = create_vcpu_fd(vcpu);
2528         if (r < 0)
2529                 goto unlink;
2530         return r;
2531
2532 unlink:
2533         mutex_lock(&kvm->lock);
2534         kvm->vcpus[n] = NULL;
2535         mutex_unlock(&kvm->lock);
2536
2537 mmu_unload:
2538         vcpu_load(vcpu);
2539         kvm_mmu_unload(vcpu);
2540         vcpu_put(vcpu);
2541
2542 free_vcpu:
2543         kvm_arch_ops->vcpu_free(vcpu);
2544         return r;
2545 }
2546
2547 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2548 {
2549         u64 efer;
2550         int i;
2551         struct kvm_cpuid_entry *e, *entry;
2552
2553         rdmsrl(MSR_EFER, efer);
2554         entry = NULL;
2555         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2556                 e = &vcpu->cpuid_entries[i];
2557                 if (e->function == 0x80000001) {
2558                         entry = e;
2559                         break;
2560                 }
2561         }
2562         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2563                 entry->edx &= ~(1 << 20);
2564                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2565         }
2566 }
2567
2568 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2569                                     struct kvm_cpuid *cpuid,
2570                                     struct kvm_cpuid_entry __user *entries)
2571 {
2572         int r;
2573
2574         r = -E2BIG;
2575         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2576                 goto out;
2577         r = -EFAULT;
2578         if (copy_from_user(&vcpu->cpuid_entries, entries,
2579                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2580                 goto out;
2581         vcpu->cpuid_nent = cpuid->nent;
2582         cpuid_fix_nx_cap(vcpu);
2583         return 0;
2584
2585 out:
2586         return r;
2587 }
2588
2589 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2590 {
2591         if (sigset) {
2592                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2593                 vcpu->sigset_active = 1;
2594                 vcpu->sigset = *sigset;
2595         } else
2596                 vcpu->sigset_active = 0;
2597         return 0;
2598 }
2599
2600 /*
2601  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2602  * we have asm/x86/processor.h
2603  */
2604 struct fxsave {
2605         u16     cwd;
2606         u16     swd;
2607         u16     twd;
2608         u16     fop;
2609         u64     rip;
2610         u64     rdp;
2611         u32     mxcsr;
2612         u32     mxcsr_mask;
2613         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2614 #ifdef CONFIG_X86_64
2615         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2616 #else
2617         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2618 #endif
2619 };
2620
2621 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2622 {
2623         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2624
2625         vcpu_load(vcpu);
2626
2627         memcpy(fpu->fpr, fxsave->st_space, 128);
2628         fpu->fcw = fxsave->cwd;
2629         fpu->fsw = fxsave->swd;
2630         fpu->ftwx = fxsave->twd;
2631         fpu->last_opcode = fxsave->fop;
2632         fpu->last_ip = fxsave->rip;
2633         fpu->last_dp = fxsave->rdp;
2634         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2635
2636         vcpu_put(vcpu);
2637
2638         return 0;
2639 }
2640
2641 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2642 {
2643         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2644
2645         vcpu_load(vcpu);
2646
2647         memcpy(fxsave->st_space, fpu->fpr, 128);
2648         fxsave->cwd = fpu->fcw;
2649         fxsave->swd = fpu->fsw;
2650         fxsave->twd = fpu->ftwx;
2651         fxsave->fop = fpu->last_opcode;
2652         fxsave->rip = fpu->last_ip;
2653         fxsave->rdp = fpu->last_dp;
2654         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2655
2656         vcpu_put(vcpu);
2657
2658         return 0;
2659 }
2660
2661 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2662                                     struct kvm_lapic_state *s)
2663 {
2664         vcpu_load(vcpu);
2665         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2666         vcpu_put(vcpu);
2667
2668         return 0;
2669 }
2670
2671 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2672                                     struct kvm_lapic_state *s)
2673 {
2674         vcpu_load(vcpu);
2675         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2676         kvm_apic_post_state_restore(vcpu);
2677         vcpu_put(vcpu);
2678
2679         return 0;
2680 }
2681
2682 static long kvm_vcpu_ioctl(struct file *filp,
2683                            unsigned int ioctl, unsigned long arg)
2684 {
2685         struct kvm_vcpu *vcpu = filp->private_data;
2686         void __user *argp = (void __user *)arg;
2687         int r = -EINVAL;
2688
2689         switch (ioctl) {
2690         case KVM_RUN:
2691                 r = -EINVAL;
2692                 if (arg)
2693                         goto out;
2694                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2695                 break;
2696         case KVM_GET_REGS: {
2697                 struct kvm_regs kvm_regs;
2698
2699                 memset(&kvm_regs, 0, sizeof kvm_regs);
2700                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2701                 if (r)
2702                         goto out;
2703                 r = -EFAULT;
2704                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2705                         goto out;
2706                 r = 0;
2707                 break;
2708         }
2709         case KVM_SET_REGS: {
2710                 struct kvm_regs kvm_regs;
2711
2712                 r = -EFAULT;
2713                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2714                         goto out;
2715                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2716                 if (r)
2717                         goto out;
2718                 r = 0;
2719                 break;
2720         }
2721         case KVM_GET_SREGS: {
2722                 struct kvm_sregs kvm_sregs;
2723
2724                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2725                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2726                 if (r)
2727                         goto out;
2728                 r = -EFAULT;
2729                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2730                         goto out;
2731                 r = 0;
2732                 break;
2733         }
2734         case KVM_SET_SREGS: {
2735                 struct kvm_sregs kvm_sregs;
2736
2737                 r = -EFAULT;
2738                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2739                         goto out;
2740                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2741                 if (r)
2742                         goto out;
2743                 r = 0;
2744                 break;
2745         }
2746         case KVM_TRANSLATE: {
2747                 struct kvm_translation tr;
2748
2749                 r = -EFAULT;
2750                 if (copy_from_user(&tr, argp, sizeof tr))
2751                         goto out;
2752                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2753                 if (r)
2754                         goto out;
2755                 r = -EFAULT;
2756                 if (copy_to_user(argp, &tr, sizeof tr))
2757                         goto out;
2758                 r = 0;
2759                 break;
2760         }
2761         case KVM_INTERRUPT: {
2762                 struct kvm_interrupt irq;
2763
2764                 r = -EFAULT;
2765                 if (copy_from_user(&irq, argp, sizeof irq))
2766                         goto out;
2767                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2768                 if (r)
2769                         goto out;
2770                 r = 0;
2771                 break;
2772         }
2773         case KVM_DEBUG_GUEST: {
2774                 struct kvm_debug_guest dbg;
2775
2776                 r = -EFAULT;
2777                 if (copy_from_user(&dbg, argp, sizeof dbg))
2778                         goto out;
2779                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2780                 if (r)
2781                         goto out;
2782                 r = 0;
2783                 break;
2784         }
2785         case KVM_GET_MSRS:
2786                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2787                 break;
2788         case KVM_SET_MSRS:
2789                 r = msr_io(vcpu, argp, do_set_msr, 0);
2790                 break;
2791         case KVM_SET_CPUID: {
2792                 struct kvm_cpuid __user *cpuid_arg = argp;
2793                 struct kvm_cpuid cpuid;
2794
2795                 r = -EFAULT;
2796                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2797                         goto out;
2798                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2799                 if (r)
2800                         goto out;
2801                 break;
2802         }
2803         case KVM_SET_SIGNAL_MASK: {
2804                 struct kvm_signal_mask __user *sigmask_arg = argp;
2805                 struct kvm_signal_mask kvm_sigmask;
2806                 sigset_t sigset, *p;
2807
2808                 p = NULL;
2809                 if (argp) {
2810                         r = -EFAULT;
2811                         if (copy_from_user(&kvm_sigmask, argp,
2812                                            sizeof kvm_sigmask))
2813                                 goto out;
2814                         r = -EINVAL;
2815                         if (kvm_sigmask.len != sizeof sigset)
2816                                 goto out;
2817                         r = -EFAULT;
2818                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2819                                            sizeof sigset))
2820                                 goto out;
2821                         p = &sigset;
2822                 }
2823                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2824                 break;
2825         }
2826         case KVM_GET_FPU: {
2827                 struct kvm_fpu fpu;
2828
2829                 memset(&fpu, 0, sizeof fpu);
2830                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2831                 if (r)
2832                         goto out;
2833                 r = -EFAULT;
2834                 if (copy_to_user(argp, &fpu, sizeof fpu))
2835                         goto out;
2836                 r = 0;
2837                 break;
2838         }
2839         case KVM_SET_FPU: {
2840                 struct kvm_fpu fpu;
2841
2842                 r = -EFAULT;
2843                 if (copy_from_user(&fpu, argp, sizeof fpu))
2844                         goto out;
2845                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2846                 if (r)
2847                         goto out;
2848                 r = 0;
2849                 break;
2850         }
2851         case KVM_GET_LAPIC: {
2852                 struct kvm_lapic_state lapic;
2853
2854                 memset(&lapic, 0, sizeof lapic);
2855                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2856                 if (r)
2857                         goto out;
2858                 r = -EFAULT;
2859                 if (copy_to_user(argp, &lapic, sizeof lapic))
2860                         goto out;
2861                 r = 0;
2862                 break;
2863         }
2864         case KVM_SET_LAPIC: {
2865                 struct kvm_lapic_state lapic;
2866
2867                 r = -EFAULT;
2868                 if (copy_from_user(&lapic, argp, sizeof lapic))
2869                         goto out;
2870                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2871                 if (r)
2872                         goto out;
2873                 r = 0;
2874                 break;
2875         }
2876         default:
2877                 ;
2878         }
2879 out:
2880         return r;
2881 }
2882
2883 static long kvm_vm_ioctl(struct file *filp,
2884                            unsigned int ioctl, unsigned long arg)
2885 {
2886         struct kvm *kvm = filp->private_data;
2887         void __user *argp = (void __user *)arg;
2888         int r = -EINVAL;
2889
2890         switch (ioctl) {
2891         case KVM_CREATE_VCPU:
2892                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2893                 if (r < 0)
2894                         goto out;
2895                 break;
2896         case KVM_SET_MEMORY_REGION: {
2897                 struct kvm_memory_region kvm_mem;
2898
2899                 r = -EFAULT;
2900                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2901                         goto out;
2902                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2903                 if (r)
2904                         goto out;
2905                 break;
2906         }
2907         case KVM_GET_DIRTY_LOG: {
2908                 struct kvm_dirty_log log;
2909
2910                 r = -EFAULT;
2911                 if (copy_from_user(&log, argp, sizeof log))
2912                         goto out;
2913                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2914                 if (r)
2915                         goto out;
2916                 break;
2917         }
2918         case KVM_SET_MEMORY_ALIAS: {
2919                 struct kvm_memory_alias alias;
2920
2921                 r = -EFAULT;
2922                 if (copy_from_user(&alias, argp, sizeof alias))
2923                         goto out;
2924                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2925                 if (r)
2926                         goto out;
2927                 break;
2928         }
2929         case KVM_CREATE_IRQCHIP:
2930                 r = -ENOMEM;
2931                 kvm->vpic = kvm_create_pic(kvm);
2932                 if (kvm->vpic) {
2933                         r = kvm_ioapic_init(kvm);
2934                         if (r) {
2935                                 kfree(kvm->vpic);
2936                                 kvm->vpic = NULL;
2937                                 goto out;
2938                         }
2939                 }
2940                 else
2941                         goto out;
2942                 break;
2943         case KVM_IRQ_LINE: {
2944                 struct kvm_irq_level irq_event;
2945
2946                 r = -EFAULT;
2947                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2948                         goto out;
2949                 if (irqchip_in_kernel(kvm)) {
2950                         mutex_lock(&kvm->lock);
2951                         if (irq_event.irq < 16)
2952                                 kvm_pic_set_irq(pic_irqchip(kvm),
2953                                         irq_event.irq,
2954                                         irq_event.level);
2955                         kvm_ioapic_set_irq(kvm->vioapic,
2956                                         irq_event.irq,
2957                                         irq_event.level);
2958                         mutex_unlock(&kvm->lock);
2959                         r = 0;
2960                 }
2961                 break;
2962         }
2963         case KVM_GET_IRQCHIP: {
2964                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2965                 struct kvm_irqchip chip;
2966
2967                 r = -EFAULT;
2968                 if (copy_from_user(&chip, argp, sizeof chip))
2969                         goto out;
2970                 r = -ENXIO;
2971                 if (!irqchip_in_kernel(kvm))
2972                         goto out;
2973                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
2974                 if (r)
2975                         goto out;
2976                 r = -EFAULT;
2977                 if (copy_to_user(argp, &chip, sizeof chip))
2978                         goto out;
2979                 r = 0;
2980                 break;
2981         }
2982         case KVM_SET_IRQCHIP: {
2983                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2984                 struct kvm_irqchip chip;
2985
2986                 r = -EFAULT;
2987                 if (copy_from_user(&chip, argp, sizeof chip))
2988                         goto out;
2989                 r = -ENXIO;
2990                 if (!irqchip_in_kernel(kvm))
2991                         goto out;
2992                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
2993                 if (r)
2994                         goto out;
2995                 r = 0;
2996                 break;
2997         }
2998         default:
2999                 ;
3000         }
3001 out:
3002         return r;
3003 }
3004
3005 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3006                                   unsigned long address,
3007                                   int *type)
3008 {
3009         struct kvm *kvm = vma->vm_file->private_data;
3010         unsigned long pgoff;
3011         struct page *page;
3012
3013         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3014         page = gfn_to_page(kvm, pgoff);
3015         if (!page)
3016                 return NOPAGE_SIGBUS;
3017         get_page(page);
3018         if (type != NULL)
3019                 *type = VM_FAULT_MINOR;
3020
3021         return page;
3022 }
3023
3024 static struct vm_operations_struct kvm_vm_vm_ops = {
3025         .nopage = kvm_vm_nopage,
3026 };
3027
3028 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3029 {
3030         vma->vm_ops = &kvm_vm_vm_ops;
3031         return 0;
3032 }
3033
3034 static struct file_operations kvm_vm_fops = {
3035         .release        = kvm_vm_release,
3036         .unlocked_ioctl = kvm_vm_ioctl,
3037         .compat_ioctl   = kvm_vm_ioctl,
3038         .mmap           = kvm_vm_mmap,
3039 };
3040
3041 static int kvm_dev_ioctl_create_vm(void)
3042 {
3043         int fd, r;
3044         struct inode *inode;
3045         struct file *file;
3046         struct kvm *kvm;
3047
3048         kvm = kvm_create_vm();
3049         if (IS_ERR(kvm))
3050                 return PTR_ERR(kvm);
3051         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3052         if (r) {
3053                 kvm_destroy_vm(kvm);
3054                 return r;
3055         }
3056
3057         kvm->filp = file;
3058
3059         return fd;
3060 }
3061
3062 static long kvm_dev_ioctl(struct file *filp,
3063                           unsigned int ioctl, unsigned long arg)
3064 {
3065         void __user *argp = (void __user *)arg;
3066         long r = -EINVAL;
3067
3068         switch (ioctl) {
3069         case KVM_GET_API_VERSION:
3070                 r = -EINVAL;
3071                 if (arg)
3072                         goto out;
3073                 r = KVM_API_VERSION;
3074                 break;
3075         case KVM_CREATE_VM:
3076                 r = -EINVAL;
3077                 if (arg)
3078                         goto out;
3079                 r = kvm_dev_ioctl_create_vm();
3080                 break;
3081         case KVM_GET_MSR_INDEX_LIST: {
3082                 struct kvm_msr_list __user *user_msr_list = argp;
3083                 struct kvm_msr_list msr_list;
3084                 unsigned n;
3085
3086                 r = -EFAULT;
3087                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3088                         goto out;
3089                 n = msr_list.nmsrs;
3090                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3091                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3092                         goto out;
3093                 r = -E2BIG;
3094                 if (n < num_msrs_to_save)
3095                         goto out;
3096                 r = -EFAULT;
3097                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3098                                  num_msrs_to_save * sizeof(u32)))
3099                         goto out;
3100                 if (copy_to_user(user_msr_list->indices
3101                                  + num_msrs_to_save * sizeof(u32),
3102                                  &emulated_msrs,
3103                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3104                         goto out;
3105                 r = 0;
3106                 break;
3107         }
3108         case KVM_CHECK_EXTENSION: {
3109                 int ext = (long)argp;
3110
3111                 switch (ext) {
3112                 case KVM_CAP_IRQCHIP:
3113                 case KVM_CAP_HLT:
3114                         r = 1;
3115                         break;
3116                 default:
3117                         r = 0;
3118                         break;
3119                 }
3120                 break;
3121         }
3122         case KVM_GET_VCPU_MMAP_SIZE:
3123                 r = -EINVAL;
3124                 if (arg)
3125                         goto out;
3126                 r = 2 * PAGE_SIZE;
3127                 break;
3128         default:
3129                 ;
3130         }
3131 out:
3132         return r;
3133 }
3134
3135 static struct file_operations kvm_chardev_ops = {
3136         .unlocked_ioctl = kvm_dev_ioctl,
3137         .compat_ioctl   = kvm_dev_ioctl,
3138 };
3139
3140 static struct miscdevice kvm_dev = {
3141         KVM_MINOR,
3142         "kvm",
3143         &kvm_chardev_ops,
3144 };
3145
3146 /*
3147  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3148  * cached on it.
3149  */
3150 static void decache_vcpus_on_cpu(int cpu)
3151 {
3152         struct kvm *vm;
3153         struct kvm_vcpu *vcpu;
3154         int i;
3155
3156         spin_lock(&kvm_lock);
3157         list_for_each_entry(vm, &vm_list, vm_list)
3158                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3159                         vcpu = vm->vcpus[i];
3160                         if (!vcpu)
3161                                 continue;
3162                         /*
3163                          * If the vcpu is locked, then it is running on some
3164                          * other cpu and therefore it is not cached on the
3165                          * cpu in question.
3166                          *
3167                          * If it's not locked, check the last cpu it executed
3168                          * on.
3169                          */
3170                         if (mutex_trylock(&vcpu->mutex)) {
3171                                 if (vcpu->cpu == cpu) {
3172                                         kvm_arch_ops->vcpu_decache(vcpu);
3173                                         vcpu->cpu = -1;
3174                                 }
3175                                 mutex_unlock(&vcpu->mutex);
3176                         }
3177                 }
3178         spin_unlock(&kvm_lock);
3179 }
3180
3181 static void hardware_enable(void *junk)
3182 {
3183         int cpu = raw_smp_processor_id();
3184
3185         if (cpu_isset(cpu, cpus_hardware_enabled))
3186                 return;
3187         cpu_set(cpu, cpus_hardware_enabled);
3188         kvm_arch_ops->hardware_enable(NULL);
3189 }
3190
3191 static void hardware_disable(void *junk)
3192 {
3193         int cpu = raw_smp_processor_id();
3194
3195         if (!cpu_isset(cpu, cpus_hardware_enabled))
3196                 return;
3197         cpu_clear(cpu, cpus_hardware_enabled);
3198         decache_vcpus_on_cpu(cpu);
3199         kvm_arch_ops->hardware_disable(NULL);
3200 }
3201
3202 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3203                            void *v)
3204 {
3205         int cpu = (long)v;
3206
3207         switch (val) {
3208         case CPU_DYING:
3209         case CPU_DYING_FROZEN:
3210                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3211                        cpu);
3212                 hardware_disable(NULL);
3213                 break;
3214         case CPU_UP_CANCELED:
3215         case CPU_UP_CANCELED_FROZEN:
3216                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3217                        cpu);
3218                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3219                 break;
3220         case CPU_ONLINE:
3221         case CPU_ONLINE_FROZEN:
3222                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3223                        cpu);
3224                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3225                 break;
3226         }
3227         return NOTIFY_OK;
3228 }
3229
3230 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3231                        void *v)
3232 {
3233         if (val == SYS_RESTART) {
3234                 /*
3235                  * Some (well, at least mine) BIOSes hang on reboot if
3236                  * in vmx root mode.
3237                  */
3238                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3239                 on_each_cpu(hardware_disable, NULL, 0, 1);
3240         }
3241         return NOTIFY_OK;
3242 }
3243
3244 static struct notifier_block kvm_reboot_notifier = {
3245         .notifier_call = kvm_reboot,
3246         .priority = 0,
3247 };
3248
3249 void kvm_io_bus_init(struct kvm_io_bus *bus)
3250 {
3251         memset(bus, 0, sizeof(*bus));
3252 }
3253
3254 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3255 {
3256         int i;
3257
3258         for (i = 0; i < bus->dev_count; i++) {
3259                 struct kvm_io_device *pos = bus->devs[i];
3260
3261                 kvm_iodevice_destructor(pos);
3262         }
3263 }
3264
3265 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3266 {
3267         int i;
3268
3269         for (i = 0; i < bus->dev_count; i++) {
3270                 struct kvm_io_device *pos = bus->devs[i];
3271
3272                 if (pos->in_range(pos, addr))
3273                         return pos;
3274         }
3275
3276         return NULL;
3277 }
3278
3279 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3280 {
3281         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3282
3283         bus->devs[bus->dev_count++] = dev;
3284 }
3285
3286 static struct notifier_block kvm_cpu_notifier = {
3287         .notifier_call = kvm_cpu_hotplug,
3288         .priority = 20, /* must be > scheduler priority */
3289 };
3290
3291 static u64 stat_get(void *_offset)
3292 {
3293         unsigned offset = (long)_offset;
3294         u64 total = 0;
3295         struct kvm *kvm;
3296         struct kvm_vcpu *vcpu;
3297         int i;
3298
3299         spin_lock(&kvm_lock);
3300         list_for_each_entry(kvm, &vm_list, vm_list)
3301                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3302                         vcpu = kvm->vcpus[i];
3303                         if (vcpu)
3304                                 total += *(u32 *)((void *)vcpu + offset);
3305                 }
3306         spin_unlock(&kvm_lock);
3307         return total;
3308 }
3309
3310 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3311
3312 static __init void kvm_init_debug(void)
3313 {
3314         struct kvm_stats_debugfs_item *p;
3315
3316         debugfs_dir = debugfs_create_dir("kvm", NULL);
3317         for (p = debugfs_entries; p->name; ++p)
3318                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3319                                                 (void *)(long)p->offset,
3320                                                 &stat_fops);
3321 }
3322
3323 static void kvm_exit_debug(void)
3324 {
3325         struct kvm_stats_debugfs_item *p;
3326
3327         for (p = debugfs_entries; p->name; ++p)
3328                 debugfs_remove(p->dentry);
3329         debugfs_remove(debugfs_dir);
3330 }
3331
3332 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3333 {
3334         hardware_disable(NULL);
3335         return 0;
3336 }
3337
3338 static int kvm_resume(struct sys_device *dev)
3339 {
3340         hardware_enable(NULL);
3341         return 0;
3342 }
3343
3344 static struct sysdev_class kvm_sysdev_class = {
3345         set_kset_name("kvm"),
3346         .suspend = kvm_suspend,
3347         .resume = kvm_resume,
3348 };
3349
3350 static struct sys_device kvm_sysdev = {
3351         .id = 0,
3352         .cls = &kvm_sysdev_class,
3353 };
3354
3355 hpa_t bad_page_address;
3356
3357 static inline
3358 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3359 {
3360         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3361 }
3362
3363 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3364 {
3365         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3366
3367         kvm_arch_ops->vcpu_load(vcpu, cpu);
3368 }
3369
3370 static void kvm_sched_out(struct preempt_notifier *pn,
3371                           struct task_struct *next)
3372 {
3373         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3374
3375         kvm_arch_ops->vcpu_put(vcpu);
3376 }
3377
3378 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3379                   struct module *module)
3380 {
3381         int r;
3382         int cpu;
3383
3384         if (kvm_arch_ops) {
3385                 printk(KERN_ERR "kvm: already loaded the other module\n");
3386                 return -EEXIST;
3387         }
3388
3389         if (!ops->cpu_has_kvm_support()) {
3390                 printk(KERN_ERR "kvm: no hardware support\n");
3391                 return -EOPNOTSUPP;
3392         }
3393         if (ops->disabled_by_bios()) {
3394                 printk(KERN_ERR "kvm: disabled by bios\n");
3395                 return -EOPNOTSUPP;
3396         }
3397
3398         kvm_arch_ops = ops;
3399
3400         r = kvm_arch_ops->hardware_setup();
3401         if (r < 0)
3402                 goto out;
3403
3404         for_each_online_cpu(cpu) {
3405                 smp_call_function_single(cpu,
3406                                 kvm_arch_ops->check_processor_compatibility,
3407                                 &r, 0, 1);
3408                 if (r < 0)
3409                         goto out_free_0;
3410         }
3411
3412         on_each_cpu(hardware_enable, NULL, 0, 1);
3413         r = register_cpu_notifier(&kvm_cpu_notifier);
3414         if (r)
3415                 goto out_free_1;
3416         register_reboot_notifier(&kvm_reboot_notifier);
3417
3418         r = sysdev_class_register(&kvm_sysdev_class);
3419         if (r)
3420                 goto out_free_2;
3421
3422         r = sysdev_register(&kvm_sysdev);
3423         if (r)
3424                 goto out_free_3;
3425
3426         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3427         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3428                                            __alignof__(struct kvm_vcpu), 0, 0);
3429         if (!kvm_vcpu_cache) {
3430                 r = -ENOMEM;
3431                 goto out_free_4;
3432         }
3433
3434         kvm_chardev_ops.owner = module;
3435
3436         r = misc_register(&kvm_dev);
3437         if (r) {
3438                 printk (KERN_ERR "kvm: misc device register failed\n");
3439                 goto out_free;
3440         }
3441
3442         kvm_preempt_ops.sched_in = kvm_sched_in;
3443         kvm_preempt_ops.sched_out = kvm_sched_out;
3444
3445         return r;
3446
3447 out_free:
3448         kmem_cache_destroy(kvm_vcpu_cache);
3449 out_free_4:
3450         sysdev_unregister(&kvm_sysdev);
3451 out_free_3:
3452         sysdev_class_unregister(&kvm_sysdev_class);
3453 out_free_2:
3454         unregister_reboot_notifier(&kvm_reboot_notifier);
3455         unregister_cpu_notifier(&kvm_cpu_notifier);
3456 out_free_1:
3457         on_each_cpu(hardware_disable, NULL, 0, 1);
3458 out_free_0:
3459         kvm_arch_ops->hardware_unsetup();
3460 out:
3461         kvm_arch_ops = NULL;
3462         return r;
3463 }
3464
3465 void kvm_exit_arch(void)
3466 {
3467         misc_deregister(&kvm_dev);
3468         kmem_cache_destroy(kvm_vcpu_cache);
3469         sysdev_unregister(&kvm_sysdev);
3470         sysdev_class_unregister(&kvm_sysdev_class);
3471         unregister_reboot_notifier(&kvm_reboot_notifier);
3472         unregister_cpu_notifier(&kvm_cpu_notifier);
3473         on_each_cpu(hardware_disable, NULL, 0, 1);
3474         kvm_arch_ops->hardware_unsetup();
3475         kvm_arch_ops = NULL;
3476 }
3477
3478 static __init int kvm_init(void)
3479 {
3480         static struct page *bad_page;
3481         int r;
3482
3483         r = kvm_mmu_module_init();
3484         if (r)
3485                 goto out4;
3486
3487         kvm_init_debug();
3488
3489         kvm_init_msr_list();
3490
3491         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3492                 r = -ENOMEM;
3493                 goto out;
3494         }
3495
3496         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3497         memset(__va(bad_page_address), 0, PAGE_SIZE);
3498
3499         return 0;
3500
3501 out:
3502         kvm_exit_debug();
3503         kvm_mmu_module_exit();
3504 out4:
3505         return r;
3506 }
3507
3508 static __exit void kvm_exit(void)
3509 {
3510         kvm_exit_debug();
3511         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3512         kvm_mmu_module_exit();
3513 }
3514
3515 module_init(kvm_init)
3516 module_exit(kvm_exit)
3517
3518 EXPORT_SYMBOL_GPL(kvm_init_arch);
3519 EXPORT_SYMBOL_GPL(kvm_exit_arch);