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
1650         /* cpuid 1.edx */
1651         const u32 kvm_supported_word0_x86_features =
1652                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1653                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1654                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1655                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1656                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1657                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1658                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1659                 0 /* HTT, TM, Reserved, PBE */;
1660         /* cpuid 0x80000001.edx */
1661         const u32 kvm_supported_word1_x86_features =
1662                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1663                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1664                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1665                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1666                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1667                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1668                 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1669                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1670         /* cpuid 1.ecx */
1671         const u32 kvm_supported_word4_x86_features =
1672                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1673                 0 /* DS-CPL, VMX, SMX, EST */ |
1674                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1675                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1676                 0 /* Reserved, DCA */ | F(XMM4_1) |
1677                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1678                 0 /* Reserved, XSAVE, OSXSAVE */;
1679         /* cpuid 0x80000001.ecx */
1680         const u32 kvm_supported_word6_x86_features =
1681                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1682                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1683                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1684                 0 /* SKINIT */ | 0 /* WDT */;
1685
1686         /* all calls to cpuid_count() should be made on the same cpu */
1687         get_cpu();
1688         do_cpuid_1_ent(entry, function, index);
1689         ++*nent;
1690
1691         switch (function) {
1692         case 0:
1693                 entry->eax = min(entry->eax, (u32)0xb);
1694                 break;
1695         case 1:
1696                 entry->edx &= kvm_supported_word0_x86_features;
1697                 entry->ecx &= kvm_supported_word4_x86_features;
1698                 /* we support x2apic emulation even if host does not support
1699                  * it since we emulate x2apic in software */
1700                 entry->ecx |= F(X2APIC);
1701                 break;
1702         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1703          * may return different values. This forces us to get_cpu() before
1704          * issuing the first command, and also to emulate this annoying behavior
1705          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1706         case 2: {
1707                 int t, times = entry->eax & 0xff;
1708
1709                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1710                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1711                 for (t = 1; t < times && *nent < maxnent; ++t) {
1712                         do_cpuid_1_ent(&entry[t], function, 0);
1713                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1714                         ++*nent;
1715                 }
1716                 break;
1717         }
1718         /* function 4 and 0xb have additional index. */
1719         case 4: {
1720                 int i, cache_type;
1721
1722                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1723                 /* read more entries until cache_type is zero */
1724                 for (i = 1; *nent < maxnent; ++i) {
1725                         cache_type = entry[i - 1].eax & 0x1f;
1726                         if (!cache_type)
1727                                 break;
1728                         do_cpuid_1_ent(&entry[i], function, i);
1729                         entry[i].flags |=
1730                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1731                         ++*nent;
1732                 }
1733                 break;
1734         }
1735         case 0xb: {
1736                 int i, level_type;
1737
1738                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1739                 /* read more entries until level_type is zero */
1740                 for (i = 1; *nent < maxnent; ++i) {
1741                         level_type = entry[i - 1].ecx & 0xff00;
1742                         if (!level_type)
1743                                 break;
1744                         do_cpuid_1_ent(&entry[i], function, i);
1745                         entry[i].flags |=
1746                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1747                         ++*nent;
1748                 }
1749                 break;
1750         }
1751         case 0x80000000:
1752                 entry->eax = min(entry->eax, 0x8000001a);
1753                 break;
1754         case 0x80000001:
1755                 entry->edx &= kvm_supported_word1_x86_features;
1756                 entry->ecx &= kvm_supported_word6_x86_features;
1757                 break;
1758         }
1759         put_cpu();
1760 }
1761
1762 #undef F
1763
1764 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1765                                      struct kvm_cpuid_entry2 __user *entries)
1766 {
1767         struct kvm_cpuid_entry2 *cpuid_entries;
1768         int limit, nent = 0, r = -E2BIG;
1769         u32 func;
1770
1771         if (cpuid->nent < 1)
1772                 goto out;
1773         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1774                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1775         r = -ENOMEM;
1776         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1777         if (!cpuid_entries)
1778                 goto out;
1779
1780         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1781         limit = cpuid_entries[0].eax;
1782         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1783                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1784                              &nent, cpuid->nent);
1785         r = -E2BIG;
1786         if (nent >= cpuid->nent)
1787                 goto out_free;
1788
1789         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1790         limit = cpuid_entries[nent - 1].eax;
1791         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1792                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1793                              &nent, cpuid->nent);
1794         r = -E2BIG;
1795         if (nent >= cpuid->nent)
1796                 goto out_free;
1797
1798         r = -EFAULT;
1799         if (copy_to_user(entries, cpuid_entries,
1800                          nent * sizeof(struct kvm_cpuid_entry2)))
1801                 goto out_free;
1802         cpuid->nent = nent;
1803         r = 0;
1804
1805 out_free:
1806         vfree(cpuid_entries);
1807 out:
1808         return r;
1809 }
1810
1811 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1812                                     struct kvm_lapic_state *s)
1813 {
1814         vcpu_load(vcpu);
1815         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1816         vcpu_put(vcpu);
1817
1818         return 0;
1819 }
1820
1821 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1822                                     struct kvm_lapic_state *s)
1823 {
1824         vcpu_load(vcpu);
1825         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1826         kvm_apic_post_state_restore(vcpu);
1827         update_cr8_intercept(vcpu);
1828         vcpu_put(vcpu);
1829
1830         return 0;
1831 }
1832
1833 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1834                                     struct kvm_interrupt *irq)
1835 {
1836         if (irq->irq < 0 || irq->irq >= 256)
1837                 return -EINVAL;
1838         if (irqchip_in_kernel(vcpu->kvm))
1839                 return -ENXIO;
1840         vcpu_load(vcpu);
1841
1842         kvm_queue_interrupt(vcpu, irq->irq, false);
1843
1844         vcpu_put(vcpu);
1845
1846         return 0;
1847 }
1848
1849 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1850 {
1851         vcpu_load(vcpu);
1852         kvm_inject_nmi(vcpu);
1853         vcpu_put(vcpu);
1854
1855         return 0;
1856 }
1857
1858 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1859                                            struct kvm_tpr_access_ctl *tac)
1860 {
1861         if (tac->flags)
1862                 return -EINVAL;
1863         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1864         return 0;
1865 }
1866
1867 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1868                                         u64 mcg_cap)
1869 {
1870         int r;
1871         unsigned bank_num = mcg_cap & 0xff, bank;
1872
1873         r = -EINVAL;
1874         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1875                 goto out;
1876         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1877                 goto out;
1878         r = 0;
1879         vcpu->arch.mcg_cap = mcg_cap;
1880         /* Init IA32_MCG_CTL to all 1s */
1881         if (mcg_cap & MCG_CTL_P)
1882                 vcpu->arch.mcg_ctl = ~(u64)0;
1883         /* Init IA32_MCi_CTL to all 1s */
1884         for (bank = 0; bank < bank_num; bank++)
1885                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1886 out:
1887         return r;
1888 }
1889
1890 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1891                                       struct kvm_x86_mce *mce)
1892 {
1893         u64 mcg_cap = vcpu->arch.mcg_cap;
1894         unsigned bank_num = mcg_cap & 0xff;
1895         u64 *banks = vcpu->arch.mce_banks;
1896
1897         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1898                 return -EINVAL;
1899         /*
1900          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1901          * reporting is disabled
1902          */
1903         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1904             vcpu->arch.mcg_ctl != ~(u64)0)
1905                 return 0;
1906         banks += 4 * mce->bank;
1907         /*
1908          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1909          * reporting is disabled for the bank
1910          */
1911         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1912                 return 0;
1913         if (mce->status & MCI_STATUS_UC) {
1914                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1915                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
1916                         printk(KERN_DEBUG "kvm: set_mce: "
1917                                "injects mce exception while "
1918                                "previous one is in progress!\n");
1919                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1920                         return 0;
1921                 }
1922                 if (banks[1] & MCI_STATUS_VAL)
1923                         mce->status |= MCI_STATUS_OVER;
1924                 banks[2] = mce->addr;
1925                 banks[3] = mce->misc;
1926                 vcpu->arch.mcg_status = mce->mcg_status;
1927                 banks[1] = mce->status;
1928                 kvm_queue_exception(vcpu, MC_VECTOR);
1929         } else if (!(banks[1] & MCI_STATUS_VAL)
1930                    || !(banks[1] & MCI_STATUS_UC)) {
1931                 if (banks[1] & MCI_STATUS_VAL)
1932                         mce->status |= MCI_STATUS_OVER;
1933                 banks[2] = mce->addr;
1934                 banks[3] = mce->misc;
1935                 banks[1] = mce->status;
1936         } else
1937                 banks[1] |= MCI_STATUS_OVER;
1938         return 0;
1939 }
1940
1941 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
1942                                                struct kvm_vcpu_events *events)
1943 {
1944         vcpu_load(vcpu);
1945
1946         events->exception.injected = vcpu->arch.exception.pending;
1947         events->exception.nr = vcpu->arch.exception.nr;
1948         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
1949         events->exception.error_code = vcpu->arch.exception.error_code;
1950
1951         events->interrupt.injected = vcpu->arch.interrupt.pending;
1952         events->interrupt.nr = vcpu->arch.interrupt.nr;
1953         events->interrupt.soft = vcpu->arch.interrupt.soft;
1954
1955         events->nmi.injected = vcpu->arch.nmi_injected;
1956         events->nmi.pending = vcpu->arch.nmi_pending;
1957         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
1958
1959         events->sipi_vector = vcpu->arch.sipi_vector;
1960
1961         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
1962                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
1963
1964         vcpu_put(vcpu);
1965 }
1966
1967 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
1968                                               struct kvm_vcpu_events *events)
1969 {
1970         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
1971                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
1972                 return -EINVAL;
1973
1974         vcpu_load(vcpu);
1975
1976         vcpu->arch.exception.pending = events->exception.injected;
1977         vcpu->arch.exception.nr = events->exception.nr;
1978         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
1979         vcpu->arch.exception.error_code = events->exception.error_code;
1980
1981         vcpu->arch.interrupt.pending = events->interrupt.injected;
1982         vcpu->arch.interrupt.nr = events->interrupt.nr;
1983         vcpu->arch.interrupt.soft = events->interrupt.soft;
1984         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
1985                 kvm_pic_clear_isr_ack(vcpu->kvm);
1986
1987         vcpu->arch.nmi_injected = events->nmi.injected;
1988         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
1989                 vcpu->arch.nmi_pending = events->nmi.pending;
1990         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
1991
1992         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
1993                 vcpu->arch.sipi_vector = events->sipi_vector;
1994
1995         vcpu_put(vcpu);
1996
1997         return 0;
1998 }
1999
2000 long kvm_arch_vcpu_ioctl(struct file *filp,
2001                          unsigned int ioctl, unsigned long arg)
2002 {
2003         struct kvm_vcpu *vcpu = filp->private_data;
2004         void __user *argp = (void __user *)arg;
2005         int r;
2006         struct kvm_lapic_state *lapic = NULL;
2007
2008         switch (ioctl) {
2009         case KVM_GET_LAPIC: {
2010                 r = -EINVAL;
2011                 if (!vcpu->arch.apic)
2012                         goto out;
2013                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2014
2015                 r = -ENOMEM;
2016                 if (!lapic)
2017                         goto out;
2018                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2019                 if (r)
2020                         goto out;
2021                 r = -EFAULT;
2022                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2023                         goto out;
2024                 r = 0;
2025                 break;
2026         }
2027         case KVM_SET_LAPIC: {
2028                 r = -EINVAL;
2029                 if (!vcpu->arch.apic)
2030                         goto out;
2031                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2032                 r = -ENOMEM;
2033                 if (!lapic)
2034                         goto out;
2035                 r = -EFAULT;
2036                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2037                         goto out;
2038                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2039                 if (r)
2040                         goto out;
2041                 r = 0;
2042                 break;
2043         }
2044         case KVM_INTERRUPT: {
2045                 struct kvm_interrupt irq;
2046
2047                 r = -EFAULT;
2048                 if (copy_from_user(&irq, argp, sizeof irq))
2049                         goto out;
2050                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2051                 if (r)
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056         case KVM_NMI: {
2057                 r = kvm_vcpu_ioctl_nmi(vcpu);
2058                 if (r)
2059                         goto out;
2060                 r = 0;
2061                 break;
2062         }
2063         case KVM_SET_CPUID: {
2064                 struct kvm_cpuid __user *cpuid_arg = argp;
2065                 struct kvm_cpuid cpuid;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2069                         goto out;
2070                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2071                 if (r)
2072                         goto out;
2073                 break;
2074         }
2075         case KVM_SET_CPUID2: {
2076                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2077                 struct kvm_cpuid2 cpuid;
2078
2079                 r = -EFAULT;
2080                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2081                         goto out;
2082                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2083                                               cpuid_arg->entries);
2084                 if (r)
2085                         goto out;
2086                 break;
2087         }
2088         case KVM_GET_CPUID2: {
2089                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2090                 struct kvm_cpuid2 cpuid;
2091
2092                 r = -EFAULT;
2093                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2094                         goto out;
2095                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2096                                               cpuid_arg->entries);
2097                 if (r)
2098                         goto out;
2099                 r = -EFAULT;
2100                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2101                         goto out;
2102                 r = 0;
2103                 break;
2104         }
2105         case KVM_GET_MSRS:
2106                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2107                 break;
2108         case KVM_SET_MSRS:
2109                 r = msr_io(vcpu, argp, do_set_msr, 0);
2110                 break;
2111         case KVM_TPR_ACCESS_REPORTING: {
2112                 struct kvm_tpr_access_ctl tac;
2113
2114                 r = -EFAULT;
2115                 if (copy_from_user(&tac, argp, sizeof tac))
2116                         goto out;
2117                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2118                 if (r)
2119                         goto out;
2120                 r = -EFAULT;
2121                 if (copy_to_user(argp, &tac, sizeof tac))
2122                         goto out;
2123                 r = 0;
2124                 break;
2125         };
2126         case KVM_SET_VAPIC_ADDR: {
2127                 struct kvm_vapic_addr va;
2128
2129                 r = -EINVAL;
2130                 if (!irqchip_in_kernel(vcpu->kvm))
2131                         goto out;
2132                 r = -EFAULT;
2133                 if (copy_from_user(&va, argp, sizeof va))
2134                         goto out;
2135                 r = 0;
2136                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2137                 break;
2138         }
2139         case KVM_X86_SETUP_MCE: {
2140                 u64 mcg_cap;
2141
2142                 r = -EFAULT;
2143                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2144                         goto out;
2145                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2146                 break;
2147         }
2148         case KVM_X86_SET_MCE: {
2149                 struct kvm_x86_mce mce;
2150
2151                 r = -EFAULT;
2152                 if (copy_from_user(&mce, argp, sizeof mce))
2153                         goto out;
2154                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2155                 break;
2156         }
2157         case KVM_GET_VCPU_EVENTS: {
2158                 struct kvm_vcpu_events events;
2159
2160                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2161
2162                 r = -EFAULT;
2163                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2164                         break;
2165                 r = 0;
2166                 break;
2167         }
2168         case KVM_SET_VCPU_EVENTS: {
2169                 struct kvm_vcpu_events events;
2170
2171                 r = -EFAULT;
2172                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2173                         break;
2174
2175                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2176                 break;
2177         }
2178         default:
2179                 r = -EINVAL;
2180         }
2181 out:
2182         kfree(lapic);
2183         return r;
2184 }
2185
2186 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2187 {
2188         int ret;
2189
2190         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2191                 return -1;
2192         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2193         return ret;
2194 }
2195
2196 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2197                                               u64 ident_addr)
2198 {
2199         kvm->arch.ept_identity_map_addr = ident_addr;
2200         return 0;
2201 }
2202
2203 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2204                                           u32 kvm_nr_mmu_pages)
2205 {
2206         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2207                 return -EINVAL;
2208
2209         down_write(&kvm->slots_lock);
2210         spin_lock(&kvm->mmu_lock);
2211
2212         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2213         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2214
2215         spin_unlock(&kvm->mmu_lock);
2216         up_write(&kvm->slots_lock);
2217         return 0;
2218 }
2219
2220 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2221 {
2222         return kvm->arch.n_alloc_mmu_pages;
2223 }
2224
2225 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2226 {
2227         int i;
2228         struct kvm_mem_alias *alias;
2229
2230         for (i = 0; i < kvm->arch.naliases; ++i) {
2231                 alias = &kvm->arch.aliases[i];
2232                 if (gfn >= alias->base_gfn
2233                     && gfn < alias->base_gfn + alias->npages)
2234                         return alias->target_gfn + gfn - alias->base_gfn;
2235         }
2236         return gfn;
2237 }
2238
2239 /*
2240  * Set a new alias region.  Aliases map a portion of physical memory into
2241  * another portion.  This is useful for memory windows, for example the PC
2242  * VGA region.
2243  */
2244 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2245                                          struct kvm_memory_alias *alias)
2246 {
2247         int r, n;
2248         struct kvm_mem_alias *p;
2249
2250         r = -EINVAL;
2251         /* General sanity checks */
2252         if (alias->memory_size & (PAGE_SIZE - 1))
2253                 goto out;
2254         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2255                 goto out;
2256         if (alias->slot >= KVM_ALIAS_SLOTS)
2257                 goto out;
2258         if (alias->guest_phys_addr + alias->memory_size
2259             < alias->guest_phys_addr)
2260                 goto out;
2261         if (alias->target_phys_addr + alias->memory_size
2262             < alias->target_phys_addr)
2263                 goto out;
2264
2265         down_write(&kvm->slots_lock);
2266         spin_lock(&kvm->mmu_lock);
2267
2268         p = &kvm->arch.aliases[alias->slot];
2269         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2270         p->npages = alias->memory_size >> PAGE_SHIFT;
2271         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2272
2273         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2274                 if (kvm->arch.aliases[n - 1].npages)
2275                         break;
2276         kvm->arch.naliases = n;
2277
2278         spin_unlock(&kvm->mmu_lock);
2279         kvm_mmu_zap_all(kvm);
2280
2281         up_write(&kvm->slots_lock);
2282
2283         return 0;
2284
2285 out:
2286         return r;
2287 }
2288
2289 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2290 {
2291         int r;
2292
2293         r = 0;
2294         switch (chip->chip_id) {
2295         case KVM_IRQCHIP_PIC_MASTER:
2296                 memcpy(&chip->chip.pic,
2297                         &pic_irqchip(kvm)->pics[0],
2298                         sizeof(struct kvm_pic_state));
2299                 break;
2300         case KVM_IRQCHIP_PIC_SLAVE:
2301                 memcpy(&chip->chip.pic,
2302                         &pic_irqchip(kvm)->pics[1],
2303                         sizeof(struct kvm_pic_state));
2304                 break;
2305         case KVM_IRQCHIP_IOAPIC:
2306                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2307                 break;
2308         default:
2309                 r = -EINVAL;
2310                 break;
2311         }
2312         return r;
2313 }
2314
2315 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2316 {
2317         int r;
2318
2319         r = 0;
2320         switch (chip->chip_id) {
2321         case KVM_IRQCHIP_PIC_MASTER:
2322                 spin_lock(&pic_irqchip(kvm)->lock);
2323                 memcpy(&pic_irqchip(kvm)->pics[0],
2324                         &chip->chip.pic,
2325                         sizeof(struct kvm_pic_state));
2326                 spin_unlock(&pic_irqchip(kvm)->lock);
2327                 break;
2328         case KVM_IRQCHIP_PIC_SLAVE:
2329                 spin_lock(&pic_irqchip(kvm)->lock);
2330                 memcpy(&pic_irqchip(kvm)->pics[1],
2331                         &chip->chip.pic,
2332                         sizeof(struct kvm_pic_state));
2333                 spin_unlock(&pic_irqchip(kvm)->lock);
2334                 break;
2335         case KVM_IRQCHIP_IOAPIC:
2336                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2337                 break;
2338         default:
2339                 r = -EINVAL;
2340                 break;
2341         }
2342         kvm_pic_update_irq(pic_irqchip(kvm));
2343         return r;
2344 }
2345
2346 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2347 {
2348         int r = 0;
2349
2350         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2351         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2352         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2353         return r;
2354 }
2355
2356 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2357 {
2358         int r = 0;
2359
2360         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2361         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2362         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2363         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2364         return r;
2365 }
2366
2367 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2368 {
2369         int r = 0;
2370
2371         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2372         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2373                 sizeof(ps->channels));
2374         ps->flags = kvm->arch.vpit->pit_state.flags;
2375         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2376         return r;
2377 }
2378
2379 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2380 {
2381         int r = 0, start = 0;
2382         u32 prev_legacy, cur_legacy;
2383         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2384         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2385         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2386         if (!prev_legacy && cur_legacy)
2387                 start = 1;
2388         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2389                sizeof(kvm->arch.vpit->pit_state.channels));
2390         kvm->arch.vpit->pit_state.flags = ps->flags;
2391         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2392         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2393         return r;
2394 }
2395
2396 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2397                                  struct kvm_reinject_control *control)
2398 {
2399         if (!kvm->arch.vpit)
2400                 return -ENXIO;
2401         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2402         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2403         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2404         return 0;
2405 }
2406
2407 /*
2408  * Get (and clear) the dirty memory log for a memory slot.
2409  */
2410 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2411                                       struct kvm_dirty_log *log)
2412 {
2413         int r;
2414         int n;
2415         struct kvm_memory_slot *memslot;
2416         int is_dirty = 0;
2417
2418         down_write(&kvm->slots_lock);
2419
2420         r = kvm_get_dirty_log(kvm, log, &is_dirty);
2421         if (r)
2422                 goto out;
2423
2424         /* If nothing is dirty, don't bother messing with page tables. */
2425         if (is_dirty) {
2426                 spin_lock(&kvm->mmu_lock);
2427                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2428                 spin_unlock(&kvm->mmu_lock);
2429                 memslot = &kvm->memslots[log->slot];
2430                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2431                 memset(memslot->dirty_bitmap, 0, n);
2432         }
2433         r = 0;
2434 out:
2435         up_write(&kvm->slots_lock);
2436         return r;
2437 }
2438
2439 long kvm_arch_vm_ioctl(struct file *filp,
2440                        unsigned int ioctl, unsigned long arg)
2441 {
2442         struct kvm *kvm = filp->private_data;
2443         void __user *argp = (void __user *)arg;
2444         int r = -ENOTTY;
2445         /*
2446          * This union makes it completely explicit to gcc-3.x
2447          * that these two variables' stack usage should be
2448          * combined, not added together.
2449          */
2450         union {
2451                 struct kvm_pit_state ps;
2452                 struct kvm_pit_state2 ps2;
2453                 struct kvm_memory_alias alias;
2454                 struct kvm_pit_config pit_config;
2455         } u;
2456
2457         switch (ioctl) {
2458         case KVM_SET_TSS_ADDR:
2459                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2460                 if (r < 0)
2461                         goto out;
2462                 break;
2463         case KVM_SET_IDENTITY_MAP_ADDR: {
2464                 u64 ident_addr;
2465
2466                 r = -EFAULT;
2467                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2468                         goto out;
2469                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2470                 if (r < 0)
2471                         goto out;
2472                 break;
2473         }
2474         case KVM_SET_MEMORY_REGION: {
2475                 struct kvm_memory_region kvm_mem;
2476                 struct kvm_userspace_memory_region kvm_userspace_mem;
2477
2478                 r = -EFAULT;
2479                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2480                         goto out;
2481                 kvm_userspace_mem.slot = kvm_mem.slot;
2482                 kvm_userspace_mem.flags = kvm_mem.flags;
2483                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2484                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2485                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2486                 if (r)
2487                         goto out;
2488                 break;
2489         }
2490         case KVM_SET_NR_MMU_PAGES:
2491                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2492                 if (r)
2493                         goto out;
2494                 break;
2495         case KVM_GET_NR_MMU_PAGES:
2496                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2497                 break;
2498         case KVM_SET_MEMORY_ALIAS:
2499                 r = -EFAULT;
2500                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2501                         goto out;
2502                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2503                 if (r)
2504                         goto out;
2505                 break;
2506         case KVM_CREATE_IRQCHIP: {
2507                 struct kvm_pic *vpic;
2508
2509                 mutex_lock(&kvm->lock);
2510                 r = -EEXIST;
2511                 if (kvm->arch.vpic)
2512                         goto create_irqchip_unlock;
2513                 r = -ENOMEM;
2514                 vpic = kvm_create_pic(kvm);
2515                 if (vpic) {
2516                         r = kvm_ioapic_init(kvm);
2517                         if (r) {
2518                                 kfree(vpic);
2519                                 goto create_irqchip_unlock;
2520                         }
2521                 } else
2522                         goto create_irqchip_unlock;
2523                 smp_wmb();
2524                 kvm->arch.vpic = vpic;
2525                 smp_wmb();
2526                 r = kvm_setup_default_irq_routing(kvm);
2527                 if (r) {
2528                         mutex_lock(&kvm->irq_lock);
2529                         kfree(kvm->arch.vpic);
2530                         kfree(kvm->arch.vioapic);
2531                         kvm->arch.vpic = NULL;
2532                         kvm->arch.vioapic = NULL;
2533                         mutex_unlock(&kvm->irq_lock);
2534                 }
2535         create_irqchip_unlock:
2536                 mutex_unlock(&kvm->lock);
2537                 break;
2538         }
2539         case KVM_CREATE_PIT:
2540                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2541                 goto create_pit;
2542         case KVM_CREATE_PIT2:
2543                 r = -EFAULT;
2544                 if (copy_from_user(&u.pit_config, argp,
2545                                    sizeof(struct kvm_pit_config)))
2546                         goto out;
2547         create_pit:
2548                 down_write(&kvm->slots_lock);
2549                 r = -EEXIST;
2550                 if (kvm->arch.vpit)
2551                         goto create_pit_unlock;
2552                 r = -ENOMEM;
2553                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2554                 if (kvm->arch.vpit)
2555                         r = 0;
2556         create_pit_unlock:
2557                 up_write(&kvm->slots_lock);
2558                 break;
2559         case KVM_IRQ_LINE_STATUS:
2560         case KVM_IRQ_LINE: {
2561                 struct kvm_irq_level irq_event;
2562
2563                 r = -EFAULT;
2564                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2565                         goto out;
2566                 if (irqchip_in_kernel(kvm)) {
2567                         __s32 status;
2568                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2569                                         irq_event.irq, irq_event.level);
2570                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2571                                 irq_event.status = status;
2572                                 if (copy_to_user(argp, &irq_event,
2573                                                         sizeof irq_event))
2574                                         goto out;
2575                         }
2576                         r = 0;
2577                 }
2578                 break;
2579         }
2580         case KVM_GET_IRQCHIP: {
2581                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2582                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2583
2584                 r = -ENOMEM;
2585                 if (!chip)
2586                         goto out;
2587                 r = -EFAULT;
2588                 if (copy_from_user(chip, argp, sizeof *chip))
2589                         goto get_irqchip_out;
2590                 r = -ENXIO;
2591                 if (!irqchip_in_kernel(kvm))
2592                         goto get_irqchip_out;
2593                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2594                 if (r)
2595                         goto get_irqchip_out;
2596                 r = -EFAULT;
2597                 if (copy_to_user(argp, chip, sizeof *chip))
2598                         goto get_irqchip_out;
2599                 r = 0;
2600         get_irqchip_out:
2601                 kfree(chip);
2602                 if (r)
2603                         goto out;
2604                 break;
2605         }
2606         case KVM_SET_IRQCHIP: {
2607                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2608                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2609
2610                 r = -ENOMEM;
2611                 if (!chip)
2612                         goto out;
2613                 r = -EFAULT;
2614                 if (copy_from_user(chip, argp, sizeof *chip))
2615                         goto set_irqchip_out;
2616                 r = -ENXIO;
2617                 if (!irqchip_in_kernel(kvm))
2618                         goto set_irqchip_out;
2619                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2620                 if (r)
2621                         goto set_irqchip_out;
2622                 r = 0;
2623         set_irqchip_out:
2624                 kfree(chip);
2625                 if (r)
2626                         goto out;
2627                 break;
2628         }
2629         case KVM_GET_PIT: {
2630                 r = -EFAULT;
2631                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2632                         goto out;
2633                 r = -ENXIO;
2634                 if (!kvm->arch.vpit)
2635                         goto out;
2636                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2637                 if (r)
2638                         goto out;
2639                 r = -EFAULT;
2640                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2641                         goto out;
2642                 r = 0;
2643                 break;
2644         }
2645         case KVM_SET_PIT: {
2646                 r = -EFAULT;
2647                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2648                         goto out;
2649                 r = -ENXIO;
2650                 if (!kvm->arch.vpit)
2651                         goto out;
2652                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2653                 if (r)
2654                         goto out;
2655                 r = 0;
2656                 break;
2657         }
2658         case KVM_GET_PIT2: {
2659                 r = -ENXIO;
2660                 if (!kvm->arch.vpit)
2661                         goto out;
2662                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2663                 if (r)
2664                         goto out;
2665                 r = -EFAULT;
2666                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2667                         goto out;
2668                 r = 0;
2669                 break;
2670         }
2671         case KVM_SET_PIT2: {
2672                 r = -EFAULT;
2673                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2674                         goto out;
2675                 r = -ENXIO;
2676                 if (!kvm->arch.vpit)
2677                         goto out;
2678                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2679                 if (r)
2680                         goto out;
2681                 r = 0;
2682                 break;
2683         }
2684         case KVM_REINJECT_CONTROL: {
2685                 struct kvm_reinject_control control;
2686                 r =  -EFAULT;
2687                 if (copy_from_user(&control, argp, sizeof(control)))
2688                         goto out;
2689                 r = kvm_vm_ioctl_reinject(kvm, &control);
2690                 if (r)
2691                         goto out;
2692                 r = 0;
2693                 break;
2694         }
2695         case KVM_XEN_HVM_CONFIG: {
2696                 r = -EFAULT;
2697                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2698                                    sizeof(struct kvm_xen_hvm_config)))
2699                         goto out;
2700                 r = -EINVAL;
2701                 if (kvm->arch.xen_hvm_config.flags)
2702                         goto out;
2703                 r = 0;
2704                 break;
2705         }
2706         case KVM_SET_CLOCK: {
2707                 struct timespec now;
2708                 struct kvm_clock_data user_ns;
2709                 u64 now_ns;
2710                 s64 delta;
2711
2712                 r = -EFAULT;
2713                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2714                         goto out;
2715
2716                 r = -EINVAL;
2717                 if (user_ns.flags)
2718                         goto out;
2719
2720                 r = 0;
2721                 ktime_get_ts(&now);
2722                 now_ns = timespec_to_ns(&now);
2723                 delta = user_ns.clock - now_ns;
2724                 kvm->arch.kvmclock_offset = delta;
2725                 break;
2726         }
2727         case KVM_GET_CLOCK: {
2728                 struct timespec now;
2729                 struct kvm_clock_data user_ns;
2730                 u64 now_ns;
2731
2732                 ktime_get_ts(&now);
2733                 now_ns = timespec_to_ns(&now);
2734                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2735                 user_ns.flags = 0;
2736
2737                 r = -EFAULT;
2738                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2739                         goto out;
2740                 r = 0;
2741                 break;
2742         }
2743
2744         default:
2745                 ;
2746         }
2747 out:
2748         return r;
2749 }
2750
2751 static void kvm_init_msr_list(void)
2752 {
2753         u32 dummy[2];
2754         unsigned i, j;
2755
2756         /* skip the first msrs in the list. KVM-specific */
2757         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2758                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2759                         continue;
2760                 if (j < i)
2761                         msrs_to_save[j] = msrs_to_save[i];
2762                 j++;
2763         }
2764         num_msrs_to_save = j;
2765 }
2766
2767 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2768                            const void *v)
2769 {
2770         if (vcpu->arch.apic &&
2771             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2772                 return 0;
2773
2774         return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2775 }
2776
2777 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2778 {
2779         if (vcpu->arch.apic &&
2780             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2781                 return 0;
2782
2783         return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2784 }
2785
2786 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2787                                struct kvm_vcpu *vcpu)
2788 {
2789         void *data = val;
2790         int r = X86EMUL_CONTINUE;
2791
2792         while (bytes) {
2793                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2794                 unsigned offset = addr & (PAGE_SIZE-1);
2795                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2796                 int ret;
2797
2798                 if (gpa == UNMAPPED_GVA) {
2799                         r = X86EMUL_PROPAGATE_FAULT;
2800                         goto out;
2801                 }
2802                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2803                 if (ret < 0) {
2804                         r = X86EMUL_UNHANDLEABLE;
2805                         goto out;
2806                 }
2807
2808                 bytes -= toread;
2809                 data += toread;
2810                 addr += toread;
2811         }
2812 out:
2813         return r;
2814 }
2815
2816 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2817                                 struct kvm_vcpu *vcpu)
2818 {
2819         void *data = val;
2820         int r = X86EMUL_CONTINUE;
2821
2822         while (bytes) {
2823                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2824                 unsigned offset = addr & (PAGE_SIZE-1);
2825                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2826                 int ret;
2827
2828                 if (gpa == UNMAPPED_GVA) {
2829                         r = X86EMUL_PROPAGATE_FAULT;
2830                         goto out;
2831                 }
2832                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2833                 if (ret < 0) {
2834                         r = X86EMUL_UNHANDLEABLE;
2835                         goto out;
2836                 }
2837
2838                 bytes -= towrite;
2839                 data += towrite;
2840                 addr += towrite;
2841         }
2842 out:
2843         return r;
2844 }
2845
2846
2847 static int emulator_read_emulated(unsigned long addr,
2848                                   void *val,
2849                                   unsigned int bytes,
2850                                   struct kvm_vcpu *vcpu)
2851 {
2852         gpa_t                 gpa;
2853
2854         if (vcpu->mmio_read_completed) {
2855                 memcpy(val, vcpu->mmio_data, bytes);
2856                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2857                                vcpu->mmio_phys_addr, *(u64 *)val);
2858                 vcpu->mmio_read_completed = 0;
2859                 return X86EMUL_CONTINUE;
2860         }
2861
2862         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2863
2864         /* For APIC access vmexit */
2865         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2866                 goto mmio;
2867
2868         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2869                                 == X86EMUL_CONTINUE)
2870                 return X86EMUL_CONTINUE;
2871         if (gpa == UNMAPPED_GVA)
2872                 return X86EMUL_PROPAGATE_FAULT;
2873
2874 mmio:
2875         /*
2876          * Is this MMIO handled locally?
2877          */
2878         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2879                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2880                 return X86EMUL_CONTINUE;
2881         }
2882
2883         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2884
2885         vcpu->mmio_needed = 1;
2886         vcpu->mmio_phys_addr = gpa;
2887         vcpu->mmio_size = bytes;
2888         vcpu->mmio_is_write = 0;
2889
2890         return X86EMUL_UNHANDLEABLE;
2891 }
2892
2893 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2894                           const void *val, int bytes)
2895 {
2896         int ret;
2897
2898         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2899         if (ret < 0)
2900                 return 0;
2901         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2902         return 1;
2903 }
2904
2905 static int emulator_write_emulated_onepage(unsigned long addr,
2906                                            const void *val,
2907                                            unsigned int bytes,
2908                                            struct kvm_vcpu *vcpu)
2909 {
2910         gpa_t                 gpa;
2911
2912         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2913
2914         if (gpa == UNMAPPED_GVA) {
2915                 kvm_inject_page_fault(vcpu, addr, 2);
2916                 return X86EMUL_PROPAGATE_FAULT;
2917         }
2918
2919         /* For APIC access vmexit */
2920         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2921                 goto mmio;
2922
2923         if (emulator_write_phys(vcpu, gpa, val, bytes))
2924                 return X86EMUL_CONTINUE;
2925
2926 mmio:
2927         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2928         /*
2929          * Is this MMIO handled locally?
2930          */
2931         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2932                 return X86EMUL_CONTINUE;
2933
2934         vcpu->mmio_needed = 1;
2935         vcpu->mmio_phys_addr = gpa;
2936         vcpu->mmio_size = bytes;
2937         vcpu->mmio_is_write = 1;
2938         memcpy(vcpu->mmio_data, val, bytes);
2939
2940         return X86EMUL_CONTINUE;
2941 }
2942
2943 int emulator_write_emulated(unsigned long addr,
2944                                    const void *val,
2945                                    unsigned int bytes,
2946                                    struct kvm_vcpu *vcpu)
2947 {
2948         /* Crossing a page boundary? */
2949         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2950                 int rc, now;
2951
2952                 now = -addr & ~PAGE_MASK;
2953                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2954                 if (rc != X86EMUL_CONTINUE)
2955                         return rc;
2956                 addr += now;
2957                 val += now;
2958                 bytes -= now;
2959         }
2960         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2961 }
2962 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2963
2964 static int emulator_cmpxchg_emulated(unsigned long addr,
2965                                      const void *old,
2966                                      const void *new,
2967                                      unsigned int bytes,
2968                                      struct kvm_vcpu *vcpu)
2969 {
2970         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2971 #ifndef CONFIG_X86_64
2972         /* guests cmpxchg8b have to be emulated atomically */
2973         if (bytes == 8) {
2974                 gpa_t gpa;
2975                 struct page *page;
2976                 char *kaddr;
2977                 u64 val;
2978
2979                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2980
2981                 if (gpa == UNMAPPED_GVA ||
2982                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2983                         goto emul_write;
2984
2985                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2986                         goto emul_write;
2987
2988                 val = *(u64 *)new;
2989
2990                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2991
2992                 kaddr = kmap_atomic(page, KM_USER0);
2993                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2994                 kunmap_atomic(kaddr, KM_USER0);
2995                 kvm_release_page_dirty(page);
2996         }
2997 emul_write:
2998 #endif
2999
3000         return emulator_write_emulated(addr, new, bytes, vcpu);
3001 }
3002
3003 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3004 {
3005         return kvm_x86_ops->get_segment_base(vcpu, seg);
3006 }
3007
3008 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3009 {
3010         kvm_mmu_invlpg(vcpu, address);
3011         return X86EMUL_CONTINUE;
3012 }
3013
3014 int emulate_clts(struct kvm_vcpu *vcpu)
3015 {
3016         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
3017         return X86EMUL_CONTINUE;
3018 }
3019
3020 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3021 {
3022         struct kvm_vcpu *vcpu = ctxt->vcpu;
3023
3024         switch (dr) {
3025         case 0 ... 3:
3026                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3027                 return X86EMUL_CONTINUE;
3028         default:
3029                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3030                 return X86EMUL_UNHANDLEABLE;
3031         }
3032 }
3033
3034 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3035 {
3036         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3037         int exception;
3038
3039         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3040         if (exception) {
3041                 /* FIXME: better handling */
3042                 return X86EMUL_UNHANDLEABLE;
3043         }
3044         return X86EMUL_CONTINUE;
3045 }
3046
3047 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3048 {
3049         u8 opcodes[4];
3050         unsigned long rip = kvm_rip_read(vcpu);
3051         unsigned long rip_linear;
3052
3053         if (!printk_ratelimit())
3054                 return;
3055
3056         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3057
3058         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3059
3060         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3061                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3062 }
3063 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3064
3065 static struct x86_emulate_ops emulate_ops = {
3066         .read_std            = kvm_read_guest_virt,
3067         .read_emulated       = emulator_read_emulated,
3068         .write_emulated      = emulator_write_emulated,
3069         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3070 };
3071
3072 static void cache_all_regs(struct kvm_vcpu *vcpu)
3073 {
3074         kvm_register_read(vcpu, VCPU_REGS_RAX);
3075         kvm_register_read(vcpu, VCPU_REGS_RSP);
3076         kvm_register_read(vcpu, VCPU_REGS_RIP);
3077         vcpu->arch.regs_dirty = ~0;
3078 }
3079
3080 int emulate_instruction(struct kvm_vcpu *vcpu,
3081                         unsigned long cr2,
3082                         u16 error_code,
3083                         int emulation_type)
3084 {
3085         int r, shadow_mask;
3086         struct decode_cache *c;
3087         struct kvm_run *run = vcpu->run;
3088
3089         kvm_clear_exception_queue(vcpu);
3090         vcpu->arch.mmio_fault_cr2 = cr2;
3091         /*
3092          * TODO: fix emulate.c to use guest_read/write_register
3093          * instead of direct ->regs accesses, can save hundred cycles
3094          * on Intel for instructions that don't read/change RSP, for
3095          * for example.
3096          */
3097         cache_all_regs(vcpu);
3098
3099         vcpu->mmio_is_write = 0;
3100         vcpu->arch.pio.string = 0;
3101
3102         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3103                 int cs_db, cs_l;
3104                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3105
3106                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3107                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3108                 vcpu->arch.emulate_ctxt.mode =
3109                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3110                         ? X86EMUL_MODE_REAL : cs_l
3111                         ? X86EMUL_MODE_PROT64 : cs_db
3112                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3113
3114                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3115
3116                 /* Only allow emulation of specific instructions on #UD
3117                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3118                 c = &vcpu->arch.emulate_ctxt.decode;
3119                 if (emulation_type & EMULTYPE_TRAP_UD) {
3120                         if (!c->twobyte)
3121                                 return EMULATE_FAIL;
3122                         switch (c->b) {
3123                         case 0x01: /* VMMCALL */
3124                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3125                                         return EMULATE_FAIL;
3126                                 break;
3127                         case 0x34: /* sysenter */
3128                         case 0x35: /* sysexit */
3129                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3130                                         return EMULATE_FAIL;
3131                                 break;
3132                         case 0x05: /* syscall */
3133                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3134                                         return EMULATE_FAIL;
3135                                 break;
3136                         default:
3137                                 return EMULATE_FAIL;
3138                         }
3139
3140                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3141                                 return EMULATE_FAIL;
3142                 }
3143
3144                 ++vcpu->stat.insn_emulation;
3145                 if (r)  {
3146                         ++vcpu->stat.insn_emulation_fail;
3147                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3148                                 return EMULATE_DONE;
3149                         return EMULATE_FAIL;
3150                 }
3151         }
3152
3153         if (emulation_type & EMULTYPE_SKIP) {
3154                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3155                 return EMULATE_DONE;
3156         }
3157
3158         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3159         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3160
3161         if (r == 0)
3162                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3163
3164         if (vcpu->arch.pio.string)
3165                 return EMULATE_DO_MMIO;
3166
3167         if ((r || vcpu->mmio_is_write) && run) {
3168                 run->exit_reason = KVM_EXIT_MMIO;
3169                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3170                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3171                 run->mmio.len = vcpu->mmio_size;
3172                 run->mmio.is_write = vcpu->mmio_is_write;
3173         }
3174
3175         if (r) {
3176                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3177                         return EMULATE_DONE;
3178                 if (!vcpu->mmio_needed) {
3179                         kvm_report_emulation_failure(vcpu, "mmio");
3180                         return EMULATE_FAIL;
3181                 }
3182                 return EMULATE_DO_MMIO;
3183         }
3184
3185         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3186
3187         if (vcpu->mmio_is_write) {
3188                 vcpu->mmio_needed = 0;
3189                 return EMULATE_DO_MMIO;
3190         }
3191
3192         return EMULATE_DONE;
3193 }
3194 EXPORT_SYMBOL_GPL(emulate_instruction);
3195
3196 static int pio_copy_data(struct kvm_vcpu *vcpu)
3197 {
3198         void *p = vcpu->arch.pio_data;
3199         gva_t q = vcpu->arch.pio.guest_gva;
3200         unsigned bytes;
3201         int ret;
3202
3203         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3204         if (vcpu->arch.pio.in)
3205                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3206         else
3207                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3208         return ret;
3209 }
3210
3211 int complete_pio(struct kvm_vcpu *vcpu)
3212 {
3213         struct kvm_pio_request *io = &vcpu->arch.pio;
3214         long delta;
3215         int r;
3216         unsigned long val;
3217
3218         if (!io->string) {
3219                 if (io->in) {
3220                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3221                         memcpy(&val, vcpu->arch.pio_data, io->size);
3222                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3223                 }
3224         } else {
3225                 if (io->in) {
3226                         r = pio_copy_data(vcpu);
3227                         if (r)
3228                                 return r;
3229                 }
3230
3231                 delta = 1;
3232                 if (io->rep) {
3233                         delta *= io->cur_count;
3234                         /*
3235                          * The size of the register should really depend on
3236                          * current address size.
3237                          */
3238                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3239                         val -= delta;
3240                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3241                 }
3242                 if (io->down)
3243                         delta = -delta;
3244                 delta *= io->size;
3245                 if (io->in) {
3246                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3247                         val += delta;
3248                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3249                 } else {
3250                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3251                         val += delta;
3252                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3253                 }
3254         }
3255
3256         io->count -= io->cur_count;
3257         io->cur_count = 0;
3258
3259         return 0;
3260 }
3261
3262 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3263 {
3264         /* TODO: String I/O for in kernel device */
3265         int r;
3266
3267         if (vcpu->arch.pio.in)
3268                 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3269                                     vcpu->arch.pio.size, pd);
3270         else
3271                 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3272                                      vcpu->arch.pio.size, pd);
3273         return r;
3274 }
3275
3276 static int pio_string_write(struct kvm_vcpu *vcpu)
3277 {
3278         struct kvm_pio_request *io = &vcpu->arch.pio;
3279         void *pd = vcpu->arch.pio_data;
3280         int i, r = 0;
3281
3282         for (i = 0; i < io->cur_count; i++) {
3283                 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
3284                                      io->port, io->size, pd)) {
3285                         r = -EOPNOTSUPP;
3286                         break;
3287                 }
3288                 pd += io->size;
3289         }
3290         return r;
3291 }
3292
3293 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3294 {
3295         unsigned long val;
3296
3297         vcpu->run->exit_reason = KVM_EXIT_IO;
3298         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3299         vcpu->run->io.size = vcpu->arch.pio.size = size;
3300         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3301         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3302         vcpu->run->io.port = vcpu->arch.pio.port = port;
3303         vcpu->arch.pio.in = in;
3304         vcpu->arch.pio.string = 0;
3305         vcpu->arch.pio.down = 0;
3306         vcpu->arch.pio.rep = 0;
3307
3308         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3309                       size, 1);
3310
3311         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3312         memcpy(vcpu->arch.pio_data, &val, 4);
3313
3314         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3315                 complete_pio(vcpu);
3316                 return 1;
3317         }
3318         return 0;
3319 }
3320 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3321
3322 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3323                   int size, unsigned long count, int down,
3324                   gva_t address, int rep, unsigned port)
3325 {
3326         unsigned now, in_page;
3327         int ret = 0;
3328
3329         vcpu->run->exit_reason = KVM_EXIT_IO;
3330         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3331         vcpu->run->io.size = vcpu->arch.pio.size = size;
3332         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3333         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3334         vcpu->run->io.port = vcpu->arch.pio.port = port;
3335         vcpu->arch.pio.in = in;
3336         vcpu->arch.pio.string = 1;
3337         vcpu->arch.pio.down = down;
3338         vcpu->arch.pio.rep = rep;
3339
3340         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3341                       size, count);
3342
3343         if (!count) {
3344                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3345                 return 1;
3346         }
3347
3348         if (!down)
3349                 in_page = PAGE_SIZE - offset_in_page(address);
3350         else
3351                 in_page = offset_in_page(address) + size;
3352         now = min(count, (unsigned long)in_page / size);
3353         if (!now)
3354                 now = 1;
3355         if (down) {
3356                 /*
3357                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3358                  */
3359                 pr_unimpl(vcpu, "guest string pio down\n");
3360                 kvm_inject_gp(vcpu, 0);
3361                 return 1;
3362         }
3363         vcpu->run->io.count = now;
3364         vcpu->arch.pio.cur_count = now;
3365
3366         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3367                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3368
3369         vcpu->arch.pio.guest_gva = address;
3370
3371         if (!vcpu->arch.pio.in) {
3372                 /* string PIO write */
3373                 ret = pio_copy_data(vcpu);
3374                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3375                         kvm_inject_gp(vcpu, 0);
3376                         return 1;
3377                 }
3378                 if (ret == 0 && !pio_string_write(vcpu)) {
3379                         complete_pio(vcpu);
3380                         if (vcpu->arch.pio.count == 0)
3381                                 ret = 1;
3382                 }
3383         }
3384         /* no string PIO read support yet */
3385
3386         return ret;
3387 }
3388 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3389
3390 static void bounce_off(void *info)
3391 {
3392         /* nothing */
3393 }
3394
3395 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3396                                      void *data)
3397 {
3398         struct cpufreq_freqs *freq = data;
3399         struct kvm *kvm;
3400         struct kvm_vcpu *vcpu;
3401         int i, send_ipi = 0;
3402
3403         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3404                 return 0;
3405         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3406                 return 0;
3407         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3408
3409         spin_lock(&kvm_lock);
3410         list_for_each_entry(kvm, &vm_list, vm_list) {
3411                 kvm_for_each_vcpu(i, vcpu, kvm) {
3412                         if (vcpu->cpu != freq->cpu)
3413                                 continue;
3414                         if (!kvm_request_guest_time_update(vcpu))
3415                                 continue;
3416                         if (vcpu->cpu != smp_processor_id())
3417                                 send_ipi++;
3418                 }
3419         }
3420         spin_unlock(&kvm_lock);
3421
3422         if (freq->old < freq->new && send_ipi) {
3423                 /*
3424                  * We upscale the frequency.  Must make the guest
3425                  * doesn't see old kvmclock values while running with
3426                  * the new frequency, otherwise we risk the guest sees
3427                  * time go backwards.
3428                  *
3429                  * In case we update the frequency for another cpu
3430                  * (which might be in guest context) send an interrupt
3431                  * to kick the cpu out of guest context.  Next time
3432                  * guest context is entered kvmclock will be updated,
3433                  * so the guest will not see stale values.
3434                  */
3435                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3436         }
3437         return 0;
3438 }
3439
3440 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3441         .notifier_call  = kvmclock_cpufreq_notifier
3442 };
3443
3444 static void kvm_timer_init(void)
3445 {
3446         int cpu;
3447
3448         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3449                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3450                                           CPUFREQ_TRANSITION_NOTIFIER);
3451                 for_each_online_cpu(cpu) {
3452                         unsigned long khz = cpufreq_get(cpu);
3453                         if (!khz)
3454                                 khz = tsc_khz;
3455                         per_cpu(cpu_tsc_khz, cpu) = khz;
3456                 }
3457         } else {
3458                 for_each_possible_cpu(cpu)
3459                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3460         }
3461 }
3462
3463 int kvm_arch_init(void *opaque)
3464 {
3465         int r;
3466         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3467
3468         if (kvm_x86_ops) {
3469                 printk(KERN_ERR "kvm: already loaded the other module\n");
3470                 r = -EEXIST;
3471                 goto out;
3472         }
3473
3474         if (!ops->cpu_has_kvm_support()) {
3475                 printk(KERN_ERR "kvm: no hardware support\n");
3476                 r = -EOPNOTSUPP;
3477                 goto out;
3478         }
3479         if (ops->disabled_by_bios()) {
3480                 printk(KERN_ERR "kvm: disabled by bios\n");
3481                 r = -EOPNOTSUPP;
3482                 goto out;
3483         }
3484
3485         r = kvm_mmu_module_init();
3486         if (r)
3487                 goto out;
3488
3489         kvm_init_msr_list();
3490
3491         kvm_x86_ops = ops;
3492         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3493         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3494         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3495                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3496
3497         kvm_timer_init();
3498
3499         return 0;
3500
3501 out:
3502         return r;
3503 }
3504
3505 void kvm_arch_exit(void)
3506 {
3507         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3508                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3509                                             CPUFREQ_TRANSITION_NOTIFIER);
3510         kvm_x86_ops = NULL;
3511         kvm_mmu_module_exit();
3512 }
3513
3514 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3515 {
3516         ++vcpu->stat.halt_exits;
3517         if (irqchip_in_kernel(vcpu->kvm)) {
3518                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3519                 return 1;
3520         } else {
3521                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3522                 return 0;
3523         }
3524 }
3525 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3526
3527 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3528                            unsigned long a1)
3529 {
3530         if (is_long_mode(vcpu))
3531                 return a0;
3532         else
3533                 return a0 | ((gpa_t)a1 << 32);
3534 }
3535
3536 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3537 {
3538         unsigned long nr, a0, a1, a2, a3, ret;
3539         int r = 1;
3540
3541         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3542         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3543         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3544         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3545         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3546
3547         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3548
3549         if (!is_long_mode(vcpu)) {
3550                 nr &= 0xFFFFFFFF;
3551                 a0 &= 0xFFFFFFFF;
3552                 a1 &= 0xFFFFFFFF;
3553                 a2 &= 0xFFFFFFFF;
3554                 a3 &= 0xFFFFFFFF;
3555         }
3556
3557         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3558                 ret = -KVM_EPERM;
3559                 goto out;
3560         }
3561
3562         switch (nr) {
3563         case KVM_HC_VAPIC_POLL_IRQ:
3564                 ret = 0;
3565                 break;
3566         case KVM_HC_MMU_OP:
3567                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3568                 break;
3569         default:
3570                 ret = -KVM_ENOSYS;
3571                 break;
3572         }
3573 out:
3574         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3575         ++vcpu->stat.hypercalls;
3576         return r;
3577 }
3578 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3579
3580 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3581 {
3582         char instruction[3];
3583         int ret = 0;
3584         unsigned long rip = kvm_rip_read(vcpu);
3585
3586
3587         /*
3588          * Blow out the MMU to ensure that no other VCPU has an active mapping
3589          * to ensure that the updated hypercall appears atomically across all
3590          * VCPUs.
3591          */
3592         kvm_mmu_zap_all(vcpu->kvm);
3593
3594         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3595         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3596             != X86EMUL_CONTINUE)
3597                 ret = -EFAULT;
3598
3599         return ret;
3600 }
3601
3602 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3603 {
3604         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3605 }
3606
3607 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3608 {
3609         struct descriptor_table dt = { limit, base };
3610
3611         kvm_x86_ops->set_gdt(vcpu, &dt);
3612 }
3613
3614 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3615 {
3616         struct descriptor_table dt = { limit, base };
3617
3618         kvm_x86_ops->set_idt(vcpu, &dt);
3619 }
3620
3621 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3622                    unsigned long *rflags)
3623 {
3624         kvm_lmsw(vcpu, msw);
3625         *rflags = kvm_get_rflags(vcpu);
3626 }
3627
3628 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3629 {
3630         unsigned long value;
3631
3632         switch (cr) {
3633         case 0:
3634                 value = vcpu->arch.cr0;
3635                 break;
3636         case 2:
3637                 value = vcpu->arch.cr2;
3638                 break;
3639         case 3:
3640                 value = vcpu->arch.cr3;
3641                 break;
3642         case 4:
3643                 value = kvm_read_cr4(vcpu);
3644                 break;
3645         case 8:
3646                 value = kvm_get_cr8(vcpu);
3647                 break;
3648         default:
3649                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3650                 return 0;
3651         }
3652
3653         return value;
3654 }
3655
3656 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3657                      unsigned long *rflags)
3658 {
3659         switch (cr) {
3660         case 0:
3661                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3662                 *rflags = kvm_get_rflags(vcpu);
3663                 break;
3664         case 2:
3665                 vcpu->arch.cr2 = val;
3666                 break;
3667         case 3:
3668                 kvm_set_cr3(vcpu, val);
3669                 break;
3670         case 4:
3671                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3672                 break;
3673         case 8:
3674                 kvm_set_cr8(vcpu, val & 0xfUL);
3675                 break;
3676         default:
3677                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3678         }
3679 }
3680
3681 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3682 {
3683         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3684         int j, nent = vcpu->arch.cpuid_nent;
3685
3686         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3687         /* when no next entry is found, the current entry[i] is reselected */
3688         for (j = i + 1; ; j = (j + 1) % nent) {
3689                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3690                 if (ej->function == e->function) {
3691                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3692                         return j;
3693                 }
3694         }
3695         return 0; /* silence gcc, even though control never reaches here */
3696 }
3697
3698 /* find an entry with matching function, matching index (if needed), and that
3699  * should be read next (if it's stateful) */
3700 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3701         u32 function, u32 index)
3702 {
3703         if (e->function != function)
3704                 return 0;
3705         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3706                 return 0;
3707         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3708             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3709                 return 0;
3710         return 1;
3711 }
3712
3713 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3714                                               u32 function, u32 index)
3715 {
3716         int i;
3717         struct kvm_cpuid_entry2 *best = NULL;
3718
3719         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3720                 struct kvm_cpuid_entry2 *e;
3721
3722                 e = &vcpu->arch.cpuid_entries[i];
3723                 if (is_matching_cpuid_entry(e, function, index)) {
3724                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3725                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3726                         best = e;
3727                         break;
3728                 }
3729                 /*
3730                  * Both basic or both extended?
3731                  */
3732                 if (((e->function ^ function) & 0x80000000) == 0)
3733                         if (!best || e->function > best->function)
3734                                 best = e;
3735         }
3736         return best;
3737 }
3738 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
3739
3740 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3741 {
3742         struct kvm_cpuid_entry2 *best;
3743
3744         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3745         if (best)
3746                 return best->eax & 0xff;
3747         return 36;
3748 }
3749
3750 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3751 {
3752         u32 function, index;
3753         struct kvm_cpuid_entry2 *best;
3754
3755         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3756         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3757         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3758         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3759         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3760         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3761         best = kvm_find_cpuid_entry(vcpu, function, index);
3762         if (best) {
3763                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3764                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3765                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3766                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3767         }
3768         kvm_x86_ops->skip_emulated_instruction(vcpu);
3769         trace_kvm_cpuid(function,
3770                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3771                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3772                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3773                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3774 }
3775 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3776
3777 /*
3778  * Check if userspace requested an interrupt window, and that the
3779  * interrupt window is open.
3780  *
3781  * No need to exit to userspace if we already have an interrupt queued.
3782  */
3783 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3784 {
3785         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3786                 vcpu->run->request_interrupt_window &&
3787                 kvm_arch_interrupt_allowed(vcpu));
3788 }
3789
3790 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3791 {
3792         struct kvm_run *kvm_run = vcpu->run;
3793
3794         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3795         kvm_run->cr8 = kvm_get_cr8(vcpu);
3796         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3797         if (irqchip_in_kernel(vcpu->kvm))
3798                 kvm_run->ready_for_interrupt_injection = 1;
3799         else
3800                 kvm_run->ready_for_interrupt_injection =
3801                         kvm_arch_interrupt_allowed(vcpu) &&
3802                         !kvm_cpu_has_interrupt(vcpu) &&
3803                         !kvm_event_needs_reinjection(vcpu);
3804 }
3805
3806 static void vapic_enter(struct kvm_vcpu *vcpu)
3807 {
3808         struct kvm_lapic *apic = vcpu->arch.apic;
3809         struct page *page;
3810
3811         if (!apic || !apic->vapic_addr)
3812                 return;
3813
3814         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3815
3816         vcpu->arch.apic->vapic_page = page;
3817 }
3818
3819 static void vapic_exit(struct kvm_vcpu *vcpu)
3820 {
3821         struct kvm_lapic *apic = vcpu->arch.apic;
3822
3823         if (!apic || !apic->vapic_addr)
3824                 return;
3825
3826         down_read(&vcpu->kvm->slots_lock);
3827         kvm_release_page_dirty(apic->vapic_page);
3828         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3829         up_read(&vcpu->kvm->slots_lock);
3830 }
3831
3832 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3833 {
3834         int max_irr, tpr;
3835
3836         if (!kvm_x86_ops->update_cr8_intercept)
3837                 return;
3838
3839         if (!vcpu->arch.apic)
3840                 return;
3841
3842         if (!vcpu->arch.apic->vapic_addr)
3843                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3844         else
3845                 max_irr = -1;
3846
3847         if (max_irr != -1)
3848                 max_irr >>= 4;
3849
3850         tpr = kvm_lapic_get_cr8(vcpu);
3851
3852         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3853 }
3854
3855 static void inject_pending_event(struct kvm_vcpu *vcpu)
3856 {
3857         /* try to reinject previous events if any */
3858         if (vcpu->arch.exception.pending) {
3859                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3860                                           vcpu->arch.exception.has_error_code,
3861                                           vcpu->arch.exception.error_code);
3862                 return;
3863         }
3864
3865         if (vcpu->arch.nmi_injected) {
3866                 kvm_x86_ops->set_nmi(vcpu);
3867                 return;
3868         }
3869
3870         if (vcpu->arch.interrupt.pending) {
3871                 kvm_x86_ops->set_irq(vcpu);
3872                 return;
3873         }
3874
3875         /* try to inject new event if pending */
3876         if (vcpu->arch.nmi_pending) {
3877                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3878                         vcpu->arch.nmi_pending = false;
3879                         vcpu->arch.nmi_injected = true;
3880                         kvm_x86_ops->set_nmi(vcpu);
3881                 }
3882         } else if (kvm_cpu_has_interrupt(vcpu)) {
3883                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3884                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3885                                             false);
3886                         kvm_x86_ops->set_irq(vcpu);
3887                 }
3888         }
3889 }
3890
3891 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3892 {
3893         int r;
3894         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3895                 vcpu->run->request_interrupt_window;
3896
3897         if (vcpu->requests)
3898                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3899                         kvm_mmu_unload(vcpu);
3900
3901         r = kvm_mmu_reload(vcpu);
3902         if (unlikely(r))
3903                 goto out;
3904
3905         if (vcpu->requests) {
3906                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3907                         __kvm_migrate_timers(vcpu);
3908                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3909                         kvm_write_guest_time(vcpu);
3910                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3911                         kvm_mmu_sync_roots(vcpu);
3912                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3913                         kvm_x86_ops->tlb_flush(vcpu);
3914                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3915                                        &vcpu->requests)) {
3916                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3917                         r = 0;
3918                         goto out;
3919                 }
3920                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3921                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3922                         r = 0;
3923                         goto out;
3924                 }
3925         }
3926
3927         preempt_disable();
3928
3929         kvm_x86_ops->prepare_guest_switch(vcpu);
3930         kvm_load_guest_fpu(vcpu);
3931
3932         local_irq_disable();
3933
3934         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3935         smp_mb__after_clear_bit();
3936
3937         if (vcpu->requests || need_resched() || signal_pending(current)) {
3938                 set_bit(KVM_REQ_KICK, &vcpu->requests);
3939                 local_irq_enable();
3940                 preempt_enable();
3941                 r = 1;
3942                 goto out;
3943         }
3944
3945         inject_pending_event(vcpu);
3946
3947         /* enable NMI/IRQ window open exits if needed */
3948         if (vcpu->arch.nmi_pending)
3949                 kvm_x86_ops->enable_nmi_window(vcpu);
3950         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3951                 kvm_x86_ops->enable_irq_window(vcpu);
3952
3953         if (kvm_lapic_enabled(vcpu)) {
3954                 update_cr8_intercept(vcpu);
3955                 kvm_lapic_sync_to_vapic(vcpu);
3956         }
3957
3958         up_read(&vcpu->kvm->slots_lock);
3959
3960         kvm_guest_enter();
3961
3962         if (unlikely(vcpu->arch.switch_db_regs)) {
3963                 set_debugreg(0, 7);
3964                 set_debugreg(vcpu->arch.eff_db[0], 0);
3965                 set_debugreg(vcpu->arch.eff_db[1], 1);
3966                 set_debugreg(vcpu->arch.eff_db[2], 2);
3967                 set_debugreg(vcpu->arch.eff_db[3], 3);
3968         }
3969
3970         trace_kvm_entry(vcpu->vcpu_id);
3971         kvm_x86_ops->run(vcpu);
3972
3973         /*
3974          * If the guest has used debug registers, at least dr7
3975          * will be disabled while returning to the host.
3976          * If we don't have active breakpoints in the host, we don't
3977          * care about the messed up debug address registers. But if
3978          * we have some of them active, restore the old state.
3979          */
3980         if (hw_breakpoint_active())
3981                 hw_breakpoint_restore();
3982
3983         set_bit(KVM_REQ_KICK, &vcpu->requests);
3984         local_irq_enable();
3985
3986         ++vcpu->stat.exits;
3987
3988         /*
3989          * We must have an instruction between local_irq_enable() and
3990          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3991          * the interrupt shadow.  The stat.exits increment will do nicely.
3992          * But we need to prevent reordering, hence this barrier():
3993          */
3994         barrier();
3995
3996         kvm_guest_exit();
3997
3998         preempt_enable();
3999
4000         down_read(&vcpu->kvm->slots_lock);
4001
4002         /*
4003          * Profile KVM exit RIPs:
4004          */
4005         if (unlikely(prof_on == KVM_PROFILING)) {
4006                 unsigned long rip = kvm_rip_read(vcpu);
4007                 profile_hit(KVM_PROFILING, (void *)rip);
4008         }
4009
4010
4011         kvm_lapic_sync_from_vapic(vcpu);
4012
4013         r = kvm_x86_ops->handle_exit(vcpu);
4014 out:
4015         return r;
4016 }
4017
4018
4019 static int __vcpu_run(struct kvm_vcpu *vcpu)
4020 {
4021         int r;
4022
4023         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4024                 pr_debug("vcpu %d received sipi with vector # %x\n",
4025                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4026                 kvm_lapic_reset(vcpu);
4027                 r = kvm_arch_vcpu_reset(vcpu);
4028                 if (r)
4029                         return r;
4030                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4031         }
4032
4033         down_read(&vcpu->kvm->slots_lock);
4034         vapic_enter(vcpu);
4035
4036         r = 1;
4037         while (r > 0) {
4038                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4039                         r = vcpu_enter_guest(vcpu);
4040                 else {
4041                         up_read(&vcpu->kvm->slots_lock);
4042                         kvm_vcpu_block(vcpu);
4043                         down_read(&vcpu->kvm->slots_lock);
4044                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4045                         {
4046                                 switch(vcpu->arch.mp_state) {
4047                                 case KVM_MP_STATE_HALTED:
4048                                         vcpu->arch.mp_state =
4049                                                 KVM_MP_STATE_RUNNABLE;
4050                                 case KVM_MP_STATE_RUNNABLE:
4051                                         break;
4052                                 case KVM_MP_STATE_SIPI_RECEIVED:
4053                                 default:
4054                                         r = -EINTR;
4055                                         break;
4056                                 }
4057                         }
4058                 }
4059
4060                 if (r <= 0)
4061                         break;
4062
4063                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4064                 if (kvm_cpu_has_pending_timer(vcpu))
4065                         kvm_inject_pending_timer_irqs(vcpu);
4066
4067                 if (dm_request_for_irq_injection(vcpu)) {
4068                         r = -EINTR;
4069                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4070                         ++vcpu->stat.request_irq_exits;
4071                 }
4072                 if (signal_pending(current)) {
4073                         r = -EINTR;
4074                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4075                         ++vcpu->stat.signal_exits;
4076                 }
4077                 if (need_resched()) {
4078                         up_read(&vcpu->kvm->slots_lock);
4079                         kvm_resched(vcpu);
4080                         down_read(&vcpu->kvm->slots_lock);
4081                 }
4082         }
4083
4084         up_read(&vcpu->kvm->slots_lock);
4085         post_kvm_run_save(vcpu);
4086
4087         vapic_exit(vcpu);
4088
4089         return r;
4090 }
4091
4092 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4093 {
4094         int r;
4095         sigset_t sigsaved;
4096
4097         vcpu_load(vcpu);
4098
4099         if (vcpu->sigset_active)
4100                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4101
4102         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4103                 kvm_vcpu_block(vcpu);
4104                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4105                 r = -EAGAIN;
4106                 goto out;
4107         }
4108
4109         /* re-sync apic's tpr */
4110         if (!irqchip_in_kernel(vcpu->kvm))
4111                 kvm_set_cr8(vcpu, kvm_run->cr8);
4112
4113         if (vcpu->arch.pio.cur_count) {
4114                 r = complete_pio(vcpu);
4115                 if (r)
4116                         goto out;
4117         }
4118         if (vcpu->mmio_needed) {
4119                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4120                 vcpu->mmio_read_completed = 1;
4121                 vcpu->mmio_needed = 0;
4122
4123                 down_read(&vcpu->kvm->slots_lock);
4124                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4125                                         EMULTYPE_NO_DECODE);
4126                 up_read(&vcpu->kvm->slots_lock);
4127                 if (r == EMULATE_DO_MMIO) {
4128                         /*
4129                          * Read-modify-write.  Back to userspace.
4130                          */
4131                         r = 0;
4132                         goto out;
4133                 }
4134         }
4135         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4136                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4137                                      kvm_run->hypercall.ret);
4138
4139         r = __vcpu_run(vcpu);
4140
4141 out:
4142         if (vcpu->sigset_active)
4143                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4144
4145         vcpu_put(vcpu);
4146         return r;
4147 }
4148
4149 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4150 {
4151         vcpu_load(vcpu);
4152
4153         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4154         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4155         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4156         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4157         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4158         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4159         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4160         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4161 #ifdef CONFIG_X86_64
4162         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4163         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4164         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4165         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4166         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4167         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4168         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4169         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4170 #endif
4171
4172         regs->rip = kvm_rip_read(vcpu);
4173         regs->rflags = kvm_get_rflags(vcpu);
4174
4175         vcpu_put(vcpu);
4176
4177         return 0;
4178 }
4179
4180 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4181 {
4182         vcpu_load(vcpu);
4183
4184         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4185         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4186         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4187         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4188         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4189         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4190         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4191         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4192 #ifdef CONFIG_X86_64
4193         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4194         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4195         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4196         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4197         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4198         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4199         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4200         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4201 #endif
4202
4203         kvm_rip_write(vcpu, regs->rip);
4204         kvm_set_rflags(vcpu, regs->rflags);
4205
4206         vcpu->arch.exception.pending = false;
4207
4208         vcpu_put(vcpu);
4209
4210         return 0;
4211 }
4212
4213 void kvm_get_segment(struct kvm_vcpu *vcpu,
4214                      struct kvm_segment *var, int seg)
4215 {
4216         kvm_x86_ops->get_segment(vcpu, var, seg);
4217 }
4218
4219 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4220 {
4221         struct kvm_segment cs;
4222
4223         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4224         *db = cs.db;
4225         *l = cs.l;
4226 }
4227 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4228
4229 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4230                                   struct kvm_sregs *sregs)
4231 {
4232         struct descriptor_table dt;
4233
4234         vcpu_load(vcpu);
4235
4236         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4237         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4238         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4239         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4240         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4241         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4242
4243         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4244         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4245
4246         kvm_x86_ops->get_idt(vcpu, &dt);
4247         sregs->idt.limit = dt.limit;
4248         sregs->idt.base = dt.base;
4249         kvm_x86_ops->get_gdt(vcpu, &dt);
4250         sregs->gdt.limit = dt.limit;
4251         sregs->gdt.base = dt.base;
4252
4253         sregs->cr0 = vcpu->arch.cr0;
4254         sregs->cr2 = vcpu->arch.cr2;
4255         sregs->cr3 = vcpu->arch.cr3;
4256         sregs->cr4 = kvm_read_cr4(vcpu);
4257         sregs->cr8 = kvm_get_cr8(vcpu);
4258         sregs->efer = vcpu->arch.shadow_efer;
4259         sregs->apic_base = kvm_get_apic_base(vcpu);
4260
4261         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4262
4263         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4264                 set_bit(vcpu->arch.interrupt.nr,
4265                         (unsigned long *)sregs->interrupt_bitmap);
4266
4267         vcpu_put(vcpu);
4268
4269         return 0;
4270 }
4271
4272 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4273                                     struct kvm_mp_state *mp_state)
4274 {
4275         vcpu_load(vcpu);
4276         mp_state->mp_state = vcpu->arch.mp_state;
4277         vcpu_put(vcpu);
4278         return 0;
4279 }
4280
4281 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4282                                     struct kvm_mp_state *mp_state)
4283 {
4284         vcpu_load(vcpu);
4285         vcpu->arch.mp_state = mp_state->mp_state;
4286         vcpu_put(vcpu);
4287         return 0;
4288 }
4289
4290 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4291                         struct kvm_segment *var, int seg)
4292 {
4293         kvm_x86_ops->set_segment(vcpu, var, seg);
4294 }
4295
4296 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4297                                    struct kvm_segment *kvm_desct)
4298 {
4299         kvm_desct->base = get_desc_base(seg_desc);
4300         kvm_desct->limit = get_desc_limit(seg_desc);
4301         if (seg_desc->g) {
4302                 kvm_desct->limit <<= 12;
4303                 kvm_desct->limit |= 0xfff;
4304         }
4305         kvm_desct->selector = selector;
4306         kvm_desct->type = seg_desc->type;
4307         kvm_desct->present = seg_desc->p;
4308         kvm_desct->dpl = seg_desc->dpl;
4309         kvm_desct->db = seg_desc->d;
4310         kvm_desct->s = seg_desc->s;
4311         kvm_desct->l = seg_desc->l;
4312         kvm_desct->g = seg_desc->g;
4313         kvm_desct->avl = seg_desc->avl;
4314         if (!selector)
4315                 kvm_desct->unusable = 1;
4316         else
4317                 kvm_desct->unusable = 0;
4318         kvm_desct->padding = 0;
4319 }
4320
4321 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4322                                           u16 selector,
4323                                           struct descriptor_table *dtable)
4324 {
4325         if (selector & 1 << 2) {
4326                 struct kvm_segment kvm_seg;
4327
4328                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4329
4330                 if (kvm_seg.unusable)
4331                         dtable->limit = 0;
4332                 else
4333                         dtable->limit = kvm_seg.limit;
4334                 dtable->base = kvm_seg.base;
4335         }
4336         else
4337                 kvm_x86_ops->get_gdt(vcpu, dtable);
4338 }
4339
4340 /* allowed just for 8 bytes segments */
4341 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4342                                          struct desc_struct *seg_desc)
4343 {
4344         struct descriptor_table dtable;
4345         u16 index = selector >> 3;
4346
4347         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4348
4349         if (dtable.limit < index * 8 + 7) {
4350                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4351                 return 1;
4352         }
4353         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4354 }
4355
4356 /* allowed just for 8 bytes segments */
4357 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4358                                          struct desc_struct *seg_desc)
4359 {
4360         struct descriptor_table dtable;
4361         u16 index = selector >> 3;
4362
4363         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4364
4365         if (dtable.limit < index * 8 + 7)
4366                 return 1;
4367         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4368 }
4369
4370 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4371                              struct desc_struct *seg_desc)
4372 {
4373         u32 base_addr = get_desc_base(seg_desc);
4374
4375         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4376 }
4377
4378 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4379 {
4380         struct kvm_segment kvm_seg;
4381
4382         kvm_get_segment(vcpu, &kvm_seg, seg);
4383         return kvm_seg.selector;
4384 }
4385
4386 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4387                                                 u16 selector,
4388                                                 struct kvm_segment *kvm_seg)
4389 {
4390         struct desc_struct seg_desc;
4391
4392         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4393                 return 1;
4394         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4395         return 0;
4396 }
4397
4398 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4399 {
4400         struct kvm_segment segvar = {
4401                 .base = selector << 4,
4402                 .limit = 0xffff,
4403                 .selector = selector,
4404                 .type = 3,
4405                 .present = 1,
4406                 .dpl = 3,
4407                 .db = 0,
4408                 .s = 1,
4409                 .l = 0,
4410                 .g = 0,
4411                 .avl = 0,
4412                 .unusable = 0,
4413         };
4414         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4415         return 0;
4416 }
4417
4418 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4419 {
4420         return (seg != VCPU_SREG_LDTR) &&
4421                 (seg != VCPU_SREG_TR) &&
4422                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4423 }
4424
4425 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4426                                          u16 selector)
4427 {
4428         /* NULL selector is not valid for CS and SS */
4429         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4430                 if (!selector)
4431                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4432 }
4433
4434 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4435                                 int type_bits, int seg)
4436 {
4437         struct kvm_segment kvm_seg;
4438
4439         if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4440                 return kvm_load_realmode_segment(vcpu, selector, seg);
4441         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4442                 return 1;
4443
4444         kvm_check_segment_descriptor(vcpu, seg, selector);
4445         kvm_seg.type |= type_bits;
4446
4447         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4448             seg != VCPU_SREG_LDTR)
4449                 if (!kvm_seg.s)
4450                         kvm_seg.unusable = 1;
4451
4452         kvm_set_segment(vcpu, &kvm_seg, seg);
4453         return 0;
4454 }
4455
4456 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4457                                 struct tss_segment_32 *tss)
4458 {
4459         tss->cr3 = vcpu->arch.cr3;
4460         tss->eip = kvm_rip_read(vcpu);
4461         tss->eflags = kvm_get_rflags(vcpu);
4462         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4463         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4464         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4465         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4466         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4467         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4468         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4469         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4470         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4471         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4472         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4473         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4474         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4475         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4476         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4477 }
4478
4479 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4480                                   struct tss_segment_32 *tss)
4481 {
4482         kvm_set_cr3(vcpu, tss->cr3);
4483
4484         kvm_rip_write(vcpu, tss->eip);
4485         kvm_set_rflags(vcpu, tss->eflags | 2);
4486
4487         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4488         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4489         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4490         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4491         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4492         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4493         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4494         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4495
4496         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4497                 return 1;
4498
4499         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4500                 return 1;
4501
4502         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4503                 return 1;
4504
4505         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4506                 return 1;
4507
4508         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4509                 return 1;
4510
4511         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4512                 return 1;
4513
4514         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4515                 return 1;
4516         return 0;
4517 }
4518
4519 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4520                                 struct tss_segment_16 *tss)
4521 {
4522         tss->ip = kvm_rip_read(vcpu);
4523         tss->flag = kvm_get_rflags(vcpu);
4524         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4525         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4526         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4527         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4528         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4529         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4530         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4531         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4532
4533         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4534         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4535         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4536         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4537         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4538 }
4539
4540 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4541                                  struct tss_segment_16 *tss)
4542 {
4543         kvm_rip_write(vcpu, tss->ip);
4544         kvm_set_rflags(vcpu, tss->flag | 2);
4545         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4546         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4547         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4548         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4549         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4550         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4551         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4552         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4553
4554         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4555                 return 1;
4556
4557         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4558                 return 1;
4559
4560         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4561                 return 1;
4562
4563         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4564                 return 1;
4565
4566         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4567                 return 1;
4568         return 0;
4569 }
4570
4571 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4572                               u16 old_tss_sel, u32 old_tss_base,
4573                               struct desc_struct *nseg_desc)
4574 {
4575         struct tss_segment_16 tss_segment_16;
4576         int ret = 0;
4577
4578         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4579                            sizeof tss_segment_16))
4580                 goto out;
4581
4582         save_state_to_tss16(vcpu, &tss_segment_16);
4583
4584         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4585                             sizeof tss_segment_16))
4586                 goto out;
4587
4588         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4589                            &tss_segment_16, sizeof tss_segment_16))
4590                 goto out;
4591
4592         if (old_tss_sel != 0xffff) {
4593                 tss_segment_16.prev_task_link = old_tss_sel;
4594
4595                 if (kvm_write_guest(vcpu->kvm,
4596                                     get_tss_base_addr(vcpu, nseg_desc),
4597                                     &tss_segment_16.prev_task_link,
4598                                     sizeof tss_segment_16.prev_task_link))
4599                         goto out;
4600         }
4601
4602         if (load_state_from_tss16(vcpu, &tss_segment_16))
4603                 goto out;
4604
4605         ret = 1;
4606 out:
4607         return ret;
4608 }
4609
4610 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4611                        u16 old_tss_sel, u32 old_tss_base,
4612                        struct desc_struct *nseg_desc)
4613 {
4614         struct tss_segment_32 tss_segment_32;
4615         int ret = 0;
4616
4617         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4618                            sizeof tss_segment_32))
4619                 goto out;
4620
4621         save_state_to_tss32(vcpu, &tss_segment_32);
4622
4623         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4624                             sizeof tss_segment_32))
4625                 goto out;
4626
4627         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4628                            &tss_segment_32, sizeof tss_segment_32))
4629                 goto out;
4630
4631         if (old_tss_sel != 0xffff) {
4632                 tss_segment_32.prev_task_link = old_tss_sel;
4633
4634                 if (kvm_write_guest(vcpu->kvm,
4635                                     get_tss_base_addr(vcpu, nseg_desc),
4636                                     &tss_segment_32.prev_task_link,
4637                                     sizeof tss_segment_32.prev_task_link))
4638                         goto out;
4639         }
4640
4641         if (load_state_from_tss32(vcpu, &tss_segment_32))
4642                 goto out;
4643
4644         ret = 1;
4645 out:
4646         return ret;
4647 }
4648
4649 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4650 {
4651         struct kvm_segment tr_seg;
4652         struct desc_struct cseg_desc;
4653         struct desc_struct nseg_desc;
4654         int ret = 0;
4655         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4656         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4657
4658         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4659
4660         /* FIXME: Handle errors. Failure to read either TSS or their
4661          * descriptors should generate a pagefault.
4662          */
4663         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4664                 goto out;
4665
4666         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4667                 goto out;
4668
4669         if (reason != TASK_SWITCH_IRET) {
4670                 int cpl;
4671
4672                 cpl = kvm_x86_ops->get_cpl(vcpu);
4673                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4674                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4675                         return 1;
4676                 }
4677         }
4678
4679         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4680                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4681                 return 1;
4682         }
4683
4684         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4685                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4686                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4687         }
4688
4689         if (reason == TASK_SWITCH_IRET) {
4690                 u32 eflags = kvm_get_rflags(vcpu);
4691                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4692         }
4693
4694         /* set back link to prev task only if NT bit is set in eflags
4695            note that old_tss_sel is not used afetr this point */
4696         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4697                 old_tss_sel = 0xffff;
4698
4699         if (nseg_desc.type & 8)
4700                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4701                                          old_tss_base, &nseg_desc);
4702         else
4703                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4704                                          old_tss_base, &nseg_desc);
4705
4706         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4707                 u32 eflags = kvm_get_rflags(vcpu);
4708                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4709         }
4710
4711         if (reason != TASK_SWITCH_IRET) {
4712                 nseg_desc.type |= (1 << 1);
4713                 save_guest_segment_descriptor(vcpu, tss_selector,
4714                                               &nseg_desc);
4715         }
4716
4717         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4718         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4719         tr_seg.type = 11;
4720         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4721 out:
4722         return ret;
4723 }
4724 EXPORT_SYMBOL_GPL(kvm_task_switch);
4725
4726 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4727                                   struct kvm_sregs *sregs)
4728 {
4729         int mmu_reset_needed = 0;
4730         int pending_vec, max_bits;
4731         struct descriptor_table dt;
4732
4733         vcpu_load(vcpu);
4734
4735         dt.limit = sregs->idt.limit;
4736         dt.base = sregs->idt.base;
4737         kvm_x86_ops->set_idt(vcpu, &dt);
4738         dt.limit = sregs->gdt.limit;
4739         dt.base = sregs->gdt.base;
4740         kvm_x86_ops->set_gdt(vcpu, &dt);
4741
4742         vcpu->arch.cr2 = sregs->cr2;
4743         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4744         vcpu->arch.cr3 = sregs->cr3;
4745
4746         kvm_set_cr8(vcpu, sregs->cr8);
4747
4748         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4749         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4750         kvm_set_apic_base(vcpu, sregs->apic_base);
4751
4752         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4753         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4754         vcpu->arch.cr0 = sregs->cr0;
4755
4756         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4757         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4758         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4759                 load_pdptrs(vcpu, vcpu->arch.cr3);
4760                 mmu_reset_needed = 1;
4761         }
4762
4763         if (mmu_reset_needed)
4764                 kvm_mmu_reset_context(vcpu);
4765
4766         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4767         pending_vec = find_first_bit(
4768                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4769         if (pending_vec < max_bits) {
4770                 kvm_queue_interrupt(vcpu, pending_vec, false);
4771                 pr_debug("Set back pending irq %d\n", pending_vec);
4772                 if (irqchip_in_kernel(vcpu->kvm))
4773                         kvm_pic_clear_isr_ack(vcpu->kvm);
4774         }
4775
4776         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4777         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4778         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4779         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4780         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4781         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4782
4783         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4784         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4785
4786         update_cr8_intercept(vcpu);
4787
4788         /* Older userspace won't unhalt the vcpu on reset. */
4789         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4790             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4791             !(vcpu->arch.cr0 & X86_CR0_PE))
4792                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4793
4794         vcpu_put(vcpu);
4795
4796         return 0;
4797 }
4798
4799 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4800                                         struct kvm_guest_debug *dbg)
4801 {
4802         unsigned long rflags;
4803         int i, r;
4804
4805         vcpu_load(vcpu);
4806
4807         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4808                 r = -EBUSY;
4809                 if (vcpu->arch.exception.pending)
4810                         goto unlock_out;
4811                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4812                         kvm_queue_exception(vcpu, DB_VECTOR);
4813                 else
4814                         kvm_queue_exception(vcpu, BP_VECTOR);
4815         }
4816
4817         /*
4818          * Read rflags as long as potentially injected trace flags are still
4819          * filtered out.
4820          */
4821         rflags = kvm_get_rflags(vcpu);
4822
4823         vcpu->guest_debug = dbg->control;
4824         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4825                 vcpu->guest_debug = 0;
4826
4827         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4828                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4829                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4830                 vcpu->arch.switch_db_regs =
4831                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4832         } else {
4833                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4834                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4835                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4836         }
4837
4838         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4839                 vcpu->arch.singlestep_cs =
4840                         get_segment_selector(vcpu, VCPU_SREG_CS);
4841                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4842         }
4843
4844         /*
4845          * Trigger an rflags update that will inject or remove the trace
4846          * flags.
4847          */
4848         kvm_set_rflags(vcpu, rflags);
4849
4850         kvm_x86_ops->set_guest_debug(vcpu, dbg);
4851
4852         r = 0;
4853
4854 unlock_out:
4855         vcpu_put(vcpu);
4856
4857         return r;
4858 }
4859
4860 /*
4861  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4862  * we have asm/x86/processor.h
4863  */
4864 struct fxsave {
4865         u16     cwd;
4866         u16     swd;
4867         u16     twd;
4868         u16     fop;
4869         u64     rip;
4870         u64     rdp;
4871         u32     mxcsr;
4872         u32     mxcsr_mask;
4873         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4874 #ifdef CONFIG_X86_64
4875         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4876 #else
4877         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4878 #endif
4879 };
4880
4881 /*
4882  * Translate a guest virtual address to a guest physical address.
4883  */
4884 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4885                                     struct kvm_translation *tr)
4886 {
4887         unsigned long vaddr = tr->linear_address;
4888         gpa_t gpa;
4889
4890         vcpu_load(vcpu);
4891         down_read(&vcpu->kvm->slots_lock);
4892         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4893         up_read(&vcpu->kvm->slots_lock);
4894         tr->physical_address = gpa;
4895         tr->valid = gpa != UNMAPPED_GVA;
4896         tr->writeable = 1;
4897         tr->usermode = 0;
4898         vcpu_put(vcpu);
4899
4900         return 0;
4901 }
4902
4903 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4904 {
4905         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4906
4907         vcpu_load(vcpu);
4908
4909         memcpy(fpu->fpr, fxsave->st_space, 128);
4910         fpu->fcw = fxsave->cwd;
4911         fpu->fsw = fxsave->swd;
4912         fpu->ftwx = fxsave->twd;
4913         fpu->last_opcode = fxsave->fop;
4914         fpu->last_ip = fxsave->rip;
4915         fpu->last_dp = fxsave->rdp;
4916         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4917
4918         vcpu_put(vcpu);
4919
4920         return 0;
4921 }
4922
4923 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4924 {
4925         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4926
4927         vcpu_load(vcpu);
4928
4929         memcpy(fxsave->st_space, fpu->fpr, 128);
4930         fxsave->cwd = fpu->fcw;
4931         fxsave->swd = fpu->fsw;
4932         fxsave->twd = fpu->ftwx;
4933         fxsave->fop = fpu->last_opcode;
4934         fxsave->rip = fpu->last_ip;
4935         fxsave->rdp = fpu->last_dp;
4936         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4937
4938         vcpu_put(vcpu);
4939
4940         return 0;
4941 }
4942
4943 void fx_init(struct kvm_vcpu *vcpu)
4944 {
4945         unsigned after_mxcsr_mask;
4946
4947         /*
4948          * Touch the fpu the first time in non atomic context as if
4949          * this is the first fpu instruction the exception handler
4950          * will fire before the instruction returns and it'll have to
4951          * allocate ram with GFP_KERNEL.
4952          */
4953         if (!used_math())
4954                 kvm_fx_save(&vcpu->arch.host_fx_image);
4955
4956         /* Initialize guest FPU by resetting ours and saving into guest's */
4957         preempt_disable();
4958         kvm_fx_save(&vcpu->arch.host_fx_image);
4959         kvm_fx_finit();
4960         kvm_fx_save(&vcpu->arch.guest_fx_image);
4961         kvm_fx_restore(&vcpu->arch.host_fx_image);
4962         preempt_enable();
4963
4964         vcpu->arch.cr0 |= X86_CR0_ET;
4965         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4966         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4967         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4968                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4969 }
4970 EXPORT_SYMBOL_GPL(fx_init);
4971
4972 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4973 {
4974         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4975                 return;
4976
4977         vcpu->guest_fpu_loaded = 1;
4978         kvm_fx_save(&vcpu->arch.host_fx_image);
4979         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4980 }
4981 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4982
4983 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4984 {
4985         if (!vcpu->guest_fpu_loaded)
4986                 return;
4987
4988         vcpu->guest_fpu_loaded = 0;
4989         kvm_fx_save(&vcpu->arch.guest_fx_image);
4990         kvm_fx_restore(&vcpu->arch.host_fx_image);
4991         ++vcpu->stat.fpu_reload;
4992 }
4993 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4994
4995 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4996 {
4997         if (vcpu->arch.time_page) {
4998                 kvm_release_page_dirty(vcpu->arch.time_page);
4999                 vcpu->arch.time_page = NULL;
5000         }
5001
5002         kvm_x86_ops->vcpu_free(vcpu);
5003 }
5004
5005 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5006                                                 unsigned int id)
5007 {
5008         return kvm_x86_ops->vcpu_create(kvm, id);
5009 }
5010
5011 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5012 {
5013         int r;
5014
5015         /* We do fxsave: this must be aligned. */
5016         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5017
5018         vcpu->arch.mtrr_state.have_fixed = 1;
5019         vcpu_load(vcpu);
5020         r = kvm_arch_vcpu_reset(vcpu);
5021         if (r == 0)
5022                 r = kvm_mmu_setup(vcpu);
5023         vcpu_put(vcpu);
5024         if (r < 0)
5025                 goto free_vcpu;
5026
5027         return 0;
5028 free_vcpu:
5029         kvm_x86_ops->vcpu_free(vcpu);
5030         return r;
5031 }
5032
5033 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5034 {
5035         vcpu_load(vcpu);
5036         kvm_mmu_unload(vcpu);
5037         vcpu_put(vcpu);
5038
5039         kvm_x86_ops->vcpu_free(vcpu);
5040 }
5041
5042 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5043 {
5044         vcpu->arch.nmi_pending = false;
5045         vcpu->arch.nmi_injected = false;
5046
5047         vcpu->arch.switch_db_regs = 0;
5048         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5049         vcpu->arch.dr6 = DR6_FIXED_1;
5050         vcpu->arch.dr7 = DR7_FIXED_1;
5051
5052         return kvm_x86_ops->vcpu_reset(vcpu);
5053 }
5054
5055 int kvm_arch_hardware_enable(void *garbage)
5056 {
5057         /*
5058          * Since this may be called from a hotplug notifcation,
5059          * we can't get the CPU frequency directly.
5060          */
5061         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5062                 int cpu = raw_smp_processor_id();
5063                 per_cpu(cpu_tsc_khz, cpu) = 0;
5064         }
5065
5066         kvm_shared_msr_cpu_online();
5067
5068         return kvm_x86_ops->hardware_enable(garbage);
5069 }
5070
5071 void kvm_arch_hardware_disable(void *garbage)
5072 {
5073         kvm_x86_ops->hardware_disable(garbage);
5074         drop_user_return_notifiers(garbage);
5075 }
5076
5077 int kvm_arch_hardware_setup(void)
5078 {
5079         return kvm_x86_ops->hardware_setup();
5080 }
5081
5082 void kvm_arch_hardware_unsetup(void)
5083 {
5084         kvm_x86_ops->hardware_unsetup();
5085 }
5086
5087 void kvm_arch_check_processor_compat(void *rtn)
5088 {
5089         kvm_x86_ops->check_processor_compatibility(rtn);
5090 }
5091
5092 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5093 {
5094         struct page *page;
5095         struct kvm *kvm;
5096         int r;
5097
5098         BUG_ON(vcpu->kvm == NULL);
5099         kvm = vcpu->kvm;
5100
5101         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5102         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5103                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5104         else
5105                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5106
5107         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5108         if (!page) {
5109                 r = -ENOMEM;
5110                 goto fail;
5111         }
5112         vcpu->arch.pio_data = page_address(page);
5113
5114         r = kvm_mmu_create(vcpu);
5115         if (r < 0)
5116                 goto fail_free_pio_data;
5117
5118         if (irqchip_in_kernel(kvm)) {
5119                 r = kvm_create_lapic(vcpu);
5120                 if (r < 0)
5121                         goto fail_mmu_destroy;
5122         }
5123
5124         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5125                                        GFP_KERNEL);
5126         if (!vcpu->arch.mce_banks) {
5127                 r = -ENOMEM;
5128                 goto fail_free_lapic;
5129         }
5130         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5131
5132         return 0;
5133 fail_free_lapic:
5134         kvm_free_lapic(vcpu);
5135 fail_mmu_destroy:
5136         kvm_mmu_destroy(vcpu);
5137 fail_free_pio_data:
5138         free_page((unsigned long)vcpu->arch.pio_data);
5139 fail:
5140         return r;
5141 }
5142
5143 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5144 {
5145         kfree(vcpu->arch.mce_banks);
5146         kvm_free_lapic(vcpu);
5147         down_read(&vcpu->kvm->slots_lock);
5148         kvm_mmu_destroy(vcpu);
5149         up_read(&vcpu->kvm->slots_lock);
5150         free_page((unsigned long)vcpu->arch.pio_data);
5151 }
5152
5153 struct  kvm *kvm_arch_create_vm(void)
5154 {
5155         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5156
5157         if (!kvm)
5158                 return ERR_PTR(-ENOMEM);
5159
5160         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5161         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5162
5163         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5164         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5165
5166         rdtscll(kvm->arch.vm_init_tsc);
5167
5168         return kvm;
5169 }
5170
5171 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5172 {
5173         vcpu_load(vcpu);
5174         kvm_mmu_unload(vcpu);
5175         vcpu_put(vcpu);
5176 }
5177
5178 static void kvm_free_vcpus(struct kvm *kvm)
5179 {
5180         unsigned int i;
5181         struct kvm_vcpu *vcpu;
5182
5183         /*
5184          * Unpin any mmu pages first.
5185          */
5186         kvm_for_each_vcpu(i, vcpu, kvm)
5187                 kvm_unload_vcpu_mmu(vcpu);
5188         kvm_for_each_vcpu(i, vcpu, kvm)
5189                 kvm_arch_vcpu_free(vcpu);
5190
5191         mutex_lock(&kvm->lock);
5192         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5193                 kvm->vcpus[i] = NULL;
5194
5195         atomic_set(&kvm->online_vcpus, 0);
5196         mutex_unlock(&kvm->lock);
5197 }
5198
5199 void kvm_arch_sync_events(struct kvm *kvm)
5200 {
5201         kvm_free_all_assigned_devices(kvm);
5202 }
5203
5204 void kvm_arch_destroy_vm(struct kvm *kvm)
5205 {
5206         kvm_iommu_unmap_guest(kvm);
5207         kvm_free_pit(kvm);
5208         kfree(kvm->arch.vpic);
5209         kfree(kvm->arch.vioapic);
5210         kvm_free_vcpus(kvm);
5211         kvm_free_physmem(kvm);
5212         if (kvm->arch.apic_access_page)
5213                 put_page(kvm->arch.apic_access_page);
5214         if (kvm->arch.ept_identity_pagetable)
5215                 put_page(kvm->arch.ept_identity_pagetable);
5216         kfree(kvm);
5217 }
5218
5219 int kvm_arch_set_memory_region(struct kvm *kvm,
5220                                 struct kvm_userspace_memory_region *mem,
5221                                 struct kvm_memory_slot old,
5222                                 int user_alloc)
5223 {
5224         int npages = mem->memory_size >> PAGE_SHIFT;
5225         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
5226
5227         /*To keep backward compatibility with older userspace,
5228          *x86 needs to hanlde !user_alloc case.
5229          */
5230         if (!user_alloc) {
5231                 if (npages && !old.rmap) {
5232                         unsigned long userspace_addr;
5233
5234                         down_write(&current->mm->mmap_sem);
5235                         userspace_addr = do_mmap(NULL, 0,
5236                                                  npages * PAGE_SIZE,
5237                                                  PROT_READ | PROT_WRITE,
5238                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5239                                                  0);
5240                         up_write(&current->mm->mmap_sem);
5241
5242                         if (IS_ERR((void *)userspace_addr))
5243                                 return PTR_ERR((void *)userspace_addr);
5244
5245                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
5246                         spin_lock(&kvm->mmu_lock);
5247                         memslot->userspace_addr = userspace_addr;
5248                         spin_unlock(&kvm->mmu_lock);
5249                 } else {
5250                         if (!old.user_alloc && old.rmap) {
5251                                 int ret;
5252
5253                                 down_write(&current->mm->mmap_sem);
5254                                 ret = do_munmap(current->mm, old.userspace_addr,
5255                                                 old.npages * PAGE_SIZE);
5256                                 up_write(&current->mm->mmap_sem);
5257                                 if (ret < 0)
5258                                         printk(KERN_WARNING
5259                                        "kvm_vm_ioctl_set_memory_region: "
5260                                        "failed to munmap memory\n");
5261                         }
5262                 }
5263         }
5264
5265         spin_lock(&kvm->mmu_lock);
5266         if (!kvm->arch.n_requested_mmu_pages) {
5267                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5268                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5269         }
5270
5271         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5272         spin_unlock(&kvm->mmu_lock);
5273
5274         return 0;
5275 }
5276
5277 void kvm_arch_flush_shadow(struct kvm *kvm)
5278 {
5279         kvm_mmu_zap_all(kvm);
5280         kvm_reload_remote_mmus(kvm);
5281 }
5282
5283 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5284 {
5285         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5286                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5287                 || vcpu->arch.nmi_pending ||
5288                 (kvm_arch_interrupt_allowed(vcpu) &&
5289                  kvm_cpu_has_interrupt(vcpu));
5290 }
5291
5292 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5293 {
5294         int me;
5295         int cpu = vcpu->cpu;
5296
5297         if (waitqueue_active(&vcpu->wq)) {
5298                 wake_up_interruptible(&vcpu->wq);
5299                 ++vcpu->stat.halt_wakeup;
5300         }
5301
5302         me = get_cpu();
5303         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5304                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5305                         smp_send_reschedule(cpu);
5306         put_cpu();
5307 }
5308
5309 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5310 {
5311         return kvm_x86_ops->interrupt_allowed(vcpu);
5312 }
5313
5314 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5315 {
5316         unsigned long rflags;
5317
5318         rflags = kvm_x86_ops->get_rflags(vcpu);
5319         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5320                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5321         return rflags;
5322 }
5323 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5324
5325 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5326 {
5327         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5328             vcpu->arch.singlestep_cs ==
5329                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5330             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5331                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5332         kvm_x86_ops->set_rflags(vcpu, rflags);
5333 }
5334 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5335
5336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5339 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5340 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5341 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5342 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5343 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5344 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5345 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5346 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);