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 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                         if (irq_event.irq < 16)
1960                                 kvm_pic_set_irq(pic_irqchip(kvm),
1961                                         irq_event.irq,
1962                                         irq_event.level);
1963                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1964                                         irq_event.irq,
1965                                         irq_event.level);
1966                         mutex_unlock(&kvm->lock);
1967                         r = 0;
1968                 }
1969                 break;
1970         }
1971         case KVM_GET_IRQCHIP: {
1972                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1973                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1974
1975                 r = -ENOMEM;
1976                 if (!chip)
1977                         goto out;
1978                 r = -EFAULT;
1979                 if (copy_from_user(chip, argp, sizeof *chip))
1980                         goto get_irqchip_out;
1981                 r = -ENXIO;
1982                 if (!irqchip_in_kernel(kvm))
1983                         goto get_irqchip_out;
1984                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1985                 if (r)
1986                         goto get_irqchip_out;
1987                 r = -EFAULT;
1988                 if (copy_to_user(argp, chip, sizeof *chip))
1989                         goto get_irqchip_out;
1990                 r = 0;
1991         get_irqchip_out:
1992                 kfree(chip);
1993                 if (r)
1994                         goto out;
1995                 break;
1996         }
1997         case KVM_SET_IRQCHIP: {
1998                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1999                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2000
2001                 r = -ENOMEM;
2002                 if (!chip)
2003                         goto out;
2004                 r = -EFAULT;
2005                 if (copy_from_user(chip, argp, sizeof *chip))
2006                         goto set_irqchip_out;
2007                 r = -ENXIO;
2008                 if (!irqchip_in_kernel(kvm))
2009                         goto set_irqchip_out;
2010                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2011                 if (r)
2012                         goto set_irqchip_out;
2013                 r = 0;
2014         set_irqchip_out:
2015                 kfree(chip);
2016                 if (r)
2017                         goto out;
2018                 break;
2019         }
2020         case KVM_ASSIGN_PCI_DEVICE: {
2021                 struct kvm_assigned_pci_dev assigned_dev;
2022
2023                 r = -EFAULT;
2024                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2025                         goto out;
2026                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2027                 if (r)
2028                         goto out;
2029                 break;
2030         }
2031         case KVM_ASSIGN_IRQ: {
2032                 struct kvm_assigned_irq assigned_irq;
2033
2034                 r = -EFAULT;
2035                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2036                         goto out;
2037                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2038                 if (r)
2039                         goto out;
2040                 break;
2041         }
2042         case KVM_GET_PIT: {
2043                 r = -EFAULT;
2044                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2045                         goto out;
2046                 r = -ENXIO;
2047                 if (!kvm->arch.vpit)
2048                         goto out;
2049                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2050                 if (r)
2051                         goto out;
2052                 r = -EFAULT;
2053                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2054                         goto out;
2055                 r = 0;
2056                 break;
2057         }
2058         case KVM_SET_PIT: {
2059                 r = -EFAULT;
2060                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2061                         goto out;
2062                 r = -ENXIO;
2063                 if (!kvm->arch.vpit)
2064                         goto out;
2065                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2066                 if (r)
2067                         goto out;
2068                 r = 0;
2069                 break;
2070         }
2071         default:
2072                 ;
2073         }
2074 out:
2075         return r;
2076 }
2077
2078 static void kvm_init_msr_list(void)
2079 {
2080         u32 dummy[2];
2081         unsigned i, j;
2082
2083         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2084                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2085                         continue;
2086                 if (j < i)
2087                         msrs_to_save[j] = msrs_to_save[i];
2088                 j++;
2089         }
2090         num_msrs_to_save = j;
2091 }
2092
2093 /*
2094  * Only apic need an MMIO device hook, so shortcut now..
2095  */
2096 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2097                                                 gpa_t addr, int len,
2098                                                 int is_write)
2099 {
2100         struct kvm_io_device *dev;
2101
2102         if (vcpu->arch.apic) {
2103                 dev = &vcpu->arch.apic->dev;
2104                 if (dev->in_range(dev, addr, len, is_write))
2105                         return dev;
2106         }
2107         return NULL;
2108 }
2109
2110
2111 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2112                                                 gpa_t addr, int len,
2113                                                 int is_write)
2114 {
2115         struct kvm_io_device *dev;
2116
2117         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2118         if (dev == NULL)
2119                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2120                                           is_write);
2121         return dev;
2122 }
2123
2124 int emulator_read_std(unsigned long addr,
2125                              void *val,
2126                              unsigned int bytes,
2127                              struct kvm_vcpu *vcpu)
2128 {
2129         void *data = val;
2130         int r = X86EMUL_CONTINUE;
2131
2132         while (bytes) {
2133                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2134                 unsigned offset = addr & (PAGE_SIZE-1);
2135                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
2136                 int ret;
2137
2138                 if (gpa == UNMAPPED_GVA) {
2139                         r = X86EMUL_PROPAGATE_FAULT;
2140                         goto out;
2141                 }
2142                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
2143                 if (ret < 0) {
2144                         r = X86EMUL_UNHANDLEABLE;
2145                         goto out;
2146                 }
2147
2148                 bytes -= tocopy;
2149                 data += tocopy;
2150                 addr += tocopy;
2151         }
2152 out:
2153         return r;
2154 }
2155 EXPORT_SYMBOL_GPL(emulator_read_std);
2156
2157 static int emulator_read_emulated(unsigned long addr,
2158                                   void *val,
2159                                   unsigned int bytes,
2160                                   struct kvm_vcpu *vcpu)
2161 {
2162         struct kvm_io_device *mmio_dev;
2163         gpa_t                 gpa;
2164
2165         if (vcpu->mmio_read_completed) {
2166                 memcpy(val, vcpu->mmio_data, bytes);
2167                 vcpu->mmio_read_completed = 0;
2168                 return X86EMUL_CONTINUE;
2169         }
2170
2171         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2172
2173         /* For APIC access vmexit */
2174         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2175                 goto mmio;
2176
2177         if (emulator_read_std(addr, val, bytes, vcpu)
2178                         == X86EMUL_CONTINUE)
2179                 return X86EMUL_CONTINUE;
2180         if (gpa == UNMAPPED_GVA)
2181                 return X86EMUL_PROPAGATE_FAULT;
2182
2183 mmio:
2184         /*
2185          * Is this MMIO handled locally?
2186          */
2187         mutex_lock(&vcpu->kvm->lock);
2188         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2189         if (mmio_dev) {
2190                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2191                 mutex_unlock(&vcpu->kvm->lock);
2192                 return X86EMUL_CONTINUE;
2193         }
2194         mutex_unlock(&vcpu->kvm->lock);
2195
2196         vcpu->mmio_needed = 1;
2197         vcpu->mmio_phys_addr = gpa;
2198         vcpu->mmio_size = bytes;
2199         vcpu->mmio_is_write = 0;
2200
2201         return X86EMUL_UNHANDLEABLE;
2202 }
2203
2204 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2205                           const void *val, int bytes)
2206 {
2207         int ret;
2208
2209         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2210         if (ret < 0)
2211                 return 0;
2212         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2213         return 1;
2214 }
2215
2216 static int emulator_write_emulated_onepage(unsigned long addr,
2217                                            const void *val,
2218                                            unsigned int bytes,
2219                                            struct kvm_vcpu *vcpu)
2220 {
2221         struct kvm_io_device *mmio_dev;
2222         gpa_t                 gpa;
2223
2224         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2225
2226         if (gpa == UNMAPPED_GVA) {
2227                 kvm_inject_page_fault(vcpu, addr, 2);
2228                 return X86EMUL_PROPAGATE_FAULT;
2229         }
2230
2231         /* For APIC access vmexit */
2232         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2233                 goto mmio;
2234
2235         if (emulator_write_phys(vcpu, gpa, val, bytes))
2236                 return X86EMUL_CONTINUE;
2237
2238 mmio:
2239         /*
2240          * Is this MMIO handled locally?
2241          */
2242         mutex_lock(&vcpu->kvm->lock);
2243         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2244         if (mmio_dev) {
2245                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2246                 mutex_unlock(&vcpu->kvm->lock);
2247                 return X86EMUL_CONTINUE;
2248         }
2249         mutex_unlock(&vcpu->kvm->lock);
2250
2251         vcpu->mmio_needed = 1;
2252         vcpu->mmio_phys_addr = gpa;
2253         vcpu->mmio_size = bytes;
2254         vcpu->mmio_is_write = 1;
2255         memcpy(vcpu->mmio_data, val, bytes);
2256
2257         return X86EMUL_CONTINUE;
2258 }
2259
2260 int emulator_write_emulated(unsigned long addr,
2261                                    const void *val,
2262                                    unsigned int bytes,
2263                                    struct kvm_vcpu *vcpu)
2264 {
2265         /* Crossing a page boundary? */
2266         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2267                 int rc, now;
2268
2269                 now = -addr & ~PAGE_MASK;
2270                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2271                 if (rc != X86EMUL_CONTINUE)
2272                         return rc;
2273                 addr += now;
2274                 val += now;
2275                 bytes -= now;
2276         }
2277         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2278 }
2279 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2280
2281 static int emulator_cmpxchg_emulated(unsigned long addr,
2282                                      const void *old,
2283                                      const void *new,
2284                                      unsigned int bytes,
2285                                      struct kvm_vcpu *vcpu)
2286 {
2287         static int reported;
2288
2289         if (!reported) {
2290                 reported = 1;
2291                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2292         }
2293 #ifndef CONFIG_X86_64
2294         /* guests cmpxchg8b have to be emulated atomically */
2295         if (bytes == 8) {
2296                 gpa_t gpa;
2297                 struct page *page;
2298                 char *kaddr;
2299                 u64 val;
2300
2301                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2302
2303                 if (gpa == UNMAPPED_GVA ||
2304                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2305                         goto emul_write;
2306
2307                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2308                         goto emul_write;
2309
2310                 val = *(u64 *)new;
2311
2312                 down_read(&current->mm->mmap_sem);
2313                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2314                 up_read(&current->mm->mmap_sem);
2315
2316                 kaddr = kmap_atomic(page, KM_USER0);
2317                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2318                 kunmap_atomic(kaddr, KM_USER0);
2319                 kvm_release_page_dirty(page);
2320         }
2321 emul_write:
2322 #endif
2323
2324         return emulator_write_emulated(addr, new, bytes, vcpu);
2325 }
2326
2327 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2328 {
2329         return kvm_x86_ops->get_segment_base(vcpu, seg);
2330 }
2331
2332 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2333 {
2334         return X86EMUL_CONTINUE;
2335 }
2336
2337 int emulate_clts(struct kvm_vcpu *vcpu)
2338 {
2339         KVMTRACE_0D(CLTS, vcpu, handler);
2340         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2341         return X86EMUL_CONTINUE;
2342 }
2343
2344 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2345 {
2346         struct kvm_vcpu *vcpu = ctxt->vcpu;
2347
2348         switch (dr) {
2349         case 0 ... 3:
2350                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2351                 return X86EMUL_CONTINUE;
2352         default:
2353                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2354                 return X86EMUL_UNHANDLEABLE;
2355         }
2356 }
2357
2358 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2359 {
2360         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2361         int exception;
2362
2363         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2364         if (exception) {
2365                 /* FIXME: better handling */
2366                 return X86EMUL_UNHANDLEABLE;
2367         }
2368         return X86EMUL_CONTINUE;
2369 }
2370
2371 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2372 {
2373         u8 opcodes[4];
2374         unsigned long rip = kvm_rip_read(vcpu);
2375         unsigned long rip_linear;
2376
2377         if (!printk_ratelimit())
2378                 return;
2379
2380         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2381
2382         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2383
2384         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2385                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2386 }
2387 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2388
2389 static struct x86_emulate_ops emulate_ops = {
2390         .read_std            = emulator_read_std,
2391         .read_emulated       = emulator_read_emulated,
2392         .write_emulated      = emulator_write_emulated,
2393         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2394 };
2395
2396 static void cache_all_regs(struct kvm_vcpu *vcpu)
2397 {
2398         kvm_register_read(vcpu, VCPU_REGS_RAX);
2399         kvm_register_read(vcpu, VCPU_REGS_RSP);
2400         kvm_register_read(vcpu, VCPU_REGS_RIP);
2401         vcpu->arch.regs_dirty = ~0;
2402 }
2403
2404 int emulate_instruction(struct kvm_vcpu *vcpu,
2405                         struct kvm_run *run,
2406                         unsigned long cr2,
2407                         u16 error_code,
2408                         int emulation_type)
2409 {
2410         int r;
2411         struct decode_cache *c;
2412
2413         kvm_clear_exception_queue(vcpu);
2414         vcpu->arch.mmio_fault_cr2 = cr2;
2415         /*
2416          * TODO: fix x86_emulate.c to use guest_read/write_register
2417          * instead of direct ->regs accesses, can save hundred cycles
2418          * on Intel for instructions that don't read/change RSP, for
2419          * for example.
2420          */
2421         cache_all_regs(vcpu);
2422
2423         vcpu->mmio_is_write = 0;
2424         vcpu->arch.pio.string = 0;
2425
2426         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2427                 int cs_db, cs_l;
2428                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2429
2430                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2431                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2432                 vcpu->arch.emulate_ctxt.mode =
2433                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2434                         ? X86EMUL_MODE_REAL : cs_l
2435                         ? X86EMUL_MODE_PROT64 : cs_db
2436                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2437
2438                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2439
2440                 /* Reject the instructions other than VMCALL/VMMCALL when
2441                  * try to emulate invalid opcode */
2442                 c = &vcpu->arch.emulate_ctxt.decode;
2443                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2444                     (!(c->twobyte && c->b == 0x01 &&
2445                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2446                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2447                         return EMULATE_FAIL;
2448
2449                 ++vcpu->stat.insn_emulation;
2450                 if (r)  {
2451                         ++vcpu->stat.insn_emulation_fail;
2452                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2453                                 return EMULATE_DONE;
2454                         return EMULATE_FAIL;
2455                 }
2456         }
2457
2458         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2459
2460         if (vcpu->arch.pio.string)
2461                 return EMULATE_DO_MMIO;
2462
2463         if ((r || vcpu->mmio_is_write) && run) {
2464                 run->exit_reason = KVM_EXIT_MMIO;
2465                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2466                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2467                 run->mmio.len = vcpu->mmio_size;
2468                 run->mmio.is_write = vcpu->mmio_is_write;
2469         }
2470
2471         if (r) {
2472                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2473                         return EMULATE_DONE;
2474                 if (!vcpu->mmio_needed) {
2475                         kvm_report_emulation_failure(vcpu, "mmio");
2476                         return EMULATE_FAIL;
2477                 }
2478                 return EMULATE_DO_MMIO;
2479         }
2480
2481         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2482
2483         if (vcpu->mmio_is_write) {
2484                 vcpu->mmio_needed = 0;
2485                 return EMULATE_DO_MMIO;
2486         }
2487
2488         return EMULATE_DONE;
2489 }
2490 EXPORT_SYMBOL_GPL(emulate_instruction);
2491
2492 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2493 {
2494         int i;
2495
2496         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2497                 if (vcpu->arch.pio.guest_pages[i]) {
2498                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2499                         vcpu->arch.pio.guest_pages[i] = NULL;
2500                 }
2501 }
2502
2503 static int pio_copy_data(struct kvm_vcpu *vcpu)
2504 {
2505         void *p = vcpu->arch.pio_data;
2506         void *q;
2507         unsigned bytes;
2508         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2509
2510         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2511                  PAGE_KERNEL);
2512         if (!q) {
2513                 free_pio_guest_pages(vcpu);
2514                 return -ENOMEM;
2515         }
2516         q += vcpu->arch.pio.guest_page_offset;
2517         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2518         if (vcpu->arch.pio.in)
2519                 memcpy(q, p, bytes);
2520         else
2521                 memcpy(p, q, bytes);
2522         q -= vcpu->arch.pio.guest_page_offset;
2523         vunmap(q);
2524         free_pio_guest_pages(vcpu);
2525         return 0;
2526 }
2527
2528 int complete_pio(struct kvm_vcpu *vcpu)
2529 {
2530         struct kvm_pio_request *io = &vcpu->arch.pio;
2531         long delta;
2532         int r;
2533         unsigned long val;
2534
2535         if (!io->string) {
2536                 if (io->in) {
2537                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2538                         memcpy(&val, vcpu->arch.pio_data, io->size);
2539                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2540                 }
2541         } else {
2542                 if (io->in) {
2543                         r = pio_copy_data(vcpu);
2544                         if (r)
2545                                 return r;
2546                 }
2547
2548                 delta = 1;
2549                 if (io->rep) {
2550                         delta *= io->cur_count;
2551                         /*
2552                          * The size of the register should really depend on
2553                          * current address size.
2554                          */
2555                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2556                         val -= delta;
2557                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2558                 }
2559                 if (io->down)
2560                         delta = -delta;
2561                 delta *= io->size;
2562                 if (io->in) {
2563                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2564                         val += delta;
2565                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2566                 } else {
2567                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2568                         val += delta;
2569                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2570                 }
2571         }
2572
2573         io->count -= io->cur_count;
2574         io->cur_count = 0;
2575
2576         return 0;
2577 }
2578
2579 static void kernel_pio(struct kvm_io_device *pio_dev,
2580                        struct kvm_vcpu *vcpu,
2581                        void *pd)
2582 {
2583         /* TODO: String I/O for in kernel device */
2584
2585         mutex_lock(&vcpu->kvm->lock);
2586         if (vcpu->arch.pio.in)
2587                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2588                                   vcpu->arch.pio.size,
2589                                   pd);
2590         else
2591                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2592                                    vcpu->arch.pio.size,
2593                                    pd);
2594         mutex_unlock(&vcpu->kvm->lock);
2595 }
2596
2597 static void pio_string_write(struct kvm_io_device *pio_dev,
2598                              struct kvm_vcpu *vcpu)
2599 {
2600         struct kvm_pio_request *io = &vcpu->arch.pio;
2601         void *pd = vcpu->arch.pio_data;
2602         int i;
2603
2604         mutex_lock(&vcpu->kvm->lock);
2605         for (i = 0; i < io->cur_count; i++) {
2606                 kvm_iodevice_write(pio_dev, io->port,
2607                                    io->size,
2608                                    pd);
2609                 pd += io->size;
2610         }
2611         mutex_unlock(&vcpu->kvm->lock);
2612 }
2613
2614 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2615                                                gpa_t addr, int len,
2616                                                int is_write)
2617 {
2618         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2619 }
2620
2621 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2622                   int size, unsigned port)
2623 {
2624         struct kvm_io_device *pio_dev;
2625         unsigned long val;
2626
2627         vcpu->run->exit_reason = KVM_EXIT_IO;
2628         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2629         vcpu->run->io.size = vcpu->arch.pio.size = size;
2630         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2631         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2632         vcpu->run->io.port = vcpu->arch.pio.port = port;
2633         vcpu->arch.pio.in = in;
2634         vcpu->arch.pio.string = 0;
2635         vcpu->arch.pio.down = 0;
2636         vcpu->arch.pio.guest_page_offset = 0;
2637         vcpu->arch.pio.rep = 0;
2638
2639         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2640                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2641                             handler);
2642         else
2643                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2644                             handler);
2645
2646         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2647         memcpy(vcpu->arch.pio_data, &val, 4);
2648
2649         kvm_x86_ops->skip_emulated_instruction(vcpu);
2650
2651         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2652         if (pio_dev) {
2653                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2654                 complete_pio(vcpu);
2655                 return 1;
2656         }
2657         return 0;
2658 }
2659 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2660
2661 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2662                   int size, unsigned long count, int down,
2663                   gva_t address, int rep, unsigned port)
2664 {
2665         unsigned now, in_page;
2666         int i, ret = 0;
2667         int nr_pages = 1;
2668         struct page *page;
2669         struct kvm_io_device *pio_dev;
2670
2671         vcpu->run->exit_reason = KVM_EXIT_IO;
2672         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2673         vcpu->run->io.size = vcpu->arch.pio.size = size;
2674         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2675         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2676         vcpu->run->io.port = vcpu->arch.pio.port = port;
2677         vcpu->arch.pio.in = in;
2678         vcpu->arch.pio.string = 1;
2679         vcpu->arch.pio.down = down;
2680         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2681         vcpu->arch.pio.rep = rep;
2682
2683         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2684                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2685                             handler);
2686         else
2687                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2688                             handler);
2689
2690         if (!count) {
2691                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2692                 return 1;
2693         }
2694
2695         if (!down)
2696                 in_page = PAGE_SIZE - offset_in_page(address);
2697         else
2698                 in_page = offset_in_page(address) + size;
2699         now = min(count, (unsigned long)in_page / size);
2700         if (!now) {
2701                 /*
2702                  * String I/O straddles page boundary.  Pin two guest pages
2703                  * so that we satisfy atomicity constraints.  Do just one
2704                  * transaction to avoid complexity.
2705                  */
2706                 nr_pages = 2;
2707                 now = 1;
2708         }
2709         if (down) {
2710                 /*
2711                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2712                  */
2713                 pr_unimpl(vcpu, "guest string pio down\n");
2714                 kvm_inject_gp(vcpu, 0);
2715                 return 1;
2716         }
2717         vcpu->run->io.count = now;
2718         vcpu->arch.pio.cur_count = now;
2719
2720         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2721                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2722
2723         for (i = 0; i < nr_pages; ++i) {
2724                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2725                 vcpu->arch.pio.guest_pages[i] = page;
2726                 if (!page) {
2727                         kvm_inject_gp(vcpu, 0);
2728                         free_pio_guest_pages(vcpu);
2729                         return 1;
2730                 }
2731         }
2732
2733         pio_dev = vcpu_find_pio_dev(vcpu, port,
2734                                     vcpu->arch.pio.cur_count,
2735                                     !vcpu->arch.pio.in);
2736         if (!vcpu->arch.pio.in) {
2737                 /* string PIO write */
2738                 ret = pio_copy_data(vcpu);
2739                 if (ret >= 0 && pio_dev) {
2740                         pio_string_write(pio_dev, vcpu);
2741                         complete_pio(vcpu);
2742                         if (vcpu->arch.pio.count == 0)
2743                                 ret = 1;
2744                 }
2745         } else if (pio_dev)
2746                 pr_unimpl(vcpu, "no string pio read support yet, "
2747                        "port %x size %d count %ld\n",
2748                         port, size, count);
2749
2750         return ret;
2751 }
2752 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2753
2754 int kvm_arch_init(void *opaque)
2755 {
2756         int r;
2757         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2758
2759         if (kvm_x86_ops) {
2760                 printk(KERN_ERR "kvm: already loaded the other module\n");
2761                 r = -EEXIST;
2762                 goto out;
2763         }
2764
2765         if (!ops->cpu_has_kvm_support()) {
2766                 printk(KERN_ERR "kvm: no hardware support\n");
2767                 r = -EOPNOTSUPP;
2768                 goto out;
2769         }
2770         if (ops->disabled_by_bios()) {
2771                 printk(KERN_ERR "kvm: disabled by bios\n");
2772                 r = -EOPNOTSUPP;
2773                 goto out;
2774         }
2775
2776         r = kvm_mmu_module_init();
2777         if (r)
2778                 goto out;
2779
2780         kvm_init_msr_list();
2781
2782         kvm_x86_ops = ops;
2783         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2784         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2785         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2786                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2787         return 0;
2788
2789 out:
2790         return r;
2791 }
2792
2793 void kvm_arch_exit(void)
2794 {
2795         kvm_x86_ops = NULL;
2796         kvm_mmu_module_exit();
2797 }
2798
2799 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2800 {
2801         ++vcpu->stat.halt_exits;
2802         KVMTRACE_0D(HLT, vcpu, handler);
2803         if (irqchip_in_kernel(vcpu->kvm)) {
2804                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2805                 up_read(&vcpu->kvm->slots_lock);
2806                 kvm_vcpu_block(vcpu);
2807                 down_read(&vcpu->kvm->slots_lock);
2808                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2809                         return -EINTR;
2810                 return 1;
2811         } else {
2812                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2813                 return 0;
2814         }
2815 }
2816 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2817
2818 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2819                            unsigned long a1)
2820 {
2821         if (is_long_mode(vcpu))
2822                 return a0;
2823         else
2824                 return a0 | ((gpa_t)a1 << 32);
2825 }
2826
2827 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2828 {
2829         unsigned long nr, a0, a1, a2, a3, ret;
2830         int r = 1;
2831
2832         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2833         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2834         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2835         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2836         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2837
2838         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2839
2840         if (!is_long_mode(vcpu)) {
2841                 nr &= 0xFFFFFFFF;
2842                 a0 &= 0xFFFFFFFF;
2843                 a1 &= 0xFFFFFFFF;
2844                 a2 &= 0xFFFFFFFF;
2845                 a3 &= 0xFFFFFFFF;
2846         }
2847
2848         switch (nr) {
2849         case KVM_HC_VAPIC_POLL_IRQ:
2850                 ret = 0;
2851                 break;
2852         case KVM_HC_MMU_OP:
2853                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2854                 break;
2855         default:
2856                 ret = -KVM_ENOSYS;
2857                 break;
2858         }
2859         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2860         ++vcpu->stat.hypercalls;
2861         return r;
2862 }
2863 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2864
2865 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2866 {
2867         char instruction[3];
2868         int ret = 0;
2869         unsigned long rip = kvm_rip_read(vcpu);
2870
2871
2872         /*
2873          * Blow out the MMU to ensure that no other VCPU has an active mapping
2874          * to ensure that the updated hypercall appears atomically across all
2875          * VCPUs.
2876          */
2877         kvm_mmu_zap_all(vcpu->kvm);
2878
2879         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2880         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2881             != X86EMUL_CONTINUE)
2882                 ret = -EFAULT;
2883
2884         return ret;
2885 }
2886
2887 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2888 {
2889         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2890 }
2891
2892 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2893 {
2894         struct descriptor_table dt = { limit, base };
2895
2896         kvm_x86_ops->set_gdt(vcpu, &dt);
2897 }
2898
2899 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2900 {
2901         struct descriptor_table dt = { limit, base };
2902
2903         kvm_x86_ops->set_idt(vcpu, &dt);
2904 }
2905
2906 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2907                    unsigned long *rflags)
2908 {
2909         kvm_lmsw(vcpu, msw);
2910         *rflags = kvm_x86_ops->get_rflags(vcpu);
2911 }
2912
2913 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2914 {
2915         unsigned long value;
2916
2917         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2918         switch (cr) {
2919         case 0:
2920                 value = vcpu->arch.cr0;
2921                 break;
2922         case 2:
2923                 value = vcpu->arch.cr2;
2924                 break;
2925         case 3:
2926                 value = vcpu->arch.cr3;
2927                 break;
2928         case 4:
2929                 value = vcpu->arch.cr4;
2930                 break;
2931         case 8:
2932                 value = kvm_get_cr8(vcpu);
2933                 break;
2934         default:
2935                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2936                 return 0;
2937         }
2938         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2939                     (u32)((u64)value >> 32), handler);
2940
2941         return value;
2942 }
2943
2944 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2945                      unsigned long *rflags)
2946 {
2947         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2948                     (u32)((u64)val >> 32), handler);
2949
2950         switch (cr) {
2951         case 0:
2952                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2953                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2954                 break;
2955         case 2:
2956                 vcpu->arch.cr2 = val;
2957                 break;
2958         case 3:
2959                 kvm_set_cr3(vcpu, val);
2960                 break;
2961         case 4:
2962                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2963                 break;
2964         case 8:
2965                 kvm_set_cr8(vcpu, val & 0xfUL);
2966                 break;
2967         default:
2968                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2969         }
2970 }
2971
2972 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2973 {
2974         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2975         int j, nent = vcpu->arch.cpuid_nent;
2976
2977         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2978         /* when no next entry is found, the current entry[i] is reselected */
2979         for (j = i + 1; j == i; j = (j + 1) % nent) {
2980                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2981                 if (ej->function == e->function) {
2982                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2983                         return j;
2984                 }
2985         }
2986         return 0; /* silence gcc, even though control never reaches here */
2987 }
2988
2989 /* find an entry with matching function, matching index (if needed), and that
2990  * should be read next (if it's stateful) */
2991 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2992         u32 function, u32 index)
2993 {
2994         if (e->function != function)
2995                 return 0;
2996         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2997                 return 0;
2998         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2999                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3000                 return 0;
3001         return 1;
3002 }
3003
3004 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3005 {
3006         int i;
3007         u32 function, index;
3008         struct kvm_cpuid_entry2 *e, *best;
3009
3010         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3011         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3012         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3013         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3014         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3015         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3016         best = NULL;
3017         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3018                 e = &vcpu->arch.cpuid_entries[i];
3019                 if (is_matching_cpuid_entry(e, function, index)) {
3020                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3021                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3022                         best = e;
3023                         break;
3024                 }
3025                 /*
3026                  * Both basic or both extended?
3027                  */
3028                 if (((e->function ^ function) & 0x80000000) == 0)
3029                         if (!best || e->function > best->function)
3030                                 best = e;
3031         }
3032         if (best) {
3033                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3034                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3035                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3036                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3037         }
3038         kvm_x86_ops->skip_emulated_instruction(vcpu);
3039         KVMTRACE_5D(CPUID, vcpu, function,
3040                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3041                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3042                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3043                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3044 }
3045 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3046
3047 /*
3048  * Check if userspace requested an interrupt window, and that the
3049  * interrupt window is open.
3050  *
3051  * No need to exit to userspace if we already have an interrupt queued.
3052  */
3053 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3054                                           struct kvm_run *kvm_run)
3055 {
3056         return (!vcpu->arch.irq_summary &&
3057                 kvm_run->request_interrupt_window &&
3058                 vcpu->arch.interrupt_window_open &&
3059                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3060 }
3061
3062 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3063                               struct kvm_run *kvm_run)
3064 {
3065         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3066         kvm_run->cr8 = kvm_get_cr8(vcpu);
3067         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3068         if (irqchip_in_kernel(vcpu->kvm))
3069                 kvm_run->ready_for_interrupt_injection = 1;
3070         else
3071                 kvm_run->ready_for_interrupt_injection =
3072                                         (vcpu->arch.interrupt_window_open &&
3073                                          vcpu->arch.irq_summary == 0);
3074 }
3075
3076 static void vapic_enter(struct kvm_vcpu *vcpu)
3077 {
3078         struct kvm_lapic *apic = vcpu->arch.apic;
3079         struct page *page;
3080
3081         if (!apic || !apic->vapic_addr)
3082                 return;
3083
3084         down_read(&current->mm->mmap_sem);
3085         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3086         up_read(&current->mm->mmap_sem);
3087
3088         vcpu->arch.apic->vapic_page = page;
3089 }
3090
3091 static void vapic_exit(struct kvm_vcpu *vcpu)
3092 {
3093         struct kvm_lapic *apic = vcpu->arch.apic;
3094
3095         if (!apic || !apic->vapic_addr)
3096                 return;
3097
3098         down_read(&vcpu->kvm->slots_lock);
3099         kvm_release_page_dirty(apic->vapic_page);
3100         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3101         up_read(&vcpu->kvm->slots_lock);
3102 }
3103
3104 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3105 {
3106         int r;
3107
3108         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3109                 pr_debug("vcpu %d received sipi with vector # %x\n",
3110                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
3111                 kvm_lapic_reset(vcpu);
3112                 r = kvm_x86_ops->vcpu_reset(vcpu);
3113                 if (r)
3114                         return r;
3115                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3116         }
3117
3118         down_read(&vcpu->kvm->slots_lock);
3119         vapic_enter(vcpu);
3120
3121 again:
3122         if (vcpu->requests)
3123                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3124                         kvm_mmu_unload(vcpu);
3125
3126         r = kvm_mmu_reload(vcpu);
3127         if (unlikely(r))
3128                 goto out;
3129
3130         if (vcpu->requests) {
3131                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3132                         __kvm_migrate_timers(vcpu);
3133                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3134                         kvm_x86_ops->tlb_flush(vcpu);
3135                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3136                                        &vcpu->requests)) {
3137                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3138                         r = 0;
3139                         goto out;
3140                 }
3141                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3142                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3143                         r = 0;
3144                         goto out;
3145                 }
3146         }
3147
3148         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3149         kvm_inject_pending_timer_irqs(vcpu);
3150
3151         preempt_disable();
3152
3153         kvm_x86_ops->prepare_guest_switch(vcpu);
3154         kvm_load_guest_fpu(vcpu);
3155
3156         local_irq_disable();
3157
3158         if (vcpu->requests || need_resched()) {
3159                 local_irq_enable();
3160                 preempt_enable();
3161                 r = 1;
3162                 goto out;
3163         }
3164
3165         if (signal_pending(current)) {
3166                 local_irq_enable();
3167                 preempt_enable();
3168                 r = -EINTR;
3169                 kvm_run->exit_reason = KVM_EXIT_INTR;
3170                 ++vcpu->stat.signal_exits;
3171                 goto out;
3172         }
3173
3174         if (vcpu->guest_debug.enabled)
3175                 kvm_x86_ops->guest_debug_pre(vcpu);
3176
3177         vcpu->guest_mode = 1;
3178         /*
3179          * Make sure that guest_mode assignment won't happen after
3180          * testing the pending IRQ vector bitmap.
3181          */
3182         smp_wmb();
3183
3184         if (vcpu->arch.exception.pending)
3185                 __queue_exception(vcpu);
3186         else if (irqchip_in_kernel(vcpu->kvm))
3187                 kvm_x86_ops->inject_pending_irq(vcpu);
3188         else
3189                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3190
3191         kvm_lapic_sync_to_vapic(vcpu);
3192
3193         up_read(&vcpu->kvm->slots_lock);
3194
3195         kvm_guest_enter();
3196
3197
3198         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3199         kvm_x86_ops->run(vcpu, kvm_run);
3200
3201         vcpu->guest_mode = 0;
3202         local_irq_enable();
3203
3204         ++vcpu->stat.exits;
3205
3206         /*
3207          * We must have an instruction between local_irq_enable() and
3208          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3209          * the interrupt shadow.  The stat.exits increment will do nicely.
3210          * But we need to prevent reordering, hence this barrier():
3211          */
3212         barrier();
3213
3214         kvm_guest_exit();
3215
3216         preempt_enable();
3217
3218         down_read(&vcpu->kvm->slots_lock);
3219
3220         /*
3221          * Profile KVM exit RIPs:
3222          */
3223         if (unlikely(prof_on == KVM_PROFILING)) {
3224                 unsigned long rip = kvm_rip_read(vcpu);
3225                 profile_hit(KVM_PROFILING, (void *)rip);
3226         }
3227
3228         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3229                 vcpu->arch.exception.pending = false;
3230
3231         kvm_lapic_sync_from_vapic(vcpu);
3232
3233         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3234
3235         if (r > 0) {
3236                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3237                         r = -EINTR;
3238                         kvm_run->exit_reason = KVM_EXIT_INTR;
3239                         ++vcpu->stat.request_irq_exits;
3240                         goto out;
3241                 }
3242                 if (!need_resched())
3243                         goto again;
3244         }
3245
3246 out:
3247         up_read(&vcpu->kvm->slots_lock);
3248         if (r > 0) {
3249                 kvm_resched(vcpu);
3250                 down_read(&vcpu->kvm->slots_lock);
3251                 goto again;
3252         }
3253
3254         post_kvm_run_save(vcpu, kvm_run);
3255
3256         vapic_exit(vcpu);
3257
3258         return r;
3259 }
3260
3261 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3262 {
3263         int r;
3264         sigset_t sigsaved;
3265
3266         vcpu_load(vcpu);
3267
3268         if (vcpu->sigset_active)
3269                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3270
3271         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3272                 kvm_vcpu_block(vcpu);
3273                 r = -EAGAIN;
3274                 goto out;
3275         }
3276
3277         /* re-sync apic's tpr */
3278         if (!irqchip_in_kernel(vcpu->kvm))
3279                 kvm_set_cr8(vcpu, kvm_run->cr8);
3280
3281         if (vcpu->arch.pio.cur_count) {
3282                 r = complete_pio(vcpu);
3283                 if (r)
3284                         goto out;
3285         }
3286 #if CONFIG_HAS_IOMEM
3287         if (vcpu->mmio_needed) {
3288                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3289                 vcpu->mmio_read_completed = 1;
3290                 vcpu->mmio_needed = 0;
3291
3292                 down_read(&vcpu->kvm->slots_lock);
3293                 r = emulate_instruction(vcpu, kvm_run,
3294                                         vcpu->arch.mmio_fault_cr2, 0,
3295                                         EMULTYPE_NO_DECODE);
3296                 up_read(&vcpu->kvm->slots_lock);
3297                 if (r == EMULATE_DO_MMIO) {
3298                         /*
3299                          * Read-modify-write.  Back to userspace.
3300                          */
3301                         r = 0;
3302                         goto out;
3303                 }
3304         }
3305 #endif
3306         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3307                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3308                                      kvm_run->hypercall.ret);
3309
3310         r = __vcpu_run(vcpu, kvm_run);
3311
3312 out:
3313         if (vcpu->sigset_active)
3314                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3315
3316         vcpu_put(vcpu);
3317         return r;
3318 }
3319
3320 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3321 {
3322         vcpu_load(vcpu);
3323
3324         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3325         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3326         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3327         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3328         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3329         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3330         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3331         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3332 #ifdef CONFIG_X86_64
3333         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3334         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3335         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3336         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3337         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3338         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3339         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3340         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3341 #endif
3342
3343         regs->rip = kvm_rip_read(vcpu);
3344         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3345
3346         /*
3347          * Don't leak debug flags in case they were set for guest debugging
3348          */
3349         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3350                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3351
3352         vcpu_put(vcpu);
3353
3354         return 0;
3355 }
3356
3357 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3358 {
3359         vcpu_load(vcpu);
3360
3361         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3362         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3363         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3364         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3365         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3366         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3367         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3368         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3369 #ifdef CONFIG_X86_64
3370         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3371         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3372         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3373         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3374         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3375         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3376         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3377         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3378
3379 #endif
3380
3381         kvm_rip_write(vcpu, regs->rip);
3382         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3383
3384
3385         vcpu->arch.exception.pending = false;
3386
3387         vcpu_put(vcpu);
3388
3389         return 0;
3390 }
3391
3392 void kvm_get_segment(struct kvm_vcpu *vcpu,
3393                      struct kvm_segment *var, int seg)
3394 {
3395         kvm_x86_ops->get_segment(vcpu, var, seg);
3396 }
3397
3398 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3399 {
3400         struct kvm_segment cs;
3401
3402         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3403         *db = cs.db;
3404         *l = cs.l;
3405 }
3406 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3407
3408 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3409                                   struct kvm_sregs *sregs)
3410 {
3411         struct descriptor_table dt;
3412         int pending_vec;
3413
3414         vcpu_load(vcpu);
3415
3416         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3417         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3418         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3419         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3420         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3421         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3422
3423         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3424         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3425
3426         kvm_x86_ops->get_idt(vcpu, &dt);
3427         sregs->idt.limit = dt.limit;
3428         sregs->idt.base = dt.base;
3429         kvm_x86_ops->get_gdt(vcpu, &dt);
3430         sregs->gdt.limit = dt.limit;
3431         sregs->gdt.base = dt.base;
3432
3433         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3434         sregs->cr0 = vcpu->arch.cr0;
3435         sregs->cr2 = vcpu->arch.cr2;
3436         sregs->cr3 = vcpu->arch.cr3;
3437         sregs->cr4 = vcpu->arch.cr4;
3438         sregs->cr8 = kvm_get_cr8(vcpu);
3439         sregs->efer = vcpu->arch.shadow_efer;
3440         sregs->apic_base = kvm_get_apic_base(vcpu);
3441
3442         if (irqchip_in_kernel(vcpu->kvm)) {
3443                 memset(sregs->interrupt_bitmap, 0,
3444                        sizeof sregs->interrupt_bitmap);
3445                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3446                 if (pending_vec >= 0)
3447                         set_bit(pending_vec,
3448                                 (unsigned long *)sregs->interrupt_bitmap);
3449         } else
3450                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3451                        sizeof sregs->interrupt_bitmap);
3452
3453         vcpu_put(vcpu);
3454
3455         return 0;
3456 }
3457
3458 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3459                                     struct kvm_mp_state *mp_state)
3460 {
3461         vcpu_load(vcpu);
3462         mp_state->mp_state = vcpu->arch.mp_state;
3463         vcpu_put(vcpu);
3464         return 0;
3465 }
3466
3467 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3468                                     struct kvm_mp_state *mp_state)
3469 {
3470         vcpu_load(vcpu);
3471         vcpu->arch.mp_state = mp_state->mp_state;
3472         vcpu_put(vcpu);
3473         return 0;
3474 }
3475
3476 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3477                         struct kvm_segment *var, int seg)
3478 {
3479         kvm_x86_ops->set_segment(vcpu, var, seg);
3480 }
3481
3482 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3483                                    struct kvm_segment *kvm_desct)
3484 {
3485         kvm_desct->base = seg_desc->base0;
3486         kvm_desct->base |= seg_desc->base1 << 16;
3487         kvm_desct->base |= seg_desc->base2 << 24;
3488         kvm_desct->limit = seg_desc->limit0;
3489         kvm_desct->limit |= seg_desc->limit << 16;
3490         if (seg_desc->g) {
3491                 kvm_desct->limit <<= 12;
3492                 kvm_desct->limit |= 0xfff;
3493         }
3494         kvm_desct->selector = selector;
3495         kvm_desct->type = seg_desc->type;
3496         kvm_desct->present = seg_desc->p;
3497         kvm_desct->dpl = seg_desc->dpl;
3498         kvm_desct->db = seg_desc->d;
3499         kvm_desct->s = seg_desc->s;
3500         kvm_desct->l = seg_desc->l;
3501         kvm_desct->g = seg_desc->g;
3502         kvm_desct->avl = seg_desc->avl;
3503         if (!selector)
3504                 kvm_desct->unusable = 1;
3505         else
3506                 kvm_desct->unusable = 0;
3507         kvm_desct->padding = 0;
3508 }
3509
3510 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3511                                            u16 selector,
3512                                            struct descriptor_table *dtable)
3513 {
3514         if (selector & 1 << 2) {
3515                 struct kvm_segment kvm_seg;
3516
3517                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3518
3519                 if (kvm_seg.unusable)
3520                         dtable->limit = 0;
3521                 else
3522                         dtable->limit = kvm_seg.limit;
3523                 dtable->base = kvm_seg.base;
3524         }
3525         else
3526                 kvm_x86_ops->get_gdt(vcpu, dtable);
3527 }
3528
3529 /* allowed just for 8 bytes segments */
3530 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3531                                          struct desc_struct *seg_desc)
3532 {
3533         gpa_t gpa;
3534         struct descriptor_table dtable;
3535         u16 index = selector >> 3;
3536
3537         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3538
3539         if (dtable.limit < index * 8 + 7) {
3540                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3541                 return 1;
3542         }
3543         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3544         gpa += index * 8;
3545         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3546 }
3547
3548 /* allowed just for 8 bytes segments */
3549 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3550                                          struct desc_struct *seg_desc)
3551 {
3552         gpa_t gpa;
3553         struct descriptor_table dtable;
3554         u16 index = selector >> 3;
3555
3556         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3557
3558         if (dtable.limit < index * 8 + 7)
3559                 return 1;
3560         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3561         gpa += index * 8;
3562         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3563 }
3564
3565 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3566                              struct desc_struct *seg_desc)
3567 {
3568         u32 base_addr;
3569
3570         base_addr = seg_desc->base0;
3571         base_addr |= (seg_desc->base1 << 16);
3572         base_addr |= (seg_desc->base2 << 24);
3573
3574         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3575 }
3576
3577 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3578 {
3579         struct kvm_segment kvm_seg;
3580
3581         kvm_get_segment(vcpu, &kvm_seg, seg);
3582         return kvm_seg.selector;
3583 }
3584
3585 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3586                                                 u16 selector,
3587                                                 struct kvm_segment *kvm_seg)
3588 {
3589         struct desc_struct seg_desc;
3590
3591         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3592                 return 1;
3593         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3594         return 0;
3595 }
3596
3597 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3598                                 int type_bits, int seg)
3599 {
3600         struct kvm_segment kvm_seg;
3601
3602         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3603                 return 1;
3604         kvm_seg.type |= type_bits;
3605
3606         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3607             seg != VCPU_SREG_LDTR)
3608                 if (!kvm_seg.s)
3609                         kvm_seg.unusable = 1;
3610
3611         kvm_set_segment(vcpu, &kvm_seg, seg);
3612         return 0;
3613 }
3614
3615 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3616                                 struct tss_segment_32 *tss)
3617 {
3618         tss->cr3 = vcpu->arch.cr3;
3619         tss->eip = kvm_rip_read(vcpu);
3620         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3621         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3622         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3623         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3624         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3625         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3626         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3627         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3628         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3629         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3630         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3631         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3632         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3633         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3634         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3635         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3636         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3637 }
3638
3639 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3640                                   struct tss_segment_32 *tss)
3641 {
3642         kvm_set_cr3(vcpu, tss->cr3);
3643
3644         kvm_rip_write(vcpu, tss->eip);
3645         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3646
3647         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3648         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3649         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3650         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3651         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3652         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3653         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3654         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3655
3656         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3657                 return 1;
3658
3659         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3660                 return 1;
3661
3662         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3663                 return 1;
3664
3665         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3666                 return 1;
3667
3668         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3669                 return 1;
3670
3671         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3672                 return 1;
3673
3674         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3675                 return 1;
3676         return 0;
3677 }
3678
3679 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3680                                 struct tss_segment_16 *tss)
3681 {
3682         tss->ip = kvm_rip_read(vcpu);
3683         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3684         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3685         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3686         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3687         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3688         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3689         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3690         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3691         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3692
3693         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3694         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3695         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3696         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3697         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3698         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3699 }
3700
3701 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3702                                  struct tss_segment_16 *tss)
3703 {
3704         kvm_rip_write(vcpu, tss->ip);
3705         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3706         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3707         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3708         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3709         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3710         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3711         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3712         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3713         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3714
3715         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3716                 return 1;
3717
3718         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3719                 return 1;
3720
3721         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3722                 return 1;
3723
3724         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3725                 return 1;
3726
3727         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3728                 return 1;
3729         return 0;
3730 }
3731
3732 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3733                        u32 old_tss_base,
3734                        struct desc_struct *nseg_desc)
3735 {
3736         struct tss_segment_16 tss_segment_16;
3737         int ret = 0;
3738
3739         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3740                            sizeof tss_segment_16))
3741                 goto out;
3742
3743         save_state_to_tss16(vcpu, &tss_segment_16);
3744
3745         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3746                             sizeof tss_segment_16))
3747                 goto out;
3748
3749         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3750                            &tss_segment_16, sizeof tss_segment_16))
3751                 goto out;
3752
3753         if (load_state_from_tss16(vcpu, &tss_segment_16))
3754                 goto out;
3755
3756         ret = 1;
3757 out:
3758         return ret;
3759 }
3760
3761 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3762                        u32 old_tss_base,
3763                        struct desc_struct *nseg_desc)
3764 {
3765         struct tss_segment_32 tss_segment_32;
3766         int ret = 0;
3767
3768         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3769                            sizeof tss_segment_32))
3770                 goto out;
3771
3772         save_state_to_tss32(vcpu, &tss_segment_32);
3773
3774         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3775                             sizeof tss_segment_32))
3776                 goto out;
3777
3778         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3779                            &tss_segment_32, sizeof tss_segment_32))
3780                 goto out;
3781
3782         if (load_state_from_tss32(vcpu, &tss_segment_32))
3783                 goto out;
3784
3785         ret = 1;
3786 out:
3787         return ret;
3788 }
3789
3790 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3791 {
3792         struct kvm_segment tr_seg;
3793         struct desc_struct cseg_desc;
3794         struct desc_struct nseg_desc;
3795         int ret = 0;
3796         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3797         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3798
3799         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3800
3801         /* FIXME: Handle errors. Failure to read either TSS or their
3802          * descriptors should generate a pagefault.
3803          */
3804         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3805                 goto out;
3806
3807         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3808                 goto out;
3809
3810         if (reason != TASK_SWITCH_IRET) {
3811                 int cpl;
3812
3813                 cpl = kvm_x86_ops->get_cpl(vcpu);
3814                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3815                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3816                         return 1;
3817                 }
3818         }
3819
3820         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3821                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3822                 return 1;
3823         }
3824
3825         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3826                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3827                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3828         }
3829
3830         if (reason == TASK_SWITCH_IRET) {
3831                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3832                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3833         }
3834
3835         kvm_x86_ops->skip_emulated_instruction(vcpu);
3836
3837         if (nseg_desc.type & 8)
3838                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3839                                          &nseg_desc);
3840         else
3841                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3842                                          &nseg_desc);
3843
3844         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3845                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3846                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3847         }
3848
3849         if (reason != TASK_SWITCH_IRET) {
3850                 nseg_desc.type |= (1 << 1);
3851                 save_guest_segment_descriptor(vcpu, tss_selector,
3852                                               &nseg_desc);
3853         }
3854
3855         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3856         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3857         tr_seg.type = 11;
3858         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3859 out:
3860         return ret;
3861 }
3862 EXPORT_SYMBOL_GPL(kvm_task_switch);
3863
3864 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3865                                   struct kvm_sregs *sregs)
3866 {
3867         int mmu_reset_needed = 0;
3868         int i, pending_vec, max_bits;
3869         struct descriptor_table dt;
3870
3871         vcpu_load(vcpu);
3872
3873         dt.limit = sregs->idt.limit;
3874         dt.base = sregs->idt.base;
3875         kvm_x86_ops->set_idt(vcpu, &dt);
3876         dt.limit = sregs->gdt.limit;
3877         dt.base = sregs->gdt.base;
3878         kvm_x86_ops->set_gdt(vcpu, &dt);
3879
3880         vcpu->arch.cr2 = sregs->cr2;
3881         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3882         vcpu->arch.cr3 = sregs->cr3;
3883
3884         kvm_set_cr8(vcpu, sregs->cr8);
3885
3886         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3887         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3888         kvm_set_apic_base(vcpu, sregs->apic_base);
3889
3890         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3891
3892         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3893         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3894         vcpu->arch.cr0 = sregs->cr0;
3895
3896         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3897         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3898         if (!is_long_mode(vcpu) && is_pae(vcpu))
3899                 load_pdptrs(vcpu, vcpu->arch.cr3);
3900
3901         if (mmu_reset_needed)
3902                 kvm_mmu_reset_context(vcpu);
3903
3904         if (!irqchip_in_kernel(vcpu->kvm)) {
3905                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3906                        sizeof vcpu->arch.irq_pending);
3907                 vcpu->arch.irq_summary = 0;
3908                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3909                         if (vcpu->arch.irq_pending[i])
3910                                 __set_bit(i, &vcpu->arch.irq_summary);
3911         } else {
3912                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3913                 pending_vec = find_first_bit(
3914                         (const unsigned long *)sregs->interrupt_bitmap,
3915                         max_bits);
3916                 /* Only pending external irq is handled here */
3917                 if (pending_vec < max_bits) {
3918                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3919                         pr_debug("Set back pending irq %d\n",
3920                                  pending_vec);
3921                 }
3922         }
3923
3924         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3925         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3926         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3927         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3928         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3929         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3930
3931         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3932         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3933
3934         vcpu_put(vcpu);
3935
3936         return 0;
3937 }
3938
3939 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3940                                     struct kvm_debug_guest *dbg)
3941 {
3942         int r;
3943
3944         vcpu_load(vcpu);
3945
3946         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3947
3948         vcpu_put(vcpu);
3949
3950         return r;
3951 }
3952
3953 /*
3954  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3955  * we have asm/x86/processor.h
3956  */
3957 struct fxsave {
3958         u16     cwd;
3959         u16     swd;
3960         u16     twd;
3961         u16     fop;
3962         u64     rip;
3963         u64     rdp;
3964         u32     mxcsr;
3965         u32     mxcsr_mask;
3966         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3967 #ifdef CONFIG_X86_64
3968         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3969 #else
3970         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3971 #endif
3972 };
3973
3974 /*
3975  * Translate a guest virtual address to a guest physical address.
3976  */
3977 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3978                                     struct kvm_translation *tr)
3979 {
3980         unsigned long vaddr = tr->linear_address;
3981         gpa_t gpa;
3982
3983         vcpu_load(vcpu);
3984         down_read(&vcpu->kvm->slots_lock);
3985         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3986         up_read(&vcpu->kvm->slots_lock);
3987         tr->physical_address = gpa;
3988         tr->valid = gpa != UNMAPPED_GVA;
3989         tr->writeable = 1;
3990         tr->usermode = 0;
3991         vcpu_put(vcpu);
3992
3993         return 0;
3994 }
3995
3996 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3997 {
3998         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3999
4000         vcpu_load(vcpu);
4001
4002         memcpy(fpu->fpr, fxsave->st_space, 128);
4003         fpu->fcw = fxsave->cwd;
4004         fpu->fsw = fxsave->swd;
4005         fpu->ftwx = fxsave->twd;
4006         fpu->last_opcode = fxsave->fop;
4007         fpu->last_ip = fxsave->rip;
4008         fpu->last_dp = fxsave->rdp;
4009         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4010
4011         vcpu_put(vcpu);
4012
4013         return 0;
4014 }
4015
4016 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4017 {
4018         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4019
4020         vcpu_load(vcpu);
4021
4022         memcpy(fxsave->st_space, fpu->fpr, 128);
4023         fxsave->cwd = fpu->fcw;
4024         fxsave->swd = fpu->fsw;
4025         fxsave->twd = fpu->ftwx;
4026         fxsave->fop = fpu->last_opcode;
4027         fxsave->rip = fpu->last_ip;
4028         fxsave->rdp = fpu->last_dp;
4029         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4030
4031         vcpu_put(vcpu);
4032
4033         return 0;
4034 }
4035
4036 void fx_init(struct kvm_vcpu *vcpu)
4037 {
4038         unsigned after_mxcsr_mask;
4039
4040         /*
4041          * Touch the fpu the first time in non atomic context as if
4042          * this is the first fpu instruction the exception handler
4043          * will fire before the instruction returns and it'll have to
4044          * allocate ram with GFP_KERNEL.
4045          */
4046         if (!used_math())
4047                 kvm_fx_save(&vcpu->arch.host_fx_image);
4048
4049         /* Initialize guest FPU by resetting ours and saving into guest's */
4050         preempt_disable();
4051         kvm_fx_save(&vcpu->arch.host_fx_image);
4052         kvm_fx_finit();
4053         kvm_fx_save(&vcpu->arch.guest_fx_image);
4054         kvm_fx_restore(&vcpu->arch.host_fx_image);
4055         preempt_enable();
4056
4057         vcpu->arch.cr0 |= X86_CR0_ET;
4058         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4059         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4060         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4061                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4062 }
4063 EXPORT_SYMBOL_GPL(fx_init);
4064
4065 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4066 {
4067         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4068                 return;
4069
4070         vcpu->guest_fpu_loaded = 1;
4071         kvm_fx_save(&vcpu->arch.host_fx_image);
4072         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4073 }
4074 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4075
4076 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4077 {
4078         if (!vcpu->guest_fpu_loaded)
4079                 return;
4080
4081         vcpu->guest_fpu_loaded = 0;
4082         kvm_fx_save(&vcpu->arch.guest_fx_image);
4083         kvm_fx_restore(&vcpu->arch.host_fx_image);
4084         ++vcpu->stat.fpu_reload;
4085 }
4086 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4087
4088 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4089 {
4090         kvm_x86_ops->vcpu_free(vcpu);
4091 }
4092
4093 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4094                                                 unsigned int id)
4095 {
4096         return kvm_x86_ops->vcpu_create(kvm, id);
4097 }
4098
4099 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4100 {
4101         int r;
4102
4103         /* We do fxsave: this must be aligned. */
4104         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4105
4106         vcpu_load(vcpu);
4107         r = kvm_arch_vcpu_reset(vcpu);
4108         if (r == 0)
4109                 r = kvm_mmu_setup(vcpu);
4110         vcpu_put(vcpu);
4111         if (r < 0)
4112                 goto free_vcpu;
4113
4114         return 0;
4115 free_vcpu:
4116         kvm_x86_ops->vcpu_free(vcpu);
4117         return r;
4118 }
4119
4120 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4121 {
4122         vcpu_load(vcpu);
4123         kvm_mmu_unload(vcpu);
4124         vcpu_put(vcpu);
4125
4126         kvm_x86_ops->vcpu_free(vcpu);
4127 }
4128
4129 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4130 {
4131         return kvm_x86_ops->vcpu_reset(vcpu);
4132 }
4133
4134 void kvm_arch_hardware_enable(void *garbage)
4135 {
4136         kvm_x86_ops->hardware_enable(garbage);
4137 }
4138
4139 void kvm_arch_hardware_disable(void *garbage)
4140 {
4141         kvm_x86_ops->hardware_disable(garbage);
4142 }
4143
4144 int kvm_arch_hardware_setup(void)
4145 {
4146         return kvm_x86_ops->hardware_setup();
4147 }
4148
4149 void kvm_arch_hardware_unsetup(void)
4150 {
4151         kvm_x86_ops->hardware_unsetup();
4152 }
4153
4154 void kvm_arch_check_processor_compat(void *rtn)
4155 {
4156         kvm_x86_ops->check_processor_compatibility(rtn);
4157 }
4158
4159 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4160 {
4161         struct page *page;
4162         struct kvm *kvm;
4163         int r;
4164
4165         BUG_ON(vcpu->kvm == NULL);
4166         kvm = vcpu->kvm;
4167
4168         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4169         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4170                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4171         else
4172                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4173
4174         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4175         if (!page) {
4176                 r = -ENOMEM;
4177                 goto fail;
4178         }
4179         vcpu->arch.pio_data = page_address(page);
4180
4181         r = kvm_mmu_create(vcpu);
4182         if (r < 0)
4183                 goto fail_free_pio_data;
4184
4185         if (irqchip_in_kernel(kvm)) {
4186                 r = kvm_create_lapic(vcpu);
4187                 if (r < 0)
4188                         goto fail_mmu_destroy;
4189         }
4190
4191         return 0;
4192
4193 fail_mmu_destroy:
4194         kvm_mmu_destroy(vcpu);
4195 fail_free_pio_data:
4196         free_page((unsigned long)vcpu->arch.pio_data);
4197 fail:
4198         return r;
4199 }
4200
4201 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4202 {
4203         kvm_free_lapic(vcpu);
4204         down_read(&vcpu->kvm->slots_lock);
4205         kvm_mmu_destroy(vcpu);
4206         up_read(&vcpu->kvm->slots_lock);
4207         free_page((unsigned long)vcpu->arch.pio_data);
4208 }
4209
4210 struct  kvm *kvm_arch_create_vm(void)
4211 {
4212         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4213
4214         if (!kvm)
4215                 return ERR_PTR(-ENOMEM);
4216
4217         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4218         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4219
4220         return kvm;
4221 }
4222
4223 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4224 {
4225         vcpu_load(vcpu);
4226         kvm_mmu_unload(vcpu);
4227         vcpu_put(vcpu);
4228 }
4229
4230 static void kvm_free_vcpus(struct kvm *kvm)
4231 {
4232         unsigned int i;
4233
4234         /*
4235          * Unpin any mmu pages first.
4236          */
4237         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4238                 if (kvm->vcpus[i])
4239                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4240         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4241                 if (kvm->vcpus[i]) {
4242                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4243                         kvm->vcpus[i] = NULL;
4244                 }
4245         }
4246
4247 }
4248
4249 void kvm_arch_destroy_vm(struct kvm *kvm)
4250 {
4251         kvm_free_assigned_devices(kvm);
4252         kvm_free_pit(kvm);
4253         kfree(kvm->arch.vpic);
4254         kfree(kvm->arch.vioapic);
4255         kvm_free_vcpus(kvm);
4256         kvm_free_physmem(kvm);
4257         if (kvm->arch.apic_access_page)
4258                 put_page(kvm->arch.apic_access_page);
4259         if (kvm->arch.ept_identity_pagetable)
4260                 put_page(kvm->arch.ept_identity_pagetable);
4261         kfree(kvm);
4262 }
4263
4264 int kvm_arch_set_memory_region(struct kvm *kvm,
4265                                 struct kvm_userspace_memory_region *mem,
4266                                 struct kvm_memory_slot old,
4267                                 int user_alloc)
4268 {
4269         int npages = mem->memory_size >> PAGE_SHIFT;
4270         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4271
4272         /*To keep backward compatibility with older userspace,
4273          *x86 needs to hanlde !user_alloc case.
4274          */
4275         if (!user_alloc) {
4276                 if (npages && !old.rmap) {
4277                         unsigned long userspace_addr;
4278
4279                         down_write(&current->mm->mmap_sem);
4280                         userspace_addr = do_mmap(NULL, 0,
4281                                                  npages * PAGE_SIZE,
4282                                                  PROT_READ | PROT_WRITE,
4283                                                  MAP_SHARED | MAP_ANONYMOUS,
4284                                                  0);
4285                         up_write(&current->mm->mmap_sem);
4286
4287                         if (IS_ERR((void *)userspace_addr))
4288                                 return PTR_ERR((void *)userspace_addr);
4289
4290                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4291                         spin_lock(&kvm->mmu_lock);
4292                         memslot->userspace_addr = userspace_addr;
4293                         spin_unlock(&kvm->mmu_lock);
4294                 } else {
4295                         if (!old.user_alloc && old.rmap) {
4296                                 int ret;
4297
4298                                 down_write(&current->mm->mmap_sem);
4299                                 ret = do_munmap(current->mm, old.userspace_addr,
4300                                                 old.npages * PAGE_SIZE);
4301                                 up_write(&current->mm->mmap_sem);
4302                                 if (ret < 0)
4303                                         printk(KERN_WARNING
4304                                        "kvm_vm_ioctl_set_memory_region: "
4305                                        "failed to munmap memory\n");
4306                         }
4307                 }
4308         }
4309
4310         if (!kvm->arch.n_requested_mmu_pages) {
4311                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4312                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4313         }
4314
4315         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4316         kvm_flush_remote_tlbs(kvm);
4317
4318         return 0;
4319 }
4320
4321 void kvm_arch_flush_shadow(struct kvm *kvm)
4322 {
4323         kvm_mmu_zap_all(kvm);
4324 }
4325
4326 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4327 {
4328         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4329                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4330 }
4331
4332 static void vcpu_kick_intr(void *info)
4333 {
4334 #ifdef DEBUG
4335         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4336         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4337 #endif
4338 }
4339
4340 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4341 {
4342         int ipi_pcpu = vcpu->cpu;
4343         int cpu = get_cpu();
4344
4345         if (waitqueue_active(&vcpu->wq)) {
4346                 wake_up_interruptible(&vcpu->wq);
4347                 ++vcpu->stat.halt_wakeup;
4348         }
4349         /*
4350          * We may be called synchronously with irqs disabled in guest mode,
4351          * So need not to call smp_call_function_single() in that case.
4352          */
4353         if (vcpu->guest_mode && vcpu->cpu != cpu)
4354                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4355         put_cpu();
4356 }