SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40 #include <linux/user-return-notifier.h>
  41 #include <trace/events/kvm.h>
  42 #undef TRACE_INCLUDE_FILE
  43 #define CREATE_TRACE_POINTS
  44 #include "trace.h"
  45
  46 #include <asm/debugreg.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/msr.h>
  49 #include <asm/desc.h>
  50 #include <asm/mtrr.h>
  51 #include <asm/mce.h>
  52
  53 #define MAX_IO_MSRS 256
  54 #define CR0_RESERVED_BITS                                               \
  55         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  56                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  57                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  58 #define CR4_RESERVED_BITS                                               \
  59         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  60                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  61                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  62                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  63
  64 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  65
  66 #define KVM_MAX_MCE_BANKS 32
  67 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  68
  69 /* EFER defaults:
  70  * - enable syscall per default because its emulated by KVM
  71  * - enable LME and LMA per default on 64 bit KVM
  72  */
  73 #ifdef CONFIG_X86_64
  74 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  75 #else
  76 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  77 #endif
  78
  79 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  80 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  81
  82 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  83 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  84                                     struct kvm_cpuid_entry2 __user *entries);
  85
  86 struct kvm_x86_ops *kvm_x86_ops;
  87 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  88
  89 int ignore_msrs = 0;
  90 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  91
  92 #define KVM_NR_SHARED_MSRS 16
  93
  94 struct kvm_shared_msrs_global {
  95         int nr;
  96         u32 msrs[KVM_NR_SHARED_MSRS];
  97 };
  98
  99 struct kvm_shared_msrs {
 100         struct user_return_notifier urn;
 101         bool registered;
 102         struct kvm_shared_msr_values {
 103                 u64 host;
 104                 u64 curr;
 105         } values[KVM_NR_SHARED_MSRS];
 106 };
 107
 108 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 109 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 110
 111 struct kvm_stats_debugfs_item debugfs_entries[] = {
 112         { "pf_fixed", VCPU_STAT(pf_fixed) },
 113         { "pf_guest", VCPU_STAT(pf_guest) },
 114         { "tlb_flush", VCPU_STAT(tlb_flush) },
 115         { "invlpg", VCPU_STAT(invlpg) },
 116         { "exits", VCPU_STAT(exits) },
 117         { "io_exits", VCPU_STAT(io_exits) },
 118         { "mmio_exits", VCPU_STAT(mmio_exits) },
 119         { "signal_exits", VCPU_STAT(signal_exits) },
 120         { "irq_window", VCPU_STAT(irq_window_exits) },
 121         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 122         { "halt_exits", VCPU_STAT(halt_exits) },
 123         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 124         { "hypercalls", VCPU_STAT(hypercalls) },
 125         { "request_irq", VCPU_STAT(request_irq_exits) },
 126         { "irq_exits", VCPU_STAT(irq_exits) },
 127         { "host_state_reload", VCPU_STAT(host_state_reload) },
 128         { "efer_reload", VCPU_STAT(efer_reload) },
 129         { "fpu_reload", VCPU_STAT(fpu_reload) },
 130         { "insn_emulation", VCPU_STAT(insn_emulation) },
 131         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 132         { "irq_injections", VCPU_STAT(irq_injections) },
 133         { "nmi_injections", VCPU_STAT(nmi_injections) },
 134         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 135         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 136         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 137         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 138         { "mmu_flooded", VM_STAT(mmu_flooded) },
 139         { "mmu_recycled", VM_STAT(mmu_recycled) },
 140         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 141         { "mmu_unsync", VM_STAT(mmu_unsync) },
 142         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 143         { "largepages", VM_STAT(lpages) },
 144         { NULL }
 145 };
 146
 147 static void kvm_on_user_return(struct user_return_notifier *urn)
 148 {
 149         unsigned slot;
 150         struct kvm_shared_msrs *locals
 151                 = container_of(urn, struct kvm_shared_msrs, urn);
 152         struct kvm_shared_msr_values *values;
 153
 154         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 155                 values = &locals->values[slot];
 156                 if (values->host != values->curr) {
 157                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 158                         values->curr = values->host;
 159                 }
 160         }
 161         locals->registered = false;
 162         user_return_notifier_unregister(urn);
 163 }
 164
 165 static void shared_msr_update(unsigned slot, u32 msr)
 166 {
 167         struct kvm_shared_msrs *smsr;
 168         u64 value;
 169
 170         smsr = &__get_cpu_var(shared_msrs);
 171         /* only read, and nobody should modify it at this time,
 172          * so don't need lock */
 173         if (slot >= shared_msrs_global.nr) {
 174                 printk(KERN_ERR "kvm: invalid MSR slot!");
 175                 return;
 176         }
 177         rdmsrl_safe(msr, &value);
 178         smsr->values[slot].host = value;
 179         smsr->values[slot].curr = value;
 180 }
 181
 182 void kvm_define_shared_msr(unsigned slot, u32 msr)
 183 {
 184         if (slot >= shared_msrs_global.nr)
 185                 shared_msrs_global.nr = slot + 1;
 186         shared_msrs_global.msrs[slot] = msr;
 187         /* we need ensured the shared_msr_global have been updated */
 188         smp_wmb();
 189 }
 190 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 191
 192 static void kvm_shared_msr_cpu_online(void)
 193 {
 194         unsigned i;
 195
 196         for (i = 0; i < shared_msrs_global.nr; ++i)
 197                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 198 }
 199
 200 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 201 {
 202         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 203
 204         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 205                 return;
 206         smsr->values[slot].curr = value;
 207         wrmsrl(shared_msrs_global.msrs[slot], value);
 208         if (!smsr->registered) {
 209                 smsr->urn.on_user_return = kvm_on_user_return;
 210                 user_return_notifier_register(&smsr->urn);
 211                 smsr->registered = true;
 212         }
 213 }
 214 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 215
 216 static void drop_user_return_notifiers(void *ignore)
 217 {
 218         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 219
 220         if (smsr->registered)
 221                 kvm_on_user_return(&smsr->urn);
 222 }
 223
 224 unsigned long segment_base(u16 selector)
 225 {
 226         struct descriptor_table gdt;
 227         struct desc_struct *d;
 228         unsigned long table_base;
 229         unsigned long v;
 230
 231         if (selector == 0)
 232                 return 0;
 233
 234         kvm_get_gdt(&gdt);
 235         table_base = gdt.base;
 236
 237         if (selector & 4) {           /* from ldt */
 238                 u16 ldt_selector = kvm_read_ldt();
 239
 240                 table_base = segment_base(ldt_selector);
 241         }
 242         d = (struct desc_struct *)(table_base + (selector & ~7));
 243         v = get_desc_base(d);
 244 #ifdef CONFIG_X86_64
 245         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 246                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 247 #endif
 248         return v;
 249 }
 250 EXPORT_SYMBOL_GPL(segment_base);
 251
 252 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 253 {
 254         if (irqchip_in_kernel(vcpu->kvm))
 255                 return vcpu->arch.apic_base;
 256         else
 257                 return vcpu->arch.apic_base;
 258 }
 259 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 260
 261 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 262 {
 263         /* TODO: reserve bits check */
 264         if (irqchip_in_kernel(vcpu->kvm))
 265                 kvm_lapic_set_base(vcpu, data);
 266         else
 267                 vcpu->arch.apic_base = data;
 268 }
 269 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 270
 271 #define EXCPT_BENIGN            0
 272 #define EXCPT_CONTRIBUTORY      1
 273 #define EXCPT_PF                2
 274
 275 static int exception_class(int vector)
 276 {
 277         switch (vector) {
 278         case PF_VECTOR:
 279                 return EXCPT_PF;
 280         case DE_VECTOR:
 281         case TS_VECTOR:
 282         case NP_VECTOR:
 283         case SS_VECTOR:
 284         case GP_VECTOR:
 285                 return EXCPT_CONTRIBUTORY;
 286         default:
 287                 break;
 288         }
 289         return EXCPT_BENIGN;
 290 }
 291
 292 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 293                 unsigned nr, bool has_error, u32 error_code)
 294 {
 295         u32 prev_nr;
 296         int class1, class2;
 297
 298         if (!vcpu->arch.exception.pending) {
 299         queue:
 300                 vcpu->arch.exception.pending = true;
 301                 vcpu->arch.exception.has_error_code = has_error;
 302                 vcpu->arch.exception.nr = nr;
 303                 vcpu->arch.exception.error_code = error_code;
 304                 return;
 305         }
 306
 307         /* to check exception */
 308         prev_nr = vcpu->arch.exception.nr;
 309         if (prev_nr == DF_VECTOR) {
 310                 /* triple fault -> shutdown */
 311                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 312                 return;
 313         }
 314         class1 = exception_class(prev_nr);
 315         class2 = exception_class(nr);
 316         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 317                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 318                 /* generate double fault per SDM Table 5-5 */
 319                 vcpu->arch.exception.pending = true;
 320                 vcpu->arch.exception.has_error_code = true;
 321                 vcpu->arch.exception.nr = DF_VECTOR;
 322                 vcpu->arch.exception.error_code = 0;
 323         } else
 324                 /* replace previous exception with a new one in a hope
 325                    that instruction re-execution will regenerate lost
 326                    exception */
 327                 goto queue;
 328 }
 329
 330 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 331 {
 332         kvm_multiple_exception(vcpu, nr, false, 0);
 333 }
 334 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 335
 336 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 337                            u32 error_code)
 338 {
 339         ++vcpu->stat.pf_guest;
 340         vcpu->arch.cr2 = addr;
 341         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 342 }
 343
 344 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 345 {
 346         vcpu->arch.nmi_pending = 1;
 347 }
 348 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 349
 350 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 351 {
 352         kvm_multiple_exception(vcpu, nr, true, error_code);
 353 }
 354 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 355
 356 /*
 357  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 358  * a #GP and return false.
 359  */
 360 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 361 {
 362         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 363                 return true;
 364         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 365         return false;
 366 }
 367 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 368
 369 /*
 370  * Load the pae pdptrs.  Return true is they are all valid.
 371  */
 372 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 373 {
 374         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 375         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 376         int i;
 377         int ret;
 378         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 379
 380         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 381                                   offset * sizeof(u64), sizeof(pdpte));
 382         if (ret < 0) {
 383                 ret = 0;
 384                 goto out;
 385         }
 386         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 387                 if (is_present_gpte(pdpte[i]) &&
 388                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 389                         ret = 0;
 390                         goto out;
 391                 }
 392         }
 393         ret = 1;
 394
 395         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 396         __set_bit(VCPU_EXREG_PDPTR,
 397                   (unsigned long *)&vcpu->arch.regs_avail);
 398         __set_bit(VCPU_EXREG_PDPTR,
 399                   (unsigned long *)&vcpu->arch.regs_dirty);
 400 out:
 401
 402         return ret;
 403 }
 404 EXPORT_SYMBOL_GPL(load_pdptrs);
 405
 406 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 407 {
 408         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 409         bool changed = true;
 410         int r;
 411
 412         if (is_long_mode(vcpu) || !is_pae(vcpu))
 413                 return false;
 414
 415         if (!test_bit(VCPU_EXREG_PDPTR,
 416                       (unsigned long *)&vcpu->arch.regs_avail))
 417                 return true;
 418
 419         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 420         if (r < 0)
 421                 goto out;
 422         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 423 out:
 424
 425         return changed;
 426 }
 427
 428 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 429 {
 430         if (cr0 & CR0_RESERVED_BITS) {
 431                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 432                        cr0, vcpu->arch.cr0);
 433                 kvm_inject_gp(vcpu, 0);
 434                 return;
 435         }
 436
 437         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 438                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 439                 kvm_inject_gp(vcpu, 0);
 440                 return;
 441         }
 442
 443         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 444                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 445                        "and a clear PE flag\n");
 446                 kvm_inject_gp(vcpu, 0);
 447                 return;
 448         }
 449
 450         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 451 #ifdef CONFIG_X86_64
 452                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 453                         int cs_db, cs_l;
 454
 455                         if (!is_pae(vcpu)) {
 456                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 457                                        "in long mode while PAE is disabled\n");
 458                                 kvm_inject_gp(vcpu, 0);
 459                                 return;
 460                         }
 461                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 462                         if (cs_l) {
 463                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 464                                        "in long mode while CS.L == 1\n");
 465                                 kvm_inject_gp(vcpu, 0);
 466                                 return;
 467
 468                         }
 469                 } else
 470 #endif
 471                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 472                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 473                                "reserved bits\n");
 474                         kvm_inject_gp(vcpu, 0);
 475                         return;
 476                 }
 477
 478         }
 479
 480         kvm_x86_ops->set_cr0(vcpu, cr0);
 481         vcpu->arch.cr0 = cr0;
 482
 483         kvm_mmu_reset_context(vcpu);
 484         return;
 485 }
 486 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 487
 488 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 489 {
 490         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 491 }
 492 EXPORT_SYMBOL_GPL(kvm_lmsw);
 493
 494 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 495 {
 496         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 497         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 498
 499         if (cr4 & CR4_RESERVED_BITS) {
 500                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 501                 kvm_inject_gp(vcpu, 0);
 502                 return;
 503         }
 504
 505         if (is_long_mode(vcpu)) {
 506                 if (!(cr4 & X86_CR4_PAE)) {
 507                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 508                                "in long mode\n");
 509                         kvm_inject_gp(vcpu, 0);
 510                         return;
 511                 }
 512         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 513                    && ((cr4 ^ old_cr4) & pdptr_bits)
 514                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 515                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 516                 kvm_inject_gp(vcpu, 0);
 517                 return;
 518         }
 519
 520         if (cr4 & X86_CR4_VMXE) {
 521                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 522                 kvm_inject_gp(vcpu, 0);
 523                 return;
 524         }
 525         kvm_x86_ops->set_cr4(vcpu, cr4);
 526         vcpu->arch.cr4 = cr4;
 527         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 528         kvm_mmu_reset_context(vcpu);
 529 }
 530 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 531
 532 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 533 {
 534         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 535                 kvm_mmu_sync_roots(vcpu);
 536                 kvm_mmu_flush_tlb(vcpu);
 537                 return;
 538         }
 539
 540         if (is_long_mode(vcpu)) {
 541                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 542                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 543                         kvm_inject_gp(vcpu, 0);
 544                         return;
 545                 }
 546         } else {
 547                 if (is_pae(vcpu)) {
 548                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 549                                 printk(KERN_DEBUG
 550                                        "set_cr3: #GP, reserved bits\n");
 551                                 kvm_inject_gp(vcpu, 0);
 552                                 return;
 553                         }
 554                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 555                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 556                                        "reserved bits\n");
 557                                 kvm_inject_gp(vcpu, 0);
 558                                 return;
 559                         }
 560                 }
 561                 /*
 562                  * We don't check reserved bits in nonpae mode, because
 563                  * this isn't enforced, and VMware depends on this.
 564                  */
 565         }
 566
 567         /*
 568          * Does the new cr3 value map to physical memory? (Note, we
 569          * catch an invalid cr3 even in real-mode, because it would
 570          * cause trouble later on when we turn on paging anyway.)
 571          *
 572          * A real CPU would silently accept an invalid cr3 and would
 573          * attempt to use it - with largely undefined (and often hard
 574          * to debug) behavior on the guest side.
 575          */
 576         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 577                 kvm_inject_gp(vcpu, 0);
 578         else {
 579                 vcpu->arch.cr3 = cr3;
 580                 vcpu->arch.mmu.new_cr3(vcpu);
 581         }
 582 }
 583 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 584
 585 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 586 {
 587         if (cr8 & CR8_RESERVED_BITS) {
 588                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 589                 kvm_inject_gp(vcpu, 0);
 590                 return;
 591         }
 592         if (irqchip_in_kernel(vcpu->kvm))
 593                 kvm_lapic_set_tpr(vcpu, cr8);
 594         else
 595                 vcpu->arch.cr8 = cr8;
 596 }
 597 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 598
 599 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 600 {
 601         if (irqchip_in_kernel(vcpu->kvm))
 602                 return kvm_lapic_get_cr8(vcpu);
 603         else
 604                 return vcpu->arch.cr8;
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 607
 608 static inline u32 bit(int bitno)
 609 {
 610         return 1 << (bitno & 31);
 611 }
 612
 613 /*
 614  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 615  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 616  *
 617  * This list is modified at module load time to reflect the
 618  * capabilities of the host cpu. This capabilities test skips MSRs that are
 619  * kvm-specific. Those are put in the beginning of the list.
 620  */
 621
 622 #define KVM_SAVE_MSRS_BEGIN     2
 623 static u32 msrs_to_save[] = {
 624         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 625         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 626         MSR_K6_STAR,
 627 #ifdef CONFIG_X86_64
 628         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 629 #endif
 630         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 631 };
 632
 633 static unsigned num_msrs_to_save;
 634
 635 static u32 emulated_msrs[] = {
 636         MSR_IA32_MISC_ENABLE,
 637 };
 638
 639 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 640 {
 641         if (efer & efer_reserved_bits) {
 642                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 643                        efer);
 644                 kvm_inject_gp(vcpu, 0);
 645                 return;
 646         }
 647
 648         if (is_paging(vcpu)
 649             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 650                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 651                 kvm_inject_gp(vcpu, 0);
 652                 return;
 653         }
 654
 655         if (efer & EFER_FFXSR) {
 656                 struct kvm_cpuid_entry2 *feat;
 657
 658                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 659                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 660                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 661                         kvm_inject_gp(vcpu, 0);
 662                         return;
 663                 }
 664         }
 665
 666         if (efer & EFER_SVME) {
 667                 struct kvm_cpuid_entry2 *feat;
 668
 669                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 670                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 671                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 672                         kvm_inject_gp(vcpu, 0);
 673                         return;
 674                 }
 675         }
 676
 677         kvm_x86_ops->set_efer(vcpu, efer);
 678
 679         efer &= ~EFER_LMA;
 680         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 681
 682         vcpu->arch.shadow_efer = efer;
 683
 684         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 685         kvm_mmu_reset_context(vcpu);
 686 }
 687
 688 void kvm_enable_efer_bits(u64 mask)
 689 {
 690        efer_reserved_bits &= ~mask;
 691 }
 692 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 693
 694
 695 /*
 696  * Writes msr value into into the appropriate "register".
 697  * Returns 0 on success, non-0 otherwise.
 698  * Assumes vcpu_load() was already called.
 699  */
 700 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 701 {
 702         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 703 }
 704
 705 /*
 706  * Adapt set_msr() to msr_io()'s calling convention
 707  */
 708 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 709 {
 710         return kvm_set_msr(vcpu, index, *data);
 711 }
 712
 713 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 714 {
 715         static int version;
 716         struct pvclock_wall_clock wc;
 717         struct timespec boot;
 718
 719         if (!wall_clock)
 720                 return;
 721
 722         version++;
 723
 724         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 725
 726         /*
 727          * The guest calculates current wall clock time by adding
 728          * system time (updated by kvm_write_guest_time below) to the
 729          * wall clock specified here.  guest system time equals host
 730          * system time for us, thus we must fill in host boot time here.
 731          */
 732         getboottime(&boot);
 733
 734         wc.sec = boot.tv_sec;
 735         wc.nsec = boot.tv_nsec;
 736         wc.version = version;
 737
 738         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 739
 740         version++;
 741         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 742 }
 743
 744 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 745 {
 746         uint32_t quotient, remainder;
 747
 748         /* Don't try to replace with do_div(), this one calculates
 749          * "(dividend << 32) / divisor" */
 750         __asm__ ( "divl %4"
 751                   : "=a" (quotient), "=d" (remainder)
 752                   : "0" (0), "1" (dividend), "r" (divisor) );
 753         return quotient;
 754 }
 755
 756 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 757 {
 758         uint64_t nsecs = 1000000000LL;
 759         int32_t  shift = 0;
 760         uint64_t tps64;
 761         uint32_t tps32;
 762
 763         tps64 = tsc_khz * 1000LL;
 764         while (tps64 > nsecs*2) {
 765                 tps64 >>= 1;
 766                 shift--;
 767         }
 768
 769         tps32 = (uint32_t)tps64;
 770         while (tps32 <= (uint32_t)nsecs) {
 771                 tps32 <<= 1;
 772                 shift++;
 773         }
 774
 775         hv_clock->tsc_shift = shift;
 776         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 777
 778         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 779                  __func__, tsc_khz, hv_clock->tsc_shift,
 780                  hv_clock->tsc_to_system_mul);
 781 }
 782
 783 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 784
 785 static void kvm_write_guest_time(struct kvm_vcpu *v)
 786 {
 787         struct timespec ts;
 788         unsigned long flags;
 789         struct kvm_vcpu_arch *vcpu = &v->arch;
 790         void *shared_kaddr;
 791         unsigned long this_tsc_khz;
 792
 793         if ((!vcpu->time_page))
 794                 return;
 795
 796         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 797         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 798                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 799                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 800         }
 801         put_cpu_var(cpu_tsc_khz);
 802
 803         /* Keep irq disabled to prevent changes to the clock */
 804         local_irq_save(flags);
 805         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 806         ktime_get_ts(&ts);
 807         monotonic_to_bootbased(&ts);
 808         local_irq_restore(flags);
 809
 810         /* With all the info we got, fill in the values */
 811
 812         vcpu->hv_clock.system_time = ts.tv_nsec +
 813                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 814
 815         /*
 816          * The interface expects us to write an even number signaling that the
 817          * update is finished. Since the guest won't see the intermediate
 818          * state, we just increase by 2 at the end.
 819          */
 820         vcpu->hv_clock.version += 2;
 821
 822         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 823
 824         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 825                sizeof(vcpu->hv_clock));
 826
 827         kunmap_atomic(shared_kaddr, KM_USER0);
 828
 829         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 830 }
 831
 832 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 833 {
 834         struct kvm_vcpu_arch *vcpu = &v->arch;
 835
 836         if (!vcpu->time_page)
 837                 return 0;
 838         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 839         return 1;
 840 }
 841
 842 static bool msr_mtrr_valid(unsigned msr)
 843 {
 844         switch (msr) {
 845         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 846         case MSR_MTRRfix64K_00000:
 847         case MSR_MTRRfix16K_80000:
 848         case MSR_MTRRfix16K_A0000:
 849         case MSR_MTRRfix4K_C0000:
 850         case MSR_MTRRfix4K_C8000:
 851         case MSR_MTRRfix4K_D0000:
 852         case MSR_MTRRfix4K_D8000:
 853         case MSR_MTRRfix4K_E0000:
 854         case MSR_MTRRfix4K_E8000:
 855         case MSR_MTRRfix4K_F0000:
 856         case MSR_MTRRfix4K_F8000:
 857         case MSR_MTRRdefType:
 858         case MSR_IA32_CR_PAT:
 859                 return true;
 860         case 0x2f8:
 861                 return true;
 862         }
 863         return false;
 864 }
 865
 866 static bool valid_pat_type(unsigned t)
 867 {
 868         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 869 }
 870
 871 static bool valid_mtrr_type(unsigned t)
 872 {
 873         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 874 }
 875
 876 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 877 {
 878         int i;
 879
 880         if (!msr_mtrr_valid(msr))
 881                 return false;
 882
 883         if (msr == MSR_IA32_CR_PAT) {
 884                 for (i = 0; i < 8; i++)
 885                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 886                                 return false;
 887                 return true;
 888         } else if (msr == MSR_MTRRdefType) {
 889                 if (data & ~0xcff)
 890                         return false;
 891                 return valid_mtrr_type(data & 0xff);
 892         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 893                 for (i = 0; i < 8 ; i++)
 894                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 895                                 return false;
 896                 return true;
 897         }
 898
 899         /* variable MTRRs */
 900         return valid_mtrr_type(data & 0xff);
 901 }
 902
 903 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 904 {
 905         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 906
 907         if (!mtrr_valid(vcpu, msr, data))
 908                 return 1;
 909
 910         if (msr == MSR_MTRRdefType) {
 911                 vcpu->arch.mtrr_state.def_type = data;
 912                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 913         } else if (msr == MSR_MTRRfix64K_00000)
 914                 p[0] = data;
 915         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 916                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 917         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 918                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 919         else if (msr == MSR_IA32_CR_PAT)
 920                 vcpu->arch.pat = data;
 921         else {  /* Variable MTRRs */
 922                 int idx, is_mtrr_mask;
 923                 u64 *pt;
 924
 925                 idx = (msr - 0x200) / 2;
 926                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 927                 if (!is_mtrr_mask)
 928                         pt =
 929                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 930                 else
 931                         pt =
 932                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 933                 *pt = data;
 934         }
 935
 936         kvm_mmu_reset_context(vcpu);
 937         return 0;
 938 }
 939
 940 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 941 {
 942         u64 mcg_cap = vcpu->arch.mcg_cap;
 943         unsigned bank_num = mcg_cap & 0xff;
 944
 945         switch (msr) {
 946         case MSR_IA32_MCG_STATUS:
 947                 vcpu->arch.mcg_status = data;
 948                 break;
 949         case MSR_IA32_MCG_CTL:
 950                 if (!(mcg_cap & MCG_CTL_P))
 951                         return 1;
 952                 if (data != 0 && data != ~(u64)0)
 953                         return -1;
 954                 vcpu->arch.mcg_ctl = data;
 955                 break;
 956         default:
 957                 if (msr >= MSR_IA32_MC0_CTL &&
 958                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 959                         u32 offset = msr - MSR_IA32_MC0_CTL;
 960                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 961                         if ((offset & 0x3) == 0 &&
 962                             data != 0 && data != ~(u64)0)
 963                                 return -1;
 964                         vcpu->arch.mce_banks[offset] = data;
 965                         break;
 966                 }
 967                 return 1;
 968         }
 969         return 0;
 970 }
 971
 972 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 973 {
 974         struct kvm *kvm = vcpu->kvm;
 975         int lm = is_long_mode(vcpu);
 976         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 977                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 978         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 979                 : kvm->arch.xen_hvm_config.blob_size_32;
 980         u32 page_num = data & ~PAGE_MASK;
 981         u64 page_addr = data & PAGE_MASK;
 982         u8 *page;
 983         int r;
 984
 985         r = -E2BIG;
 986         if (page_num >= blob_size)
 987                 goto out;
 988         r = -ENOMEM;
 989         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 990         if (!page)
 991                 goto out;
 992         r = -EFAULT;
 993         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 994                 goto out_free;
 995         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 996                 goto out_free;
 997         r = 0;
 998 out_free:
 999         kfree(page);
1000 out:
1001         return r;
1002 }
1003
1004 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1005 {
1006         switch (msr) {
1007         case MSR_EFER:
1008                 set_efer(vcpu, data);
1009                 break;
1010         case MSR_K7_HWCR:
1011                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1012                 if (data != 0) {
1013                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1014                                 data);
1015                         return 1;
1016                 }
1017                 break;
1018         case MSR_FAM10H_MMIO_CONF_BASE:
1019                 if (data != 0) {
1020                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1021                                 "0x%llx\n", data);
1022                         return 1;
1023                 }
1024                 break;
1025         case MSR_AMD64_NB_CFG:
1026                 break;
1027         case MSR_IA32_DEBUGCTLMSR:
1028                 if (!data) {
1029                         /* We support the non-activated case already */
1030                         break;
1031                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1032                         /* Values other than LBR and BTF are vendor-specific,
1033                            thus reserved and should throw a #GP */
1034                         return 1;
1035                 }
1036                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1037                         __func__, data);
1038                 break;
1039         case MSR_IA32_UCODE_REV:
1040         case MSR_IA32_UCODE_WRITE:
1041         case MSR_VM_HSAVE_PA:
1042         case MSR_AMD64_PATCH_LOADER:
1043                 break;
1044         case 0x200 ... 0x2ff:
1045                 return set_msr_mtrr(vcpu, msr, data);
1046         case MSR_IA32_APICBASE:
1047                 kvm_set_apic_base(vcpu, data);
1048                 break;
1049         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1050                 return kvm_x2apic_msr_write(vcpu, msr, data);
1051         case MSR_IA32_MISC_ENABLE:
1052                 vcpu->arch.ia32_misc_enable_msr = data;
1053                 break;
1054         case MSR_KVM_WALL_CLOCK:
1055                 vcpu->kvm->arch.wall_clock = data;
1056                 kvm_write_wall_clock(vcpu->kvm, data);
1057                 break;
1058         case MSR_KVM_SYSTEM_TIME: {
1059                 if (vcpu->arch.time_page) {
1060                         kvm_release_page_dirty(vcpu->arch.time_page);
1061                         vcpu->arch.time_page = NULL;
1062                 }
1063
1064                 vcpu->arch.time = data;
1065
1066                 /* we verify if the enable bit is set... */
1067                 if (!(data & 1))
1068                         break;
1069
1070                 /* ...but clean it before doing the actual write */
1071                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1072
1073                 vcpu->arch.time_page =
1074                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1075
1076                 if (is_error_page(vcpu->arch.time_page)) {
1077                         kvm_release_page_clean(vcpu->arch.time_page);
1078                         vcpu->arch.time_page = NULL;
1079                 }
1080
1081                 kvm_request_guest_time_update(vcpu);
1082                 break;
1083         }
1084         case MSR_IA32_MCG_CTL:
1085         case MSR_IA32_MCG_STATUS:
1086         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1087                 return set_msr_mce(vcpu, msr, data);
1088
1089         /* Performance counters are not protected by a CPUID bit,
1090          * so we should check all of them in the generic path for the sake of
1091          * cross vendor migration.
1092          * Writing a zero into the event select MSRs disables them,
1093          * which we perfectly emulate ;-). Any other value should be at least
1094          * reported, some guests depend on them.
1095          */
1096         case MSR_P6_EVNTSEL0:
1097         case MSR_P6_EVNTSEL1:
1098         case MSR_K7_EVNTSEL0:
1099         case MSR_K7_EVNTSEL1:
1100         case MSR_K7_EVNTSEL2:
1101         case MSR_K7_EVNTSEL3:
1102                 if (data != 0)
1103                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1104                                 "0x%x data 0x%llx\n", msr, data);
1105                 break;
1106         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1107          * so we ignore writes to make it happy.
1108          */
1109         case MSR_P6_PERFCTR0:
1110         case MSR_P6_PERFCTR1:
1111         case MSR_K7_PERFCTR0:
1112         case MSR_K7_PERFCTR1:
1113         case MSR_K7_PERFCTR2:
1114         case MSR_K7_PERFCTR3:
1115                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1116                         "0x%x data 0x%llx\n", msr, data);
1117                 break;
1118         default:
1119                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1120                         return xen_hvm_config(vcpu, data);
1121                 if (!ignore_msrs) {
1122                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1123                                 msr, data);
1124                         return 1;
1125                 } else {
1126                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1127                                 msr, data);
1128                         break;
1129                 }
1130         }
1131         return 0;
1132 }
1133 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1134
1135
1136 /*
1137  * Reads an msr value (of 'msr_index') into 'pdata'.
1138  * Returns 0 on success, non-0 otherwise.
1139  * Assumes vcpu_load() was already called.
1140  */
1141 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1142 {
1143         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1144 }
1145
1146 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1147 {
1148         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1149
1150         if (!msr_mtrr_valid(msr))
1151                 return 1;
1152
1153         if (msr == MSR_MTRRdefType)
1154                 *pdata = vcpu->arch.mtrr_state.def_type +
1155                          (vcpu->arch.mtrr_state.enabled << 10);
1156         else if (msr == MSR_MTRRfix64K_00000)
1157                 *pdata = p[0];
1158         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1159                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1160         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1161                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1162         else if (msr == MSR_IA32_CR_PAT)
1163                 *pdata = vcpu->arch.pat;
1164         else {  /* Variable MTRRs */
1165                 int idx, is_mtrr_mask;
1166                 u64 *pt;
1167
1168                 idx = (msr - 0x200) / 2;
1169                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1170                 if (!is_mtrr_mask)
1171                         pt =
1172                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1173                 else
1174                         pt =
1175                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1176                 *pdata = *pt;
1177         }
1178
1179         return 0;
1180 }
1181
1182 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1183 {
1184         u64 data;
1185         u64 mcg_cap = vcpu->arch.mcg_cap;
1186         unsigned bank_num = mcg_cap & 0xff;
1187
1188         switch (msr) {
1189         case MSR_IA32_P5_MC_ADDR:
1190         case MSR_IA32_P5_MC_TYPE:
1191                 data = 0;
1192                 break;
1193         case MSR_IA32_MCG_CAP:
1194                 data = vcpu->arch.mcg_cap;
1195                 break;
1196         case MSR_IA32_MCG_CTL:
1197                 if (!(mcg_cap & MCG_CTL_P))
1198                         return 1;
1199                 data = vcpu->arch.mcg_ctl;
1200                 break;
1201         case MSR_IA32_MCG_STATUS:
1202                 data = vcpu->arch.mcg_status;
1203                 break;
1204         default:
1205                 if (msr >= MSR_IA32_MC0_CTL &&
1206                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1207                         u32 offset = msr - MSR_IA32_MC0_CTL;
1208                         data = vcpu->arch.mce_banks[offset];
1209                         break;
1210                 }
1211                 return 1;
1212         }
1213         *pdata = data;
1214         return 0;
1215 }
1216
1217 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1218 {
1219         u64 data;
1220
1221         switch (msr) {
1222         case MSR_IA32_PLATFORM_ID:
1223         case MSR_IA32_UCODE_REV:
1224         case MSR_IA32_EBL_CR_POWERON:
1225         case MSR_IA32_DEBUGCTLMSR:
1226         case MSR_IA32_LASTBRANCHFROMIP:
1227         case MSR_IA32_LASTBRANCHTOIP:
1228         case MSR_IA32_LASTINTFROMIP:
1229         case MSR_IA32_LASTINTTOIP:
1230         case MSR_K8_SYSCFG:
1231         case MSR_K7_HWCR:
1232         case MSR_VM_HSAVE_PA:
1233         case MSR_P6_PERFCTR0:
1234         case MSR_P6_PERFCTR1:
1235         case MSR_P6_EVNTSEL0:
1236         case MSR_P6_EVNTSEL1:
1237         case MSR_K7_EVNTSEL0:
1238         case MSR_K7_PERFCTR0:
1239         case MSR_K8_INT_PENDING_MSG:
1240         case MSR_AMD64_NB_CFG:
1241         case MSR_FAM10H_MMIO_CONF_BASE:
1242                 data = 0;
1243                 break;
1244         case MSR_MTRRcap:
1245                 data = 0x500 | KVM_NR_VAR_MTRR;
1246                 break;
1247         case 0x200 ... 0x2ff:
1248                 return get_msr_mtrr(vcpu, msr, pdata);
1249         case 0xcd: /* fsb frequency */
1250                 data = 3;
1251                 break;
1252         case MSR_IA32_APICBASE:
1253                 data = kvm_get_apic_base(vcpu);
1254                 break;
1255         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1256                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1257                 break;
1258         case MSR_IA32_MISC_ENABLE:
1259                 data = vcpu->arch.ia32_misc_enable_msr;
1260                 break;
1261         case MSR_IA32_PERF_STATUS:
1262                 /* TSC increment by tick */
1263                 data = 1000ULL;
1264                 /* CPU multiplier */
1265                 data |= (((uint64_t)4ULL) << 40);
1266                 break;
1267         case MSR_EFER:
1268                 data = vcpu->arch.shadow_efer;
1269                 break;
1270         case MSR_KVM_WALL_CLOCK:
1271                 data = vcpu->kvm->arch.wall_clock;
1272                 break;
1273         case MSR_KVM_SYSTEM_TIME:
1274                 data = vcpu->arch.time;
1275                 break;
1276         case MSR_IA32_P5_MC_ADDR:
1277         case MSR_IA32_P5_MC_TYPE:
1278         case MSR_IA32_MCG_CAP:
1279         case MSR_IA32_MCG_CTL:
1280         case MSR_IA32_MCG_STATUS:
1281         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1282                 return get_msr_mce(vcpu, msr, pdata);
1283         default:
1284                 if (!ignore_msrs) {
1285                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1286                         return 1;
1287                 } else {
1288                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1289                         data = 0;
1290                 }
1291                 break;
1292         }
1293         *pdata = data;
1294         return 0;
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1297
1298 /*
1299  * Read or write a bunch of msrs. All parameters are kernel addresses.
1300  *
1301  * @return number of msrs set successfully.
1302  */
1303 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1304                     struct kvm_msr_entry *entries,
1305                     int (*do_msr)(struct kvm_vcpu *vcpu,
1306                                   unsigned index, u64 *data))
1307 {
1308         int i;
1309
1310         vcpu_load(vcpu);
1311
1312         down_read(&vcpu->kvm->slots_lock);
1313         for (i = 0; i < msrs->nmsrs; ++i)
1314                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1315                         break;
1316         up_read(&vcpu->kvm->slots_lock);
1317
1318         vcpu_put(vcpu);
1319
1320         return i;
1321 }
1322
1323 /*
1324  * Read or write a bunch of msrs. Parameters are user addresses.
1325  *
1326  * @return number of msrs set successfully.
1327  */
1328 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1329                   int (*do_msr)(struct kvm_vcpu *vcpu,
1330                                 unsigned index, u64 *data),
1331                   int writeback)
1332 {
1333         struct kvm_msrs msrs;
1334         struct kvm_msr_entry *entries;
1335         int r, n;
1336         unsigned size;
1337
1338         r = -EFAULT;
1339         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1340                 goto out;
1341
1342         r = -E2BIG;
1343         if (msrs.nmsrs >= MAX_IO_MSRS)
1344                 goto out;
1345
1346         r = -ENOMEM;
1347         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1348         entries = vmalloc(size);
1349         if (!entries)
1350                 goto out;
1351
1352         r = -EFAULT;
1353         if (copy_from_user(entries, user_msrs->entries, size))
1354                 goto out_free;
1355
1356         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1357         if (r < 0)
1358                 goto out_free;
1359
1360         r = -EFAULT;
1361         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1362                 goto out_free;
1363
1364         r = n;
1365
1366 out_free:
1367         vfree(entries);
1368 out:
1369         return r;
1370 }
1371
1372 int kvm_dev_ioctl_check_extension(long ext)
1373 {
1374         int r;
1375
1376         switch (ext) {
1377         case KVM_CAP_IRQCHIP:
1378         case KVM_CAP_HLT:
1379         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1380         case KVM_CAP_SET_TSS_ADDR:
1381         case KVM_CAP_EXT_CPUID:
1382         case KVM_CAP_CLOCKSOURCE:
1383         case KVM_CAP_PIT:
1384         case KVM_CAP_NOP_IO_DELAY:
1385         case KVM_CAP_MP_STATE:
1386         case KVM_CAP_SYNC_MMU:
1387         case KVM_CAP_REINJECT_CONTROL:
1388         case KVM_CAP_IRQ_INJECT_STATUS:
1389         case KVM_CAP_ASSIGN_DEV_IRQ:
1390         case KVM_CAP_IRQFD:
1391         case KVM_CAP_IOEVENTFD:
1392         case KVM_CAP_PIT2:
1393         case KVM_CAP_PIT_STATE2:
1394         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1395         case KVM_CAP_XEN_HVM:
1396         case KVM_CAP_ADJUST_CLOCK:
1397         case KVM_CAP_VCPU_EVENTS:
1398                 r = 1;
1399                 break;
1400         case KVM_CAP_COALESCED_MMIO:
1401                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1402                 break;
1403         case KVM_CAP_VAPIC:
1404                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1405                 break;
1406         case KVM_CAP_NR_VCPUS:
1407                 r = KVM_MAX_VCPUS;
1408                 break;
1409         case KVM_CAP_NR_MEMSLOTS:
1410                 r = KVM_MEMORY_SLOTS;
1411                 break;
1412         case KVM_CAP_PV_MMU:    /* obsolete */
1413                 r = 0;
1414                 break;
1415         case KVM_CAP_IOMMU:
1416                 r = iommu_found();
1417                 break;
1418         case KVM_CAP_MCE:
1419                 r = KVM_MAX_MCE_BANKS;
1420                 break;
1421         default:
1422                 r = 0;
1423                 break;
1424         }
1425         return r;
1426
1427 }
1428
1429 long kvm_arch_dev_ioctl(struct file *filp,
1430                         unsigned int ioctl, unsigned long arg)
1431 {
1432         void __user *argp = (void __user *)arg;
1433         long r;
1434
1435         switch (ioctl) {
1436         case KVM_GET_MSR_INDEX_LIST: {
1437                 struct kvm_msr_list __user *user_msr_list = argp;
1438                 struct kvm_msr_list msr_list;
1439                 unsigned n;
1440
1441                 r = -EFAULT;
1442                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1443                         goto out;
1444                 n = msr_list.nmsrs;
1445                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1446                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1447                         goto out;
1448                 r = -E2BIG;
1449                 if (n < msr_list.nmsrs)
1450                         goto out;
1451                 r = -EFAULT;
1452                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1453                                  num_msrs_to_save * sizeof(u32)))
1454                         goto out;
1455                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1456                                  &emulated_msrs,
1457                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1458                         goto out;
1459                 r = 0;
1460                 break;
1461         }
1462         case KVM_GET_SUPPORTED_CPUID: {
1463                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1464                 struct kvm_cpuid2 cpuid;
1465
1466                 r = -EFAULT;
1467                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1468                         goto out;
1469                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1470                                                       cpuid_arg->entries);
1471                 if (r)
1472                         goto out;
1473
1474                 r = -EFAULT;
1475                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1476                         goto out;
1477                 r = 0;
1478                 break;
1479         }
1480         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1481                 u64 mce_cap;
1482
1483                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1484                 r = -EFAULT;
1485                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1486                         goto out;
1487                 r = 0;
1488                 break;
1489         }
1490         default:
1491                 r = -EINVAL;
1492         }
1493 out:
1494         return r;
1495 }
1496
1497 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1498 {
1499         kvm_x86_ops->vcpu_load(vcpu, cpu);
1500         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1501                 unsigned long khz = cpufreq_quick_get(cpu);
1502                 if (!khz)
1503                         khz = tsc_khz;
1504                 per_cpu(cpu_tsc_khz, cpu) = khz;
1505         }
1506         kvm_request_guest_time_update(vcpu);
1507 }
1508
1509 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1510 {
1511         kvm_x86_ops->vcpu_put(vcpu);
1512         kvm_put_guest_fpu(vcpu);
1513 }
1514
1515 static int is_efer_nx(void)
1516 {
1517         unsigned long long efer = 0;
1518
1519         rdmsrl_safe(MSR_EFER, &efer);
1520         return efer & EFER_NX;
1521 }
1522
1523 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1524 {
1525         int i;
1526         struct kvm_cpuid_entry2 *e, *entry;
1527
1528         entry = NULL;
1529         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1530                 e = &vcpu->arch.cpuid_entries[i];
1531                 if (e->function == 0x80000001) {
1532                         entry = e;
1533                         break;
1534                 }
1535         }
1536         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1537                 entry->edx &= ~(1 << 20);
1538                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1539         }
1540 }
1541
1542 /* when an old userspace process fills a new kernel module */
1543 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1544                                     struct kvm_cpuid *cpuid,
1545                                     struct kvm_cpuid_entry __user *entries)
1546 {
1547         int r, i;
1548         struct kvm_cpuid_entry *cpuid_entries;
1549
1550         r = -E2BIG;
1551         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1552                 goto out;
1553         r = -ENOMEM;
1554         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1555         if (!cpuid_entries)
1556                 goto out;
1557         r = -EFAULT;
1558         if (copy_from_user(cpuid_entries, entries,
1559                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1560                 goto out_free;
1561         for (i = 0; i < cpuid->nent; i++) {
1562                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1563                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1564                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1565                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1566                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1567                 vcpu->arch.cpuid_entries[i].index = 0;
1568                 vcpu->arch.cpuid_entries[i].flags = 0;
1569                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1570                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1571                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1572         }
1573         vcpu->arch.cpuid_nent = cpuid->nent;
1574         cpuid_fix_nx_cap(vcpu);
1575         r = 0;
1576         kvm_apic_set_version(vcpu);
1577         kvm_x86_ops->cpuid_update(vcpu);
1578
1579 out_free:
1580         vfree(cpuid_entries);
1581 out:
1582         return r;
1583 }
1584
1585 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1586                                      struct kvm_cpuid2 *cpuid,
1587                                      struct kvm_cpuid_entry2 __user *entries)
1588 {
1589         int r;
1590
1591         r = -E2BIG;
1592         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1593                 goto out;
1594         r = -EFAULT;
1595         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1596                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1597                 goto out;
1598         vcpu->arch.cpuid_nent = cpuid->nent;
1599         kvm_apic_set_version(vcpu);
1600         kvm_x86_ops->cpuid_update(vcpu);
1601         return 0;
1602
1603 out:
1604         return r;
1605 }
1606
1607 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1608                                      struct kvm_cpuid2 *cpuid,
1609                                      struct kvm_cpuid_entry2 __user *entries)
1610 {
1611         int r;
1612
1613         r = -E2BIG;
1614         if (cpuid->nent < vcpu->arch.cpuid_nent)
1615                 goto out;
1616         r = -EFAULT;
1617         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1618                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1619                 goto out;
1620         return 0;
1621
1622 out:
1623         cpuid->nent = vcpu->arch.cpuid_nent;
1624         return r;
1625 }
1626
1627 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1628                            u32 index)
1629 {
1630         entry->function = function;
1631         entry->index = index;
1632         cpuid_count(entry->function, entry->index,
1633                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1634         entry->flags = 0;
1635 }
1636
1637 #define F(x) bit(X86_FEATURE_##x)
1638
1639 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1640                          u32 index, int *nent, int maxnent)
1641 {
1642         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1643         unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1644 #ifdef CONFIG_X86_64
1645         unsigned f_lm = F(LM);
1646 #else
1647         unsigned f_lm = 0;
1648 #endif
1649         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1650
1651         /* cpuid 1.edx */
1652         const u32 kvm_supported_word0_x86_features =
1653                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1654                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1655                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1656                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1657                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1658                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1659                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1660                 0 /* HTT, TM, Reserved, PBE */;
1661         /* cpuid 0x80000001.edx */
1662         const u32 kvm_supported_word1_x86_features =
1663                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1664                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1665                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1666                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1667                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1668                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1669                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1670                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1671         /* cpuid 1.ecx */
1672         const u32 kvm_supported_word4_x86_features =
1673                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1674                 0 /* DS-CPL, VMX, SMX, EST */ |
1675                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1676                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1677                 0 /* Reserved, DCA */ | F(XMM4_1) |
1678                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1679                 0 /* Reserved, XSAVE, OSXSAVE */;
1680         /* cpuid 0x80000001.ecx */
1681         const u32 kvm_supported_word6_x86_features =
1682                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1683                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1684                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1685                 0 /* SKINIT */ | 0 /* WDT */;
1686
1687         /* all calls to cpuid_count() should be made on the same cpu */
1688         get_cpu();
1689         do_cpuid_1_ent(entry, function, index);
1690         ++*nent;
1691
1692         switch (function) {
1693         case 0:
1694                 entry->eax = min(entry->eax, (u32)0xb);
1695                 break;
1696         case 1:
1697                 entry->edx &= kvm_supported_word0_x86_features;
1698                 entry->ecx &= kvm_supported_word4_x86_features;
1699                 /* we support x2apic emulation even if host does not support
1700                  * it since we emulate x2apic in software */
1701                 entry->ecx |= F(X2APIC);
1702                 break;
1703         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1704          * may return different values. This forces us to get_cpu() before
1705          * issuing the first command, and also to emulate this annoying behavior
1706          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1707         case 2: {
1708                 int t, times = entry->eax & 0xff;
1709
1710                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1711                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1712                 for (t = 1; t < times && *nent < maxnent; ++t) {
1713                         do_cpuid_1_ent(&entry[t], function, 0);
1714                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1715                         ++*nent;
1716                 }
1717                 break;
1718         }
1719         /* function 4 and 0xb have additional index. */
1720         case 4: {
1721                 int i, cache_type;
1722
1723                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1724                 /* read more entries until cache_type is zero */
1725                 for (i = 1; *nent < maxnent; ++i) {
1726                         cache_type = entry[i - 1].eax & 0x1f;
1727                         if (!cache_type)
1728                                 break;
1729                         do_cpuid_1_ent(&entry[i], function, i);
1730                         entry[i].flags |=
1731                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1732                         ++*nent;
1733                 }
1734                 break;
1735         }
1736         case 0xb: {
1737                 int i, level_type;
1738
1739                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1740                 /* read more entries until level_type is zero */
1741                 for (i = 1; *nent < maxnent; ++i) {
1742                         level_type = entry[i - 1].ecx & 0xff00;
1743                         if (!level_type)
1744                                 break;
1745                         do_cpuid_1_ent(&entry[i], function, i);
1746                         entry[i].flags |=
1747                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1748                         ++*nent;
1749                 }
1750                 break;
1751         }
1752         case 0x80000000:
1753                 entry->eax = min(entry->eax, 0x8000001a);
1754                 break;
1755         case 0x80000001:
1756                 entry->edx &= kvm_supported_word1_x86_features;
1757                 entry->ecx &= kvm_supported_word6_x86_features;
1758                 break;
1759         }
1760         put_cpu();
1761 }
1762
1763 #undef F
1764
1765 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1766                                      struct kvm_cpuid_entry2 __user *entries)
1767 {
1768         struct kvm_cpuid_entry2 *cpuid_entries;
1769         int limit, nent = 0, r = -E2BIG;
1770         u32 func;
1771
1772         if (cpuid->nent < 1)
1773                 goto out;
1774         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1775                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1776         r = -ENOMEM;
1777         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1778         if (!cpuid_entries)
1779                 goto out;
1780
1781         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1782         limit = cpuid_entries[0].eax;
1783         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1784                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1785                              &nent, cpuid->nent);
1786         r = -E2BIG;
1787         if (nent >= cpuid->nent)
1788                 goto out_free;
1789
1790         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1791         limit = cpuid_entries[nent - 1].eax;
1792         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1793                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1794                              &nent, cpuid->nent);
1795         r = -E2BIG;
1796         if (nent >= cpuid->nent)
1797                 goto out_free;
1798
1799         r = -EFAULT;
1800         if (copy_to_user(entries, cpuid_entries,
1801                          nent * sizeof(struct kvm_cpuid_entry2)))
1802                 goto out_free;
1803         cpuid->nent = nent;
1804         r = 0;
1805
1806 out_free:
1807         vfree(cpuid_entries);
1808 out:
1809         return r;
1810 }
1811
1812 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1813                                     struct kvm_lapic_state *s)
1814 {
1815         vcpu_load(vcpu);
1816         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1817         vcpu_put(vcpu);
1818
1819         return 0;
1820 }
1821
1822 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1823                                     struct kvm_lapic_state *s)
1824 {
1825         vcpu_load(vcpu);
1826         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1827         kvm_apic_post_state_restore(vcpu);
1828         update_cr8_intercept(vcpu);
1829         vcpu_put(vcpu);
1830
1831         return 0;
1832 }
1833
1834 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1835                                     struct kvm_interrupt *irq)
1836 {
1837         if (irq->irq < 0 || irq->irq >= 256)
1838                 return -EINVAL;
1839         if (irqchip_in_kernel(vcpu->kvm))
1840                 return -ENXIO;
1841         vcpu_load(vcpu);
1842
1843         kvm_queue_interrupt(vcpu, irq->irq, false);
1844
1845         vcpu_put(vcpu);
1846
1847         return 0;
1848 }
1849
1850 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1851 {
1852         vcpu_load(vcpu);
1853         kvm_inject_nmi(vcpu);
1854         vcpu_put(vcpu);
1855
1856         return 0;
1857 }
1858
1859 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1860                                            struct kvm_tpr_access_ctl *tac)
1861 {
1862         if (tac->flags)
1863                 return -EINVAL;
1864         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1865         return 0;
1866 }
1867
1868 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1869                                         u64 mcg_cap)
1870 {
1871         int r;
1872         unsigned bank_num = mcg_cap & 0xff, bank;
1873
1874         r = -EINVAL;
1875         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1876                 goto out;
1877         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1878                 goto out;
1879         r = 0;
1880         vcpu->arch.mcg_cap = mcg_cap;
1881         /* Init IA32_MCG_CTL to all 1s */
1882         if (mcg_cap & MCG_CTL_P)
1883                 vcpu->arch.mcg_ctl = ~(u64)0;
1884         /* Init IA32_MCi_CTL to all 1s */
1885         for (bank = 0; bank < bank_num; bank++)
1886                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1887 out:
1888         return r;
1889 }
1890
1891 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1892                                       struct kvm_x86_mce *mce)
1893 {
1894         u64 mcg_cap = vcpu->arch.mcg_cap;
1895         unsigned bank_num = mcg_cap & 0xff;
1896         u64 *banks = vcpu->arch.mce_banks;
1897
1898         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1899                 return -EINVAL;
1900         /*
1901          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1902          * reporting is disabled
1903          */
1904         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1905             vcpu->arch.mcg_ctl != ~(u64)0)
1906                 return 0;
1907         banks += 4 * mce->bank;
1908         /*
1909          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1910          * reporting is disabled for the bank
1911          */
1912         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1913                 return 0;
1914         if (mce->status & MCI_STATUS_UC) {
1915                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1916                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
1917                         printk(KERN_DEBUG "kvm: set_mce: "
1918                                "injects mce exception while "
1919                                "previous one is in progress!\n");
1920                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1921                         return 0;
1922                 }
1923                 if (banks[1] & MCI_STATUS_VAL)
1924                         mce->status |= MCI_STATUS_OVER;
1925                 banks[2] = mce->addr;
1926                 banks[3] = mce->misc;
1927                 vcpu->arch.mcg_status = mce->mcg_status;
1928                 banks[1] = mce->status;
1929                 kvm_queue_exception(vcpu, MC_VECTOR);
1930         } else if (!(banks[1] & MCI_STATUS_VAL)
1931                    || !(banks[1] & MCI_STATUS_UC)) {
1932                 if (banks[1] & MCI_STATUS_VAL)
1933                         mce->status |= MCI_STATUS_OVER;
1934                 banks[2] = mce->addr;
1935                 banks[3] = mce->misc;
1936                 banks[1] = mce->status;
1937         } else
1938                 banks[1] |= MCI_STATUS_OVER;
1939         return 0;
1940 }
1941
1942 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
1943                                                struct kvm_vcpu_events *events)
1944 {
1945         vcpu_load(vcpu);
1946
1947         events->exception.injected = vcpu->arch.exception.pending;
1948         events->exception.nr = vcpu->arch.exception.nr;
1949         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
1950         events->exception.error_code = vcpu->arch.exception.error_code;
1951
1952         events->interrupt.injected = vcpu->arch.interrupt.pending;
1953         events->interrupt.nr = vcpu->arch.interrupt.nr;
1954         events->interrupt.soft = vcpu->arch.interrupt.soft;
1955
1956         events->nmi.injected = vcpu->arch.nmi_injected;
1957         events->nmi.pending = vcpu->arch.nmi_pending;
1958         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
1959
1960         events->sipi_vector = vcpu->arch.sipi_vector;
1961
1962         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
1963                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
1964
1965         vcpu_put(vcpu);
1966 }
1967
1968 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
1969                                               struct kvm_vcpu_events *events)
1970 {
1971         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
1972                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
1973                 return -EINVAL;
1974
1975         vcpu_load(vcpu);
1976
1977         vcpu->arch.exception.pending = events->exception.injected;
1978         vcpu->arch.exception.nr = events->exception.nr;
1979         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
1980         vcpu->arch.exception.error_code = events->exception.error_code;
1981
1982         vcpu->arch.interrupt.pending = events->interrupt.injected;
1983         vcpu->arch.interrupt.nr = events->interrupt.nr;
1984         vcpu->arch.interrupt.soft = events->interrupt.soft;
1985         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
1986                 kvm_pic_clear_isr_ack(vcpu->kvm);
1987
1988         vcpu->arch.nmi_injected = events->nmi.injected;
1989         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
1990                 vcpu->arch.nmi_pending = events->nmi.pending;
1991         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
1992
1993         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
1994                 vcpu->arch.sipi_vector = events->sipi_vector;
1995
1996         vcpu_put(vcpu);
1997
1998         return 0;
1999 }
2000
2001 long kvm_arch_vcpu_ioctl(struct file *filp,
2002                          unsigned int ioctl, unsigned long arg)
2003 {
2004         struct kvm_vcpu *vcpu = filp->private_data;
2005         void __user *argp = (void __user *)arg;
2006         int r;
2007         struct kvm_lapic_state *lapic = NULL;
2008
2009         switch (ioctl) {
2010         case KVM_GET_LAPIC: {
2011                 r = -EINVAL;
2012                 if (!vcpu->arch.apic)
2013                         goto out;
2014                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2015
2016                 r = -ENOMEM;
2017                 if (!lapic)
2018                         goto out;
2019                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2020                 if (r)
2021                         goto out;
2022                 r = -EFAULT;
2023                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2024                         goto out;
2025                 r = 0;
2026                 break;
2027         }
2028         case KVM_SET_LAPIC: {
2029                 r = -EINVAL;
2030                 if (!vcpu->arch.apic)
2031                         goto out;
2032                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2033                 r = -ENOMEM;
2034                 if (!lapic)
2035                         goto out;
2036                 r = -EFAULT;
2037                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2038                         goto out;
2039                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2040                 if (r)
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         case KVM_INTERRUPT: {
2046                 struct kvm_interrupt irq;
2047
2048                 r = -EFAULT;
2049                 if (copy_from_user(&irq, argp, sizeof irq))
2050                         goto out;
2051                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2052                 if (r)
2053                         goto out;
2054                 r = 0;
2055                 break;
2056         }
2057         case KVM_NMI: {
2058                 r = kvm_vcpu_ioctl_nmi(vcpu);
2059                 if (r)
2060                         goto out;
2061                 r = 0;
2062                 break;
2063         }
2064         case KVM_SET_CPUID: {
2065                 struct kvm_cpuid __user *cpuid_arg = argp;
2066                 struct kvm_cpuid cpuid;
2067
2068                 r = -EFAULT;
2069                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2070                         goto out;
2071                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2072                 if (r)
2073                         goto out;
2074                 break;
2075         }
2076         case KVM_SET_CPUID2: {
2077                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2078                 struct kvm_cpuid2 cpuid;
2079
2080                 r = -EFAULT;
2081                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2082                         goto out;
2083                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2084                                               cpuid_arg->entries);
2085                 if (r)
2086                         goto out;
2087                 break;
2088         }
2089         case KVM_GET_CPUID2: {
2090                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2091                 struct kvm_cpuid2 cpuid;
2092
2093                 r = -EFAULT;
2094                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2095                         goto out;
2096                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2097                                               cpuid_arg->entries);
2098                 if (r)
2099                         goto out;
2100                 r = -EFAULT;
2101                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2102                         goto out;
2103                 r = 0;
2104                 break;
2105         }
2106         case KVM_GET_MSRS:
2107                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2108                 break;
2109         case KVM_SET_MSRS:
2110                 r = msr_io(vcpu, argp, do_set_msr, 0);
2111                 break;
2112         case KVM_TPR_ACCESS_REPORTING: {
2113                 struct kvm_tpr_access_ctl tac;
2114
2115                 r = -EFAULT;
2116                 if (copy_from_user(&tac, argp, sizeof tac))
2117                         goto out;
2118                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2119                 if (r)
2120                         goto out;
2121                 r = -EFAULT;
2122                 if (copy_to_user(argp, &tac, sizeof tac))
2123                         goto out;
2124                 r = 0;
2125                 break;
2126         };
2127         case KVM_SET_VAPIC_ADDR: {
2128                 struct kvm_vapic_addr va;
2129
2130                 r = -EINVAL;
2131                 if (!irqchip_in_kernel(vcpu->kvm))
2132                         goto out;
2133                 r = -EFAULT;
2134                 if (copy_from_user(&va, argp, sizeof va))
2135                         goto out;
2136                 r = 0;
2137                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2138                 break;
2139         }
2140         case KVM_X86_SETUP_MCE: {
2141                 u64 mcg_cap;
2142
2143                 r = -EFAULT;
2144                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2145                         goto out;
2146                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2147                 break;
2148         }
2149         case KVM_X86_SET_MCE: {
2150                 struct kvm_x86_mce mce;
2151
2152                 r = -EFAULT;
2153                 if (copy_from_user(&mce, argp, sizeof mce))
2154                         goto out;
2155                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2156                 break;
2157         }
2158         case KVM_GET_VCPU_EVENTS: {
2159                 struct kvm_vcpu_events events;
2160
2161                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2162
2163                 r = -EFAULT;
2164                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2165                         break;
2166                 r = 0;
2167                 break;
2168         }
2169         case KVM_SET_VCPU_EVENTS: {
2170                 struct kvm_vcpu_events events;
2171
2172                 r = -EFAULT;
2173                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2174                         break;
2175
2176                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2177                 break;
2178         }
2179         default:
2180                 r = -EINVAL;
2181         }
2182 out:
2183         kfree(lapic);
2184         return r;
2185 }
2186
2187 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2188 {
2189         int ret;
2190
2191         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2192                 return -1;
2193         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2194         return ret;
2195 }
2196
2197 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2198                                               u64 ident_addr)
2199 {
2200         kvm->arch.ept_identity_map_addr = ident_addr;
2201         return 0;
2202 }
2203
2204 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2205                                           u32 kvm_nr_mmu_pages)
2206 {
2207         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2208                 return -EINVAL;
2209
2210         down_write(&kvm->slots_lock);
2211         spin_lock(&kvm->mmu_lock);
2212
2213         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2214         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2215
2216         spin_unlock(&kvm->mmu_lock);
2217         up_write(&kvm->slots_lock);
2218         return 0;
2219 }
2220
2221 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2222 {
2223         return kvm->arch.n_alloc_mmu_pages;
2224 }
2225
2226 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2227 {
2228         int i;
2229         struct kvm_mem_alias *alias;
2230         struct kvm_mem_aliases *aliases = kvm->arch.aliases;
2231
2232         for (i = 0; i < aliases->naliases; ++i) {
2233                 alias = &aliases->aliases[i];
2234                 if (gfn >= alias->base_gfn
2235                     && gfn < alias->base_gfn + alias->npages)
2236                         return alias->target_gfn + gfn - alias->base_gfn;
2237         }
2238         return gfn;
2239 }
2240
2241 /*
2242  * Set a new alias region.  Aliases map a portion of physical memory into
2243  * another portion.  This is useful for memory windows, for example the PC
2244  * VGA region.
2245  */
2246 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2247                                          struct kvm_memory_alias *alias)
2248 {
2249         int r, n;
2250         struct kvm_mem_alias *p;
2251         struct kvm_mem_aliases *aliases;
2252
2253         r = -EINVAL;
2254         /* General sanity checks */
2255         if (alias->memory_size & (PAGE_SIZE - 1))
2256                 goto out;
2257         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2258                 goto out;
2259         if (alias->slot >= KVM_ALIAS_SLOTS)
2260                 goto out;
2261         if (alias->guest_phys_addr + alias->memory_size
2262             < alias->guest_phys_addr)
2263                 goto out;
2264         if (alias->target_phys_addr + alias->memory_size
2265             < alias->target_phys_addr)
2266                 goto out;
2267
2268         down_write(&kvm->slots_lock);
2269         spin_lock(&kvm->mmu_lock);
2270
2271         aliases = kvm->arch.aliases;
2272
2273         p = &aliases->aliases[alias->slot];
2274         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2275         p->npages = alias->memory_size >> PAGE_SHIFT;
2276         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2277
2278         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2279                 if (aliases->aliases[n - 1].npages)
2280                         break;
2281         aliases->naliases = n;
2282
2283         spin_unlock(&kvm->mmu_lock);
2284         kvm_mmu_zap_all(kvm);
2285
2286         up_write(&kvm->slots_lock);
2287
2288         return 0;
2289
2290 out:
2291         return r;
2292 }
2293
2294 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2295 {
2296         int r;
2297
2298         r = 0;
2299         switch (chip->chip_id) {
2300         case KVM_IRQCHIP_PIC_MASTER:
2301                 memcpy(&chip->chip.pic,
2302                         &pic_irqchip(kvm)->pics[0],
2303                         sizeof(struct kvm_pic_state));
2304                 break;
2305         case KVM_IRQCHIP_PIC_SLAVE:
2306                 memcpy(&chip->chip.pic,
2307                         &pic_irqchip(kvm)->pics[1],
2308                         sizeof(struct kvm_pic_state));
2309                 break;
2310         case KVM_IRQCHIP_IOAPIC:
2311                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2312                 break;
2313         default:
2314                 r = -EINVAL;
2315                 break;
2316         }
2317         return r;
2318 }
2319
2320 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2321 {
2322         int r;
2323
2324         r = 0;
2325         switch (chip->chip_id) {
2326         case KVM_IRQCHIP_PIC_MASTER:
2327                 spin_lock(&pic_irqchip(kvm)->lock);
2328                 memcpy(&pic_irqchip(kvm)->pics[0],
2329                         &chip->chip.pic,
2330                         sizeof(struct kvm_pic_state));
2331                 spin_unlock(&pic_irqchip(kvm)->lock);
2332                 break;
2333         case KVM_IRQCHIP_PIC_SLAVE:
2334                 spin_lock(&pic_irqchip(kvm)->lock);
2335                 memcpy(&pic_irqchip(kvm)->pics[1],
2336                         &chip->chip.pic,
2337                         sizeof(struct kvm_pic_state));
2338                 spin_unlock(&pic_irqchip(kvm)->lock);
2339                 break;
2340         case KVM_IRQCHIP_IOAPIC:
2341                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2342                 break;
2343         default:
2344                 r = -EINVAL;
2345                 break;
2346         }
2347         kvm_pic_update_irq(pic_irqchip(kvm));
2348         return r;
2349 }
2350
2351 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2352 {
2353         int r = 0;
2354
2355         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2356         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2357         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2358         return r;
2359 }
2360
2361 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2362 {
2363         int r = 0;
2364
2365         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2366         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2367         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2368         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2369         return r;
2370 }
2371
2372 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2373 {
2374         int r = 0;
2375
2376         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2377         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2378                 sizeof(ps->channels));
2379         ps->flags = kvm->arch.vpit->pit_state.flags;
2380         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2381         return r;
2382 }
2383
2384 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2385 {
2386         int r = 0, start = 0;
2387         u32 prev_legacy, cur_legacy;
2388         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2389         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2390         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2391         if (!prev_legacy && cur_legacy)
2392                 start = 1;
2393         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2394                sizeof(kvm->arch.vpit->pit_state.channels));
2395         kvm->arch.vpit->pit_state.flags = ps->flags;
2396         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2397         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2398         return r;
2399 }
2400
2401 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2402                                  struct kvm_reinject_control *control)
2403 {
2404         if (!kvm->arch.vpit)
2405                 return -ENXIO;
2406         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2407         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2408         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2409         return 0;
2410 }
2411
2412 /*
2413  * Get (and clear) the dirty memory log for a memory slot.
2414  */
2415 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2416                                       struct kvm_dirty_log *log)
2417 {
2418         int r, n, i;
2419         struct kvm_memory_slot *memslot;
2420         unsigned long is_dirty = 0;
2421         unsigned long *dirty_bitmap = NULL;
2422
2423         down_write(&kvm->slots_lock);
2424
2425         r = -EINVAL;
2426         if (log->slot >= KVM_MEMORY_SLOTS)
2427                 goto out;
2428
2429         memslot = &kvm->memslots->memslots[log->slot];
2430         r = -ENOENT;
2431         if (!memslot->dirty_bitmap)
2432                 goto out;
2433
2434         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2435
2436         r = -ENOMEM;
2437         dirty_bitmap = vmalloc(n);
2438         if (!dirty_bitmap)
2439                 goto out;
2440         memset(dirty_bitmap, 0, n);
2441
2442         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2443                 is_dirty = memslot->dirty_bitmap[i];
2444
2445         /* If nothing is dirty, don't bother messing with page tables. */
2446         if (is_dirty) {
2447                 struct kvm_memslots *slots, *old_slots;
2448
2449                 spin_lock(&kvm->mmu_lock);
2450                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2451                 spin_unlock(&kvm->mmu_lock);
2452
2453                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2454                 if (!slots)
2455                         goto out_free;
2456
2457                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2458                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2459
2460                 old_slots = kvm->memslots;
2461                 rcu_assign_pointer(kvm->memslots, slots);
2462                 synchronize_srcu_expedited(&kvm->srcu);
2463                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2464                 kfree(old_slots);
2465         }
2466
2467         r = 0;
2468         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2469                 r = -EFAULT;
2470 out_free:
2471         vfree(dirty_bitmap);
2472 out:
2473         up_write(&kvm->slots_lock);
2474         return r;
2475 }
2476
2477 long kvm_arch_vm_ioctl(struct file *filp,
2478                        unsigned int ioctl, unsigned long arg)
2479 {
2480         struct kvm *kvm = filp->private_data;
2481         void __user *argp = (void __user *)arg;
2482         int r = -ENOTTY;
2483         /*
2484          * This union makes it completely explicit to gcc-3.x
2485          * that these two variables' stack usage should be
2486          * combined, not added together.
2487          */
2488         union {
2489                 struct kvm_pit_state ps;
2490                 struct kvm_pit_state2 ps2;
2491                 struct kvm_memory_alias alias;
2492                 struct kvm_pit_config pit_config;
2493         } u;
2494
2495         switch (ioctl) {
2496         case KVM_SET_TSS_ADDR:
2497                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2498                 if (r < 0)
2499                         goto out;
2500                 break;
2501         case KVM_SET_IDENTITY_MAP_ADDR: {
2502                 u64 ident_addr;
2503
2504                 r = -EFAULT;
2505                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2506                         goto out;
2507                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2508                 if (r < 0)
2509                         goto out;
2510                 break;
2511         }
2512         case KVM_SET_MEMORY_REGION: {
2513                 struct kvm_memory_region kvm_mem;
2514                 struct kvm_userspace_memory_region kvm_userspace_mem;
2515
2516                 r = -EFAULT;
2517                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2518                         goto out;
2519                 kvm_userspace_mem.slot = kvm_mem.slot;
2520                 kvm_userspace_mem.flags = kvm_mem.flags;
2521                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2522                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2523                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2524                 if (r)
2525                         goto out;
2526                 break;
2527         }
2528         case KVM_SET_NR_MMU_PAGES:
2529                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2530                 if (r)
2531                         goto out;
2532                 break;
2533         case KVM_GET_NR_MMU_PAGES:
2534                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2535                 break;
2536         case KVM_SET_MEMORY_ALIAS:
2537                 r = -EFAULT;
2538                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2539                         goto out;
2540                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2541                 if (r)
2542                         goto out;
2543                 break;
2544         case KVM_CREATE_IRQCHIP: {
2545                 struct kvm_pic *vpic;
2546
2547                 mutex_lock(&kvm->lock);
2548                 r = -EEXIST;
2549                 if (kvm->arch.vpic)
2550                         goto create_irqchip_unlock;
2551                 r = -ENOMEM;
2552                 vpic = kvm_create_pic(kvm);
2553                 if (vpic) {
2554                         r = kvm_ioapic_init(kvm);
2555                         if (r) {
2556                                 kfree(vpic);
2557                                 goto create_irqchip_unlock;
2558                         }
2559                 } else
2560                         goto create_irqchip_unlock;
2561                 smp_wmb();
2562                 kvm->arch.vpic = vpic;
2563                 smp_wmb();
2564                 r = kvm_setup_default_irq_routing(kvm);
2565                 if (r) {
2566                         mutex_lock(&kvm->irq_lock);
2567                         kfree(kvm->arch.vpic);
2568                         kfree(kvm->arch.vioapic);
2569                         kvm->arch.vpic = NULL;
2570                         kvm->arch.vioapic = NULL;
2571                         mutex_unlock(&kvm->irq_lock);
2572                 }
2573         create_irqchip_unlock:
2574                 mutex_unlock(&kvm->lock);
2575                 break;
2576         }
2577         case KVM_CREATE_PIT:
2578                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2579                 goto create_pit;
2580         case KVM_CREATE_PIT2:
2581                 r = -EFAULT;
2582                 if (copy_from_user(&u.pit_config, argp,
2583                                    sizeof(struct kvm_pit_config)))
2584                         goto out;
2585         create_pit:
2586                 down_write(&kvm->slots_lock);
2587                 r = -EEXIST;
2588                 if (kvm->arch.vpit)
2589                         goto create_pit_unlock;
2590                 r = -ENOMEM;
2591                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2592                 if (kvm->arch.vpit)
2593                         r = 0;
2594         create_pit_unlock:
2595                 up_write(&kvm->slots_lock);
2596                 break;
2597         case KVM_IRQ_LINE_STATUS:
2598         case KVM_IRQ_LINE: {
2599                 struct kvm_irq_level irq_event;
2600
2601                 r = -EFAULT;
2602                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2603                         goto out;
2604                 if (irqchip_in_kernel(kvm)) {
2605                         __s32 status;
2606                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2607                                         irq_event.irq, irq_event.level);
2608                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2609                                 irq_event.status = status;
2610                                 if (copy_to_user(argp, &irq_event,
2611                                                         sizeof irq_event))
2612                                         goto out;
2613                         }
2614                         r = 0;
2615                 }
2616                 break;
2617         }
2618         case KVM_GET_IRQCHIP: {
2619                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2620                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2621
2622                 r = -ENOMEM;
2623                 if (!chip)
2624                         goto out;
2625                 r = -EFAULT;
2626                 if (copy_from_user(chip, argp, sizeof *chip))
2627                         goto get_irqchip_out;
2628                 r = -ENXIO;
2629                 if (!irqchip_in_kernel(kvm))
2630                         goto get_irqchip_out;
2631                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2632                 if (r)
2633                         goto get_irqchip_out;
2634                 r = -EFAULT;
2635                 if (copy_to_user(argp, chip, sizeof *chip))
2636                         goto get_irqchip_out;
2637                 r = 0;
2638         get_irqchip_out:
2639                 kfree(chip);
2640                 if (r)
2641                         goto out;
2642                 break;
2643         }
2644         case KVM_SET_IRQCHIP: {
2645                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2646                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2647
2648                 r = -ENOMEM;
2649                 if (!chip)
2650                         goto out;
2651                 r = -EFAULT;
2652                 if (copy_from_user(chip, argp, sizeof *chip))
2653                         goto set_irqchip_out;
2654                 r = -ENXIO;
2655                 if (!irqchip_in_kernel(kvm))
2656                         goto set_irqchip_out;
2657                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2658                 if (r)
2659                         goto set_irqchip_out;
2660                 r = 0;
2661         set_irqchip_out:
2662                 kfree(chip);
2663                 if (r)
2664                         goto out;
2665                 break;
2666         }
2667         case KVM_GET_PIT: {
2668                 r = -EFAULT;
2669                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2670                         goto out;
2671                 r = -ENXIO;
2672                 if (!kvm->arch.vpit)
2673                         goto out;
2674                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2675                 if (r)
2676                         goto out;
2677                 r = -EFAULT;
2678                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2679                         goto out;
2680                 r = 0;
2681                 break;
2682         }
2683         case KVM_SET_PIT: {
2684                 r = -EFAULT;
2685                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2686                         goto out;
2687                 r = -ENXIO;
2688                 if (!kvm->arch.vpit)
2689                         goto out;
2690                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2691                 if (r)
2692                         goto out;
2693                 r = 0;
2694                 break;
2695         }
2696         case KVM_GET_PIT2: {
2697                 r = -ENXIO;
2698                 if (!kvm->arch.vpit)
2699                         goto out;
2700                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2701                 if (r)
2702                         goto out;
2703                 r = -EFAULT;
2704                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2705                         goto out;
2706                 r = 0;
2707                 break;
2708         }
2709         case KVM_SET_PIT2: {
2710                 r = -EFAULT;
2711                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2712                         goto out;
2713                 r = -ENXIO;
2714                 if (!kvm->arch.vpit)
2715                         goto out;
2716                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2717                 if (r)
2718                         goto out;
2719                 r = 0;
2720                 break;
2721         }
2722         case KVM_REINJECT_CONTROL: {
2723                 struct kvm_reinject_control control;
2724                 r =  -EFAULT;
2725                 if (copy_from_user(&control, argp, sizeof(control)))
2726                         goto out;
2727                 r = kvm_vm_ioctl_reinject(kvm, &control);
2728                 if (r)
2729                         goto out;
2730                 r = 0;
2731                 break;
2732         }
2733         case KVM_XEN_HVM_CONFIG: {
2734                 r = -EFAULT;
2735                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2736                                    sizeof(struct kvm_xen_hvm_config)))
2737                         goto out;
2738                 r = -EINVAL;
2739                 if (kvm->arch.xen_hvm_config.flags)
2740                         goto out;
2741                 r = 0;
2742                 break;
2743         }
2744         case KVM_SET_CLOCK: {
2745                 struct timespec now;
2746                 struct kvm_clock_data user_ns;
2747                 u64 now_ns;
2748                 s64 delta;
2749
2750                 r = -EFAULT;
2751                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2752                         goto out;
2753
2754                 r = -EINVAL;
2755                 if (user_ns.flags)
2756                         goto out;
2757
2758                 r = 0;
2759                 ktime_get_ts(&now);
2760                 now_ns = timespec_to_ns(&now);
2761                 delta = user_ns.clock - now_ns;
2762                 kvm->arch.kvmclock_offset = delta;
2763                 break;
2764         }
2765         case KVM_GET_CLOCK: {
2766                 struct timespec now;
2767                 struct kvm_clock_data user_ns;
2768                 u64 now_ns;
2769
2770                 ktime_get_ts(&now);
2771                 now_ns = timespec_to_ns(&now);
2772                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2773                 user_ns.flags = 0;
2774
2775                 r = -EFAULT;
2776                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2777                         goto out;
2778                 r = 0;
2779                 break;
2780         }
2781
2782         default:
2783                 ;
2784         }
2785 out:
2786         return r;
2787 }
2788
2789 static void kvm_init_msr_list(void)
2790 {
2791         u32 dummy[2];
2792         unsigned i, j;
2793
2794         /* skip the first msrs in the list. KVM-specific */
2795         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2796                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2797                         continue;
2798                 if (j < i)
2799                         msrs_to_save[j] = msrs_to_save[i];
2800                 j++;
2801         }
2802         num_msrs_to_save = j;
2803 }
2804
2805 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2806                            const void *v)
2807 {
2808         if (vcpu->arch.apic &&
2809             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2810                 return 0;
2811
2812         return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2813 }
2814
2815 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2816 {
2817         if (vcpu->arch.apic &&
2818             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2819                 return 0;
2820
2821         return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2822 }
2823
2824 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2825                                struct kvm_vcpu *vcpu)
2826 {
2827         void *data = val;
2828         int r = X86EMUL_CONTINUE;
2829
2830         while (bytes) {
2831                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2832                 unsigned offset = addr & (PAGE_SIZE-1);
2833                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2834                 int ret;
2835
2836                 if (gpa == UNMAPPED_GVA) {
2837                         r = X86EMUL_PROPAGATE_FAULT;
2838                         goto out;
2839                 }
2840                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2841                 if (ret < 0) {
2842                         r = X86EMUL_UNHANDLEABLE;
2843                         goto out;
2844                 }
2845
2846                 bytes -= toread;
2847                 data += toread;
2848                 addr += toread;
2849         }
2850 out:
2851         return r;
2852 }
2853
2854 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2855                                 struct kvm_vcpu *vcpu)
2856 {
2857         void *data = val;
2858         int r = X86EMUL_CONTINUE;
2859
2860         while (bytes) {
2861                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2862                 unsigned offset = addr & (PAGE_SIZE-1);
2863                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2864                 int ret;
2865
2866                 if (gpa == UNMAPPED_GVA) {
2867                         r = X86EMUL_PROPAGATE_FAULT;
2868                         goto out;
2869                 }
2870                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2871                 if (ret < 0) {
2872                         r = X86EMUL_UNHANDLEABLE;
2873                         goto out;
2874                 }
2875
2876                 bytes -= towrite;
2877                 data += towrite;
2878                 addr += towrite;
2879         }
2880 out:
2881         return r;
2882 }
2883
2884
2885 static int emulator_read_emulated(unsigned long addr,
2886                                   void *val,
2887                                   unsigned int bytes,
2888                                   struct kvm_vcpu *vcpu)
2889 {
2890         gpa_t                 gpa;
2891
2892         if (vcpu->mmio_read_completed) {
2893                 memcpy(val, vcpu->mmio_data, bytes);
2894                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2895                                vcpu->mmio_phys_addr, *(u64 *)val);
2896                 vcpu->mmio_read_completed = 0;
2897                 return X86EMUL_CONTINUE;
2898         }
2899
2900         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2901
2902         /* For APIC access vmexit */
2903         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2904                 goto mmio;
2905
2906         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2907                                 == X86EMUL_CONTINUE)
2908                 return X86EMUL_CONTINUE;
2909         if (gpa == UNMAPPED_GVA)
2910                 return X86EMUL_PROPAGATE_FAULT;
2911
2912 mmio:
2913         /*
2914          * Is this MMIO handled locally?
2915          */
2916         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2917                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2918                 return X86EMUL_CONTINUE;
2919         }
2920
2921         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2922
2923         vcpu->mmio_needed = 1;
2924         vcpu->mmio_phys_addr = gpa;
2925         vcpu->mmio_size = bytes;
2926         vcpu->mmio_is_write = 0;
2927
2928         return X86EMUL_UNHANDLEABLE;
2929 }
2930
2931 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2932                           const void *val, int bytes)
2933 {
2934         int ret;
2935
2936         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2937         if (ret < 0)
2938                 return 0;
2939         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2940         return 1;
2941 }
2942
2943 static int emulator_write_emulated_onepage(unsigned long addr,
2944                                            const void *val,
2945                                            unsigned int bytes,
2946                                            struct kvm_vcpu *vcpu)
2947 {
2948         gpa_t                 gpa;
2949
2950         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2951
2952         if (gpa == UNMAPPED_GVA) {
2953                 kvm_inject_page_fault(vcpu, addr, 2);
2954                 return X86EMUL_PROPAGATE_FAULT;
2955         }
2956
2957         /* For APIC access vmexit */
2958         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2959                 goto mmio;
2960
2961         if (emulator_write_phys(vcpu, gpa, val, bytes))
2962                 return X86EMUL_CONTINUE;
2963
2964 mmio:
2965         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2966         /*
2967          * Is this MMIO handled locally?
2968          */
2969         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2970                 return X86EMUL_CONTINUE;
2971
2972         vcpu->mmio_needed = 1;
2973         vcpu->mmio_phys_addr = gpa;
2974         vcpu->mmio_size = bytes;
2975         vcpu->mmio_is_write = 1;
2976         memcpy(vcpu->mmio_data, val, bytes);
2977
2978         return X86EMUL_CONTINUE;
2979 }
2980
2981 int emulator_write_emulated(unsigned long addr,
2982                                    const void *val,
2983                                    unsigned int bytes,
2984                                    struct kvm_vcpu *vcpu)
2985 {
2986         /* Crossing a page boundary? */
2987         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2988                 int rc, now;
2989
2990                 now = -addr & ~PAGE_MASK;
2991                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2992                 if (rc != X86EMUL_CONTINUE)
2993                         return rc;
2994                 addr += now;
2995                 val += now;
2996                 bytes -= now;
2997         }
2998         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2999 }
3000 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3001
3002 static int emulator_cmpxchg_emulated(unsigned long addr,
3003                                      const void *old,
3004                                      const void *new,
3005                                      unsigned int bytes,
3006                                      struct kvm_vcpu *vcpu)
3007 {
3008         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3009 #ifndef CONFIG_X86_64
3010         /* guests cmpxchg8b have to be emulated atomically */
3011         if (bytes == 8) {
3012                 gpa_t gpa;
3013                 struct page *page;
3014                 char *kaddr;
3015                 u64 val;
3016
3017                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3018
3019                 if (gpa == UNMAPPED_GVA ||
3020                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3021                         goto emul_write;
3022
3023                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3024                         goto emul_write;
3025
3026                 val = *(u64 *)new;
3027
3028                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3029
3030                 kaddr = kmap_atomic(page, KM_USER0);
3031                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3032                 kunmap_atomic(kaddr, KM_USER0);
3033                 kvm_release_page_dirty(page);
3034         }
3035 emul_write:
3036 #endif
3037
3038         return emulator_write_emulated(addr, new, bytes, vcpu);
3039 }
3040
3041 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3042 {
3043         return kvm_x86_ops->get_segment_base(vcpu, seg);
3044 }
3045
3046 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3047 {
3048         kvm_mmu_invlpg(vcpu, address);
3049         return X86EMUL_CONTINUE;
3050 }
3051
3052 int emulate_clts(struct kvm_vcpu *vcpu)
3053 {
3054         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
3055         return X86EMUL_CONTINUE;
3056 }
3057
3058 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3059 {
3060         struct kvm_vcpu *vcpu = ctxt->vcpu;
3061
3062         switch (dr) {
3063         case 0 ... 3:
3064                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3065                 return X86EMUL_CONTINUE;
3066         default:
3067                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3068                 return X86EMUL_UNHANDLEABLE;
3069         }
3070 }
3071
3072 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3073 {
3074         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3075         int exception;
3076
3077         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3078         if (exception) {
3079                 /* FIXME: better handling */
3080                 return X86EMUL_UNHANDLEABLE;
3081         }
3082         return X86EMUL_CONTINUE;
3083 }
3084
3085 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3086 {
3087         u8 opcodes[4];
3088         unsigned long rip = kvm_rip_read(vcpu);
3089         unsigned long rip_linear;
3090
3091         if (!printk_ratelimit())
3092                 return;
3093
3094         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3095
3096         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3097
3098         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3099                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3100 }
3101 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3102
3103 static struct x86_emulate_ops emulate_ops = {
3104         .read_std            = kvm_read_guest_virt,
3105         .read_emulated       = emulator_read_emulated,
3106         .write_emulated      = emulator_write_emulated,
3107         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3108 };
3109
3110 static void cache_all_regs(struct kvm_vcpu *vcpu)
3111 {
3112         kvm_register_read(vcpu, VCPU_REGS_RAX);
3113         kvm_register_read(vcpu, VCPU_REGS_RSP);
3114         kvm_register_read(vcpu, VCPU_REGS_RIP);
3115         vcpu->arch.regs_dirty = ~0;
3116 }
3117
3118 int emulate_instruction(struct kvm_vcpu *vcpu,
3119                         unsigned long cr2,
3120                         u16 error_code,
3121                         int emulation_type)
3122 {
3123         int r, shadow_mask;
3124         struct decode_cache *c;
3125         struct kvm_run *run = vcpu->run;
3126
3127         kvm_clear_exception_queue(vcpu);
3128         vcpu->arch.mmio_fault_cr2 = cr2;
3129         /*
3130          * TODO: fix emulate.c to use guest_read/write_register
3131          * instead of direct ->regs accesses, can save hundred cycles
3132          * on Intel for instructions that don't read/change RSP, for
3133          * for example.
3134          */
3135         cache_all_regs(vcpu);
3136
3137         vcpu->mmio_is_write = 0;
3138         vcpu->arch.pio.string = 0;
3139
3140         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3141                 int cs_db, cs_l;
3142                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3143
3144                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3145                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3146                 vcpu->arch.emulate_ctxt.mode =
3147                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3148                         ? X86EMUL_MODE_REAL : cs_l
3149                         ? X86EMUL_MODE_PROT64 : cs_db
3150                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3151
3152                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3153
3154                 /* Only allow emulation of specific instructions on #UD
3155                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3156                 c = &vcpu->arch.emulate_ctxt.decode;
3157                 if (emulation_type & EMULTYPE_TRAP_UD) {
3158                         if (!c->twobyte)
3159                                 return EMULATE_FAIL;
3160                         switch (c->b) {
3161                         case 0x01: /* VMMCALL */
3162                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3163                                         return EMULATE_FAIL;
3164                                 break;
3165                         case 0x34: /* sysenter */
3166                         case 0x35: /* sysexit */
3167                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3168                                         return EMULATE_FAIL;
3169                                 break;
3170                         case 0x05: /* syscall */
3171                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3172                                         return EMULATE_FAIL;
3173                                 break;
3174                         default:
3175                                 return EMULATE_FAIL;
3176                         }
3177
3178                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3179                                 return EMULATE_FAIL;
3180                 }
3181
3182                 ++vcpu->stat.insn_emulation;
3183                 if (r)  {
3184                         ++vcpu->stat.insn_emulation_fail;
3185                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3186                                 return EMULATE_DONE;
3187                         return EMULATE_FAIL;
3188                 }
3189         }
3190
3191         if (emulation_type & EMULTYPE_SKIP) {
3192                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3193                 return EMULATE_DONE;
3194         }
3195
3196         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3197         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3198
3199         if (r == 0)
3200                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3201
3202         if (vcpu->arch.pio.string)
3203                 return EMULATE_DO_MMIO;
3204
3205         if ((r || vcpu->mmio_is_write) && run) {
3206                 run->exit_reason = KVM_EXIT_MMIO;
3207                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3208                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3209                 run->mmio.len = vcpu->mmio_size;
3210                 run->mmio.is_write = vcpu->mmio_is_write;
3211         }
3212
3213         if (r) {
3214                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3215                         return EMULATE_DONE;
3216                 if (!vcpu->mmio_needed) {
3217                         kvm_report_emulation_failure(vcpu, "mmio");
3218                         return EMULATE_FAIL;
3219                 }
3220                 return EMULATE_DO_MMIO;
3221         }
3222
3223         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3224
3225         if (vcpu->mmio_is_write) {
3226                 vcpu->mmio_needed = 0;
3227                 return EMULATE_DO_MMIO;
3228         }
3229
3230         return EMULATE_DONE;
3231 }
3232 EXPORT_SYMBOL_GPL(emulate_instruction);
3233
3234 static int pio_copy_data(struct kvm_vcpu *vcpu)
3235 {
3236         void *p = vcpu->arch.pio_data;
3237         gva_t q = vcpu->arch.pio.guest_gva;
3238         unsigned bytes;
3239         int ret;
3240
3241         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3242         if (vcpu->arch.pio.in)
3243                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3244         else
3245                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3246         return ret;
3247 }
3248
3249 int complete_pio(struct kvm_vcpu *vcpu)
3250 {
3251         struct kvm_pio_request *io = &vcpu->arch.pio;
3252         long delta;
3253         int r;
3254         unsigned long val;
3255
3256         if (!io->string) {
3257                 if (io->in) {
3258                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3259                         memcpy(&val, vcpu->arch.pio_data, io->size);
3260                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3261                 }
3262         } else {
3263                 if (io->in) {
3264                         r = pio_copy_data(vcpu);
3265                         if (r)
3266                                 return r;
3267                 }
3268
3269                 delta = 1;
3270                 if (io->rep) {
3271                         delta *= io->cur_count;
3272                         /*
3273                          * The size of the register should really depend on
3274                          * current address size.
3275                          */
3276                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3277                         val -= delta;
3278                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3279                 }
3280                 if (io->down)
3281                         delta = -delta;
3282                 delta *= io->size;
3283                 if (io->in) {
3284                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3285                         val += delta;
3286                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3287                 } else {
3288                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3289                         val += delta;
3290                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3291                 }
3292         }
3293
3294         io->count -= io->cur_count;
3295         io->cur_count = 0;
3296
3297         return 0;
3298 }
3299
3300 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3301 {
3302         /* TODO: String I/O for in kernel device */
3303         int r;
3304
3305         if (vcpu->arch.pio.in)
3306                 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3307                                     vcpu->arch.pio.size, pd);
3308         else
3309                 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3310                                      vcpu->arch.pio.size, pd);
3311         return r;
3312 }
3313
3314 static int pio_string_write(struct kvm_vcpu *vcpu)
3315 {
3316         struct kvm_pio_request *io = &vcpu->arch.pio;
3317         void *pd = vcpu->arch.pio_data;
3318         int i, r = 0;
3319
3320         for (i = 0; i < io->cur_count; i++) {
3321                 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
3322                                      io->port, io->size, pd)) {
3323                         r = -EOPNOTSUPP;
3324                         break;
3325                 }
3326                 pd += io->size;
3327         }
3328         return r;
3329 }
3330
3331 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3332 {
3333         unsigned long val;
3334
3335         vcpu->run->exit_reason = KVM_EXIT_IO;
3336         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3337         vcpu->run->io.size = vcpu->arch.pio.size = size;
3338         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3339         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3340         vcpu->run->io.port = vcpu->arch.pio.port = port;
3341         vcpu->arch.pio.in = in;
3342         vcpu->arch.pio.string = 0;
3343         vcpu->arch.pio.down = 0;
3344         vcpu->arch.pio.rep = 0;
3345
3346         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3347                       size, 1);
3348
3349         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3350         memcpy(vcpu->arch.pio_data, &val, 4);
3351
3352         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3353                 complete_pio(vcpu);
3354                 return 1;
3355         }
3356         return 0;
3357 }
3358 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3359
3360 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3361                   int size, unsigned long count, int down,
3362                   gva_t address, int rep, unsigned port)
3363 {
3364         unsigned now, in_page;
3365         int ret = 0;
3366
3367         vcpu->run->exit_reason = KVM_EXIT_IO;
3368         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3369         vcpu->run->io.size = vcpu->arch.pio.size = size;
3370         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3371         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3372         vcpu->run->io.port = vcpu->arch.pio.port = port;
3373         vcpu->arch.pio.in = in;
3374         vcpu->arch.pio.string = 1;
3375         vcpu->arch.pio.down = down;
3376         vcpu->arch.pio.rep = rep;
3377
3378         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3379                       size, count);
3380
3381         if (!count) {
3382                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3383                 return 1;
3384         }
3385
3386         if (!down)
3387                 in_page = PAGE_SIZE - offset_in_page(address);
3388         else
3389                 in_page = offset_in_page(address) + size;
3390         now = min(count, (unsigned long)in_page / size);
3391         if (!now)
3392                 now = 1;
3393         if (down) {
3394                 /*
3395                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3396                  */
3397                 pr_unimpl(vcpu, "guest string pio down\n");
3398                 kvm_inject_gp(vcpu, 0);
3399                 return 1;
3400         }
3401         vcpu->run->io.count = now;
3402         vcpu->arch.pio.cur_count = now;
3403
3404         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3405                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3406
3407         vcpu->arch.pio.guest_gva = address;
3408
3409         if (!vcpu->arch.pio.in) {
3410                 /* string PIO write */
3411                 ret = pio_copy_data(vcpu);
3412                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3413                         kvm_inject_gp(vcpu, 0);
3414                         return 1;
3415                 }
3416                 if (ret == 0 && !pio_string_write(vcpu)) {
3417                         complete_pio(vcpu);
3418                         if (vcpu->arch.pio.count == 0)
3419                                 ret = 1;
3420                 }
3421         }
3422         /* no string PIO read support yet */
3423
3424         return ret;
3425 }
3426 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3427
3428 static void bounce_off(void *info)
3429 {
3430         /* nothing */
3431 }
3432
3433 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3434                                      void *data)
3435 {
3436         struct cpufreq_freqs *freq = data;
3437         struct kvm *kvm;
3438         struct kvm_vcpu *vcpu;
3439         int i, send_ipi = 0;
3440
3441         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3442                 return 0;
3443         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3444                 return 0;
3445         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3446
3447         spin_lock(&kvm_lock);
3448         list_for_each_entry(kvm, &vm_list, vm_list) {
3449                 kvm_for_each_vcpu(i, vcpu, kvm) {
3450                         if (vcpu->cpu != freq->cpu)
3451                                 continue;
3452                         if (!kvm_request_guest_time_update(vcpu))
3453                                 continue;
3454                         if (vcpu->cpu != smp_processor_id())
3455                                 send_ipi++;
3456                 }
3457         }
3458         spin_unlock(&kvm_lock);
3459
3460         if (freq->old < freq->new && send_ipi) {
3461                 /*
3462                  * We upscale the frequency.  Must make the guest
3463                  * doesn't see old kvmclock values while running with
3464                  * the new frequency, otherwise we risk the guest sees
3465                  * time go backwards.
3466                  *
3467                  * In case we update the frequency for another cpu
3468                  * (which might be in guest context) send an interrupt
3469                  * to kick the cpu out of guest context.  Next time
3470                  * guest context is entered kvmclock will be updated,
3471                  * so the guest will not see stale values.
3472                  */
3473                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3474         }
3475         return 0;
3476 }
3477
3478 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3479         .notifier_call  = kvmclock_cpufreq_notifier
3480 };
3481
3482 static void kvm_timer_init(void)
3483 {
3484         int cpu;
3485
3486         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3487                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3488                                           CPUFREQ_TRANSITION_NOTIFIER);
3489                 for_each_online_cpu(cpu) {
3490                         unsigned long khz = cpufreq_get(cpu);
3491                         if (!khz)
3492                                 khz = tsc_khz;
3493                         per_cpu(cpu_tsc_khz, cpu) = khz;
3494                 }
3495         } else {
3496                 for_each_possible_cpu(cpu)
3497                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3498         }
3499 }
3500
3501 int kvm_arch_init(void *opaque)
3502 {
3503         int r;
3504         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3505
3506         if (kvm_x86_ops) {
3507                 printk(KERN_ERR "kvm: already loaded the other module\n");
3508                 r = -EEXIST;
3509                 goto out;
3510         }
3511
3512         if (!ops->cpu_has_kvm_support()) {
3513                 printk(KERN_ERR "kvm: no hardware support\n");
3514                 r = -EOPNOTSUPP;
3515                 goto out;
3516         }
3517         if (ops->disabled_by_bios()) {
3518                 printk(KERN_ERR "kvm: disabled by bios\n");
3519                 r = -EOPNOTSUPP;
3520                 goto out;
3521         }
3522
3523         r = kvm_mmu_module_init();
3524         if (r)
3525                 goto out;
3526
3527         kvm_init_msr_list();
3528
3529         kvm_x86_ops = ops;
3530         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3531         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3532         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3533                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3534
3535         kvm_timer_init();
3536
3537         return 0;
3538
3539 out:
3540         return r;
3541 }
3542
3543 void kvm_arch_exit(void)
3544 {
3545         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3546                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3547                                             CPUFREQ_TRANSITION_NOTIFIER);
3548         kvm_x86_ops = NULL;
3549         kvm_mmu_module_exit();
3550 }
3551
3552 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3553 {
3554         ++vcpu->stat.halt_exits;
3555         if (irqchip_in_kernel(vcpu->kvm)) {
3556                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3557                 return 1;
3558         } else {
3559                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3560                 return 0;
3561         }
3562 }
3563 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3564
3565 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3566                            unsigned long a1)
3567 {
3568         if (is_long_mode(vcpu))
3569                 return a0;
3570         else
3571                 return a0 | ((gpa_t)a1 << 32);
3572 }
3573
3574 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3575 {
3576         unsigned long nr, a0, a1, a2, a3, ret;
3577         int r = 1;
3578
3579         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3580         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3581         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3582         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3583         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3584
3585         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3586
3587         if (!is_long_mode(vcpu)) {
3588                 nr &= 0xFFFFFFFF;
3589                 a0 &= 0xFFFFFFFF;
3590                 a1 &= 0xFFFFFFFF;
3591                 a2 &= 0xFFFFFFFF;
3592                 a3 &= 0xFFFFFFFF;
3593         }
3594
3595         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3596                 ret = -KVM_EPERM;
3597                 goto out;
3598         }
3599
3600         switch (nr) {
3601         case KVM_HC_VAPIC_POLL_IRQ:
3602                 ret = 0;
3603                 break;
3604         case KVM_HC_MMU_OP:
3605                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3606                 break;
3607         default:
3608                 ret = -KVM_ENOSYS;
3609                 break;
3610         }
3611 out:
3612         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3613         ++vcpu->stat.hypercalls;
3614         return r;
3615 }
3616 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3617
3618 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3619 {
3620         char instruction[3];
3621         int ret = 0;
3622         unsigned long rip = kvm_rip_read(vcpu);
3623
3624
3625         /*
3626          * Blow out the MMU to ensure that no other VCPU has an active mapping
3627          * to ensure that the updated hypercall appears atomically across all
3628          * VCPUs.
3629          */
3630         kvm_mmu_zap_all(vcpu->kvm);
3631
3632         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3633         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3634             != X86EMUL_CONTINUE)
3635                 ret = -EFAULT;
3636
3637         return ret;
3638 }
3639
3640 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3641 {
3642         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3643 }
3644
3645 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3646 {
3647         struct descriptor_table dt = { limit, base };
3648
3649         kvm_x86_ops->set_gdt(vcpu, &dt);
3650 }
3651
3652 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3653 {
3654         struct descriptor_table dt = { limit, base };
3655
3656         kvm_x86_ops->set_idt(vcpu, &dt);
3657 }
3658
3659 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3660                    unsigned long *rflags)
3661 {
3662         kvm_lmsw(vcpu, msw);
3663         *rflags = kvm_get_rflags(vcpu);
3664 }
3665
3666 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3667 {
3668         unsigned long value;
3669
3670         switch (cr) {
3671         case 0:
3672                 value = vcpu->arch.cr0;
3673                 break;
3674         case 2:
3675                 value = vcpu->arch.cr2;
3676                 break;
3677         case 3:
3678                 value = vcpu->arch.cr3;
3679                 break;
3680         case 4:
3681                 value = kvm_read_cr4(vcpu);
3682                 break;
3683         case 8:
3684                 value = kvm_get_cr8(vcpu);
3685                 break;
3686         default:
3687                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3688                 return 0;
3689         }
3690
3691         return value;
3692 }
3693
3694 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3695                      unsigned long *rflags)
3696 {
3697         switch (cr) {
3698         case 0:
3699                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3700                 *rflags = kvm_get_rflags(vcpu);
3701                 break;
3702         case 2:
3703                 vcpu->arch.cr2 = val;
3704                 break;
3705         case 3:
3706                 kvm_set_cr3(vcpu, val);
3707                 break;
3708         case 4:
3709                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3710                 break;
3711         case 8:
3712                 kvm_set_cr8(vcpu, val & 0xfUL);
3713                 break;
3714         default:
3715                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3716         }
3717 }
3718
3719 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3720 {
3721         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3722         int j, nent = vcpu->arch.cpuid_nent;
3723
3724         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3725         /* when no next entry is found, the current entry[i] is reselected */
3726         for (j = i + 1; ; j = (j + 1) % nent) {
3727                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3728                 if (ej->function == e->function) {
3729                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3730                         return j;
3731                 }
3732         }
3733         return 0; /* silence gcc, even though control never reaches here */
3734 }
3735
3736 /* find an entry with matching function, matching index (if needed), and that
3737  * should be read next (if it's stateful) */
3738 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3739         u32 function, u32 index)
3740 {
3741         if (e->function != function)
3742                 return 0;
3743         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3744                 return 0;
3745         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3746             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3747                 return 0;
3748         return 1;
3749 }
3750
3751 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3752                                               u32 function, u32 index)
3753 {
3754         int i;
3755         struct kvm_cpuid_entry2 *best = NULL;
3756
3757         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3758                 struct kvm_cpuid_entry2 *e;
3759
3760                 e = &vcpu->arch.cpuid_entries[i];
3761                 if (is_matching_cpuid_entry(e, function, index)) {
3762                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3763                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3764                         best = e;
3765                         break;
3766                 }
3767                 /*
3768                  * Both basic or both extended?
3769                  */
3770                 if (((e->function ^ function) & 0x80000000) == 0)
3771                         if (!best || e->function > best->function)
3772                                 best = e;
3773         }
3774         return best;
3775 }
3776 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
3777
3778 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3779 {
3780         struct kvm_cpuid_entry2 *best;
3781
3782         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3783         if (best)
3784                 return best->eax & 0xff;
3785         return 36;
3786 }
3787
3788 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3789 {
3790         u32 function, index;
3791         struct kvm_cpuid_entry2 *best;
3792
3793         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3794         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3795         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3796         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3797         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3798         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3799         best = kvm_find_cpuid_entry(vcpu, function, index);
3800         if (best) {
3801                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3802                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3803                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3804                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3805         }
3806         kvm_x86_ops->skip_emulated_instruction(vcpu);
3807         trace_kvm_cpuid(function,
3808                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3809                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3810                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3811                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3812 }
3813 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3814
3815 /*
3816  * Check if userspace requested an interrupt window, and that the
3817  * interrupt window is open.
3818  *
3819  * No need to exit to userspace if we already have an interrupt queued.
3820  */
3821 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3822 {
3823         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3824                 vcpu->run->request_interrupt_window &&
3825                 kvm_arch_interrupt_allowed(vcpu));
3826 }
3827
3828 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3829 {
3830         struct kvm_run *kvm_run = vcpu->run;
3831
3832         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3833         kvm_run->cr8 = kvm_get_cr8(vcpu);
3834         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3835         if (irqchip_in_kernel(vcpu->kvm))
3836                 kvm_run->ready_for_interrupt_injection = 1;
3837         else
3838                 kvm_run->ready_for_interrupt_injection =
3839                         kvm_arch_interrupt_allowed(vcpu) &&
3840                         !kvm_cpu_has_interrupt(vcpu) &&
3841                         !kvm_event_needs_reinjection(vcpu);
3842 }
3843
3844 static void vapic_enter(struct kvm_vcpu *vcpu)
3845 {
3846         struct kvm_lapic *apic = vcpu->arch.apic;
3847         struct page *page;
3848
3849         if (!apic || !apic->vapic_addr)
3850                 return;
3851
3852         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3853
3854         vcpu->arch.apic->vapic_page = page;
3855 }
3856
3857 static void vapic_exit(struct kvm_vcpu *vcpu)
3858 {
3859         struct kvm_lapic *apic = vcpu->arch.apic;
3860
3861         if (!apic || !apic->vapic_addr)
3862                 return;
3863
3864         down_read(&vcpu->kvm->slots_lock);
3865         kvm_release_page_dirty(apic->vapic_page);
3866         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3867         up_read(&vcpu->kvm->slots_lock);
3868 }
3869
3870 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3871 {
3872         int max_irr, tpr;
3873
3874         if (!kvm_x86_ops->update_cr8_intercept)
3875                 return;
3876
3877         if (!vcpu->arch.apic)
3878                 return;
3879
3880         if (!vcpu->arch.apic->vapic_addr)
3881                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3882         else
3883                 max_irr = -1;
3884
3885         if (max_irr != -1)
3886                 max_irr >>= 4;
3887
3888         tpr = kvm_lapic_get_cr8(vcpu);
3889
3890         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3891 }
3892
3893 static void inject_pending_event(struct kvm_vcpu *vcpu)
3894 {
3895         /* try to reinject previous events if any */
3896         if (vcpu->arch.exception.pending) {
3897                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3898                                           vcpu->arch.exception.has_error_code,
3899                                           vcpu->arch.exception.error_code);
3900                 return;
3901         }
3902
3903         if (vcpu->arch.nmi_injected) {
3904                 kvm_x86_ops->set_nmi(vcpu);
3905                 return;
3906         }
3907
3908         if (vcpu->arch.interrupt.pending) {
3909                 kvm_x86_ops->set_irq(vcpu);
3910                 return;
3911         }
3912
3913         /* try to inject new event if pending */
3914         if (vcpu->arch.nmi_pending) {
3915                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3916                         vcpu->arch.nmi_pending = false;
3917                         vcpu->arch.nmi_injected = true;
3918                         kvm_x86_ops->set_nmi(vcpu);
3919                 }
3920         } else if (kvm_cpu_has_interrupt(vcpu)) {
3921                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3922                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3923                                             false);
3924                         kvm_x86_ops->set_irq(vcpu);
3925                 }
3926         }
3927 }
3928
3929 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3930 {
3931         int r;
3932         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3933                 vcpu->run->request_interrupt_window;
3934
3935         if (vcpu->requests)
3936                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3937                         kvm_mmu_unload(vcpu);
3938
3939         r = kvm_mmu_reload(vcpu);
3940         if (unlikely(r))
3941                 goto out;
3942
3943         if (vcpu->requests) {
3944                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3945                         __kvm_migrate_timers(vcpu);
3946                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3947                         kvm_write_guest_time(vcpu);
3948                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3949                         kvm_mmu_sync_roots(vcpu);
3950                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3951                         kvm_x86_ops->tlb_flush(vcpu);
3952                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3953                                        &vcpu->requests)) {
3954                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3955                         r = 0;
3956                         goto out;
3957                 }
3958                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3959                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3960                         r = 0;
3961                         goto out;
3962                 }
3963         }
3964
3965         preempt_disable();
3966
3967         kvm_x86_ops->prepare_guest_switch(vcpu);
3968         kvm_load_guest_fpu(vcpu);
3969
3970         local_irq_disable();
3971
3972         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3973         smp_mb__after_clear_bit();
3974
3975         if (vcpu->requests || need_resched() || signal_pending(current)) {
3976                 set_bit(KVM_REQ_KICK, &vcpu->requests);
3977                 local_irq_enable();
3978                 preempt_enable();
3979                 r = 1;
3980                 goto out;
3981         }
3982
3983         inject_pending_event(vcpu);
3984
3985         /* enable NMI/IRQ window open exits if needed */
3986         if (vcpu->arch.nmi_pending)
3987                 kvm_x86_ops->enable_nmi_window(vcpu);
3988         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3989                 kvm_x86_ops->enable_irq_window(vcpu);
3990
3991         if (kvm_lapic_enabled(vcpu)) {
3992                 update_cr8_intercept(vcpu);
3993                 kvm_lapic_sync_to_vapic(vcpu);
3994         }
3995
3996         up_read(&vcpu->kvm->slots_lock);
3997
3998         kvm_guest_enter();
3999
4000         if (unlikely(vcpu->arch.switch_db_regs)) {
4001                 set_debugreg(0, 7);
4002                 set_debugreg(vcpu->arch.eff_db[0], 0);
4003                 set_debugreg(vcpu->arch.eff_db[1], 1);
4004                 set_debugreg(vcpu->arch.eff_db[2], 2);
4005                 set_debugreg(vcpu->arch.eff_db[3], 3);
4006         }
4007
4008         trace_kvm_entry(vcpu->vcpu_id);
4009         kvm_x86_ops->run(vcpu);
4010
4011         /*
4012          * If the guest has used debug registers, at least dr7
4013          * will be disabled while returning to the host.
4014          * If we don't have active breakpoints in the host, we don't
4015          * care about the messed up debug address registers. But if
4016          * we have some of them active, restore the old state.
4017          */
4018         if (hw_breakpoint_active())
4019                 hw_breakpoint_restore();
4020
4021         set_bit(KVM_REQ_KICK, &vcpu->requests);
4022         local_irq_enable();
4023
4024         ++vcpu->stat.exits;
4025
4026         /*
4027          * We must have an instruction between local_irq_enable() and
4028          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4029          * the interrupt shadow.  The stat.exits increment will do nicely.
4030          * But we need to prevent reordering, hence this barrier():
4031          */
4032         barrier();
4033
4034         kvm_guest_exit();
4035
4036         preempt_enable();
4037
4038         down_read(&vcpu->kvm->slots_lock);
4039
4040         /*
4041          * Profile KVM exit RIPs:
4042          */
4043         if (unlikely(prof_on == KVM_PROFILING)) {
4044                 unsigned long rip = kvm_rip_read(vcpu);
4045                 profile_hit(KVM_PROFILING, (void *)rip);
4046         }
4047
4048
4049         kvm_lapic_sync_from_vapic(vcpu);
4050
4051         r = kvm_x86_ops->handle_exit(vcpu);
4052 out:
4053         return r;
4054 }
4055
4056
4057 static int __vcpu_run(struct kvm_vcpu *vcpu)
4058 {
4059         int r;
4060
4061         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4062                 pr_debug("vcpu %d received sipi with vector # %x\n",
4063                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4064                 kvm_lapic_reset(vcpu);
4065                 r = kvm_arch_vcpu_reset(vcpu);
4066                 if (r)
4067                         return r;
4068                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4069         }
4070
4071         down_read(&vcpu->kvm->slots_lock);
4072         vapic_enter(vcpu);
4073
4074         r = 1;
4075         while (r > 0) {
4076                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4077                         r = vcpu_enter_guest(vcpu);
4078                 else {
4079                         up_read(&vcpu->kvm->slots_lock);
4080                         kvm_vcpu_block(vcpu);
4081                         down_read(&vcpu->kvm->slots_lock);
4082                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4083                         {
4084                                 switch(vcpu->arch.mp_state) {
4085                                 case KVM_MP_STATE_HALTED:
4086                                         vcpu->arch.mp_state =
4087                                                 KVM_MP_STATE_RUNNABLE;
4088                                 case KVM_MP_STATE_RUNNABLE:
4089                                         break;
4090                                 case KVM_MP_STATE_SIPI_RECEIVED:
4091                                 default:
4092                                         r = -EINTR;
4093                                         break;
4094                                 }
4095                         }
4096                 }
4097
4098                 if (r <= 0)
4099                         break;
4100
4101                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4102                 if (kvm_cpu_has_pending_timer(vcpu))
4103                         kvm_inject_pending_timer_irqs(vcpu);
4104
4105                 if (dm_request_for_irq_injection(vcpu)) {
4106                         r = -EINTR;
4107                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4108                         ++vcpu->stat.request_irq_exits;
4109                 }
4110                 if (signal_pending(current)) {
4111                         r = -EINTR;
4112                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4113                         ++vcpu->stat.signal_exits;
4114                 }
4115                 if (need_resched()) {
4116                         up_read(&vcpu->kvm->slots_lock);
4117                         kvm_resched(vcpu);
4118                         down_read(&vcpu->kvm->slots_lock);
4119                 }
4120         }
4121
4122         up_read(&vcpu->kvm->slots_lock);
4123         post_kvm_run_save(vcpu);
4124
4125         vapic_exit(vcpu);
4126
4127         return r;
4128 }
4129
4130 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4131 {
4132         int r;
4133         sigset_t sigsaved;
4134
4135         vcpu_load(vcpu);
4136
4137         if (vcpu->sigset_active)
4138                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4139
4140         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4141                 kvm_vcpu_block(vcpu);
4142                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4143                 r = -EAGAIN;
4144                 goto out;
4145         }
4146
4147         /* re-sync apic's tpr */
4148         if (!irqchip_in_kernel(vcpu->kvm))
4149                 kvm_set_cr8(vcpu, kvm_run->cr8);
4150
4151         if (vcpu->arch.pio.cur_count) {
4152                 r = complete_pio(vcpu);
4153                 if (r)
4154                         goto out;
4155         }
4156         if (vcpu->mmio_needed) {
4157                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4158                 vcpu->mmio_read_completed = 1;
4159                 vcpu->mmio_needed = 0;
4160
4161                 down_read(&vcpu->kvm->slots_lock);
4162                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4163                                         EMULTYPE_NO_DECODE);
4164                 up_read(&vcpu->kvm->slots_lock);
4165                 if (r == EMULATE_DO_MMIO) {
4166                         /*
4167                          * Read-modify-write.  Back to userspace.
4168                          */
4169                         r = 0;
4170                         goto out;
4171                 }
4172         }
4173         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4174                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4175                                      kvm_run->hypercall.ret);
4176
4177         r = __vcpu_run(vcpu);
4178
4179 out:
4180         if (vcpu->sigset_active)
4181                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4182
4183         vcpu_put(vcpu);
4184         return r;
4185 }
4186
4187 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4188 {
4189         vcpu_load(vcpu);
4190
4191         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4192         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4193         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4194         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4195         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4196         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4197         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4198         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4199 #ifdef CONFIG_X86_64
4200         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4201         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4202         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4203         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4204         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4205         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4206         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4207         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4208 #endif
4209
4210         regs->rip = kvm_rip_read(vcpu);
4211         regs->rflags = kvm_get_rflags(vcpu);
4212
4213         vcpu_put(vcpu);
4214
4215         return 0;
4216 }
4217
4218 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4219 {
4220         vcpu_load(vcpu);
4221
4222         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4223         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4224         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4225         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4226         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4227         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4228         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4229         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4230 #ifdef CONFIG_X86_64
4231         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4232         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4233         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4234         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4235         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4236         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4237         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4238         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4239 #endif
4240
4241         kvm_rip_write(vcpu, regs->rip);
4242         kvm_set_rflags(vcpu, regs->rflags);
4243
4244         vcpu->arch.exception.pending = false;
4245
4246         vcpu_put(vcpu);
4247
4248         return 0;
4249 }
4250
4251 void kvm_get_segment(struct kvm_vcpu *vcpu,
4252                      struct kvm_segment *var, int seg)
4253 {
4254         kvm_x86_ops->get_segment(vcpu, var, seg);
4255 }
4256
4257 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4258 {
4259         struct kvm_segment cs;
4260
4261         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4262         *db = cs.db;
4263         *l = cs.l;
4264 }
4265 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4266
4267 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4268                                   struct kvm_sregs *sregs)
4269 {
4270         struct descriptor_table dt;
4271
4272         vcpu_load(vcpu);
4273
4274         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4275         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4276         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4277         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4278         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4279         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4280
4281         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4282         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4283
4284         kvm_x86_ops->get_idt(vcpu, &dt);
4285         sregs->idt.limit = dt.limit;
4286         sregs->idt.base = dt.base;
4287         kvm_x86_ops->get_gdt(vcpu, &dt);
4288         sregs->gdt.limit = dt.limit;
4289         sregs->gdt.base = dt.base;
4290
4291         sregs->cr0 = vcpu->arch.cr0;
4292         sregs->cr2 = vcpu->arch.cr2;
4293         sregs->cr3 = vcpu->arch.cr3;
4294         sregs->cr4 = kvm_read_cr4(vcpu);
4295         sregs->cr8 = kvm_get_cr8(vcpu);
4296         sregs->efer = vcpu->arch.shadow_efer;
4297         sregs->apic_base = kvm_get_apic_base(vcpu);
4298
4299         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4300
4301         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4302                 set_bit(vcpu->arch.interrupt.nr,
4303                         (unsigned long *)sregs->interrupt_bitmap);
4304
4305         vcpu_put(vcpu);
4306
4307         return 0;
4308 }
4309
4310 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4311                                     struct kvm_mp_state *mp_state)
4312 {
4313         vcpu_load(vcpu);
4314         mp_state->mp_state = vcpu->arch.mp_state;
4315         vcpu_put(vcpu);
4316         return 0;
4317 }
4318
4319 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4320                                     struct kvm_mp_state *mp_state)
4321 {
4322         vcpu_load(vcpu);
4323         vcpu->arch.mp_state = mp_state->mp_state;
4324         vcpu_put(vcpu);
4325         return 0;
4326 }
4327
4328 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4329                         struct kvm_segment *var, int seg)
4330 {
4331         kvm_x86_ops->set_segment(vcpu, var, seg);
4332 }
4333
4334 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4335                                    struct kvm_segment *kvm_desct)
4336 {
4337         kvm_desct->base = get_desc_base(seg_desc);
4338         kvm_desct->limit = get_desc_limit(seg_desc);
4339         if (seg_desc->g) {
4340                 kvm_desct->limit <<= 12;
4341                 kvm_desct->limit |= 0xfff;
4342         }
4343         kvm_desct->selector = selector;
4344         kvm_desct->type = seg_desc->type;
4345         kvm_desct->present = seg_desc->p;
4346         kvm_desct->dpl = seg_desc->dpl;
4347         kvm_desct->db = seg_desc->d;
4348         kvm_desct->s = seg_desc->s;
4349         kvm_desct->l = seg_desc->l;
4350         kvm_desct->g = seg_desc->g;
4351         kvm_desct->avl = seg_desc->avl;
4352         if (!selector)
4353                 kvm_desct->unusable = 1;
4354         else
4355                 kvm_desct->unusable = 0;
4356         kvm_desct->padding = 0;
4357 }
4358
4359 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4360                                           u16 selector,
4361                                           struct descriptor_table *dtable)
4362 {
4363         if (selector & 1 << 2) {
4364                 struct kvm_segment kvm_seg;
4365
4366                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4367
4368                 if (kvm_seg.unusable)
4369                         dtable->limit = 0;
4370                 else
4371                         dtable->limit = kvm_seg.limit;
4372                 dtable->base = kvm_seg.base;
4373         }
4374         else
4375                 kvm_x86_ops->get_gdt(vcpu, dtable);
4376 }
4377
4378 /* allowed just for 8 bytes segments */
4379 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4380                                          struct desc_struct *seg_desc)
4381 {
4382         struct descriptor_table dtable;
4383         u16 index = selector >> 3;
4384
4385         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4386
4387         if (dtable.limit < index * 8 + 7) {
4388                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4389                 return 1;
4390         }
4391         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4392 }
4393
4394 /* allowed just for 8 bytes segments */
4395 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4396                                          struct desc_struct *seg_desc)
4397 {
4398         struct descriptor_table dtable;
4399         u16 index = selector >> 3;
4400
4401         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4402
4403         if (dtable.limit < index * 8 + 7)
4404                 return 1;
4405         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4406 }
4407
4408 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4409                              struct desc_struct *seg_desc)
4410 {
4411         u32 base_addr = get_desc_base(seg_desc);
4412
4413         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4414 }
4415
4416 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4417 {
4418         struct kvm_segment kvm_seg;
4419
4420         kvm_get_segment(vcpu, &kvm_seg, seg);
4421         return kvm_seg.selector;
4422 }
4423
4424 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4425                                                 u16 selector,
4426                                                 struct kvm_segment *kvm_seg)
4427 {
4428         struct desc_struct seg_desc;
4429
4430         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4431                 return 1;
4432         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4433         return 0;
4434 }
4435
4436 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4437 {
4438         struct kvm_segment segvar = {
4439                 .base = selector << 4,
4440                 .limit = 0xffff,
4441                 .selector = selector,
4442                 .type = 3,
4443                 .present = 1,
4444                 .dpl = 3,
4445                 .db = 0,
4446                 .s = 1,
4447                 .l = 0,
4448                 .g = 0,
4449                 .avl = 0,
4450                 .unusable = 0,
4451         };
4452         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4453         return 0;
4454 }
4455
4456 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4457 {
4458         return (seg != VCPU_SREG_LDTR) &&
4459                 (seg != VCPU_SREG_TR) &&
4460                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4461 }
4462
4463 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4464                                          u16 selector)
4465 {
4466         /* NULL selector is not valid for CS and SS */
4467         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4468                 if (!selector)
4469                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4470 }
4471
4472 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4473                                 int type_bits, int seg)
4474 {
4475         struct kvm_segment kvm_seg;
4476
4477         if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4478                 return kvm_load_realmode_segment(vcpu, selector, seg);
4479         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4480                 return 1;
4481
4482         kvm_check_segment_descriptor(vcpu, seg, selector);
4483         kvm_seg.type |= type_bits;
4484
4485         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4486             seg != VCPU_SREG_LDTR)
4487                 if (!kvm_seg.s)
4488                         kvm_seg.unusable = 1;
4489
4490         kvm_set_segment(vcpu, &kvm_seg, seg);
4491         return 0;
4492 }
4493
4494 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4495                                 struct tss_segment_32 *tss)
4496 {
4497         tss->cr3 = vcpu->arch.cr3;
4498         tss->eip = kvm_rip_read(vcpu);
4499         tss->eflags = kvm_get_rflags(vcpu);
4500         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4501         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4502         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4503         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4504         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4505         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4506         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4507         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4508         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4509         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4510         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4511         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4512         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4513         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4514         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4515 }
4516
4517 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4518                                   struct tss_segment_32 *tss)
4519 {
4520         kvm_set_cr3(vcpu, tss->cr3);
4521
4522         kvm_rip_write(vcpu, tss->eip);
4523         kvm_set_rflags(vcpu, tss->eflags | 2);
4524
4525         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4526         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4527         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4528         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4529         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4530         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4531         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4532         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4533
4534         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4535                 return 1;
4536
4537         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4538                 return 1;
4539
4540         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4541                 return 1;
4542
4543         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4544                 return 1;
4545
4546         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4547                 return 1;
4548
4549         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4550                 return 1;
4551
4552         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4553                 return 1;
4554         return 0;
4555 }
4556
4557 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4558                                 struct tss_segment_16 *tss)
4559 {
4560         tss->ip = kvm_rip_read(vcpu);
4561         tss->flag = kvm_get_rflags(vcpu);
4562         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4563         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4564         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4565         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4566         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4567         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4568         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4569         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4570
4571         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4572         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4573         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4574         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4575         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4576 }
4577
4578 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4579                                  struct tss_segment_16 *tss)
4580 {
4581         kvm_rip_write(vcpu, tss->ip);
4582         kvm_set_rflags(vcpu, tss->flag | 2);
4583         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4584         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4585         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4586         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4587         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4588         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4589         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4590         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4591
4592         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4593                 return 1;
4594
4595         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4596                 return 1;
4597
4598         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4599                 return 1;
4600
4601         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4602                 return 1;
4603
4604         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4605                 return 1;
4606         return 0;
4607 }
4608
4609 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4610                               u16 old_tss_sel, u32 old_tss_base,
4611                               struct desc_struct *nseg_desc)
4612 {
4613         struct tss_segment_16 tss_segment_16;
4614         int ret = 0;
4615
4616         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4617                            sizeof tss_segment_16))
4618                 goto out;
4619
4620         save_state_to_tss16(vcpu, &tss_segment_16);
4621
4622         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4623                             sizeof tss_segment_16))
4624                 goto out;
4625
4626         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4627                            &tss_segment_16, sizeof tss_segment_16))
4628                 goto out;
4629
4630         if (old_tss_sel != 0xffff) {
4631                 tss_segment_16.prev_task_link = old_tss_sel;
4632
4633                 if (kvm_write_guest(vcpu->kvm,
4634                                     get_tss_base_addr(vcpu, nseg_desc),
4635                                     &tss_segment_16.prev_task_link,
4636                                     sizeof tss_segment_16.prev_task_link))
4637                         goto out;
4638         }
4639
4640         if (load_state_from_tss16(vcpu, &tss_segment_16))
4641                 goto out;
4642
4643         ret = 1;
4644 out:
4645         return ret;
4646 }
4647
4648 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4649                        u16 old_tss_sel, u32 old_tss_base,
4650                        struct desc_struct *nseg_desc)
4651 {
4652         struct tss_segment_32 tss_segment_32;
4653         int ret = 0;
4654
4655         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4656                            sizeof tss_segment_32))
4657                 goto out;
4658
4659         save_state_to_tss32(vcpu, &tss_segment_32);
4660
4661         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4662                             sizeof tss_segment_32))
4663                 goto out;
4664
4665         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4666                            &tss_segment_32, sizeof tss_segment_32))
4667                 goto out;
4668
4669         if (old_tss_sel != 0xffff) {
4670                 tss_segment_32.prev_task_link = old_tss_sel;
4671
4672                 if (kvm_write_guest(vcpu->kvm,
4673                                     get_tss_base_addr(vcpu, nseg_desc),
4674                                     &tss_segment_32.prev_task_link,
4675                                     sizeof tss_segment_32.prev_task_link))
4676                         goto out;
4677         }
4678
4679         if (load_state_from_tss32(vcpu, &tss_segment_32))
4680                 goto out;
4681
4682         ret = 1;
4683 out:
4684         return ret;
4685 }
4686
4687 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4688 {
4689         struct kvm_segment tr_seg;
4690         struct desc_struct cseg_desc;
4691         struct desc_struct nseg_desc;
4692         int ret = 0;
4693         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4694         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4695
4696         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4697
4698         /* FIXME: Handle errors. Failure to read either TSS or their
4699          * descriptors should generate a pagefault.
4700          */
4701         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4702                 goto out;
4703
4704         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4705                 goto out;
4706
4707         if (reason != TASK_SWITCH_IRET) {
4708                 int cpl;
4709
4710                 cpl = kvm_x86_ops->get_cpl(vcpu);
4711                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4712                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4713                         return 1;
4714                 }
4715         }
4716
4717         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4718                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4719                 return 1;
4720         }
4721
4722         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4723                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4724                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4725         }
4726
4727         if (reason == TASK_SWITCH_IRET) {
4728                 u32 eflags = kvm_get_rflags(vcpu);
4729                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4730         }
4731
4732         /* set back link to prev task only if NT bit is set in eflags
4733            note that old_tss_sel is not used afetr this point */
4734         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4735                 old_tss_sel = 0xffff;
4736
4737         if (nseg_desc.type & 8)
4738                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4739                                          old_tss_base, &nseg_desc);
4740         else
4741                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4742                                          old_tss_base, &nseg_desc);
4743
4744         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4745                 u32 eflags = kvm_get_rflags(vcpu);
4746                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4747         }
4748
4749         if (reason != TASK_SWITCH_IRET) {
4750                 nseg_desc.type |= (1 << 1);
4751                 save_guest_segment_descriptor(vcpu, tss_selector,
4752                                               &nseg_desc);
4753         }
4754
4755         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4756         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4757         tr_seg.type = 11;
4758         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4759 out:
4760         return ret;
4761 }
4762 EXPORT_SYMBOL_GPL(kvm_task_switch);
4763
4764 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4765                                   struct kvm_sregs *sregs)
4766 {
4767         int mmu_reset_needed = 0;
4768         int pending_vec, max_bits;
4769         struct descriptor_table dt;
4770
4771         vcpu_load(vcpu);
4772
4773         dt.limit = sregs->idt.limit;
4774         dt.base = sregs->idt.base;
4775         kvm_x86_ops->set_idt(vcpu, &dt);
4776         dt.limit = sregs->gdt.limit;
4777         dt.base = sregs->gdt.base;
4778         kvm_x86_ops->set_gdt(vcpu, &dt);
4779
4780         vcpu->arch.cr2 = sregs->cr2;
4781         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4782         vcpu->arch.cr3 = sregs->cr3;
4783
4784         kvm_set_cr8(vcpu, sregs->cr8);
4785
4786         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4787         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4788         kvm_set_apic_base(vcpu, sregs->apic_base);
4789
4790         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4791         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4792         vcpu->arch.cr0 = sregs->cr0;
4793
4794         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4795         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4796         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4797                 load_pdptrs(vcpu, vcpu->arch.cr3);
4798                 mmu_reset_needed = 1;
4799         }
4800
4801         if (mmu_reset_needed)
4802                 kvm_mmu_reset_context(vcpu);
4803
4804         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4805         pending_vec = find_first_bit(
4806                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4807         if (pending_vec < max_bits) {
4808                 kvm_queue_interrupt(vcpu, pending_vec, false);
4809                 pr_debug("Set back pending irq %d\n", pending_vec);
4810                 if (irqchip_in_kernel(vcpu->kvm))
4811                         kvm_pic_clear_isr_ack(vcpu->kvm);
4812         }
4813
4814         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4815         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4816         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4817         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4818         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4819         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4820
4821         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4822         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4823
4824         update_cr8_intercept(vcpu);
4825
4826         /* Older userspace won't unhalt the vcpu on reset. */
4827         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4828             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4829             !(vcpu->arch.cr0 & X86_CR0_PE))
4830                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4831
4832         vcpu_put(vcpu);
4833
4834         return 0;
4835 }
4836
4837 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4838                                         struct kvm_guest_debug *dbg)
4839 {
4840         unsigned long rflags;
4841         int i, r;
4842
4843         vcpu_load(vcpu);
4844
4845         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4846                 r = -EBUSY;
4847                 if (vcpu->arch.exception.pending)
4848                         goto unlock_out;
4849                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4850                         kvm_queue_exception(vcpu, DB_VECTOR);
4851                 else
4852                         kvm_queue_exception(vcpu, BP_VECTOR);
4853         }
4854
4855         /*
4856          * Read rflags as long as potentially injected trace flags are still
4857          * filtered out.
4858          */
4859         rflags = kvm_get_rflags(vcpu);
4860
4861         vcpu->guest_debug = dbg->control;
4862         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4863                 vcpu->guest_debug = 0;
4864
4865         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4866                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4867                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4868                 vcpu->arch.switch_db_regs =
4869                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4870         } else {
4871                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4872                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4873                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4874         }
4875
4876         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4877                 vcpu->arch.singlestep_cs =
4878                         get_segment_selector(vcpu, VCPU_SREG_CS);
4879                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4880         }
4881
4882         /*
4883          * Trigger an rflags update that will inject or remove the trace
4884          * flags.
4885          */
4886         kvm_set_rflags(vcpu, rflags);
4887
4888         kvm_x86_ops->set_guest_debug(vcpu, dbg);
4889
4890         r = 0;
4891
4892 unlock_out:
4893         vcpu_put(vcpu);
4894
4895         return r;
4896 }
4897
4898 /*
4899  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4900  * we have asm/x86/processor.h
4901  */
4902 struct fxsave {
4903         u16     cwd;
4904         u16     swd;
4905         u16     twd;
4906         u16     fop;
4907         u64     rip;
4908         u64     rdp;
4909         u32     mxcsr;
4910         u32     mxcsr_mask;
4911         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4912 #ifdef CONFIG_X86_64
4913         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4914 #else
4915         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4916 #endif
4917 };
4918
4919 /*
4920  * Translate a guest virtual address to a guest physical address.
4921  */
4922 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4923                                     struct kvm_translation *tr)
4924 {
4925         unsigned long vaddr = tr->linear_address;
4926         gpa_t gpa;
4927
4928         vcpu_load(vcpu);
4929         down_read(&vcpu->kvm->slots_lock);
4930         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4931         up_read(&vcpu->kvm->slots_lock);
4932         tr->physical_address = gpa;
4933         tr->valid = gpa != UNMAPPED_GVA;
4934         tr->writeable = 1;
4935         tr->usermode = 0;
4936         vcpu_put(vcpu);
4937
4938         return 0;
4939 }
4940
4941 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4942 {
4943         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4944
4945         vcpu_load(vcpu);
4946
4947         memcpy(fpu->fpr, fxsave->st_space, 128);
4948         fpu->fcw = fxsave->cwd;
4949         fpu->fsw = fxsave->swd;
4950         fpu->ftwx = fxsave->twd;
4951         fpu->last_opcode = fxsave->fop;
4952         fpu->last_ip = fxsave->rip;
4953         fpu->last_dp = fxsave->rdp;
4954         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4955
4956         vcpu_put(vcpu);
4957
4958         return 0;
4959 }
4960
4961 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4962 {
4963         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4964
4965         vcpu_load(vcpu);
4966
4967         memcpy(fxsave->st_space, fpu->fpr, 128);
4968         fxsave->cwd = fpu->fcw;
4969         fxsave->swd = fpu->fsw;
4970         fxsave->twd = fpu->ftwx;
4971         fxsave->fop = fpu->last_opcode;
4972         fxsave->rip = fpu->last_ip;
4973         fxsave->rdp = fpu->last_dp;
4974         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4975
4976         vcpu_put(vcpu);
4977
4978         return 0;
4979 }
4980
4981 void fx_init(struct kvm_vcpu *vcpu)
4982 {
4983         unsigned after_mxcsr_mask;
4984
4985         /*
4986          * Touch the fpu the first time in non atomic context as if
4987          * this is the first fpu instruction the exception handler
4988          * will fire before the instruction returns and it'll have to
4989          * allocate ram with GFP_KERNEL.
4990          */
4991         if (!used_math())
4992                 kvm_fx_save(&vcpu->arch.host_fx_image);
4993
4994         /* Initialize guest FPU by resetting ours and saving into guest's */
4995         preempt_disable();
4996         kvm_fx_save(&vcpu->arch.host_fx_image);
4997         kvm_fx_finit();
4998         kvm_fx_save(&vcpu->arch.guest_fx_image);
4999         kvm_fx_restore(&vcpu->arch.host_fx_image);
5000         preempt_enable();
5001
5002         vcpu->arch.cr0 |= X86_CR0_ET;
5003         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5004         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5005         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5006                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5007 }
5008 EXPORT_SYMBOL_GPL(fx_init);
5009
5010 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5011 {
5012         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
5013                 return;
5014
5015         vcpu->guest_fpu_loaded = 1;
5016         kvm_fx_save(&vcpu->arch.host_fx_image);
5017         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5018 }
5019 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
5020
5021 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5022 {
5023         if (!vcpu->guest_fpu_loaded)
5024                 return;
5025
5026         vcpu->guest_fpu_loaded = 0;
5027         kvm_fx_save(&vcpu->arch.guest_fx_image);
5028         kvm_fx_restore(&vcpu->arch.host_fx_image);
5029         ++vcpu->stat.fpu_reload;
5030 }
5031 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
5032
5033 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5034 {
5035         if (vcpu->arch.time_page) {
5036                 kvm_release_page_dirty(vcpu->arch.time_page);
5037                 vcpu->arch.time_page = NULL;
5038         }
5039
5040         kvm_x86_ops->vcpu_free(vcpu);
5041 }
5042
5043 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5044                                                 unsigned int id)
5045 {
5046         return kvm_x86_ops->vcpu_create(kvm, id);
5047 }
5048
5049 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5050 {
5051         int r;
5052
5053         /* We do fxsave: this must be aligned. */
5054         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5055
5056         vcpu->arch.mtrr_state.have_fixed = 1;
5057         vcpu_load(vcpu);
5058         r = kvm_arch_vcpu_reset(vcpu);
5059         if (r == 0)
5060                 r = kvm_mmu_setup(vcpu);
5061         vcpu_put(vcpu);
5062         if (r < 0)
5063                 goto free_vcpu;
5064
5065         return 0;
5066 free_vcpu:
5067         kvm_x86_ops->vcpu_free(vcpu);
5068         return r;
5069 }
5070
5071 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5072 {
5073         vcpu_load(vcpu);
5074         kvm_mmu_unload(vcpu);
5075         vcpu_put(vcpu);
5076
5077         kvm_x86_ops->vcpu_free(vcpu);
5078 }
5079
5080 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5081 {
5082         vcpu->arch.nmi_pending = false;
5083         vcpu->arch.nmi_injected = false;
5084
5085         vcpu->arch.switch_db_regs = 0;
5086         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5087         vcpu->arch.dr6 = DR6_FIXED_1;
5088         vcpu->arch.dr7 = DR7_FIXED_1;
5089
5090         return kvm_x86_ops->vcpu_reset(vcpu);
5091 }
5092
5093 int kvm_arch_hardware_enable(void *garbage)
5094 {
5095         /*
5096          * Since this may be called from a hotplug notifcation,
5097          * we can't get the CPU frequency directly.
5098          */
5099         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5100                 int cpu = raw_smp_processor_id();
5101                 per_cpu(cpu_tsc_khz, cpu) = 0;
5102         }
5103
5104         kvm_shared_msr_cpu_online();
5105
5106         return kvm_x86_ops->hardware_enable(garbage);
5107 }
5108
5109 void kvm_arch_hardware_disable(void *garbage)
5110 {
5111         kvm_x86_ops->hardware_disable(garbage);
5112         drop_user_return_notifiers(garbage);
5113 }
5114
5115 int kvm_arch_hardware_setup(void)
5116 {
5117         return kvm_x86_ops->hardware_setup();
5118 }
5119
5120 void kvm_arch_hardware_unsetup(void)
5121 {
5122         kvm_x86_ops->hardware_unsetup();
5123 }
5124
5125 void kvm_arch_check_processor_compat(void *rtn)
5126 {
5127         kvm_x86_ops->check_processor_compatibility(rtn);
5128 }
5129
5130 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5131 {
5132         struct page *page;
5133         struct kvm *kvm;
5134         int r;
5135
5136         BUG_ON(vcpu->kvm == NULL);
5137         kvm = vcpu->kvm;
5138
5139         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5140         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5141                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5142         else
5143                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5144
5145         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5146         if (!page) {
5147                 r = -ENOMEM;
5148                 goto fail;
5149         }
5150         vcpu->arch.pio_data = page_address(page);
5151
5152         r = kvm_mmu_create(vcpu);
5153         if (r < 0)
5154                 goto fail_free_pio_data;
5155
5156         if (irqchip_in_kernel(kvm)) {
5157                 r = kvm_create_lapic(vcpu);
5158                 if (r < 0)
5159                         goto fail_mmu_destroy;
5160         }
5161
5162         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5163                                        GFP_KERNEL);
5164         if (!vcpu->arch.mce_banks) {
5165                 r = -ENOMEM;
5166                 goto fail_free_lapic;
5167         }
5168         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5169
5170         return 0;
5171 fail_free_lapic:
5172         kvm_free_lapic(vcpu);
5173 fail_mmu_destroy:
5174         kvm_mmu_destroy(vcpu);
5175 fail_free_pio_data:
5176         free_page((unsigned long)vcpu->arch.pio_data);
5177 fail:
5178         return r;
5179 }
5180
5181 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5182 {
5183         kfree(vcpu->arch.mce_banks);
5184         kvm_free_lapic(vcpu);
5185         down_read(&vcpu->kvm->slots_lock);
5186         kvm_mmu_destroy(vcpu);
5187         up_read(&vcpu->kvm->slots_lock);
5188         free_page((unsigned long)vcpu->arch.pio_data);
5189 }
5190
5191 struct  kvm *kvm_arch_create_vm(void)
5192 {
5193         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5194
5195         if (!kvm)
5196                 return ERR_PTR(-ENOMEM);
5197
5198         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5199         if (!kvm->arch.aliases) {
5200                 kfree(kvm);
5201                 return ERR_PTR(-ENOMEM);
5202         }
5203
5204         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5205         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5206
5207         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5208         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5209
5210         rdtscll(kvm->arch.vm_init_tsc);
5211
5212         return kvm;
5213 }
5214
5215 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5216 {
5217         vcpu_load(vcpu);
5218         kvm_mmu_unload(vcpu);
5219         vcpu_put(vcpu);
5220 }
5221
5222 static void kvm_free_vcpus(struct kvm *kvm)
5223 {
5224         unsigned int i;
5225         struct kvm_vcpu *vcpu;
5226
5227         /*
5228          * Unpin any mmu pages first.
5229          */
5230         kvm_for_each_vcpu(i, vcpu, kvm)
5231                 kvm_unload_vcpu_mmu(vcpu);
5232         kvm_for_each_vcpu(i, vcpu, kvm)
5233                 kvm_arch_vcpu_free(vcpu);
5234
5235         mutex_lock(&kvm->lock);
5236         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5237                 kvm->vcpus[i] = NULL;
5238
5239         atomic_set(&kvm->online_vcpus, 0);
5240         mutex_unlock(&kvm->lock);
5241 }
5242
5243 void kvm_arch_sync_events(struct kvm *kvm)
5244 {
5245         kvm_free_all_assigned_devices(kvm);
5246 }
5247
5248 void kvm_arch_destroy_vm(struct kvm *kvm)
5249 {
5250         kvm_iommu_unmap_guest(kvm);
5251         kvm_free_pit(kvm);
5252         kfree(kvm->arch.vpic);
5253         kfree(kvm->arch.vioapic);
5254         kvm_free_vcpus(kvm);
5255         kvm_free_physmem(kvm);
5256         if (kvm->arch.apic_access_page)
5257                 put_page(kvm->arch.apic_access_page);
5258         if (kvm->arch.ept_identity_pagetable)
5259                 put_page(kvm->arch.ept_identity_pagetable);
5260         kfree(kvm->arch.aliases);
5261         kfree(kvm);
5262 }
5263
5264 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5265                                 struct kvm_memory_slot *memslot,
5266                                 struct kvm_memory_slot old,
5267                                 struct kvm_userspace_memory_region *mem,
5268                                 int user_alloc)
5269 {
5270         int npages = memslot->npages;
5271
5272         /*To keep backward compatibility with older userspace,
5273          *x86 needs to hanlde !user_alloc case.
5274          */
5275         if (!user_alloc) {
5276                 if (npages && !old.rmap) {
5277                         unsigned long userspace_addr;
5278
5279                         down_write(&current->mm->mmap_sem);
5280                         userspace_addr = do_mmap(NULL, 0,
5281                                                  npages * PAGE_SIZE,
5282                                                  PROT_READ | PROT_WRITE,
5283                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5284                                                  0);
5285                         up_write(&current->mm->mmap_sem);
5286
5287                         if (IS_ERR((void *)userspace_addr))
5288                                 return PTR_ERR((void *)userspace_addr);
5289
5290                         memslot->userspace_addr = userspace_addr;
5291                 }
5292         }
5293
5294
5295         return 0;
5296 }
5297
5298 void kvm_arch_commit_memory_region(struct kvm *kvm,
5299                                 struct kvm_userspace_memory_region *mem,
5300                                 struct kvm_memory_slot old,
5301                                 int user_alloc)
5302 {
5303
5304         int npages = mem->memory_size >> PAGE_SHIFT;
5305
5306         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5307                 int ret;
5308
5309                 down_write(&current->mm->mmap_sem);
5310                 ret = do_munmap(current->mm, old.userspace_addr,
5311                                 old.npages * PAGE_SIZE);
5312                 up_write(&current->mm->mmap_sem);
5313                 if (ret < 0)
5314                         printk(KERN_WARNING
5315                                "kvm_vm_ioctl_set_memory_region: "
5316                                "failed to munmap memory\n");
5317         }
5318
5319         spin_lock(&kvm->mmu_lock);
5320         if (!kvm->arch.n_requested_mmu_pages) {
5321                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5322                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5323         }
5324
5325         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5326         spin_unlock(&kvm->mmu_lock);
5327 }
5328
5329 void kvm_arch_flush_shadow(struct kvm *kvm)
5330 {
5331         kvm_mmu_zap_all(kvm);
5332         kvm_reload_remote_mmus(kvm);
5333 }
5334
5335 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5336 {
5337         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5338                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5339                 || vcpu->arch.nmi_pending ||
5340                 (kvm_arch_interrupt_allowed(vcpu) &&
5341                  kvm_cpu_has_interrupt(vcpu));
5342 }
5343
5344 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5345 {
5346         int me;
5347         int cpu = vcpu->cpu;
5348
5349         if (waitqueue_active(&vcpu->wq)) {
5350                 wake_up_interruptible(&vcpu->wq);
5351                 ++vcpu->stat.halt_wakeup;
5352         }
5353
5354         me = get_cpu();
5355         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5356                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5357                         smp_send_reschedule(cpu);
5358         put_cpu();
5359 }
5360
5361 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5362 {
5363         return kvm_x86_ops->interrupt_allowed(vcpu);
5364 }
5365
5366 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5367 {
5368         unsigned long rflags;
5369
5370         rflags = kvm_x86_ops->get_rflags(vcpu);
5371         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5372                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5373         return rflags;
5374 }
5375 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5376
5377 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5378 {
5379         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5380             vcpu->arch.singlestep_cs ==
5381                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5382             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5383                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5384         kvm_x86_ops->set_rflags(vcpu, rflags);
5385 }
5386 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5387
5388 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5389 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5390 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5391 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5392 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5393 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);