arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/pci.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/module.h>
  36 #include <linux/mman.h>
  37 #include <linux/highmem.h>
  38 #include <linux/intel-iommu.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/msr.h>
  42 #include <asm/desc.h>
  43
  44 #define MAX_IO_MSRS 256
  45 #define CR0_RESERVED_BITS                                               \
  46         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  47                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  48                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  49 #define CR4_RESERVED_BITS                                               \
  50         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  51                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  52                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  53                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  54
  55 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  56 /* EFER defaults:
  57  * - enable syscall per default because its emulated by KVM
  58  * - enable LME and LMA per default on 64 bit KVM
  59  */
  60 #ifdef CONFIG_X86_64
  61 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  62 #else
  63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  64 #endif
  65
  66 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  67 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  68
  69 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  70                                     struct kvm_cpuid_entry2 __user *entries);
  71
  72 struct kvm_x86_ops *kvm_x86_ops;
  73 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  74
  75 struct kvm_stats_debugfs_item debugfs_entries[] = {
  76         { "pf_fixed", VCPU_STAT(pf_fixed) },
  77         { "pf_guest", VCPU_STAT(pf_guest) },
  78         { "tlb_flush", VCPU_STAT(tlb_flush) },
  79         { "invlpg", VCPU_STAT(invlpg) },
  80         { "exits", VCPU_STAT(exits) },
  81         { "io_exits", VCPU_STAT(io_exits) },
  82         { "mmio_exits", VCPU_STAT(mmio_exits) },
  83         { "signal_exits", VCPU_STAT(signal_exits) },
  84         { "irq_window", VCPU_STAT(irq_window_exits) },
  85         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  86         { "halt_exits", VCPU_STAT(halt_exits) },
  87         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  88         { "hypercalls", VCPU_STAT(hypercalls) },
  89         { "request_irq", VCPU_STAT(request_irq_exits) },
  90         { "irq_exits", VCPU_STAT(irq_exits) },
  91         { "host_state_reload", VCPU_STAT(host_state_reload) },
  92         { "efer_reload", VCPU_STAT(efer_reload) },
  93         { "fpu_reload", VCPU_STAT(fpu_reload) },
  94         { "insn_emulation", VCPU_STAT(insn_emulation) },
  95         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  96         { "irq_injections", VCPU_STAT(irq_injections) },
  97         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  98         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  99         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 100         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 101         { "mmu_flooded", VM_STAT(mmu_flooded) },
 102         { "mmu_recycled", VM_STAT(mmu_recycled) },
 103         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 104         { "mmu_unsync", VM_STAT(mmu_unsync) },
 105         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 106         { "largepages", VM_STAT(lpages) },
 107         { NULL }
 108 };
 109
 110 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
 111                                                       int assigned_dev_id)
 112 {
 113         struct list_head *ptr;
 114         struct kvm_assigned_dev_kernel *match;
 115
 116         list_for_each(ptr, head) {
 117                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
 118                 if (match->assigned_dev_id == assigned_dev_id)
 119                         return match;
 120         }
 121         return NULL;
 122 }
 123
 124 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 125 {
 126         struct kvm_assigned_dev_kernel *assigned_dev;
 127
 128         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 129                                     interrupt_work);
 130
 131         /* This is taken to safely inject irq inside the guest. When
 132          * the interrupt injection (or the ioapic code) uses a
 133          * finer-grained lock, update this
 134          */
 135         mutex_lock(&assigned_dev->kvm->lock);
 136         kvm_set_irq(assigned_dev->kvm,
 137                     assigned_dev->guest_irq, 1);
 138         mutex_unlock(&assigned_dev->kvm->lock);
 139         kvm_put_kvm(assigned_dev->kvm);
 140 }
 141
 142 /* FIXME: Implement the OR logic needed to make shared interrupts on
 143  * this line behave properly
 144  */
 145 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 146 {
 147         struct kvm_assigned_dev_kernel *assigned_dev =
 148                 (struct kvm_assigned_dev_kernel *) dev_id;
 149
 150         kvm_get_kvm(assigned_dev->kvm);
 151         schedule_work(&assigned_dev->interrupt_work);
 152         disable_irq_nosync(irq);
 153         return IRQ_HANDLED;
 154 }
 155
 156 /* Ack the irq line for an assigned device */
 157 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 158 {
 159         struct kvm_assigned_dev_kernel *dev;
 160
 161         if (kian->gsi == -1)
 162                 return;
 163
 164         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 165                            ack_notifier);
 166         kvm_set_irq(dev->kvm, dev->guest_irq, 0);
 167         enable_irq(dev->host_irq);
 168 }
 169
 170 static void kvm_free_assigned_device(struct kvm *kvm,
 171                                      struct kvm_assigned_dev_kernel
 172                                      *assigned_dev)
 173 {
 174         if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested)
 175                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 176
 177         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 178
 179         if (cancel_work_sync(&assigned_dev->interrupt_work))
 180                 /* We had pending work. That means we will have to take
 181                  * care of kvm_put_kvm.
 182                  */
 183                 kvm_put_kvm(kvm);
 184
 185         pci_release_regions(assigned_dev->dev);
 186         pci_disable_device(assigned_dev->dev);
 187         pci_dev_put(assigned_dev->dev);
 188
 189         list_del(&assigned_dev->list);
 190         kfree(assigned_dev);
 191 }
 192
 193 static void kvm_free_all_assigned_devices(struct kvm *kvm)
 194 {
 195         struct list_head *ptr, *ptr2;
 196         struct kvm_assigned_dev_kernel *assigned_dev;
 197
 198         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 199                 assigned_dev = list_entry(ptr,
 200                                           struct kvm_assigned_dev_kernel,
 201                                           list);
 202
 203                 kvm_free_assigned_device(kvm, assigned_dev);
 204         }
 205 }
 206
 207 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 208                                    struct kvm_assigned_irq
 209                                    *assigned_irq)
 210 {
 211         int r = 0;
 212         struct kvm_assigned_dev_kernel *match;
 213
 214         mutex_lock(&kvm->lock);
 215
 216         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 217                                       assigned_irq->assigned_dev_id);
 218         if (!match) {
 219                 mutex_unlock(&kvm->lock);
 220                 return -EINVAL;
 221         }
 222
 223         if (match->irq_requested) {
 224                 match->guest_irq = assigned_irq->guest_irq;
 225                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 226                 mutex_unlock(&kvm->lock);
 227                 return 0;
 228         }
 229
 230         INIT_WORK(&match->interrupt_work,
 231                   kvm_assigned_dev_interrupt_work_handler);
 232
 233         if (irqchip_in_kernel(kvm)) {
 234                 if (!capable(CAP_SYS_RAWIO)) {
 235                         r = -EPERM;
 236                         goto out_release;
 237                 }
 238
 239                 if (assigned_irq->host_irq)
 240                         match->host_irq = assigned_irq->host_irq;
 241                 else
 242                         match->host_irq = match->dev->irq;
 243                 match->guest_irq = assigned_irq->guest_irq;
 244                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 245                 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 246                 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
 247
 248                 /* Even though this is PCI, we don't want to use shared
 249                  * interrupts. Sharing host devices with guest-assigned devices
 250                  * on the same interrupt line is not a happy situation: there
 251                  * are going to be long delays in accepting, acking, etc.
 252                  */
 253                 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
 254                                 "kvm_assigned_device", (void *)match)) {
 255                         r = -EIO;
 256                         goto out_release;
 257                 }
 258         }
 259
 260         match->irq_requested = true;
 261         mutex_unlock(&kvm->lock);
 262         return r;
 263 out_release:
 264         mutex_unlock(&kvm->lock);
 265         kvm_free_assigned_device(kvm, match);
 266         return r;
 267 }
 268
 269 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 270                                       struct kvm_assigned_pci_dev *assigned_dev)
 271 {
 272         int r = 0;
 273         struct kvm_assigned_dev_kernel *match;
 274         struct pci_dev *dev;
 275
 276         mutex_lock(&kvm->lock);
 277
 278         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 279                                       assigned_dev->assigned_dev_id);
 280         if (match) {
 281                 /* device already assigned */
 282                 r = -EINVAL;
 283                 goto out;
 284         }
 285
 286         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 287         if (match == NULL) {
 288                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 289                        __func__);
 290                 r = -ENOMEM;
 291                 goto out;
 292         }
 293         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 294                                    assigned_dev->devfn);
 295         if (!dev) {
 296                 printk(KERN_INFO "%s: host device not found\n", __func__);
 297                 r = -EINVAL;
 298                 goto out_free;
 299         }
 300         if (pci_enable_device(dev)) {
 301                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 302                 r = -EBUSY;
 303                 goto out_put;
 304         }
 305         r = pci_request_regions(dev, "kvm_assigned_device");
 306         if (r) {
 307                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 308                        __func__);
 309                 goto out_disable;
 310         }
 311         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 312         match->host_busnr = assigned_dev->busnr;
 313         match->host_devfn = assigned_dev->devfn;
 314         match->dev = dev;
 315
 316         match->kvm = kvm;
 317
 318         list_add(&match->list, &kvm->arch.assigned_dev_head);
 319
 320         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 321                 r = kvm_iommu_map_guest(kvm, match);
 322                 if (r)
 323                         goto out_list_del;
 324         }
 325
 326 out:
 327         mutex_unlock(&kvm->lock);
 328         return r;
 329 out_list_del:
 330         list_del(&match->list);
 331         pci_release_regions(dev);
 332 out_disable:
 333         pci_disable_device(dev);
 334 out_put:
 335         pci_dev_put(dev);
 336 out_free:
 337         kfree(match);
 338         mutex_unlock(&kvm->lock);
 339         return r;
 340 }
 341
 342 unsigned long segment_base(u16 selector)
 343 {
 344         struct descriptor_table gdt;
 345         struct desc_struct *d;
 346         unsigned long table_base;
 347         unsigned long v;
 348
 349         if (selector == 0)
 350                 return 0;
 351
 352         asm("sgdt %0" : "=m"(gdt));
 353         table_base = gdt.base;
 354
 355         if (selector & 4) {           /* from ldt */
 356                 u16 ldt_selector;
 357
 358                 asm("sldt %0" : "=g"(ldt_selector));
 359                 table_base = segment_base(ldt_selector);
 360         }
 361         d = (struct desc_struct *)(table_base + (selector & ~7));
 362         v = d->base0 | ((unsigned long)d->base1 << 16) |
 363                 ((unsigned long)d->base2 << 24);
 364 #ifdef CONFIG_X86_64
 365         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 366                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 367 #endif
 368         return v;
 369 }
 370 EXPORT_SYMBOL_GPL(segment_base);
 371
 372 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 373 {
 374         if (irqchip_in_kernel(vcpu->kvm))
 375                 return vcpu->arch.apic_base;
 376         else
 377                 return vcpu->arch.apic_base;
 378 }
 379 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 380
 381 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 382 {
 383         /* TODO: reserve bits check */
 384         if (irqchip_in_kernel(vcpu->kvm))
 385                 kvm_lapic_set_base(vcpu, data);
 386         else
 387                 vcpu->arch.apic_base = data;
 388 }
 389 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 390
 391 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 392 {
 393         WARN_ON(vcpu->arch.exception.pending);
 394         vcpu->arch.exception.pending = true;
 395         vcpu->arch.exception.has_error_code = false;
 396         vcpu->arch.exception.nr = nr;
 397 }
 398 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 399
 400 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 401                            u32 error_code)
 402 {
 403         ++vcpu->stat.pf_guest;
 404         if (vcpu->arch.exception.pending) {
 405                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 406                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 407                                         " double fault 0x%lx\n", addr);
 408                         vcpu->arch.exception.nr = DF_VECTOR;
 409                         vcpu->arch.exception.error_code = 0;
 410                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 411                         /* triple fault -> shutdown */
 412                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 413                 }
 414                 return;
 415         }
 416         vcpu->arch.cr2 = addr;
 417         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 418 }
 419
 420 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 421 {
 422         vcpu->arch.nmi_pending = 1;
 423 }
 424 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 425
 426 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 427 {
 428         WARN_ON(vcpu->arch.exception.pending);
 429         vcpu->arch.exception.pending = true;
 430         vcpu->arch.exception.has_error_code = true;
 431         vcpu->arch.exception.nr = nr;
 432         vcpu->arch.exception.error_code = error_code;
 433 }
 434 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 435
 436 static void __queue_exception(struct kvm_vcpu *vcpu)
 437 {
 438         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 439                                      vcpu->arch.exception.has_error_code,
 440                                      vcpu->arch.exception.error_code);
 441 }
 442
 443 /*
 444  * Load the pae pdptrs.  Return true is they are all valid.
 445  */
 446 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 447 {
 448         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 449         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 450         int i;
 451         int ret;
 452         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 453
 454         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 455                                   offset * sizeof(u64), sizeof(pdpte));
 456         if (ret < 0) {
 457                 ret = 0;
 458                 goto out;
 459         }
 460         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 461                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 462                         ret = 0;
 463                         goto out;
 464                 }
 465         }
 466         ret = 1;
 467
 468         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 469 out:
 470
 471         return ret;
 472 }
 473 EXPORT_SYMBOL_GPL(load_pdptrs);
 474
 475 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 476 {
 477         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 478         bool changed = true;
 479         int r;
 480
 481         if (is_long_mode(vcpu) || !is_pae(vcpu))
 482                 return false;
 483
 484         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 485         if (r < 0)
 486                 goto out;
 487         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 488 out:
 489
 490         return changed;
 491 }
 492
 493 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 494 {
 495         if (cr0 & CR0_RESERVED_BITS) {
 496                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 497                        cr0, vcpu->arch.cr0);
 498                 kvm_inject_gp(vcpu, 0);
 499                 return;
 500         }
 501
 502         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 503                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 504                 kvm_inject_gp(vcpu, 0);
 505                 return;
 506         }
 507
 508         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 509                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 510                        "and a clear PE flag\n");
 511                 kvm_inject_gp(vcpu, 0);
 512                 return;
 513         }
 514
 515         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 516 #ifdef CONFIG_X86_64
 517                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 518                         int cs_db, cs_l;
 519
 520                         if (!is_pae(vcpu)) {
 521                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 522                                        "in long mode while PAE is disabled\n");
 523                                 kvm_inject_gp(vcpu, 0);
 524                                 return;
 525                         }
 526                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 527                         if (cs_l) {
 528                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 529                                        "in long mode while CS.L == 1\n");
 530                                 kvm_inject_gp(vcpu, 0);
 531                                 return;
 532
 533                         }
 534                 } else
 535 #endif
 536                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 537                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 538                                "reserved bits\n");
 539                         kvm_inject_gp(vcpu, 0);
 540                         return;
 541                 }
 542
 543         }
 544
 545         kvm_x86_ops->set_cr0(vcpu, cr0);
 546         vcpu->arch.cr0 = cr0;
 547
 548         kvm_mmu_reset_context(vcpu);
 549         return;
 550 }
 551 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 552
 553 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 554 {
 555         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 556         KVMTRACE_1D(LMSW, vcpu,
 557                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 558                     handler);
 559 }
 560 EXPORT_SYMBOL_GPL(kvm_lmsw);
 561
 562 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 563 {
 564         if (cr4 & CR4_RESERVED_BITS) {
 565                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 566                 kvm_inject_gp(vcpu, 0);
 567                 return;
 568         }
 569
 570         if (is_long_mode(vcpu)) {
 571                 if (!(cr4 & X86_CR4_PAE)) {
 572                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 573                                "in long mode\n");
 574                         kvm_inject_gp(vcpu, 0);
 575                         return;
 576                 }
 577         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 578                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 579                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 580                 kvm_inject_gp(vcpu, 0);
 581                 return;
 582         }
 583
 584         if (cr4 & X86_CR4_VMXE) {
 585                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 586                 kvm_inject_gp(vcpu, 0);
 587                 return;
 588         }
 589         kvm_x86_ops->set_cr4(vcpu, cr4);
 590         vcpu->arch.cr4 = cr4;
 591         kvm_mmu_reset_context(vcpu);
 592 }
 593 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 594
 595 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 596 {
 597         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 598                 kvm_mmu_sync_roots(vcpu);
 599                 kvm_mmu_flush_tlb(vcpu);
 600                 return;
 601         }
 602
 603         if (is_long_mode(vcpu)) {
 604                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 605                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 606                         kvm_inject_gp(vcpu, 0);
 607                         return;
 608                 }
 609         } else {
 610                 if (is_pae(vcpu)) {
 611                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 612                                 printk(KERN_DEBUG
 613                                        "set_cr3: #GP, reserved bits\n");
 614                                 kvm_inject_gp(vcpu, 0);
 615                                 return;
 616                         }
 617                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 618                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 619                                        "reserved bits\n");
 620                                 kvm_inject_gp(vcpu, 0);
 621                                 return;
 622                         }
 623                 }
 624                 /*
 625                  * We don't check reserved bits in nonpae mode, because
 626                  * this isn't enforced, and VMware depends on this.
 627                  */
 628         }
 629
 630         /*
 631          * Does the new cr3 value map to physical memory? (Note, we
 632          * catch an invalid cr3 even in real-mode, because it would
 633          * cause trouble later on when we turn on paging anyway.)
 634          *
 635          * A real CPU would silently accept an invalid cr3 and would
 636          * attempt to use it - with largely undefined (and often hard
 637          * to debug) behavior on the guest side.
 638          */
 639         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 640                 kvm_inject_gp(vcpu, 0);
 641         else {
 642                 vcpu->arch.cr3 = cr3;
 643                 vcpu->arch.mmu.new_cr3(vcpu);
 644         }
 645 }
 646 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 647
 648 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 649 {
 650         if (cr8 & CR8_RESERVED_BITS) {
 651                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 652                 kvm_inject_gp(vcpu, 0);
 653                 return;
 654         }
 655         if (irqchip_in_kernel(vcpu->kvm))
 656                 kvm_lapic_set_tpr(vcpu, cr8);
 657         else
 658                 vcpu->arch.cr8 = cr8;
 659 }
 660 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 661
 662 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 663 {
 664         if (irqchip_in_kernel(vcpu->kvm))
 665                 return kvm_lapic_get_cr8(vcpu);
 666         else
 667                 return vcpu->arch.cr8;
 668 }
 669 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 670
 671 /*
 672  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 673  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 674  *
 675  * This list is modified at module load time to reflect the
 676  * capabilities of the host cpu.
 677  */
 678 static u32 msrs_to_save[] = {
 679         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 680         MSR_K6_STAR,
 681 #ifdef CONFIG_X86_64
 682         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 683 #endif
 684         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 685         MSR_IA32_PERF_STATUS,
 686 };
 687
 688 static unsigned num_msrs_to_save;
 689
 690 static u32 emulated_msrs[] = {
 691         MSR_IA32_MISC_ENABLE,
 692 };
 693
 694 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 695 {
 696         if (efer & efer_reserved_bits) {
 697                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 698                        efer);
 699                 kvm_inject_gp(vcpu, 0);
 700                 return;
 701         }
 702
 703         if (is_paging(vcpu)
 704             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 705                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 706                 kvm_inject_gp(vcpu, 0);
 707                 return;
 708         }
 709
 710         kvm_x86_ops->set_efer(vcpu, efer);
 711
 712         efer &= ~EFER_LMA;
 713         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 714
 715         vcpu->arch.shadow_efer = efer;
 716 }
 717
 718 void kvm_enable_efer_bits(u64 mask)
 719 {
 720        efer_reserved_bits &= ~mask;
 721 }
 722 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 723
 724
 725 /*
 726  * Writes msr value into into the appropriate "register".
 727  * Returns 0 on success, non-0 otherwise.
 728  * Assumes vcpu_load() was already called.
 729  */
 730 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 731 {
 732         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 733 }
 734
 735 /*
 736  * Adapt set_msr() to msr_io()'s calling convention
 737  */
 738 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 739 {
 740         return kvm_set_msr(vcpu, index, *data);
 741 }
 742
 743 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 744 {
 745         static int version;
 746         struct pvclock_wall_clock wc;
 747         struct timespec now, sys, boot;
 748
 749         if (!wall_clock)
 750                 return;
 751
 752         version++;
 753
 754         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 755
 756         /*
 757          * The guest calculates current wall clock time by adding
 758          * system time (updated by kvm_write_guest_time below) to the
 759          * wall clock specified here.  guest system time equals host
 760          * system time for us, thus we must fill in host boot time here.
 761          */
 762         now = current_kernel_time();
 763         ktime_get_ts(&sys);
 764         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 765
 766         wc.sec = boot.tv_sec;
 767         wc.nsec = boot.tv_nsec;
 768         wc.version = version;
 769
 770         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 771
 772         version++;
 773         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 774 }
 775
 776 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 777 {
 778         uint32_t quotient, remainder;
 779
 780         /* Don't try to replace with do_div(), this one calculates
 781          * "(dividend << 32) / divisor" */
 782         __asm__ ( "divl %4"
 783                   : "=a" (quotient), "=d" (remainder)
 784                   : "0" (0), "1" (dividend), "r" (divisor) );
 785         return quotient;
 786 }
 787
 788 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 789 {
 790         uint64_t nsecs = 1000000000LL;
 791         int32_t  shift = 0;
 792         uint64_t tps64;
 793         uint32_t tps32;
 794
 795         tps64 = tsc_khz * 1000LL;
 796         while (tps64 > nsecs*2) {
 797                 tps64 >>= 1;
 798                 shift--;
 799         }
 800
 801         tps32 = (uint32_t)tps64;
 802         while (tps32 <= (uint32_t)nsecs) {
 803                 tps32 <<= 1;
 804                 shift++;
 805         }
 806
 807         hv_clock->tsc_shift = shift;
 808         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 809
 810         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 811                  __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
 812                  hv_clock->tsc_to_system_mul);
 813 }
 814
 815 static void kvm_write_guest_time(struct kvm_vcpu *v)
 816 {
 817         struct timespec ts;
 818         unsigned long flags;
 819         struct kvm_vcpu_arch *vcpu = &v->arch;
 820         void *shared_kaddr;
 821
 822         if ((!vcpu->time_page))
 823                 return;
 824
 825         if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
 826                 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
 827                 vcpu->hv_clock_tsc_khz = tsc_khz;
 828         }
 829
 830         /* Keep irq disabled to prevent changes to the clock */
 831         local_irq_save(flags);
 832         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 833                           &vcpu->hv_clock.tsc_timestamp);
 834         ktime_get_ts(&ts);
 835         local_irq_restore(flags);
 836
 837         /* With all the info we got, fill in the values */
 838
 839         vcpu->hv_clock.system_time = ts.tv_nsec +
 840                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 841         /*
 842          * The interface expects us to write an even number signaling that the
 843          * update is finished. Since the guest won't see the intermediate
 844          * state, we just increase by 2 at the end.
 845          */
 846         vcpu->hv_clock.version += 2;
 847
 848         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 849
 850         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 851                sizeof(vcpu->hv_clock));
 852
 853         kunmap_atomic(shared_kaddr, KM_USER0);
 854
 855         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 856 }
 857
 858 static bool msr_mtrr_valid(unsigned msr)
 859 {
 860         switch (msr) {
 861         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 862         case MSR_MTRRfix64K_00000:
 863         case MSR_MTRRfix16K_80000:
 864         case MSR_MTRRfix16K_A0000:
 865         case MSR_MTRRfix4K_C0000:
 866         case MSR_MTRRfix4K_C8000:
 867         case MSR_MTRRfix4K_D0000:
 868         case MSR_MTRRfix4K_D8000:
 869         case MSR_MTRRfix4K_E0000:
 870         case MSR_MTRRfix4K_E8000:
 871         case MSR_MTRRfix4K_F0000:
 872         case MSR_MTRRfix4K_F8000:
 873         case MSR_MTRRdefType:
 874         case MSR_IA32_CR_PAT:
 875                 return true;
 876         case 0x2f8:
 877                 return true;
 878         }
 879         return false;
 880 }
 881
 882 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 883 {
 884         if (!msr_mtrr_valid(msr))
 885                 return 1;
 886
 887         vcpu->arch.mtrr[msr - 0x200] = data;
 888         return 0;
 889 }
 890
 891 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 892 {
 893         switch (msr) {
 894         case MSR_EFER:
 895                 set_efer(vcpu, data);
 896                 break;
 897         case MSR_IA32_MC0_STATUS:
 898                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 899                        __func__, data);
 900                 break;
 901         case MSR_IA32_MCG_STATUS:
 902                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 903                         __func__, data);
 904                 break;
 905         case MSR_IA32_MCG_CTL:
 906                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 907                         __func__, data);
 908                 break;
 909         case MSR_IA32_DEBUGCTLMSR:
 910                 if (!data) {
 911                         /* We support the non-activated case already */
 912                         break;
 913                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 914                         /* Values other than LBR and BTF are vendor-specific,
 915                            thus reserved and should throw a #GP */
 916                         return 1;
 917                 }
 918                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 919                         __func__, data);
 920                 break;
 921         case MSR_IA32_UCODE_REV:
 922         case MSR_IA32_UCODE_WRITE:
 923                 break;
 924         case 0x200 ... 0x2ff:
 925                 return set_msr_mtrr(vcpu, msr, data);
 926         case MSR_IA32_APICBASE:
 927                 kvm_set_apic_base(vcpu, data);
 928                 break;
 929         case MSR_IA32_MISC_ENABLE:
 930                 vcpu->arch.ia32_misc_enable_msr = data;
 931                 break;
 932         case MSR_KVM_WALL_CLOCK:
 933                 vcpu->kvm->arch.wall_clock = data;
 934                 kvm_write_wall_clock(vcpu->kvm, data);
 935                 break;
 936         case MSR_KVM_SYSTEM_TIME: {
 937                 if (vcpu->arch.time_page) {
 938                         kvm_release_page_dirty(vcpu->arch.time_page);
 939                         vcpu->arch.time_page = NULL;
 940                 }
 941
 942                 vcpu->arch.time = data;
 943
 944                 /* we verify if the enable bit is set... */
 945                 if (!(data & 1))
 946                         break;
 947
 948                 /* ...but clean it before doing the actual write */
 949                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 950
 951                 vcpu->arch.time_page =
 952                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 953
 954                 if (is_error_page(vcpu->arch.time_page)) {
 955                         kvm_release_page_clean(vcpu->arch.time_page);
 956                         vcpu->arch.time_page = NULL;
 957                 }
 958
 959                 kvm_write_guest_time(vcpu);
 960                 break;
 961         }
 962         default:
 963                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 964                 return 1;
 965         }
 966         return 0;
 967 }
 968 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 969
 970
 971 /*
 972  * Reads an msr value (of 'msr_index') into 'pdata'.
 973  * Returns 0 on success, non-0 otherwise.
 974  * Assumes vcpu_load() was already called.
 975  */
 976 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 977 {
 978         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 979 }
 980
 981 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 982 {
 983         if (!msr_mtrr_valid(msr))
 984                 return 1;
 985
 986         *pdata = vcpu->arch.mtrr[msr - 0x200];
 987         return 0;
 988 }
 989
 990 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 991 {
 992         u64 data;
 993
 994         switch (msr) {
 995         case 0xc0010010: /* SYSCFG */
 996         case 0xc0010015: /* HWCR */
 997         case MSR_IA32_PLATFORM_ID:
 998         case MSR_IA32_P5_MC_ADDR:
 999         case MSR_IA32_P5_MC_TYPE:
1000         case MSR_IA32_MC0_CTL:
1001         case MSR_IA32_MCG_STATUS:
1002         case MSR_IA32_MCG_CAP:
1003         case MSR_IA32_MCG_CTL:
1004         case MSR_IA32_MC0_MISC:
1005         case MSR_IA32_MC0_MISC+4:
1006         case MSR_IA32_MC0_MISC+8:
1007         case MSR_IA32_MC0_MISC+12:
1008         case MSR_IA32_MC0_MISC+16:
1009         case MSR_IA32_MC0_MISC+20:
1010         case MSR_IA32_UCODE_REV:
1011         case MSR_IA32_EBL_CR_POWERON:
1012         case MSR_IA32_DEBUGCTLMSR:
1013         case MSR_IA32_LASTBRANCHFROMIP:
1014         case MSR_IA32_LASTBRANCHTOIP:
1015         case MSR_IA32_LASTINTFROMIP:
1016         case MSR_IA32_LASTINTTOIP:
1017                 data = 0;
1018                 break;
1019         case MSR_MTRRcap:
1020                 data = 0x500 | KVM_NR_VAR_MTRR;
1021                 break;
1022         case 0x200 ... 0x2ff:
1023                 return get_msr_mtrr(vcpu, msr, pdata);
1024         case 0xcd: /* fsb frequency */
1025                 data = 3;
1026                 break;
1027         case MSR_IA32_APICBASE:
1028                 data = kvm_get_apic_base(vcpu);
1029                 break;
1030         case MSR_IA32_MISC_ENABLE:
1031                 data = vcpu->arch.ia32_misc_enable_msr;
1032                 break;
1033         case MSR_IA32_PERF_STATUS:
1034                 /* TSC increment by tick */
1035                 data = 1000ULL;
1036                 /* CPU multiplier */
1037                 data |= (((uint64_t)4ULL) << 40);
1038                 break;
1039         case MSR_EFER:
1040                 data = vcpu->arch.shadow_efer;
1041                 break;
1042         case MSR_KVM_WALL_CLOCK:
1043                 data = vcpu->kvm->arch.wall_clock;
1044                 break;
1045         case MSR_KVM_SYSTEM_TIME:
1046                 data = vcpu->arch.time;
1047                 break;
1048         default:
1049                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1050                 return 1;
1051         }
1052         *pdata = data;
1053         return 0;
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1056
1057 /*
1058  * Read or write a bunch of msrs. All parameters are kernel addresses.
1059  *
1060  * @return number of msrs set successfully.
1061  */
1062 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1063                     struct kvm_msr_entry *entries,
1064                     int (*do_msr)(struct kvm_vcpu *vcpu,
1065                                   unsigned index, u64 *data))
1066 {
1067         int i;
1068
1069         vcpu_load(vcpu);
1070
1071         down_read(&vcpu->kvm->slots_lock);
1072         for (i = 0; i < msrs->nmsrs; ++i)
1073                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1074                         break;
1075         up_read(&vcpu->kvm->slots_lock);
1076
1077         vcpu_put(vcpu);
1078
1079         return i;
1080 }
1081
1082 /*
1083  * Read or write a bunch of msrs. Parameters are user addresses.
1084  *
1085  * @return number of msrs set successfully.
1086  */
1087 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1088                   int (*do_msr)(struct kvm_vcpu *vcpu,
1089                                 unsigned index, u64 *data),
1090                   int writeback)
1091 {
1092         struct kvm_msrs msrs;
1093         struct kvm_msr_entry *entries;
1094         int r, n;
1095         unsigned size;
1096
1097         r = -EFAULT;
1098         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1099                 goto out;
1100
1101         r = -E2BIG;
1102         if (msrs.nmsrs >= MAX_IO_MSRS)
1103                 goto out;
1104
1105         r = -ENOMEM;
1106         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1107         entries = vmalloc(size);
1108         if (!entries)
1109                 goto out;
1110
1111         r = -EFAULT;
1112         if (copy_from_user(entries, user_msrs->entries, size))
1113                 goto out_free;
1114
1115         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1116         if (r < 0)
1117                 goto out_free;
1118
1119         r = -EFAULT;
1120         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1121                 goto out_free;
1122
1123         r = n;
1124
1125 out_free:
1126         vfree(entries);
1127 out:
1128         return r;
1129 }
1130
1131 int kvm_dev_ioctl_check_extension(long ext)
1132 {
1133         int r;
1134
1135         switch (ext) {
1136         case KVM_CAP_IRQCHIP:
1137         case KVM_CAP_HLT:
1138         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1139         case KVM_CAP_USER_MEMORY:
1140         case KVM_CAP_SET_TSS_ADDR:
1141         case KVM_CAP_EXT_CPUID:
1142         case KVM_CAP_CLOCKSOURCE:
1143         case KVM_CAP_PIT:
1144         case KVM_CAP_NOP_IO_DELAY:
1145         case KVM_CAP_MP_STATE:
1146         case KVM_CAP_SYNC_MMU:
1147                 r = 1;
1148                 break;
1149         case KVM_CAP_COALESCED_MMIO:
1150                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1151                 break;
1152         case KVM_CAP_VAPIC:
1153                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1154                 break;
1155         case KVM_CAP_NR_VCPUS:
1156                 r = KVM_MAX_VCPUS;
1157                 break;
1158         case KVM_CAP_NR_MEMSLOTS:
1159                 r = KVM_MEMORY_SLOTS;
1160                 break;
1161         case KVM_CAP_PV_MMU:
1162                 r = !tdp_enabled;
1163                 break;
1164         case KVM_CAP_IOMMU:
1165                 r = intel_iommu_found();
1166                 break;
1167         default:
1168                 r = 0;
1169                 break;
1170         }
1171         return r;
1172
1173 }
1174
1175 long kvm_arch_dev_ioctl(struct file *filp,
1176                         unsigned int ioctl, unsigned long arg)
1177 {
1178         void __user *argp = (void __user *)arg;
1179         long r;
1180
1181         switch (ioctl) {
1182         case KVM_GET_MSR_INDEX_LIST: {
1183                 struct kvm_msr_list __user *user_msr_list = argp;
1184                 struct kvm_msr_list msr_list;
1185                 unsigned n;
1186
1187                 r = -EFAULT;
1188                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1189                         goto out;
1190                 n = msr_list.nmsrs;
1191                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1192                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1193                         goto out;
1194                 r = -E2BIG;
1195                 if (n < num_msrs_to_save)
1196                         goto out;
1197                 r = -EFAULT;
1198                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1199                                  num_msrs_to_save * sizeof(u32)))
1200                         goto out;
1201                 if (copy_to_user(user_msr_list->indices
1202                                  + num_msrs_to_save * sizeof(u32),
1203                                  &emulated_msrs,
1204                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1205                         goto out;
1206                 r = 0;
1207                 break;
1208         }
1209         case KVM_GET_SUPPORTED_CPUID: {
1210                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1211                 struct kvm_cpuid2 cpuid;
1212
1213                 r = -EFAULT;
1214                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1215                         goto out;
1216                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1217                         cpuid_arg->entries);
1218                 if (r)
1219                         goto out;
1220
1221                 r = -EFAULT;
1222                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1223                         goto out;
1224                 r = 0;
1225                 break;
1226         }
1227         default:
1228                 r = -EINVAL;
1229         }
1230 out:
1231         return r;
1232 }
1233
1234 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1235 {
1236         kvm_x86_ops->vcpu_load(vcpu, cpu);
1237         kvm_write_guest_time(vcpu);
1238 }
1239
1240 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1241 {
1242         kvm_x86_ops->vcpu_put(vcpu);
1243         kvm_put_guest_fpu(vcpu);
1244 }
1245
1246 static int is_efer_nx(void)
1247 {
1248         u64 efer;
1249
1250         rdmsrl(MSR_EFER, efer);
1251         return efer & EFER_NX;
1252 }
1253
1254 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1255 {
1256         int i;
1257         struct kvm_cpuid_entry2 *e, *entry;
1258
1259         entry = NULL;
1260         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1261                 e = &vcpu->arch.cpuid_entries[i];
1262                 if (e->function == 0x80000001) {
1263                         entry = e;
1264                         break;
1265                 }
1266         }
1267         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1268                 entry->edx &= ~(1 << 20);
1269                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1270         }
1271 }
1272
1273 /* when an old userspace process fills a new kernel module */
1274 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1275                                     struct kvm_cpuid *cpuid,
1276                                     struct kvm_cpuid_entry __user *entries)
1277 {
1278         int r, i;
1279         struct kvm_cpuid_entry *cpuid_entries;
1280
1281         r = -E2BIG;
1282         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1283                 goto out;
1284         r = -ENOMEM;
1285         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1286         if (!cpuid_entries)
1287                 goto out;
1288         r = -EFAULT;
1289         if (copy_from_user(cpuid_entries, entries,
1290                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1291                 goto out_free;
1292         for (i = 0; i < cpuid->nent; i++) {
1293                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1294                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1295                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1296                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1297                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1298                 vcpu->arch.cpuid_entries[i].index = 0;
1299                 vcpu->arch.cpuid_entries[i].flags = 0;
1300                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1301                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1302                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1303         }
1304         vcpu->arch.cpuid_nent = cpuid->nent;
1305         cpuid_fix_nx_cap(vcpu);
1306         r = 0;
1307
1308 out_free:
1309         vfree(cpuid_entries);
1310 out:
1311         return r;
1312 }
1313
1314 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1315                                     struct kvm_cpuid2 *cpuid,
1316                                     struct kvm_cpuid_entry2 __user *entries)
1317 {
1318         int r;
1319
1320         r = -E2BIG;
1321         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1322                 goto out;
1323         r = -EFAULT;
1324         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1325                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1326                 goto out;
1327         vcpu->arch.cpuid_nent = cpuid->nent;
1328         return 0;
1329
1330 out:
1331         return r;
1332 }
1333
1334 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1335                                     struct kvm_cpuid2 *cpuid,
1336                                     struct kvm_cpuid_entry2 __user *entries)
1337 {
1338         int r;
1339
1340         r = -E2BIG;
1341         if (cpuid->nent < vcpu->arch.cpuid_nent)
1342                 goto out;
1343         r = -EFAULT;
1344         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1345                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1346                 goto out;
1347         return 0;
1348
1349 out:
1350         cpuid->nent = vcpu->arch.cpuid_nent;
1351         return r;
1352 }
1353
1354 static inline u32 bit(int bitno)
1355 {
1356         return 1 << (bitno & 31);
1357 }
1358
1359 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1360                           u32 index)
1361 {
1362         entry->function = function;
1363         entry->index = index;
1364         cpuid_count(entry->function, entry->index,
1365                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1366         entry->flags = 0;
1367 }
1368
1369 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1370                          u32 index, int *nent, int maxnent)
1371 {
1372         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1373                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1374                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1375                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1376                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1377                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1378                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1379                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1380                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1381                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1382         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1383                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1384                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1385                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1386                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1387                 bit(X86_FEATURE_PGE) |
1388                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1389                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1390                 bit(X86_FEATURE_SYSCALL) |
1391                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1392 #ifdef CONFIG_X86_64
1393                 bit(X86_FEATURE_LM) |
1394 #endif
1395                 bit(X86_FEATURE_MMXEXT) |
1396                 bit(X86_FEATURE_3DNOWEXT) |
1397                 bit(X86_FEATURE_3DNOW);
1398         const u32 kvm_supported_word3_x86_features =
1399                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1400         const u32 kvm_supported_word6_x86_features =
1401                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1402
1403         /* all func 2 cpuid_count() should be called on the same cpu */
1404         get_cpu();
1405         do_cpuid_1_ent(entry, function, index);
1406         ++*nent;
1407
1408         switch (function) {
1409         case 0:
1410                 entry->eax = min(entry->eax, (u32)0xb);
1411                 break;
1412         case 1:
1413                 entry->edx &= kvm_supported_word0_x86_features;
1414                 entry->ecx &= kvm_supported_word3_x86_features;
1415                 break;
1416         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1417          * may return different values. This forces us to get_cpu() before
1418          * issuing the first command, and also to emulate this annoying behavior
1419          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1420         case 2: {
1421                 int t, times = entry->eax & 0xff;
1422
1423                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1424                 for (t = 1; t < times && *nent < maxnent; ++t) {
1425                         do_cpuid_1_ent(&entry[t], function, 0);
1426                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1427                         ++*nent;
1428                 }
1429                 break;
1430         }
1431         /* function 4 and 0xb have additional index. */
1432         case 4: {
1433                 int i, cache_type;
1434
1435                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1436                 /* read more entries until cache_type is zero */
1437                 for (i = 1; *nent < maxnent; ++i) {
1438                         cache_type = entry[i - 1].eax & 0x1f;
1439                         if (!cache_type)
1440                                 break;
1441                         do_cpuid_1_ent(&entry[i], function, i);
1442                         entry[i].flags |=
1443                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1444                         ++*nent;
1445                 }
1446                 break;
1447         }
1448         case 0xb: {
1449                 int i, level_type;
1450
1451                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1452                 /* read more entries until level_type is zero */
1453                 for (i = 1; *nent < maxnent; ++i) {
1454                         level_type = entry[i - 1].ecx & 0xff;
1455                         if (!level_type)
1456                                 break;
1457                         do_cpuid_1_ent(&entry[i], function, i);
1458                         entry[i].flags |=
1459                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1460                         ++*nent;
1461                 }
1462                 break;
1463         }
1464         case 0x80000000:
1465                 entry->eax = min(entry->eax, 0x8000001a);
1466                 break;
1467         case 0x80000001:
1468                 entry->edx &= kvm_supported_word1_x86_features;
1469                 entry->ecx &= kvm_supported_word6_x86_features;
1470                 break;
1471         }
1472         put_cpu();
1473 }
1474
1475 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1476                                     struct kvm_cpuid_entry2 __user *entries)
1477 {
1478         struct kvm_cpuid_entry2 *cpuid_entries;
1479         int limit, nent = 0, r = -E2BIG;
1480         u32 func;
1481
1482         if (cpuid->nent < 1)
1483                 goto out;
1484         r = -ENOMEM;
1485         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1486         if (!cpuid_entries)
1487                 goto out;
1488
1489         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1490         limit = cpuid_entries[0].eax;
1491         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1492                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1493                                 &nent, cpuid->nent);
1494         r = -E2BIG;
1495         if (nent >= cpuid->nent)
1496                 goto out_free;
1497
1498         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1499         limit = cpuid_entries[nent - 1].eax;
1500         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1501                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1502                                &nent, cpuid->nent);
1503         r = -EFAULT;
1504         if (copy_to_user(entries, cpuid_entries,
1505                         nent * sizeof(struct kvm_cpuid_entry2)))
1506                 goto out_free;
1507         cpuid->nent = nent;
1508         r = 0;
1509
1510 out_free:
1511         vfree(cpuid_entries);
1512 out:
1513         return r;
1514 }
1515
1516 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1517                                     struct kvm_lapic_state *s)
1518 {
1519         vcpu_load(vcpu);
1520         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1521         vcpu_put(vcpu);
1522
1523         return 0;
1524 }
1525
1526 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1527                                     struct kvm_lapic_state *s)
1528 {
1529         vcpu_load(vcpu);
1530         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1531         kvm_apic_post_state_restore(vcpu);
1532         vcpu_put(vcpu);
1533
1534         return 0;
1535 }
1536
1537 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1538                                     struct kvm_interrupt *irq)
1539 {
1540         if (irq->irq < 0 || irq->irq >= 256)
1541                 return -EINVAL;
1542         if (irqchip_in_kernel(vcpu->kvm))
1543                 return -ENXIO;
1544         vcpu_load(vcpu);
1545
1546         set_bit(irq->irq, vcpu->arch.irq_pending);
1547         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1548
1549         vcpu_put(vcpu);
1550
1551         return 0;
1552 }
1553
1554 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1555                                            struct kvm_tpr_access_ctl *tac)
1556 {
1557         if (tac->flags)
1558                 return -EINVAL;
1559         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1560         return 0;
1561 }
1562
1563 long kvm_arch_vcpu_ioctl(struct file *filp,
1564                          unsigned int ioctl, unsigned long arg)
1565 {
1566         struct kvm_vcpu *vcpu = filp->private_data;
1567         void __user *argp = (void __user *)arg;
1568         int r;
1569         struct kvm_lapic_state *lapic = NULL;
1570
1571         switch (ioctl) {
1572         case KVM_GET_LAPIC: {
1573                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1574
1575                 r = -ENOMEM;
1576                 if (!lapic)
1577                         goto out;
1578                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1579                 if (r)
1580                         goto out;
1581                 r = -EFAULT;
1582                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1583                         goto out;
1584                 r = 0;
1585                 break;
1586         }
1587         case KVM_SET_LAPIC: {
1588                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1589                 r = -ENOMEM;
1590                 if (!lapic)
1591                         goto out;
1592                 r = -EFAULT;
1593                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1594                         goto out;
1595                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1596                 if (r)
1597                         goto out;
1598                 r = 0;
1599                 break;
1600         }
1601         case KVM_INTERRUPT: {
1602                 struct kvm_interrupt irq;
1603
1604                 r = -EFAULT;
1605                 if (copy_from_user(&irq, argp, sizeof irq))
1606                         goto out;
1607                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1608                 if (r)
1609                         goto out;
1610                 r = 0;
1611                 break;
1612         }
1613         case KVM_SET_CPUID: {
1614                 struct kvm_cpuid __user *cpuid_arg = argp;
1615                 struct kvm_cpuid cpuid;
1616
1617                 r = -EFAULT;
1618                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1619                         goto out;
1620                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1621                 if (r)
1622                         goto out;
1623                 break;
1624         }
1625         case KVM_SET_CPUID2: {
1626                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1627                 struct kvm_cpuid2 cpuid;
1628
1629                 r = -EFAULT;
1630                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1631                         goto out;
1632                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1633                                 cpuid_arg->entries);
1634                 if (r)
1635                         goto out;
1636                 break;
1637         }
1638         case KVM_GET_CPUID2: {
1639                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1640                 struct kvm_cpuid2 cpuid;
1641
1642                 r = -EFAULT;
1643                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1644                         goto out;
1645                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1646                                 cpuid_arg->entries);
1647                 if (r)
1648                         goto out;
1649                 r = -EFAULT;
1650                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1651                         goto out;
1652                 r = 0;
1653                 break;
1654         }
1655         case KVM_GET_MSRS:
1656                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1657                 break;
1658         case KVM_SET_MSRS:
1659                 r = msr_io(vcpu, argp, do_set_msr, 0);
1660                 break;
1661         case KVM_TPR_ACCESS_REPORTING: {
1662                 struct kvm_tpr_access_ctl tac;
1663
1664                 r = -EFAULT;
1665                 if (copy_from_user(&tac, argp, sizeof tac))
1666                         goto out;
1667                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1668                 if (r)
1669                         goto out;
1670                 r = -EFAULT;
1671                 if (copy_to_user(argp, &tac, sizeof tac))
1672                         goto out;
1673                 r = 0;
1674                 break;
1675         };
1676         case KVM_SET_VAPIC_ADDR: {
1677                 struct kvm_vapic_addr va;
1678
1679                 r = -EINVAL;
1680                 if (!irqchip_in_kernel(vcpu->kvm))
1681                         goto out;
1682                 r = -EFAULT;
1683                 if (copy_from_user(&va, argp, sizeof va))
1684                         goto out;
1685                 r = 0;
1686                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1687                 break;
1688         }
1689         default:
1690                 r = -EINVAL;
1691         }
1692 out:
1693         if (lapic)
1694                 kfree(lapic);
1695         return r;
1696 }
1697
1698 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1699 {
1700         int ret;
1701
1702         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1703                 return -1;
1704         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1705         return ret;
1706 }
1707
1708 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1709                                           u32 kvm_nr_mmu_pages)
1710 {
1711         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1712                 return -EINVAL;
1713
1714         down_write(&kvm->slots_lock);
1715
1716         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1717         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1718
1719         up_write(&kvm->slots_lock);
1720         return 0;
1721 }
1722
1723 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1724 {
1725         return kvm->arch.n_alloc_mmu_pages;
1726 }
1727
1728 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1729 {
1730         int i;
1731         struct kvm_mem_alias *alias;
1732
1733         for (i = 0; i < kvm->arch.naliases; ++i) {
1734                 alias = &kvm->arch.aliases[i];
1735                 if (gfn >= alias->base_gfn
1736                     && gfn < alias->base_gfn + alias->npages)
1737                         return alias->target_gfn + gfn - alias->base_gfn;
1738         }
1739         return gfn;
1740 }
1741
1742 /*
1743  * Set a new alias region.  Aliases map a portion of physical memory into
1744  * another portion.  This is useful for memory windows, for example the PC
1745  * VGA region.
1746  */
1747 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1748                                          struct kvm_memory_alias *alias)
1749 {
1750         int r, n;
1751         struct kvm_mem_alias *p;
1752
1753         r = -EINVAL;
1754         /* General sanity checks */
1755         if (alias->memory_size & (PAGE_SIZE - 1))
1756                 goto out;
1757         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1758                 goto out;
1759         if (alias->slot >= KVM_ALIAS_SLOTS)
1760                 goto out;
1761         if (alias->guest_phys_addr + alias->memory_size
1762             < alias->guest_phys_addr)
1763                 goto out;
1764         if (alias->target_phys_addr + alias->memory_size
1765             < alias->target_phys_addr)
1766                 goto out;
1767
1768         down_write(&kvm->slots_lock);
1769         spin_lock(&kvm->mmu_lock);
1770
1771         p = &kvm->arch.aliases[alias->slot];
1772         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1773         p->npages = alias->memory_size >> PAGE_SHIFT;
1774         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1775
1776         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1777                 if (kvm->arch.aliases[n - 1].npages)
1778                         break;
1779         kvm->arch.naliases = n;
1780
1781         spin_unlock(&kvm->mmu_lock);
1782         kvm_mmu_zap_all(kvm);
1783
1784         up_write(&kvm->slots_lock);
1785
1786         return 0;
1787
1788 out:
1789         return r;
1790 }
1791
1792 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1793 {
1794         int r;
1795
1796         r = 0;
1797         switch (chip->chip_id) {
1798         case KVM_IRQCHIP_PIC_MASTER:
1799                 memcpy(&chip->chip.pic,
1800                         &pic_irqchip(kvm)->pics[0],
1801                         sizeof(struct kvm_pic_state));
1802                 break;
1803         case KVM_IRQCHIP_PIC_SLAVE:
1804                 memcpy(&chip->chip.pic,
1805                         &pic_irqchip(kvm)->pics[1],
1806                         sizeof(struct kvm_pic_state));
1807                 break;
1808         case KVM_IRQCHIP_IOAPIC:
1809                 memcpy(&chip->chip.ioapic,
1810                         ioapic_irqchip(kvm),
1811                         sizeof(struct kvm_ioapic_state));
1812                 break;
1813         default:
1814                 r = -EINVAL;
1815                 break;
1816         }
1817         return r;
1818 }
1819
1820 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1821 {
1822         int r;
1823
1824         r = 0;
1825         switch (chip->chip_id) {
1826         case KVM_IRQCHIP_PIC_MASTER:
1827                 memcpy(&pic_irqchip(kvm)->pics[0],
1828                         &chip->chip.pic,
1829                         sizeof(struct kvm_pic_state));
1830                 break;
1831         case KVM_IRQCHIP_PIC_SLAVE:
1832                 memcpy(&pic_irqchip(kvm)->pics[1],
1833                         &chip->chip.pic,
1834                         sizeof(struct kvm_pic_state));
1835                 break;
1836         case KVM_IRQCHIP_IOAPIC:
1837                 memcpy(ioapic_irqchip(kvm),
1838                         &chip->chip.ioapic,
1839                         sizeof(struct kvm_ioapic_state));
1840                 break;
1841         default:
1842                 r = -EINVAL;
1843                 break;
1844         }
1845         kvm_pic_update_irq(pic_irqchip(kvm));
1846         return r;
1847 }
1848
1849 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1850 {
1851         int r = 0;
1852
1853         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1854         return r;
1855 }
1856
1857 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1858 {
1859         int r = 0;
1860
1861         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1862         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1863         return r;
1864 }
1865
1866 /*
1867  * Get (and clear) the dirty memory log for a memory slot.
1868  */
1869 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1870                                       struct kvm_dirty_log *log)
1871 {
1872         int r;
1873         int n;
1874         struct kvm_memory_slot *memslot;
1875         int is_dirty = 0;
1876
1877         down_write(&kvm->slots_lock);
1878
1879         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1880         if (r)
1881                 goto out;
1882
1883         /* If nothing is dirty, don't bother messing with page tables. */
1884         if (is_dirty) {
1885                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1886                 kvm_flush_remote_tlbs(kvm);
1887                 memslot = &kvm->memslots[log->slot];
1888                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1889                 memset(memslot->dirty_bitmap, 0, n);
1890         }
1891         r = 0;
1892 out:
1893         up_write(&kvm->slots_lock);
1894         return r;
1895 }
1896
1897 long kvm_arch_vm_ioctl(struct file *filp,
1898                        unsigned int ioctl, unsigned long arg)
1899 {
1900         struct kvm *kvm = filp->private_data;
1901         void __user *argp = (void __user *)arg;
1902         int r = -EINVAL;
1903         /*
1904          * This union makes it completely explicit to gcc-3.x
1905          * that these two variables' stack usage should be
1906          * combined, not added together.
1907          */
1908         union {
1909                 struct kvm_pit_state ps;
1910                 struct kvm_memory_alias alias;
1911         } u;
1912
1913         switch (ioctl) {
1914         case KVM_SET_TSS_ADDR:
1915                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1916                 if (r < 0)
1917                         goto out;
1918                 break;
1919         case KVM_SET_MEMORY_REGION: {
1920                 struct kvm_memory_region kvm_mem;
1921                 struct kvm_userspace_memory_region kvm_userspace_mem;
1922
1923                 r = -EFAULT;
1924                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1925                         goto out;
1926                 kvm_userspace_mem.slot = kvm_mem.slot;
1927                 kvm_userspace_mem.flags = kvm_mem.flags;
1928                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1929                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1930                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1931                 if (r)
1932                         goto out;
1933                 break;
1934         }
1935         case KVM_SET_NR_MMU_PAGES:
1936                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1937                 if (r)
1938                         goto out;
1939                 break;
1940         case KVM_GET_NR_MMU_PAGES:
1941                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1942                 break;
1943         case KVM_SET_MEMORY_ALIAS:
1944                 r = -EFAULT;
1945                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1946                         goto out;
1947                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1948                 if (r)
1949                         goto out;
1950                 break;
1951         case KVM_CREATE_IRQCHIP:
1952                 r = -ENOMEM;
1953                 kvm->arch.vpic = kvm_create_pic(kvm);
1954                 if (kvm->arch.vpic) {
1955                         r = kvm_ioapic_init(kvm);
1956                         if (r) {
1957                                 kfree(kvm->arch.vpic);
1958                                 kvm->arch.vpic = NULL;
1959                                 goto out;
1960                         }
1961                 } else
1962                         goto out;
1963                 break;
1964         case KVM_CREATE_PIT:
1965                 r = -ENOMEM;
1966                 kvm->arch.vpit = kvm_create_pit(kvm);
1967                 if (kvm->arch.vpit)
1968                         r = 0;
1969                 break;
1970         case KVM_IRQ_LINE: {
1971                 struct kvm_irq_level irq_event;
1972
1973                 r = -EFAULT;
1974                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1975                         goto out;
1976                 if (irqchip_in_kernel(kvm)) {
1977                         mutex_lock(&kvm->lock);
1978                         kvm_set_irq(kvm, irq_event.irq, irq_event.level);
1979                         mutex_unlock(&kvm->lock);
1980                         r = 0;
1981                 }
1982                 break;
1983         }
1984         case KVM_GET_IRQCHIP: {
1985                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1986                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1987
1988                 r = -ENOMEM;
1989                 if (!chip)
1990                         goto out;
1991                 r = -EFAULT;
1992                 if (copy_from_user(chip, argp, sizeof *chip))
1993                         goto get_irqchip_out;
1994                 r = -ENXIO;
1995                 if (!irqchip_in_kernel(kvm))
1996                         goto get_irqchip_out;
1997                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1998                 if (r)
1999                         goto get_irqchip_out;
2000                 r = -EFAULT;
2001                 if (copy_to_user(argp, chip, sizeof *chip))
2002                         goto get_irqchip_out;
2003                 r = 0;
2004         get_irqchip_out:
2005                 kfree(chip);
2006                 if (r)
2007                         goto out;
2008                 break;
2009         }
2010         case KVM_SET_IRQCHIP: {
2011                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2012                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2013
2014                 r = -ENOMEM;
2015                 if (!chip)
2016                         goto out;
2017                 r = -EFAULT;
2018                 if (copy_from_user(chip, argp, sizeof *chip))
2019                         goto set_irqchip_out;
2020                 r = -ENXIO;
2021                 if (!irqchip_in_kernel(kvm))
2022                         goto set_irqchip_out;
2023                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2024                 if (r)
2025                         goto set_irqchip_out;
2026                 r = 0;
2027         set_irqchip_out:
2028                 kfree(chip);
2029                 if (r)
2030                         goto out;
2031                 break;
2032         }
2033         case KVM_ASSIGN_PCI_DEVICE: {
2034                 struct kvm_assigned_pci_dev assigned_dev;
2035
2036                 r = -EFAULT;
2037                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2038                         goto out;
2039                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2040                 if (r)
2041                         goto out;
2042                 break;
2043         }
2044         case KVM_ASSIGN_IRQ: {
2045                 struct kvm_assigned_irq assigned_irq;
2046
2047                 r = -EFAULT;
2048                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2049                         goto out;
2050                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2051                 if (r)
2052                         goto out;
2053                 break;
2054         }
2055         case KVM_GET_PIT: {
2056                 r = -EFAULT;
2057                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2058                         goto out;
2059                 r = -ENXIO;
2060                 if (!kvm->arch.vpit)
2061                         goto out;
2062                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2063                 if (r)
2064                         goto out;
2065                 r = -EFAULT;
2066                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2067                         goto out;
2068                 r = 0;
2069                 break;
2070         }
2071         case KVM_SET_PIT: {
2072                 r = -EFAULT;
2073                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2074                         goto out;
2075                 r = -ENXIO;
2076                 if (!kvm->arch.vpit)
2077                         goto out;
2078                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2079                 if (r)
2080                         goto out;
2081                 r = 0;
2082                 break;
2083         }
2084         default:
2085                 ;
2086         }
2087 out:
2088         return r;
2089 }
2090
2091 static void kvm_init_msr_list(void)
2092 {
2093         u32 dummy[2];
2094         unsigned i, j;
2095
2096         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2097                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2098                         continue;
2099                 if (j < i)
2100                         msrs_to_save[j] = msrs_to_save[i];
2101                 j++;
2102         }
2103         num_msrs_to_save = j;
2104 }
2105
2106 /*
2107  * Only apic need an MMIO device hook, so shortcut now..
2108  */
2109 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2110                                                 gpa_t addr, int len,
2111                                                 int is_write)
2112 {
2113         struct kvm_io_device *dev;
2114
2115         if (vcpu->arch.apic) {
2116                 dev = &vcpu->arch.apic->dev;
2117                 if (dev->in_range(dev, addr, len, is_write))
2118                         return dev;
2119         }
2120         return NULL;
2121 }
2122
2123
2124 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2125                                                 gpa_t addr, int len,
2126                                                 int is_write)
2127 {
2128         struct kvm_io_device *dev;
2129
2130         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2131         if (dev == NULL)
2132                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2133                                           is_write);
2134         return dev;
2135 }
2136
2137 int emulator_read_std(unsigned long addr,
2138                              void *val,
2139                              unsigned int bytes,
2140                              struct kvm_vcpu *vcpu)
2141 {
2142         void *data = val;
2143         int r = X86EMUL_CONTINUE;
2144
2145         while (bytes) {
2146                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2147                 unsigned offset = addr & (PAGE_SIZE-1);
2148                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
2149                 int ret;
2150
2151                 if (gpa == UNMAPPED_GVA) {
2152                         r = X86EMUL_PROPAGATE_FAULT;
2153                         goto out;
2154                 }
2155                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
2156                 if (ret < 0) {
2157                         r = X86EMUL_UNHANDLEABLE;
2158                         goto out;
2159                 }
2160
2161                 bytes -= tocopy;
2162                 data += tocopy;
2163                 addr += tocopy;
2164         }
2165 out:
2166         return r;
2167 }
2168 EXPORT_SYMBOL_GPL(emulator_read_std);
2169
2170 static int emulator_read_emulated(unsigned long addr,
2171                                   void *val,
2172                                   unsigned int bytes,
2173                                   struct kvm_vcpu *vcpu)
2174 {
2175         struct kvm_io_device *mmio_dev;
2176         gpa_t                 gpa;
2177
2178         if (vcpu->mmio_read_completed) {
2179                 memcpy(val, vcpu->mmio_data, bytes);
2180                 vcpu->mmio_read_completed = 0;
2181                 return X86EMUL_CONTINUE;
2182         }
2183
2184         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2185
2186         /* For APIC access vmexit */
2187         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2188                 goto mmio;
2189
2190         if (emulator_read_std(addr, val, bytes, vcpu)
2191                         == X86EMUL_CONTINUE)
2192                 return X86EMUL_CONTINUE;
2193         if (gpa == UNMAPPED_GVA)
2194                 return X86EMUL_PROPAGATE_FAULT;
2195
2196 mmio:
2197         /*
2198          * Is this MMIO handled locally?
2199          */
2200         mutex_lock(&vcpu->kvm->lock);
2201         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2202         if (mmio_dev) {
2203                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2204                 mutex_unlock(&vcpu->kvm->lock);
2205                 return X86EMUL_CONTINUE;
2206         }
2207         mutex_unlock(&vcpu->kvm->lock);
2208
2209         vcpu->mmio_needed = 1;
2210         vcpu->mmio_phys_addr = gpa;
2211         vcpu->mmio_size = bytes;
2212         vcpu->mmio_is_write = 0;
2213
2214         return X86EMUL_UNHANDLEABLE;
2215 }
2216
2217 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2218                           const void *val, int bytes)
2219 {
2220         int ret;
2221
2222         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2223         if (ret < 0)
2224                 return 0;
2225         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2226         return 1;
2227 }
2228
2229 static int emulator_write_emulated_onepage(unsigned long addr,
2230                                            const void *val,
2231                                            unsigned int bytes,
2232                                            struct kvm_vcpu *vcpu)
2233 {
2234         struct kvm_io_device *mmio_dev;
2235         gpa_t                 gpa;
2236
2237         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2238
2239         if (gpa == UNMAPPED_GVA) {
2240                 kvm_inject_page_fault(vcpu, addr, 2);
2241                 return X86EMUL_PROPAGATE_FAULT;
2242         }
2243
2244         /* For APIC access vmexit */
2245         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2246                 goto mmio;
2247
2248         if (emulator_write_phys(vcpu, gpa, val, bytes))
2249                 return X86EMUL_CONTINUE;
2250
2251 mmio:
2252         /*
2253          * Is this MMIO handled locally?
2254          */
2255         mutex_lock(&vcpu->kvm->lock);
2256         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2257         if (mmio_dev) {
2258                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2259                 mutex_unlock(&vcpu->kvm->lock);
2260                 return X86EMUL_CONTINUE;
2261         }
2262         mutex_unlock(&vcpu->kvm->lock);
2263
2264         vcpu->mmio_needed = 1;
2265         vcpu->mmio_phys_addr = gpa;
2266         vcpu->mmio_size = bytes;
2267         vcpu->mmio_is_write = 1;
2268         memcpy(vcpu->mmio_data, val, bytes);
2269
2270         return X86EMUL_CONTINUE;
2271 }
2272
2273 int emulator_write_emulated(unsigned long addr,
2274                                    const void *val,
2275                                    unsigned int bytes,
2276                                    struct kvm_vcpu *vcpu)
2277 {
2278         /* Crossing a page boundary? */
2279         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2280                 int rc, now;
2281
2282                 now = -addr & ~PAGE_MASK;
2283                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2284                 if (rc != X86EMUL_CONTINUE)
2285                         return rc;
2286                 addr += now;
2287                 val += now;
2288                 bytes -= now;
2289         }
2290         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2291 }
2292 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2293
2294 static int emulator_cmpxchg_emulated(unsigned long addr,
2295                                      const void *old,
2296                                      const void *new,
2297                                      unsigned int bytes,
2298                                      struct kvm_vcpu *vcpu)
2299 {
2300         static int reported;
2301
2302         if (!reported) {
2303                 reported = 1;
2304                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2305         }
2306 #ifndef CONFIG_X86_64
2307         /* guests cmpxchg8b have to be emulated atomically */
2308         if (bytes == 8) {
2309                 gpa_t gpa;
2310                 struct page *page;
2311                 char *kaddr;
2312                 u64 val;
2313
2314                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2315
2316                 if (gpa == UNMAPPED_GVA ||
2317                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2318                         goto emul_write;
2319
2320                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2321                         goto emul_write;
2322
2323                 val = *(u64 *)new;
2324
2325                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2326
2327                 kaddr = kmap_atomic(page, KM_USER0);
2328                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2329                 kunmap_atomic(kaddr, KM_USER0);
2330                 kvm_release_page_dirty(page);
2331         }
2332 emul_write:
2333 #endif
2334
2335         return emulator_write_emulated(addr, new, bytes, vcpu);
2336 }
2337
2338 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2339 {
2340         return kvm_x86_ops->get_segment_base(vcpu, seg);
2341 }
2342
2343 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2344 {
2345         kvm_mmu_invlpg(vcpu, address);
2346         return X86EMUL_CONTINUE;
2347 }
2348
2349 int emulate_clts(struct kvm_vcpu *vcpu)
2350 {
2351         KVMTRACE_0D(CLTS, vcpu, handler);
2352         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2353         return X86EMUL_CONTINUE;
2354 }
2355
2356 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2357 {
2358         struct kvm_vcpu *vcpu = ctxt->vcpu;
2359
2360         switch (dr) {
2361         case 0 ... 3:
2362                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2363                 return X86EMUL_CONTINUE;
2364         default:
2365                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2366                 return X86EMUL_UNHANDLEABLE;
2367         }
2368 }
2369
2370 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2371 {
2372         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2373         int exception;
2374
2375         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2376         if (exception) {
2377                 /* FIXME: better handling */
2378                 return X86EMUL_UNHANDLEABLE;
2379         }
2380         return X86EMUL_CONTINUE;
2381 }
2382
2383 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2384 {
2385         u8 opcodes[4];
2386         unsigned long rip = kvm_rip_read(vcpu);
2387         unsigned long rip_linear;
2388
2389         if (!printk_ratelimit())
2390                 return;
2391
2392         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2393
2394         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2395
2396         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2397                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2398 }
2399 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2400
2401 static struct x86_emulate_ops emulate_ops = {
2402         .read_std            = emulator_read_std,
2403         .read_emulated       = emulator_read_emulated,
2404         .write_emulated      = emulator_write_emulated,
2405         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2406 };
2407
2408 static void cache_all_regs(struct kvm_vcpu *vcpu)
2409 {
2410         kvm_register_read(vcpu, VCPU_REGS_RAX);
2411         kvm_register_read(vcpu, VCPU_REGS_RSP);
2412         kvm_register_read(vcpu, VCPU_REGS_RIP);
2413         vcpu->arch.regs_dirty = ~0;
2414 }
2415
2416 int emulate_instruction(struct kvm_vcpu *vcpu,
2417                         struct kvm_run *run,
2418                         unsigned long cr2,
2419                         u16 error_code,
2420                         int emulation_type)
2421 {
2422         int r;
2423         struct decode_cache *c;
2424
2425         kvm_clear_exception_queue(vcpu);
2426         vcpu->arch.mmio_fault_cr2 = cr2;
2427         /*
2428          * TODO: fix x86_emulate.c to use guest_read/write_register
2429          * instead of direct ->regs accesses, can save hundred cycles
2430          * on Intel for instructions that don't read/change RSP, for
2431          * for example.
2432          */
2433         cache_all_regs(vcpu);
2434
2435         vcpu->mmio_is_write = 0;
2436         vcpu->arch.pio.string = 0;
2437
2438         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2439                 int cs_db, cs_l;
2440                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2441
2442                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2443                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2444                 vcpu->arch.emulate_ctxt.mode =
2445                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2446                         ? X86EMUL_MODE_REAL : cs_l
2447                         ? X86EMUL_MODE_PROT64 : cs_db
2448                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2449
2450                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2451
2452                 /* Reject the instructions other than VMCALL/VMMCALL when
2453                  * try to emulate invalid opcode */
2454                 c = &vcpu->arch.emulate_ctxt.decode;
2455                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2456                     (!(c->twobyte && c->b == 0x01 &&
2457                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2458                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2459                         return EMULATE_FAIL;
2460
2461                 ++vcpu->stat.insn_emulation;
2462                 if (r)  {
2463                         ++vcpu->stat.insn_emulation_fail;
2464                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2465                                 return EMULATE_DONE;
2466                         return EMULATE_FAIL;
2467                 }
2468         }
2469
2470         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2471
2472         if (vcpu->arch.pio.string)
2473                 return EMULATE_DO_MMIO;
2474
2475         if ((r || vcpu->mmio_is_write) && run) {
2476                 run->exit_reason = KVM_EXIT_MMIO;
2477                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2478                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2479                 run->mmio.len = vcpu->mmio_size;
2480                 run->mmio.is_write = vcpu->mmio_is_write;
2481         }
2482
2483         if (r) {
2484                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2485                         return EMULATE_DONE;
2486                 if (!vcpu->mmio_needed) {
2487                         kvm_report_emulation_failure(vcpu, "mmio");
2488                         return EMULATE_FAIL;
2489                 }
2490                 return EMULATE_DO_MMIO;
2491         }
2492
2493         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2494
2495         if (vcpu->mmio_is_write) {
2496                 vcpu->mmio_needed = 0;
2497                 return EMULATE_DO_MMIO;
2498         }
2499
2500         return EMULATE_DONE;
2501 }
2502 EXPORT_SYMBOL_GPL(emulate_instruction);
2503
2504 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2505 {
2506         int i;
2507
2508         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2509                 if (vcpu->arch.pio.guest_pages[i]) {
2510                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2511                         vcpu->arch.pio.guest_pages[i] = NULL;
2512                 }
2513 }
2514
2515 static int pio_copy_data(struct kvm_vcpu *vcpu)
2516 {
2517         void *p = vcpu->arch.pio_data;
2518         void *q;
2519         unsigned bytes;
2520         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2521
2522         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2523                  PAGE_KERNEL);
2524         if (!q) {
2525                 free_pio_guest_pages(vcpu);
2526                 return -ENOMEM;
2527         }
2528         q += vcpu->arch.pio.guest_page_offset;
2529         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2530         if (vcpu->arch.pio.in)
2531                 memcpy(q, p, bytes);
2532         else
2533                 memcpy(p, q, bytes);
2534         q -= vcpu->arch.pio.guest_page_offset;
2535         vunmap(q);
2536         free_pio_guest_pages(vcpu);
2537         return 0;
2538 }
2539
2540 int complete_pio(struct kvm_vcpu *vcpu)
2541 {
2542         struct kvm_pio_request *io = &vcpu->arch.pio;
2543         long delta;
2544         int r;
2545         unsigned long val;
2546
2547         if (!io->string) {
2548                 if (io->in) {
2549                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2550                         memcpy(&val, vcpu->arch.pio_data, io->size);
2551                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2552                 }
2553         } else {
2554                 if (io->in) {
2555                         r = pio_copy_data(vcpu);
2556                         if (r)
2557                                 return r;
2558                 }
2559
2560                 delta = 1;
2561                 if (io->rep) {
2562                         delta *= io->cur_count;
2563                         /*
2564                          * The size of the register should really depend on
2565                          * current address size.
2566                          */
2567                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2568                         val -= delta;
2569                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2570                 }
2571                 if (io->down)
2572                         delta = -delta;
2573                 delta *= io->size;
2574                 if (io->in) {
2575                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2576                         val += delta;
2577                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2578                 } else {
2579                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2580                         val += delta;
2581                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2582                 }
2583         }
2584
2585         io->count -= io->cur_count;
2586         io->cur_count = 0;
2587
2588         return 0;
2589 }
2590
2591 static void kernel_pio(struct kvm_io_device *pio_dev,
2592                        struct kvm_vcpu *vcpu,
2593                        void *pd)
2594 {
2595         /* TODO: String I/O for in kernel device */
2596
2597         mutex_lock(&vcpu->kvm->lock);
2598         if (vcpu->arch.pio.in)
2599                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2600                                   vcpu->arch.pio.size,
2601                                   pd);
2602         else
2603                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2604                                    vcpu->arch.pio.size,
2605                                    pd);
2606         mutex_unlock(&vcpu->kvm->lock);
2607 }
2608
2609 static void pio_string_write(struct kvm_io_device *pio_dev,
2610                              struct kvm_vcpu *vcpu)
2611 {
2612         struct kvm_pio_request *io = &vcpu->arch.pio;
2613         void *pd = vcpu->arch.pio_data;
2614         int i;
2615
2616         mutex_lock(&vcpu->kvm->lock);
2617         for (i = 0; i < io->cur_count; i++) {
2618                 kvm_iodevice_write(pio_dev, io->port,
2619                                    io->size,
2620                                    pd);
2621                 pd += io->size;
2622         }
2623         mutex_unlock(&vcpu->kvm->lock);
2624 }
2625
2626 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2627                                                gpa_t addr, int len,
2628                                                int is_write)
2629 {
2630         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2631 }
2632
2633 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2634                   int size, unsigned port)
2635 {
2636         struct kvm_io_device *pio_dev;
2637         unsigned long val;
2638
2639         vcpu->run->exit_reason = KVM_EXIT_IO;
2640         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2641         vcpu->run->io.size = vcpu->arch.pio.size = size;
2642         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2643         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2644         vcpu->run->io.port = vcpu->arch.pio.port = port;
2645         vcpu->arch.pio.in = in;
2646         vcpu->arch.pio.string = 0;
2647         vcpu->arch.pio.down = 0;
2648         vcpu->arch.pio.guest_page_offset = 0;
2649         vcpu->arch.pio.rep = 0;
2650
2651         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2652                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2653                             handler);
2654         else
2655                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2656                             handler);
2657
2658         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2659         memcpy(vcpu->arch.pio_data, &val, 4);
2660
2661         kvm_x86_ops->skip_emulated_instruction(vcpu);
2662
2663         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2664         if (pio_dev) {
2665                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2666                 complete_pio(vcpu);
2667                 return 1;
2668         }
2669         return 0;
2670 }
2671 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2672
2673 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2674                   int size, unsigned long count, int down,
2675                   gva_t address, int rep, unsigned port)
2676 {
2677         unsigned now, in_page;
2678         int i, ret = 0;
2679         int nr_pages = 1;
2680         struct page *page;
2681         struct kvm_io_device *pio_dev;
2682
2683         vcpu->run->exit_reason = KVM_EXIT_IO;
2684         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2685         vcpu->run->io.size = vcpu->arch.pio.size = size;
2686         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2687         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2688         vcpu->run->io.port = vcpu->arch.pio.port = port;
2689         vcpu->arch.pio.in = in;
2690         vcpu->arch.pio.string = 1;
2691         vcpu->arch.pio.down = down;
2692         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2693         vcpu->arch.pio.rep = rep;
2694
2695         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2696                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2697                             handler);
2698         else
2699                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2700                             handler);
2701
2702         if (!count) {
2703                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2704                 return 1;
2705         }
2706
2707         if (!down)
2708                 in_page = PAGE_SIZE - offset_in_page(address);
2709         else
2710                 in_page = offset_in_page(address) + size;
2711         now = min(count, (unsigned long)in_page / size);
2712         if (!now) {
2713                 /*
2714                  * String I/O straddles page boundary.  Pin two guest pages
2715                  * so that we satisfy atomicity constraints.  Do just one
2716                  * transaction to avoid complexity.
2717                  */
2718                 nr_pages = 2;
2719                 now = 1;
2720         }
2721         if (down) {
2722                 /*
2723                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2724                  */
2725                 pr_unimpl(vcpu, "guest string pio down\n");
2726                 kvm_inject_gp(vcpu, 0);
2727                 return 1;
2728         }
2729         vcpu->run->io.count = now;
2730         vcpu->arch.pio.cur_count = now;
2731
2732         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2733                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2734
2735         for (i = 0; i < nr_pages; ++i) {
2736                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2737                 vcpu->arch.pio.guest_pages[i] = page;
2738                 if (!page) {
2739                         kvm_inject_gp(vcpu, 0);
2740                         free_pio_guest_pages(vcpu);
2741                         return 1;
2742                 }
2743         }
2744
2745         pio_dev = vcpu_find_pio_dev(vcpu, port,
2746                                     vcpu->arch.pio.cur_count,
2747                                     !vcpu->arch.pio.in);
2748         if (!vcpu->arch.pio.in) {
2749                 /* string PIO write */
2750                 ret = pio_copy_data(vcpu);
2751                 if (ret >= 0 && pio_dev) {
2752                         pio_string_write(pio_dev, vcpu);
2753                         complete_pio(vcpu);
2754                         if (vcpu->arch.pio.count == 0)
2755                                 ret = 1;
2756                 }
2757         } else if (pio_dev)
2758                 pr_unimpl(vcpu, "no string pio read support yet, "
2759                        "port %x size %d count %ld\n",
2760                         port, size, count);
2761
2762         return ret;
2763 }
2764 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2765
2766 int kvm_arch_init(void *opaque)
2767 {
2768         int r;
2769         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2770
2771         if (kvm_x86_ops) {
2772                 printk(KERN_ERR "kvm: already loaded the other module\n");
2773                 r = -EEXIST;
2774                 goto out;
2775         }
2776
2777         if (!ops->cpu_has_kvm_support()) {
2778                 printk(KERN_ERR "kvm: no hardware support\n");
2779                 r = -EOPNOTSUPP;
2780                 goto out;
2781         }
2782         if (ops->disabled_by_bios()) {
2783                 printk(KERN_ERR "kvm: disabled by bios\n");
2784                 r = -EOPNOTSUPP;
2785                 goto out;
2786         }
2787
2788         r = kvm_mmu_module_init();
2789         if (r)
2790                 goto out;
2791
2792         kvm_init_msr_list();
2793
2794         kvm_x86_ops = ops;
2795         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2796         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2797         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2798                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2799         return 0;
2800
2801 out:
2802         return r;
2803 }
2804
2805 void kvm_arch_exit(void)
2806 {
2807         kvm_x86_ops = NULL;
2808         kvm_mmu_module_exit();
2809 }
2810
2811 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2812 {
2813         ++vcpu->stat.halt_exits;
2814         KVMTRACE_0D(HLT, vcpu, handler);
2815         if (irqchip_in_kernel(vcpu->kvm)) {
2816                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2817                 return 1;
2818         } else {
2819                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2820                 return 0;
2821         }
2822 }
2823 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2824
2825 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2826                            unsigned long a1)
2827 {
2828         if (is_long_mode(vcpu))
2829                 return a0;
2830         else
2831                 return a0 | ((gpa_t)a1 << 32);
2832 }
2833
2834 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2835 {
2836         unsigned long nr, a0, a1, a2, a3, ret;
2837         int r = 1;
2838
2839         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2840         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2841         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2842         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2843         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2844
2845         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2846
2847         if (!is_long_mode(vcpu)) {
2848                 nr &= 0xFFFFFFFF;
2849                 a0 &= 0xFFFFFFFF;
2850                 a1 &= 0xFFFFFFFF;
2851                 a2 &= 0xFFFFFFFF;
2852                 a3 &= 0xFFFFFFFF;
2853         }
2854
2855         switch (nr) {
2856         case KVM_HC_VAPIC_POLL_IRQ:
2857                 ret = 0;
2858                 break;
2859         case KVM_HC_MMU_OP:
2860                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2861                 break;
2862         default:
2863                 ret = -KVM_ENOSYS;
2864                 break;
2865         }
2866         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2867         ++vcpu->stat.hypercalls;
2868         return r;
2869 }
2870 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2871
2872 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2873 {
2874         char instruction[3];
2875         int ret = 0;
2876         unsigned long rip = kvm_rip_read(vcpu);
2877
2878
2879         /*
2880          * Blow out the MMU to ensure that no other VCPU has an active mapping
2881          * to ensure that the updated hypercall appears atomically across all
2882          * VCPUs.
2883          */
2884         kvm_mmu_zap_all(vcpu->kvm);
2885
2886         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2887         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2888             != X86EMUL_CONTINUE)
2889                 ret = -EFAULT;
2890
2891         return ret;
2892 }
2893
2894 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2895 {
2896         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2897 }
2898
2899 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2900 {
2901         struct descriptor_table dt = { limit, base };
2902
2903         kvm_x86_ops->set_gdt(vcpu, &dt);
2904 }
2905
2906 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2907 {
2908         struct descriptor_table dt = { limit, base };
2909
2910         kvm_x86_ops->set_idt(vcpu, &dt);
2911 }
2912
2913 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2914                    unsigned long *rflags)
2915 {
2916         kvm_lmsw(vcpu, msw);
2917         *rflags = kvm_x86_ops->get_rflags(vcpu);
2918 }
2919
2920 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2921 {
2922         unsigned long value;
2923
2924         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2925         switch (cr) {
2926         case 0:
2927                 value = vcpu->arch.cr0;
2928                 break;
2929         case 2:
2930                 value = vcpu->arch.cr2;
2931                 break;
2932         case 3:
2933                 value = vcpu->arch.cr3;
2934                 break;
2935         case 4:
2936                 value = vcpu->arch.cr4;
2937                 break;
2938         case 8:
2939                 value = kvm_get_cr8(vcpu);
2940                 break;
2941         default:
2942                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2943                 return 0;
2944         }
2945         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2946                     (u32)((u64)value >> 32), handler);
2947
2948         return value;
2949 }
2950
2951 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2952                      unsigned long *rflags)
2953 {
2954         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2955                     (u32)((u64)val >> 32), handler);
2956
2957         switch (cr) {
2958         case 0:
2959                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2960                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2961                 break;
2962         case 2:
2963                 vcpu->arch.cr2 = val;
2964                 break;
2965         case 3:
2966                 kvm_set_cr3(vcpu, val);
2967                 break;
2968         case 4:
2969                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2970                 break;
2971         case 8:
2972                 kvm_set_cr8(vcpu, val & 0xfUL);
2973                 break;
2974         default:
2975                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2976         }
2977 }
2978
2979 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2980 {
2981         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2982         int j, nent = vcpu->arch.cpuid_nent;
2983
2984         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2985         /* when no next entry is found, the current entry[i] is reselected */
2986         for (j = i + 1; j == i; j = (j + 1) % nent) {
2987                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2988                 if (ej->function == e->function) {
2989                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2990                         return j;
2991                 }
2992         }
2993         return 0; /* silence gcc, even though control never reaches here */
2994 }
2995
2996 /* find an entry with matching function, matching index (if needed), and that
2997  * should be read next (if it's stateful) */
2998 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2999         u32 function, u32 index)
3000 {
3001         if (e->function != function)
3002                 return 0;
3003         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3004                 return 0;
3005         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3006                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3007                 return 0;
3008         return 1;
3009 }
3010
3011 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3012 {
3013         int i;
3014         u32 function, index;
3015         struct kvm_cpuid_entry2 *e, *best;
3016
3017         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3018         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3019         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3020         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3021         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3022         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3023         best = NULL;
3024         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3025                 e = &vcpu->arch.cpuid_entries[i];
3026                 if (is_matching_cpuid_entry(e, function, index)) {
3027                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3028                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3029                         best = e;
3030                         break;
3031                 }
3032                 /*
3033                  * Both basic or both extended?
3034                  */
3035                 if (((e->function ^ function) & 0x80000000) == 0)
3036                         if (!best || e->function > best->function)
3037                                 best = e;
3038         }
3039         if (best) {
3040                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3041                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3042                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3043                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3044         }
3045         kvm_x86_ops->skip_emulated_instruction(vcpu);
3046         KVMTRACE_5D(CPUID, vcpu, function,
3047                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3048                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3049                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3050                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3051 }
3052 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3053
3054 /*
3055  * Check if userspace requested an interrupt window, and that the
3056  * interrupt window is open.
3057  *
3058  * No need to exit to userspace if we already have an interrupt queued.
3059  */
3060 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3061                                           struct kvm_run *kvm_run)
3062 {
3063         return (!vcpu->arch.irq_summary &&
3064                 kvm_run->request_interrupt_window &&
3065                 vcpu->arch.interrupt_window_open &&
3066                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3067 }
3068
3069 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3070                               struct kvm_run *kvm_run)
3071 {
3072         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3073         kvm_run->cr8 = kvm_get_cr8(vcpu);
3074         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3075         if (irqchip_in_kernel(vcpu->kvm))
3076                 kvm_run->ready_for_interrupt_injection = 1;
3077         else
3078                 kvm_run->ready_for_interrupt_injection =
3079                                         (vcpu->arch.interrupt_window_open &&
3080                                          vcpu->arch.irq_summary == 0);
3081 }
3082
3083 static void vapic_enter(struct kvm_vcpu *vcpu)
3084 {
3085         struct kvm_lapic *apic = vcpu->arch.apic;
3086         struct page *page;
3087
3088         if (!apic || !apic->vapic_addr)
3089                 return;
3090
3091         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3092
3093         vcpu->arch.apic->vapic_page = page;
3094 }
3095
3096 static void vapic_exit(struct kvm_vcpu *vcpu)
3097 {
3098         struct kvm_lapic *apic = vcpu->arch.apic;
3099
3100         if (!apic || !apic->vapic_addr)
3101                 return;
3102
3103         down_read(&vcpu->kvm->slots_lock);
3104         kvm_release_page_dirty(apic->vapic_page);
3105         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3106         up_read(&vcpu->kvm->slots_lock);
3107 }
3108
3109 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3110 {
3111         int r;
3112
3113         if (vcpu->requests)
3114                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3115                         kvm_mmu_unload(vcpu);
3116
3117         r = kvm_mmu_reload(vcpu);
3118         if (unlikely(r))
3119                 goto out;
3120
3121         if (vcpu->requests) {
3122                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3123                         __kvm_migrate_timers(vcpu);
3124                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3125                         kvm_mmu_sync_roots(vcpu);
3126                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3127                         kvm_x86_ops->tlb_flush(vcpu);
3128                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3129                                        &vcpu->requests)) {
3130                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3131                         r = 0;
3132                         goto out;
3133                 }
3134                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3135                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3136                         r = 0;
3137                         goto out;
3138                 }
3139         }
3140
3141         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3142         kvm_inject_pending_timer_irqs(vcpu);
3143
3144         preempt_disable();
3145
3146         kvm_x86_ops->prepare_guest_switch(vcpu);
3147         kvm_load_guest_fpu(vcpu);
3148
3149         local_irq_disable();
3150
3151         if (vcpu->requests || need_resched() || signal_pending(current)) {
3152                 local_irq_enable();
3153                 preempt_enable();
3154                 r = 1;
3155                 goto out;
3156         }
3157
3158         if (vcpu->guest_debug.enabled)
3159                 kvm_x86_ops->guest_debug_pre(vcpu);
3160
3161         vcpu->guest_mode = 1;
3162         /*
3163          * Make sure that guest_mode assignment won't happen after
3164          * testing the pending IRQ vector bitmap.
3165          */
3166         smp_wmb();
3167
3168         if (vcpu->arch.exception.pending)
3169                 __queue_exception(vcpu);
3170         else if (irqchip_in_kernel(vcpu->kvm))
3171                 kvm_x86_ops->inject_pending_irq(vcpu);
3172         else
3173                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3174
3175         kvm_lapic_sync_to_vapic(vcpu);
3176
3177         up_read(&vcpu->kvm->slots_lock);
3178
3179         kvm_guest_enter();
3180
3181
3182         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3183         kvm_x86_ops->run(vcpu, kvm_run);
3184
3185         vcpu->guest_mode = 0;
3186         local_irq_enable();
3187
3188         ++vcpu->stat.exits;
3189
3190         /*
3191          * We must have an instruction between local_irq_enable() and
3192          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3193          * the interrupt shadow.  The stat.exits increment will do nicely.
3194          * But we need to prevent reordering, hence this barrier():
3195          */
3196         barrier();
3197
3198         kvm_guest_exit();
3199
3200         preempt_enable();
3201
3202         down_read(&vcpu->kvm->slots_lock);
3203
3204         /*
3205          * Profile KVM exit RIPs:
3206          */
3207         if (unlikely(prof_on == KVM_PROFILING)) {
3208                 unsigned long rip = kvm_rip_read(vcpu);
3209                 profile_hit(KVM_PROFILING, (void *)rip);
3210         }
3211
3212         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3213                 vcpu->arch.exception.pending = false;
3214
3215         kvm_lapic_sync_from_vapic(vcpu);
3216
3217         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3218 out:
3219         return r;
3220 }
3221
3222 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3223 {
3224         int r;
3225
3226         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3227                 printk("vcpu %d received sipi with vector # %x\n",
3228                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
3229                 kvm_lapic_reset(vcpu);
3230                 r = kvm_x86_ops->vcpu_reset(vcpu);
3231                 if (r)
3232                         return r;
3233                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3234         }
3235
3236         down_read(&vcpu->kvm->slots_lock);
3237         vapic_enter(vcpu);
3238
3239         r = 1;
3240         while (r > 0) {
3241                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3242                         r = vcpu_enter_guest(vcpu, kvm_run);
3243                 else {
3244                         up_read(&vcpu->kvm->slots_lock);
3245                         kvm_vcpu_block(vcpu);
3246                         down_read(&vcpu->kvm->slots_lock);
3247                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3248                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3249                                         vcpu->arch.mp_state =
3250                                                         KVM_MP_STATE_RUNNABLE;
3251                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3252                                 r = -EINTR;
3253                 }
3254
3255                 if (r > 0) {
3256                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3257                                 r = -EINTR;
3258                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3259                                 ++vcpu->stat.request_irq_exits;
3260                         }
3261                         if (signal_pending(current)) {
3262                                 r = -EINTR;
3263                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3264                                 ++vcpu->stat.signal_exits;
3265                         }
3266                         if (need_resched()) {
3267                                 up_read(&vcpu->kvm->slots_lock);
3268                                 kvm_resched(vcpu);
3269                                 down_read(&vcpu->kvm->slots_lock);
3270                         }
3271                 }
3272         }
3273
3274         up_read(&vcpu->kvm->slots_lock);
3275         post_kvm_run_save(vcpu, kvm_run);
3276
3277         vapic_exit(vcpu);
3278
3279         return r;
3280 }
3281
3282 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3283 {
3284         int r;
3285         sigset_t sigsaved;
3286
3287         vcpu_load(vcpu);
3288
3289         if (vcpu->sigset_active)
3290                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3291
3292         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3293                 kvm_vcpu_block(vcpu);
3294                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3295                 r = -EAGAIN;
3296                 goto out;
3297         }
3298
3299         /* re-sync apic's tpr */
3300         if (!irqchip_in_kernel(vcpu->kvm))
3301                 kvm_set_cr8(vcpu, kvm_run->cr8);
3302
3303         if (vcpu->arch.pio.cur_count) {
3304                 r = complete_pio(vcpu);
3305                 if (r)
3306                         goto out;
3307         }
3308 #if CONFIG_HAS_IOMEM
3309         if (vcpu->mmio_needed) {
3310                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3311                 vcpu->mmio_read_completed = 1;
3312                 vcpu->mmio_needed = 0;
3313
3314                 down_read(&vcpu->kvm->slots_lock);
3315                 r = emulate_instruction(vcpu, kvm_run,
3316                                         vcpu->arch.mmio_fault_cr2, 0,
3317                                         EMULTYPE_NO_DECODE);
3318                 up_read(&vcpu->kvm->slots_lock);
3319                 if (r == EMULATE_DO_MMIO) {
3320                         /*
3321                          * Read-modify-write.  Back to userspace.
3322                          */
3323                         r = 0;
3324                         goto out;
3325                 }
3326         }
3327 #endif
3328         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3329                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3330                                      kvm_run->hypercall.ret);
3331
3332         r = __vcpu_run(vcpu, kvm_run);
3333
3334 out:
3335         if (vcpu->sigset_active)
3336                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3337
3338         vcpu_put(vcpu);
3339         return r;
3340 }
3341
3342 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3343 {
3344         vcpu_load(vcpu);
3345
3346         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3347         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3348         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3349         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3350         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3351         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3352         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3353         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3354 #ifdef CONFIG_X86_64
3355         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3356         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3357         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3358         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3359         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3360         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3361         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3362         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3363 #endif
3364
3365         regs->rip = kvm_rip_read(vcpu);
3366         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3367
3368         /*
3369          * Don't leak debug flags in case they were set for guest debugging
3370          */
3371         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3372                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3373
3374         vcpu_put(vcpu);
3375
3376         return 0;
3377 }
3378
3379 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3380 {
3381         vcpu_load(vcpu);
3382
3383         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3384         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3385         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3386         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3387         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3388         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3389         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3390         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3391 #ifdef CONFIG_X86_64
3392         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3393         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3394         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3395         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3396         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3397         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3398         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3399         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3400
3401 #endif
3402
3403         kvm_rip_write(vcpu, regs->rip);
3404         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3405
3406
3407         vcpu->arch.exception.pending = false;
3408
3409         vcpu_put(vcpu);
3410
3411         return 0;
3412 }
3413
3414 void kvm_get_segment(struct kvm_vcpu *vcpu,
3415                      struct kvm_segment *var, int seg)
3416 {
3417         kvm_x86_ops->get_segment(vcpu, var, seg);
3418 }
3419
3420 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3421 {
3422         struct kvm_segment cs;
3423
3424         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3425         *db = cs.db;
3426         *l = cs.l;
3427 }
3428 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3429
3430 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3431                                   struct kvm_sregs *sregs)
3432 {
3433         struct descriptor_table dt;
3434         int pending_vec;
3435
3436         vcpu_load(vcpu);
3437
3438         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3439         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3440         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3441         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3442         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3443         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3444
3445         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3446         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3447
3448         kvm_x86_ops->get_idt(vcpu, &dt);
3449         sregs->idt.limit = dt.limit;
3450         sregs->idt.base = dt.base;
3451         kvm_x86_ops->get_gdt(vcpu, &dt);
3452         sregs->gdt.limit = dt.limit;
3453         sregs->gdt.base = dt.base;
3454
3455         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3456         sregs->cr0 = vcpu->arch.cr0;
3457         sregs->cr2 = vcpu->arch.cr2;
3458         sregs->cr3 = vcpu->arch.cr3;
3459         sregs->cr4 = vcpu->arch.cr4;
3460         sregs->cr8 = kvm_get_cr8(vcpu);
3461         sregs->efer = vcpu->arch.shadow_efer;
3462         sregs->apic_base = kvm_get_apic_base(vcpu);
3463
3464         if (irqchip_in_kernel(vcpu->kvm)) {
3465                 memset(sregs->interrupt_bitmap, 0,
3466                        sizeof sregs->interrupt_bitmap);
3467                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3468                 if (pending_vec >= 0)
3469                         set_bit(pending_vec,
3470                                 (unsigned long *)sregs->interrupt_bitmap);
3471         } else
3472                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3473                        sizeof sregs->interrupt_bitmap);
3474
3475         vcpu_put(vcpu);
3476
3477         return 0;
3478 }
3479
3480 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3481                                     struct kvm_mp_state *mp_state)
3482 {
3483         vcpu_load(vcpu);
3484         mp_state->mp_state = vcpu->arch.mp_state;
3485         vcpu_put(vcpu);
3486         return 0;
3487 }
3488
3489 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3490                                     struct kvm_mp_state *mp_state)
3491 {
3492         vcpu_load(vcpu);
3493         vcpu->arch.mp_state = mp_state->mp_state;
3494         vcpu_put(vcpu);
3495         return 0;
3496 }
3497
3498 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3499                         struct kvm_segment *var, int seg)
3500 {
3501         kvm_x86_ops->set_segment(vcpu, var, seg);
3502 }
3503
3504 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3505                                    struct kvm_segment *kvm_desct)
3506 {
3507         kvm_desct->base = seg_desc->base0;
3508         kvm_desct->base |= seg_desc->base1 << 16;
3509         kvm_desct->base |= seg_desc->base2 << 24;
3510         kvm_desct->limit = seg_desc->limit0;
3511         kvm_desct->limit |= seg_desc->limit << 16;
3512         if (seg_desc->g) {
3513                 kvm_desct->limit <<= 12;
3514                 kvm_desct->limit |= 0xfff;
3515         }
3516         kvm_desct->selector = selector;
3517         kvm_desct->type = seg_desc->type;
3518         kvm_desct->present = seg_desc->p;
3519         kvm_desct->dpl = seg_desc->dpl;
3520         kvm_desct->db = seg_desc->d;
3521         kvm_desct->s = seg_desc->s;
3522         kvm_desct->l = seg_desc->l;
3523         kvm_desct->g = seg_desc->g;
3524         kvm_desct->avl = seg_desc->avl;
3525         if (!selector)
3526                 kvm_desct->unusable = 1;
3527         else
3528                 kvm_desct->unusable = 0;
3529         kvm_desct->padding = 0;
3530 }
3531
3532 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3533                                            u16 selector,
3534                                            struct descriptor_table *dtable)
3535 {
3536         if (selector & 1 << 2) {
3537                 struct kvm_segment kvm_seg;
3538
3539                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3540
3541                 if (kvm_seg.unusable)
3542                         dtable->limit = 0;
3543                 else
3544                         dtable->limit = kvm_seg.limit;
3545                 dtable->base = kvm_seg.base;
3546         }
3547         else
3548                 kvm_x86_ops->get_gdt(vcpu, dtable);
3549 }
3550
3551 /* allowed just for 8 bytes segments */
3552 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3553                                          struct desc_struct *seg_desc)
3554 {
3555         gpa_t gpa;
3556         struct descriptor_table dtable;
3557         u16 index = selector >> 3;
3558
3559         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3560
3561         if (dtable.limit < index * 8 + 7) {
3562                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3563                 return 1;
3564         }
3565         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3566         gpa += index * 8;
3567         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3568 }
3569
3570 /* allowed just for 8 bytes segments */
3571 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3572                                          struct desc_struct *seg_desc)
3573 {
3574         gpa_t gpa;
3575         struct descriptor_table dtable;
3576         u16 index = selector >> 3;
3577
3578         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3579
3580         if (dtable.limit < index * 8 + 7)
3581                 return 1;
3582         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3583         gpa += index * 8;
3584         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3585 }
3586
3587 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3588                              struct desc_struct *seg_desc)
3589 {
3590         u32 base_addr;
3591
3592         base_addr = seg_desc->base0;
3593         base_addr |= (seg_desc->base1 << 16);
3594         base_addr |= (seg_desc->base2 << 24);
3595
3596         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3597 }
3598
3599 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3600 {
3601         struct kvm_segment kvm_seg;
3602
3603         kvm_get_segment(vcpu, &kvm_seg, seg);
3604         return kvm_seg.selector;
3605 }
3606
3607 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3608                                                 u16 selector,
3609                                                 struct kvm_segment *kvm_seg)
3610 {
3611         struct desc_struct seg_desc;
3612
3613         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3614                 return 1;
3615         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3616         return 0;
3617 }
3618
3619 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3620 {
3621         struct kvm_segment segvar = {
3622                 .base = selector << 4,
3623                 .limit = 0xffff,
3624                 .selector = selector,
3625                 .type = 3,
3626                 .present = 1,
3627                 .dpl = 3,
3628                 .db = 0,
3629                 .s = 1,
3630                 .l = 0,
3631                 .g = 0,
3632                 .avl = 0,
3633                 .unusable = 0,
3634         };
3635         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3636         return 0;
3637 }
3638
3639 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3640                                 int type_bits, int seg)
3641 {
3642         struct kvm_segment kvm_seg;
3643
3644         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3645                 return kvm_load_realmode_segment(vcpu, selector, seg);
3646         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3647                 return 1;
3648         kvm_seg.type |= type_bits;
3649
3650         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3651             seg != VCPU_SREG_LDTR)
3652                 if (!kvm_seg.s)
3653                         kvm_seg.unusable = 1;
3654
3655         kvm_set_segment(vcpu, &kvm_seg, seg);
3656         return 0;
3657 }
3658
3659 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3660                                 struct tss_segment_32 *tss)
3661 {
3662         tss->cr3 = vcpu->arch.cr3;
3663         tss->eip = kvm_rip_read(vcpu);
3664         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3665         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3666         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3667         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3668         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3669         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3670         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3671         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3672         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3673         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3674         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3675         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3676         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3677         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3678         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3679         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3680         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3681 }
3682
3683 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3684                                   struct tss_segment_32 *tss)
3685 {
3686         kvm_set_cr3(vcpu, tss->cr3);
3687
3688         kvm_rip_write(vcpu, tss->eip);
3689         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3690
3691         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3692         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3693         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3694         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3695         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3696         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3697         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3698         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3699
3700         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3701                 return 1;
3702
3703         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3704                 return 1;
3705
3706         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3707                 return 1;
3708
3709         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3710                 return 1;
3711
3712         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3713                 return 1;
3714
3715         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3716                 return 1;
3717
3718         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3719                 return 1;
3720         return 0;
3721 }
3722
3723 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3724                                 struct tss_segment_16 *tss)
3725 {
3726         tss->ip = kvm_rip_read(vcpu);
3727         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3728         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3729         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3730         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3731         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3732         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3733         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3734         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3735         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3736
3737         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3738         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3739         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3740         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3741         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3742         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3743 }
3744
3745 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3746                                  struct tss_segment_16 *tss)
3747 {
3748         kvm_rip_write(vcpu, tss->ip);
3749         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3750         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3751         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3752         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3753         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3754         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3755         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3756         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3757         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3758
3759         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3760                 return 1;
3761
3762         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3763                 return 1;
3764
3765         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3766                 return 1;
3767
3768         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3769                 return 1;
3770
3771         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3772                 return 1;
3773         return 0;
3774 }
3775
3776 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3777                        u32 old_tss_base,
3778                        struct desc_struct *nseg_desc)
3779 {
3780         struct tss_segment_16 tss_segment_16;
3781         int ret = 0;
3782
3783         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3784                            sizeof tss_segment_16))
3785                 goto out;
3786
3787         save_state_to_tss16(vcpu, &tss_segment_16);
3788
3789         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3790                             sizeof tss_segment_16))
3791                 goto out;
3792
3793         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3794                            &tss_segment_16, sizeof tss_segment_16))
3795                 goto out;
3796
3797         if (load_state_from_tss16(vcpu, &tss_segment_16))
3798                 goto out;
3799
3800         ret = 1;
3801 out:
3802         return ret;
3803 }
3804
3805 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3806                        u32 old_tss_base,
3807                        struct desc_struct *nseg_desc)
3808 {
3809         struct tss_segment_32 tss_segment_32;
3810         int ret = 0;
3811
3812         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3813                            sizeof tss_segment_32))
3814                 goto out;
3815
3816         save_state_to_tss32(vcpu, &tss_segment_32);
3817
3818         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3819                             sizeof tss_segment_32))
3820                 goto out;
3821
3822         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3823                            &tss_segment_32, sizeof tss_segment_32))
3824                 goto out;
3825
3826         if (load_state_from_tss32(vcpu, &tss_segment_32))
3827                 goto out;
3828
3829         ret = 1;
3830 out:
3831         return ret;
3832 }
3833
3834 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3835 {
3836         struct kvm_segment tr_seg;
3837         struct desc_struct cseg_desc;
3838         struct desc_struct nseg_desc;
3839         int ret = 0;
3840         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3841         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3842
3843         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3844
3845         /* FIXME: Handle errors. Failure to read either TSS or their
3846          * descriptors should generate a pagefault.
3847          */
3848         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3849                 goto out;
3850
3851         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3852                 goto out;
3853
3854         if (reason != TASK_SWITCH_IRET) {
3855                 int cpl;
3856
3857                 cpl = kvm_x86_ops->get_cpl(vcpu);
3858                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3859                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3860                         return 1;
3861                 }
3862         }
3863
3864         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3865                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3866                 return 1;
3867         }
3868
3869         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3870                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3871                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3872         }
3873
3874         if (reason == TASK_SWITCH_IRET) {
3875                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3876                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3877         }
3878
3879         kvm_x86_ops->skip_emulated_instruction(vcpu);
3880
3881         if (nseg_desc.type & 8)
3882                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3883                                          &nseg_desc);
3884         else
3885                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3886                                          &nseg_desc);
3887
3888         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3889                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3890                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3891         }
3892
3893         if (reason != TASK_SWITCH_IRET) {
3894                 nseg_desc.type |= (1 << 1);
3895                 save_guest_segment_descriptor(vcpu, tss_selector,
3896                                               &nseg_desc);
3897         }
3898
3899         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3900         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3901         tr_seg.type = 11;
3902         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3903 out:
3904         return ret;
3905 }
3906 EXPORT_SYMBOL_GPL(kvm_task_switch);
3907
3908 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3909                                   struct kvm_sregs *sregs)
3910 {
3911         int mmu_reset_needed = 0;
3912         int i, pending_vec, max_bits;
3913         struct descriptor_table dt;
3914
3915         vcpu_load(vcpu);
3916
3917         dt.limit = sregs->idt.limit;
3918         dt.base = sregs->idt.base;
3919         kvm_x86_ops->set_idt(vcpu, &dt);
3920         dt.limit = sregs->gdt.limit;
3921         dt.base = sregs->gdt.base;
3922         kvm_x86_ops->set_gdt(vcpu, &dt);
3923
3924         vcpu->arch.cr2 = sregs->cr2;
3925         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3926         vcpu->arch.cr3 = sregs->cr3;
3927
3928         kvm_set_cr8(vcpu, sregs->cr8);
3929
3930         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3931         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3932         kvm_set_apic_base(vcpu, sregs->apic_base);
3933
3934         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3935
3936         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3937         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3938         vcpu->arch.cr0 = sregs->cr0;
3939
3940         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3941         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3942         if (!is_long_mode(vcpu) && is_pae(vcpu))
3943                 load_pdptrs(vcpu, vcpu->arch.cr3);
3944
3945         if (mmu_reset_needed)
3946                 kvm_mmu_reset_context(vcpu);
3947
3948         if (!irqchip_in_kernel(vcpu->kvm)) {
3949                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3950                        sizeof vcpu->arch.irq_pending);
3951                 vcpu->arch.irq_summary = 0;
3952                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3953                         if (vcpu->arch.irq_pending[i])
3954                                 __set_bit(i, &vcpu->arch.irq_summary);
3955         } else {
3956                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3957                 pending_vec = find_first_bit(
3958                         (const unsigned long *)sregs->interrupt_bitmap,
3959                         max_bits);
3960                 /* Only pending external irq is handled here */
3961                 if (pending_vec < max_bits) {
3962                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3963                         pr_debug("Set back pending irq %d\n",
3964                                  pending_vec);
3965                 }
3966                 kvm_pic_clear_isr_ack(vcpu->kvm);
3967         }
3968
3969         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3970         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3971         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3972         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3973         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3974         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3975
3976         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3977         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3978
3979         /* Older userspace won't unhalt the vcpu on reset. */
3980         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3981             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3982             !(vcpu->arch.cr0 & X86_CR0_PE))
3983                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3984
3985         vcpu_put(vcpu);
3986
3987         return 0;
3988 }
3989
3990 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3991                                     struct kvm_debug_guest *dbg)
3992 {
3993         int r;
3994
3995         vcpu_load(vcpu);
3996
3997         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3998
3999         vcpu_put(vcpu);
4000
4001         return r;
4002 }
4003
4004 /*
4005  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4006  * we have asm/x86/processor.h
4007  */
4008 struct fxsave {
4009         u16     cwd;
4010         u16     swd;
4011         u16     twd;
4012         u16     fop;
4013         u64     rip;
4014         u64     rdp;
4015         u32     mxcsr;
4016         u32     mxcsr_mask;
4017         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4018 #ifdef CONFIG_X86_64
4019         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4020 #else
4021         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4022 #endif
4023 };
4024
4025 /*
4026  * Translate a guest virtual address to a guest physical address.
4027  */
4028 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4029                                     struct kvm_translation *tr)
4030 {
4031         unsigned long vaddr = tr->linear_address;
4032         gpa_t gpa;
4033
4034         vcpu_load(vcpu);
4035         down_read(&vcpu->kvm->slots_lock);
4036         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4037         up_read(&vcpu->kvm->slots_lock);
4038         tr->physical_address = gpa;
4039         tr->valid = gpa != UNMAPPED_GVA;
4040         tr->writeable = 1;
4041         tr->usermode = 0;
4042         vcpu_put(vcpu);
4043
4044         return 0;
4045 }
4046
4047 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4048 {
4049         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4050
4051         vcpu_load(vcpu);
4052
4053         memcpy(fpu->fpr, fxsave->st_space, 128);
4054         fpu->fcw = fxsave->cwd;
4055         fpu->fsw = fxsave->swd;
4056         fpu->ftwx = fxsave->twd;
4057         fpu->last_opcode = fxsave->fop;
4058         fpu->last_ip = fxsave->rip;
4059         fpu->last_dp = fxsave->rdp;
4060         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4061
4062         vcpu_put(vcpu);
4063
4064         return 0;
4065 }
4066
4067 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4068 {
4069         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4070
4071         vcpu_load(vcpu);
4072
4073         memcpy(fxsave->st_space, fpu->fpr, 128);
4074         fxsave->cwd = fpu->fcw;
4075         fxsave->swd = fpu->fsw;
4076         fxsave->twd = fpu->ftwx;
4077         fxsave->fop = fpu->last_opcode;
4078         fxsave->rip = fpu->last_ip;
4079         fxsave->rdp = fpu->last_dp;
4080         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4081
4082         vcpu_put(vcpu);
4083
4084         return 0;
4085 }
4086
4087 void fx_init(struct kvm_vcpu *vcpu)
4088 {
4089         unsigned after_mxcsr_mask;
4090
4091         /*
4092          * Touch the fpu the first time in non atomic context as if
4093          * this is the first fpu instruction the exception handler
4094          * will fire before the instruction returns and it'll have to
4095          * allocate ram with GFP_KERNEL.
4096          */
4097         if (!used_math())
4098                 kvm_fx_save(&vcpu->arch.host_fx_image);
4099
4100         /* Initialize guest FPU by resetting ours and saving into guest's */
4101         preempt_disable();
4102         kvm_fx_save(&vcpu->arch.host_fx_image);
4103         kvm_fx_finit();
4104         kvm_fx_save(&vcpu->arch.guest_fx_image);
4105         kvm_fx_restore(&vcpu->arch.host_fx_image);
4106         preempt_enable();
4107
4108         vcpu->arch.cr0 |= X86_CR0_ET;
4109         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4110         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4111         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4112                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4113 }
4114 EXPORT_SYMBOL_GPL(fx_init);
4115
4116 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4117 {
4118         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4119                 return;
4120
4121         vcpu->guest_fpu_loaded = 1;
4122         kvm_fx_save(&vcpu->arch.host_fx_image);
4123         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4124 }
4125 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4126
4127 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4128 {
4129         if (!vcpu->guest_fpu_loaded)
4130                 return;
4131
4132         vcpu->guest_fpu_loaded = 0;
4133         kvm_fx_save(&vcpu->arch.guest_fx_image);
4134         kvm_fx_restore(&vcpu->arch.host_fx_image);
4135         ++vcpu->stat.fpu_reload;
4136 }
4137 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4138
4139 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4140 {
4141         kvm_x86_ops->vcpu_free(vcpu);
4142 }
4143
4144 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4145                                                 unsigned int id)
4146 {
4147         return kvm_x86_ops->vcpu_create(kvm, id);
4148 }
4149
4150 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4151 {
4152         int r;
4153
4154         /* We do fxsave: this must be aligned. */
4155         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4156
4157         vcpu_load(vcpu);
4158         r = kvm_arch_vcpu_reset(vcpu);
4159         if (r == 0)
4160                 r = kvm_mmu_setup(vcpu);
4161         vcpu_put(vcpu);
4162         if (r < 0)
4163                 goto free_vcpu;
4164
4165         return 0;
4166 free_vcpu:
4167         kvm_x86_ops->vcpu_free(vcpu);
4168         return r;
4169 }
4170
4171 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4172 {
4173         vcpu_load(vcpu);
4174         kvm_mmu_unload(vcpu);
4175         vcpu_put(vcpu);
4176
4177         kvm_x86_ops->vcpu_free(vcpu);
4178 }
4179
4180 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4181 {
4182         return kvm_x86_ops->vcpu_reset(vcpu);
4183 }
4184
4185 void kvm_arch_hardware_enable(void *garbage)
4186 {
4187         kvm_x86_ops->hardware_enable(garbage);
4188 }
4189
4190 void kvm_arch_hardware_disable(void *garbage)
4191 {
4192         kvm_x86_ops->hardware_disable(garbage);
4193 }
4194
4195 int kvm_arch_hardware_setup(void)
4196 {
4197         return kvm_x86_ops->hardware_setup();
4198 }
4199
4200 void kvm_arch_hardware_unsetup(void)
4201 {
4202         kvm_x86_ops->hardware_unsetup();
4203 }
4204
4205 void kvm_arch_check_processor_compat(void *rtn)
4206 {
4207         kvm_x86_ops->check_processor_compatibility(rtn);
4208 }
4209
4210 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4211 {
4212         struct page *page;
4213         struct kvm *kvm;
4214         int r;
4215
4216         BUG_ON(vcpu->kvm == NULL);
4217         kvm = vcpu->kvm;
4218
4219         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4220         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4221                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4222         else
4223                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4224
4225         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4226         if (!page) {
4227                 r = -ENOMEM;
4228                 goto fail;
4229         }
4230         vcpu->arch.pio_data = page_address(page);
4231
4232         r = kvm_mmu_create(vcpu);
4233         if (r < 0)
4234                 goto fail_free_pio_data;
4235
4236         if (irqchip_in_kernel(kvm)) {
4237                 r = kvm_create_lapic(vcpu);
4238                 if (r < 0)
4239                         goto fail_mmu_destroy;
4240         }
4241
4242         return 0;
4243
4244 fail_mmu_destroy:
4245         kvm_mmu_destroy(vcpu);
4246 fail_free_pio_data:
4247         free_page((unsigned long)vcpu->arch.pio_data);
4248 fail:
4249         return r;
4250 }
4251
4252 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4253 {
4254         kvm_free_lapic(vcpu);
4255         down_read(&vcpu->kvm->slots_lock);
4256         kvm_mmu_destroy(vcpu);
4257         up_read(&vcpu->kvm->slots_lock);
4258         free_page((unsigned long)vcpu->arch.pio_data);
4259 }
4260
4261 struct  kvm *kvm_arch_create_vm(void)
4262 {
4263         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4264
4265         if (!kvm)
4266                 return ERR_PTR(-ENOMEM);
4267
4268         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4269         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4270
4271         return kvm;
4272 }
4273
4274 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4275 {
4276         vcpu_load(vcpu);
4277         kvm_mmu_unload(vcpu);
4278         vcpu_put(vcpu);
4279 }
4280
4281 static void kvm_free_vcpus(struct kvm *kvm)
4282 {
4283         unsigned int i;
4284
4285         /*
4286          * Unpin any mmu pages first.
4287          */
4288         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4289                 if (kvm->vcpus[i])
4290                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4291         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4292                 if (kvm->vcpus[i]) {
4293                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4294                         kvm->vcpus[i] = NULL;
4295                 }
4296         }
4297
4298 }
4299
4300 void kvm_arch_destroy_vm(struct kvm *kvm)
4301 {
4302         kvm_iommu_unmap_guest(kvm);
4303         kvm_free_all_assigned_devices(kvm);
4304         kvm_free_pit(kvm);
4305         kfree(kvm->arch.vpic);
4306         kfree(kvm->arch.vioapic);
4307         kvm_free_vcpus(kvm);
4308         kvm_free_physmem(kvm);
4309         if (kvm->arch.apic_access_page)
4310                 put_page(kvm->arch.apic_access_page);
4311         if (kvm->arch.ept_identity_pagetable)
4312                 put_page(kvm->arch.ept_identity_pagetable);
4313         kfree(kvm);
4314 }
4315
4316 int kvm_arch_set_memory_region(struct kvm *kvm,
4317                                 struct kvm_userspace_memory_region *mem,
4318                                 struct kvm_memory_slot old,
4319                                 int user_alloc)
4320 {
4321         int npages = mem->memory_size >> PAGE_SHIFT;
4322         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4323
4324         /*To keep backward compatibility with older userspace,
4325          *x86 needs to hanlde !user_alloc case.
4326          */
4327         if (!user_alloc) {
4328                 if (npages && !old.rmap) {
4329                         unsigned long userspace_addr;
4330
4331                         down_write(&current->mm->mmap_sem);
4332                         userspace_addr = do_mmap(NULL, 0,
4333                                                  npages * PAGE_SIZE,
4334                                                  PROT_READ | PROT_WRITE,
4335                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4336                                                  0);
4337                         up_write(&current->mm->mmap_sem);
4338
4339                         if (IS_ERR((void *)userspace_addr))
4340                                 return PTR_ERR((void *)userspace_addr);
4341
4342                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4343                         spin_lock(&kvm->mmu_lock);
4344                         memslot->userspace_addr = userspace_addr;
4345                         spin_unlock(&kvm->mmu_lock);
4346                 } else {
4347                         if (!old.user_alloc && old.rmap) {
4348                                 int ret;
4349
4350                                 down_write(&current->mm->mmap_sem);
4351                                 ret = do_munmap(current->mm, old.userspace_addr,
4352                                                 old.npages * PAGE_SIZE);
4353                                 up_write(&current->mm->mmap_sem);
4354                                 if (ret < 0)
4355                                         printk(KERN_WARNING
4356                                        "kvm_vm_ioctl_set_memory_region: "
4357                                        "failed to munmap memory\n");
4358                         }
4359                 }
4360         }
4361
4362         if (!kvm->arch.n_requested_mmu_pages) {
4363                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4364                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4365         }
4366
4367         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4368         kvm_flush_remote_tlbs(kvm);
4369
4370         return 0;
4371 }
4372
4373 void kvm_arch_flush_shadow(struct kvm *kvm)
4374 {
4375         kvm_mmu_zap_all(kvm);
4376 }
4377
4378 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4379 {
4380         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4381                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4382 }
4383
4384 static void vcpu_kick_intr(void *info)
4385 {
4386 #ifdef DEBUG
4387         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4388         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4389 #endif
4390 }
4391
4392 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4393 {
4394         int ipi_pcpu = vcpu->cpu;
4395         int cpu = get_cpu();
4396
4397         if (waitqueue_active(&vcpu->wq)) {
4398                 wake_up_interruptible(&vcpu->wq);
4399                 ++vcpu->stat.halt_wakeup;
4400         }
4401         /*
4402          * We may be called synchronously with irqs disabled in guest mode,
4403          * So need not to call smp_call_function_single() in that case.
4404          */
4405         if (vcpu->guest_mode && vcpu->cpu != cpu)
4406                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4407         put_cpu();
4408 }