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