SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40 #include <linux/user-return-notifier.h>
  41 #include <linux/srcu.h>
  42 #include <linux/slab.h>
  43 #include <trace/events/kvm.h>
  44 #undef TRACE_INCLUDE_FILE
  45 #define CREATE_TRACE_POINTS
  46 #include "trace.h"
  47
  48 #include <asm/debugreg.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/msr.h>
  51 #include <asm/desc.h>
  52 #include <asm/mtrr.h>
  53 #include <asm/mce.h>
  54
  55 #define MAX_IO_MSRS 256
  56 #define CR0_RESERVED_BITS                                               \
  57         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  58                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  59                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  60 #define CR4_RESERVED_BITS                                               \
  61         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  62                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  63                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  64                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  65
  66 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  67
  68 #define KVM_MAX_MCE_BANKS 32
  69 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  70
  71 /* EFER defaults:
  72  * - enable syscall per default because its emulated by KVM
  73  * - enable LME and LMA per default on 64 bit KVM
  74  */
  75 #ifdef CONFIG_X86_64
  76 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  77 #else
  78 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  79 #endif
  80
  81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  83
  84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  86                                     struct kvm_cpuid_entry2 __user *entries);
  87
  88 struct kvm_x86_ops *kvm_x86_ops;
  89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  90
  91 int ignore_msrs = 0;
  92 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  93
  94 #define KVM_NR_SHARED_MSRS 16
  95
  96 struct kvm_shared_msrs_global {
  97         int nr;
  98         u32 msrs[KVM_NR_SHARED_MSRS];
  99 };
 100
 101 struct kvm_shared_msrs {
 102         struct user_return_notifier urn;
 103         bool registered;
 104         struct kvm_shared_msr_values {
 105                 u64 host;
 106                 u64 curr;
 107         } values[KVM_NR_SHARED_MSRS];
 108 };
 109
 110 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 111 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 112
 113 struct kvm_stats_debugfs_item debugfs_entries[] = {
 114         { "pf_fixed", VCPU_STAT(pf_fixed) },
 115         { "pf_guest", VCPU_STAT(pf_guest) },
 116         { "tlb_flush", VCPU_STAT(tlb_flush) },
 117         { "invlpg", VCPU_STAT(invlpg) },
 118         { "exits", VCPU_STAT(exits) },
 119         { "io_exits", VCPU_STAT(io_exits) },
 120         { "mmio_exits", VCPU_STAT(mmio_exits) },
 121         { "signal_exits", VCPU_STAT(signal_exits) },
 122         { "irq_window", VCPU_STAT(irq_window_exits) },
 123         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 124         { "halt_exits", VCPU_STAT(halt_exits) },
 125         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 126         { "hypercalls", VCPU_STAT(hypercalls) },
 127         { "request_irq", VCPU_STAT(request_irq_exits) },
 128         { "irq_exits", VCPU_STAT(irq_exits) },
 129         { "host_state_reload", VCPU_STAT(host_state_reload) },
 130         { "efer_reload", VCPU_STAT(efer_reload) },
 131         { "fpu_reload", VCPU_STAT(fpu_reload) },
 132         { "insn_emulation", VCPU_STAT(insn_emulation) },
 133         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 134         { "irq_injections", VCPU_STAT(irq_injections) },
 135         { "nmi_injections", VCPU_STAT(nmi_injections) },
 136         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 137         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 138         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 139         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 140         { "mmu_flooded", VM_STAT(mmu_flooded) },
 141         { "mmu_recycled", VM_STAT(mmu_recycled) },
 142         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 143         { "mmu_unsync", VM_STAT(mmu_unsync) },
 144         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 145         { "largepages", VM_STAT(lpages) },
 146         { NULL }
 147 };
 148
 149 static void kvm_on_user_return(struct user_return_notifier *urn)
 150 {
 151         unsigned slot;
 152         struct kvm_shared_msrs *locals
 153                 = container_of(urn, struct kvm_shared_msrs, urn);
 154         struct kvm_shared_msr_values *values;
 155
 156         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 157                 values = &locals->values[slot];
 158                 if (values->host != values->curr) {
 159                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 160                         values->curr = values->host;
 161                 }
 162         }
 163         locals->registered = false;
 164         user_return_notifier_unregister(urn);
 165 }
 166
 167 static void shared_msr_update(unsigned slot, u32 msr)
 168 {
 169         struct kvm_shared_msrs *smsr;
 170         u64 value;
 171
 172         smsr = &__get_cpu_var(shared_msrs);
 173         /* only read, and nobody should modify it at this time,
 174          * so don't need lock */
 175         if (slot >= shared_msrs_global.nr) {
 176                 printk(KERN_ERR "kvm: invalid MSR slot!");
 177                 return;
 178         }
 179         rdmsrl_safe(msr, &value);
 180         smsr->values[slot].host = value;
 181         smsr->values[slot].curr = value;
 182 }
 183
 184 void kvm_define_shared_msr(unsigned slot, u32 msr)
 185 {
 186         if (slot >= shared_msrs_global.nr)
 187                 shared_msrs_global.nr = slot + 1;
 188         shared_msrs_global.msrs[slot] = msr;
 189         /* we need ensured the shared_msr_global have been updated */
 190         smp_wmb();
 191 }
 192 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 193
 194 static void kvm_shared_msr_cpu_online(void)
 195 {
 196         unsigned i;
 197
 198         for (i = 0; i < shared_msrs_global.nr; ++i)
 199                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 200 }
 201
 202 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 203 {
 204         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 205
 206         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 207                 return;
 208         smsr->values[slot].curr = value;
 209         wrmsrl(shared_msrs_global.msrs[slot], value);
 210         if (!smsr->registered) {
 211                 smsr->urn.on_user_return = kvm_on_user_return;
 212                 user_return_notifier_register(&smsr->urn);
 213                 smsr->registered = true;
 214         }
 215 }
 216 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 217
 218 static void drop_user_return_notifiers(void *ignore)
 219 {
 220         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 221
 222         if (smsr->registered)
 223                 kvm_on_user_return(&smsr->urn);
 224 }
 225
 226 unsigned long segment_base(u16 selector)
 227 {
 228         struct desc_ptr gdt;
 229         struct desc_struct *d;
 230         unsigned long table_base;
 231         unsigned long v;
 232
 233         if (!(selector & ~3))
 234                 return 0;
 235
 236         native_store_gdt(&gdt);
 237         table_base = gdt.address;
 238
 239         if (selector & 4) {           /* from ldt */
 240                 u16 ldt_selector = kvm_read_ldt();
 241
 242                 if (!(ldt_selector & ~3))
 243                         return 0;
 244                 table_base = segment_base(ldt_selector);
 245         }
 246         d = (struct desc_struct *)(table_base + (selector & ~7));
 247         v = get_desc_base(d);
 248 #ifdef CONFIG_X86_64
 249         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 250                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 251 #endif
 252         return v;
 253 }
 254 EXPORT_SYMBOL_GPL(segment_base);
 255
 256 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 257 {
 258         if (irqchip_in_kernel(vcpu->kvm))
 259                 return vcpu->arch.apic_base;
 260         else
 261                 return vcpu->arch.apic_base;
 262 }
 263 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 264
 265 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 266 {
 267         /* TODO: reserve bits check */
 268         if (irqchip_in_kernel(vcpu->kvm))
 269                 kvm_lapic_set_base(vcpu, data);
 270         else
 271                 vcpu->arch.apic_base = data;
 272 }
 273 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 274
 275 #define EXCPT_BENIGN            0
 276 #define EXCPT_CONTRIBUTORY      1
 277 #define EXCPT_PF                2
 278
 279 static int exception_class(int vector)
 280 {
 281         switch (vector) {
 282         case PF_VECTOR:
 283                 return EXCPT_PF;
 284         case DE_VECTOR:
 285         case TS_VECTOR:
 286         case NP_VECTOR:
 287         case SS_VECTOR:
 288         case GP_VECTOR:
 289                 return EXCPT_CONTRIBUTORY;
 290         default:
 291                 break;
 292         }
 293         return EXCPT_BENIGN;
 294 }
 295
 296 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 297                 unsigned nr, bool has_error, u32 error_code)
 298 {
 299         u32 prev_nr;
 300         int class1, class2;
 301
 302         if (!vcpu->arch.exception.pending) {
 303         queue:
 304                 vcpu->arch.exception.pending = true;
 305                 vcpu->arch.exception.has_error_code = has_error;
 306                 vcpu->arch.exception.nr = nr;
 307                 vcpu->arch.exception.error_code = error_code;
 308                 return;
 309         }
 310
 311         /* to check exception */
 312         prev_nr = vcpu->arch.exception.nr;
 313         if (prev_nr == DF_VECTOR) {
 314                 /* triple fault -> shutdown */
 315                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 316                 return;
 317         }
 318         class1 = exception_class(prev_nr);
 319         class2 = exception_class(nr);
 320         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 321                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 322                 /* generate double fault per SDM Table 5-5 */
 323                 vcpu->arch.exception.pending = true;
 324                 vcpu->arch.exception.has_error_code = true;
 325                 vcpu->arch.exception.nr = DF_VECTOR;
 326                 vcpu->arch.exception.error_code = 0;
 327         } else
 328                 /* replace previous exception with a new one in a hope
 329                    that instruction re-execution will regenerate lost
 330                    exception */
 331                 goto queue;
 332 }
 333
 334 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 335 {
 336         kvm_multiple_exception(vcpu, nr, false, 0);
 337 }
 338 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 339
 340 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 341                            u32 error_code)
 342 {
 343         ++vcpu->stat.pf_guest;
 344         vcpu->arch.cr2 = addr;
 345         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 346 }
 347
 348 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 349 {
 350         vcpu->arch.nmi_pending = 1;
 351 }
 352 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 353
 354 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 355 {
 356         kvm_multiple_exception(vcpu, nr, true, error_code);
 357 }
 358 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 359
 360 /*
 361  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 362  * a #GP and return false.
 363  */
 364 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 365 {
 366         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 367                 return true;
 368         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 369         return false;
 370 }
 371 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 372
 373 /*
 374  * Load the pae pdptrs.  Return true is they are all valid.
 375  */
 376 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 377 {
 378         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 379         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 380         int i;
 381         int ret;
 382         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 383
 384         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 385                                   offset * sizeof(u64), sizeof(pdpte));
 386         if (ret < 0) {
 387                 ret = 0;
 388                 goto out;
 389         }
 390         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 391                 if (is_present_gpte(pdpte[i]) &&
 392                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 393                         ret = 0;
 394                         goto out;
 395                 }
 396         }
 397         ret = 1;
 398
 399         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 400         __set_bit(VCPU_EXREG_PDPTR,
 401                   (unsigned long *)&vcpu->arch.regs_avail);
 402         __set_bit(VCPU_EXREG_PDPTR,
 403                   (unsigned long *)&vcpu->arch.regs_dirty);
 404 out:
 405
 406         return ret;
 407 }
 408 EXPORT_SYMBOL_GPL(load_pdptrs);
 409
 410 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 411 {
 412         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 413         bool changed = true;
 414         int r;
 415
 416         if (is_long_mode(vcpu) || !is_pae(vcpu))
 417                 return false;
 418
 419         if (!test_bit(VCPU_EXREG_PDPTR,
 420                       (unsigned long *)&vcpu->arch.regs_avail))
 421                 return true;
 422
 423         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 424         if (r < 0)
 425                 goto out;
 426         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 427 out:
 428
 429         return changed;
 430 }
 431
 432 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 433 {
 434         cr0 |= X86_CR0_ET;
 435
 436 #ifdef CONFIG_X86_64
 437         if (cr0 & 0xffffffff00000000UL) {
 438                 kvm_inject_gp(vcpu, 0);
 439                 return;
 440         }
 441 #endif
 442
 443         cr0 &= ~CR0_RESERVED_BITS;
 444
 445         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 446                 kvm_inject_gp(vcpu, 0);
 447                 return;
 448         }
 449
 450         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 451                 kvm_inject_gp(vcpu, 0);
 452                 return;
 453         }
 454
 455         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 456 #ifdef CONFIG_X86_64
 457                 if ((vcpu->arch.efer & EFER_LME)) {
 458                         int cs_db, cs_l;
 459
 460                         if (!is_pae(vcpu)) {
 461                                 kvm_inject_gp(vcpu, 0);
 462                                 return;
 463                         }
 464                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 465                         if (cs_l) {
 466                                 kvm_inject_gp(vcpu, 0);
 467                                 return;
 468
 469                         }
 470                 } else
 471 #endif
 472                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 473                         kvm_inject_gp(vcpu, 0);
 474                         return;
 475                 }
 476
 477         }
 478
 479         kvm_x86_ops->set_cr0(vcpu, cr0);
 480
 481         kvm_mmu_reset_context(vcpu);
 482         return;
 483 }
 484 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 485
 486 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 487 {
 488         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 489 }
 490 EXPORT_SYMBOL_GPL(kvm_lmsw);
 491
 492 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 493 {
 494         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 495         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 496
 497         if (cr4 & CR4_RESERVED_BITS) {
 498                 kvm_inject_gp(vcpu, 0);
 499                 return;
 500         }
 501
 502         if (is_long_mode(vcpu)) {
 503                 if (!(cr4 & X86_CR4_PAE)) {
 504                         kvm_inject_gp(vcpu, 0);
 505                         return;
 506                 }
 507         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 508                    && ((cr4 ^ old_cr4) & pdptr_bits)
 509                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 510                 kvm_inject_gp(vcpu, 0);
 511                 return;
 512         }
 513
 514         if (cr4 & X86_CR4_VMXE) {
 515                 kvm_inject_gp(vcpu, 0);
 516                 return;
 517         }
 518         kvm_x86_ops->set_cr4(vcpu, cr4);
 519         vcpu->arch.cr4 = cr4;
 520         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 521         kvm_mmu_reset_context(vcpu);
 522 }
 523 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 524
 525 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 526 {
 527         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 528                 kvm_mmu_sync_roots(vcpu);
 529                 kvm_mmu_flush_tlb(vcpu);
 530                 return;
 531         }
 532
 533         if (is_long_mode(vcpu)) {
 534                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 535                         kvm_inject_gp(vcpu, 0);
 536                         return;
 537                 }
 538         } else {
 539                 if (is_pae(vcpu)) {
 540                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 541                                 kvm_inject_gp(vcpu, 0);
 542                                 return;
 543                         }
 544                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 545                                 kvm_inject_gp(vcpu, 0);
 546                                 return;
 547                         }
 548                 }
 549                 /*
 550                  * We don't check reserved bits in nonpae mode, because
 551                  * this isn't enforced, and VMware depends on this.
 552                  */
 553         }
 554
 555         /*
 556          * Does the new cr3 value map to physical memory? (Note, we
 557          * catch an invalid cr3 even in real-mode, because it would
 558          * cause trouble later on when we turn on paging anyway.)
 559          *
 560          * A real CPU would silently accept an invalid cr3 and would
 561          * attempt to use it - with largely undefined (and often hard
 562          * to debug) behavior on the guest side.
 563          */
 564         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 565                 kvm_inject_gp(vcpu, 0);
 566         else {
 567                 vcpu->arch.cr3 = cr3;
 568                 vcpu->arch.mmu.new_cr3(vcpu);
 569         }
 570 }
 571 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 572
 573 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 574 {
 575         if (cr8 & CR8_RESERVED_BITS) {
 576                 kvm_inject_gp(vcpu, 0);
 577                 return;
 578         }
 579         if (irqchip_in_kernel(vcpu->kvm))
 580                 kvm_lapic_set_tpr(vcpu, cr8);
 581         else
 582                 vcpu->arch.cr8 = cr8;
 583 }
 584 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 585
 586 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 587 {
 588         if (irqchip_in_kernel(vcpu->kvm))
 589                 return kvm_lapic_get_cr8(vcpu);
 590         else
 591                 return vcpu->arch.cr8;
 592 }
 593 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 594
 595 static inline u32 bit(int bitno)
 596 {
 597         return 1 << (bitno & 31);
 598 }
 599
 600 /*
 601  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 602  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 603  *
 604  * This list is modified at module load time to reflect the
 605  * capabilities of the host cpu. This capabilities test skips MSRs that are
 606  * kvm-specific. Those are put in the beginning of the list.
 607  */
 608
 609 #define KVM_SAVE_MSRS_BEGIN     5
 610 static u32 msrs_to_save[] = {
 611         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 612         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 613         HV_X64_MSR_APIC_ASSIST_PAGE,
 614         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 615         MSR_K6_STAR,
 616 #ifdef CONFIG_X86_64
 617         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 618 #endif
 619         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 620 };
 621
 622 static unsigned num_msrs_to_save;
 623
 624 static u32 emulated_msrs[] = {
 625         MSR_IA32_MISC_ENABLE,
 626 };
 627
 628 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 629 {
 630         if (efer & efer_reserved_bits) {
 631                 kvm_inject_gp(vcpu, 0);
 632                 return;
 633         }
 634
 635         if (is_paging(vcpu)
 636             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
 637                 kvm_inject_gp(vcpu, 0);
 638                 return;
 639         }
 640
 641         if (efer & EFER_FFXSR) {
 642                 struct kvm_cpuid_entry2 *feat;
 643
 644                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 645                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 646                         kvm_inject_gp(vcpu, 0);
 647                         return;
 648                 }
 649         }
 650
 651         if (efer & EFER_SVME) {
 652                 struct kvm_cpuid_entry2 *feat;
 653
 654                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 655                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 656                         kvm_inject_gp(vcpu, 0);
 657                         return;
 658                 }
 659         }
 660
 661         kvm_x86_ops->set_efer(vcpu, efer);
 662
 663         efer &= ~EFER_LMA;
 664         efer |= vcpu->arch.efer & EFER_LMA;
 665
 666         vcpu->arch.efer = efer;
 667
 668         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 669         kvm_mmu_reset_context(vcpu);
 670 }
 671
 672 void kvm_enable_efer_bits(u64 mask)
 673 {
 674        efer_reserved_bits &= ~mask;
 675 }
 676 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 677
 678
 679 /*
 680  * Writes msr value into into the appropriate "register".
 681  * Returns 0 on success, non-0 otherwise.
 682  * Assumes vcpu_load() was already called.
 683  */
 684 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 685 {
 686         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 687 }
 688
 689 /*
 690  * Adapt set_msr() to msr_io()'s calling convention
 691  */
 692 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 693 {
 694         return kvm_set_msr(vcpu, index, *data);
 695 }
 696
 697 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 698 {
 699         static int version;
 700         struct pvclock_wall_clock wc;
 701         struct timespec boot;
 702
 703         if (!wall_clock)
 704                 return;
 705
 706         version++;
 707
 708         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 709
 710         /*
 711          * The guest calculates current wall clock time by adding
 712          * system time (updated by kvm_write_guest_time below) to the
 713          * wall clock specified here.  guest system time equals host
 714          * system time for us, thus we must fill in host boot time here.
 715          */
 716         getboottime(&boot);
 717
 718         wc.sec = boot.tv_sec;
 719         wc.nsec = boot.tv_nsec;
 720         wc.version = version;
 721
 722         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 723
 724         version++;
 725         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 726 }
 727
 728 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 729 {
 730         uint32_t quotient, remainder;
 731
 732         /* Don't try to replace with do_div(), this one calculates
 733          * "(dividend << 32) / divisor" */
 734         __asm__ ( "divl %4"
 735                   : "=a" (quotient), "=d" (remainder)
 736                   : "0" (0), "1" (dividend), "r" (divisor) );
 737         return quotient;
 738 }
 739
 740 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 741 {
 742         uint64_t nsecs = 1000000000LL;
 743         int32_t  shift = 0;
 744         uint64_t tps64;
 745         uint32_t tps32;
 746
 747         tps64 = tsc_khz * 1000LL;
 748         while (tps64 > nsecs*2) {
 749                 tps64 >>= 1;
 750                 shift--;
 751         }
 752
 753         tps32 = (uint32_t)tps64;
 754         while (tps32 <= (uint32_t)nsecs) {
 755                 tps32 <<= 1;
 756                 shift++;
 757         }
 758
 759         hv_clock->tsc_shift = shift;
 760         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 761
 762         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 763                  __func__, tsc_khz, hv_clock->tsc_shift,
 764                  hv_clock->tsc_to_system_mul);
 765 }
 766
 767 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 768
 769 static void kvm_write_guest_time(struct kvm_vcpu *v)
 770 {
 771         struct timespec ts;
 772         unsigned long flags;
 773         struct kvm_vcpu_arch *vcpu = &v->arch;
 774         void *shared_kaddr;
 775         unsigned long this_tsc_khz;
 776
 777         if ((!vcpu->time_page))
 778                 return;
 779
 780         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 781         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 782                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 783                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 784         }
 785         put_cpu_var(cpu_tsc_khz);
 786
 787         /* Keep irq disabled to prevent changes to the clock */
 788         local_irq_save(flags);
 789         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 790         ktime_get_ts(&ts);
 791         monotonic_to_bootbased(&ts);
 792         local_irq_restore(flags);
 793
 794         /* With all the info we got, fill in the values */
 795
 796         vcpu->hv_clock.system_time = ts.tv_nsec +
 797                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 798
 799         /*
 800          * The interface expects us to write an even number signaling that the
 801          * update is finished. Since the guest won't see the intermediate
 802          * state, we just increase by 2 at the end.
 803          */
 804         vcpu->hv_clock.version += 2;
 805
 806         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 807
 808         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 809                sizeof(vcpu->hv_clock));
 810
 811         kunmap_atomic(shared_kaddr, KM_USER0);
 812
 813         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 814 }
 815
 816 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 817 {
 818         struct kvm_vcpu_arch *vcpu = &v->arch;
 819
 820         if (!vcpu->time_page)
 821                 return 0;
 822         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 823         return 1;
 824 }
 825
 826 static bool msr_mtrr_valid(unsigned msr)
 827 {
 828         switch (msr) {
 829         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 830         case MSR_MTRRfix64K_00000:
 831         case MSR_MTRRfix16K_80000:
 832         case MSR_MTRRfix16K_A0000:
 833         case MSR_MTRRfix4K_C0000:
 834         case MSR_MTRRfix4K_C8000:
 835         case MSR_MTRRfix4K_D0000:
 836         case MSR_MTRRfix4K_D8000:
 837         case MSR_MTRRfix4K_E0000:
 838         case MSR_MTRRfix4K_E8000:
 839         case MSR_MTRRfix4K_F0000:
 840         case MSR_MTRRfix4K_F8000:
 841         case MSR_MTRRdefType:
 842         case MSR_IA32_CR_PAT:
 843                 return true;
 844         case 0x2f8:
 845                 return true;
 846         }
 847         return false;
 848 }
 849
 850 static bool valid_pat_type(unsigned t)
 851 {
 852         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 853 }
 854
 855 static bool valid_mtrr_type(unsigned t)
 856 {
 857         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 858 }
 859
 860 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 861 {
 862         int i;
 863
 864         if (!msr_mtrr_valid(msr))
 865                 return false;
 866
 867         if (msr == MSR_IA32_CR_PAT) {
 868                 for (i = 0; i < 8; i++)
 869                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 870                                 return false;
 871                 return true;
 872         } else if (msr == MSR_MTRRdefType) {
 873                 if (data & ~0xcff)
 874                         return false;
 875                 return valid_mtrr_type(data & 0xff);
 876         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 877                 for (i = 0; i < 8 ; i++)
 878                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 879                                 return false;
 880                 return true;
 881         }
 882
 883         /* variable MTRRs */
 884         return valid_mtrr_type(data & 0xff);
 885 }
 886
 887 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 888 {
 889         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 890
 891         if (!mtrr_valid(vcpu, msr, data))
 892                 return 1;
 893
 894         if (msr == MSR_MTRRdefType) {
 895                 vcpu->arch.mtrr_state.def_type = data;
 896                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 897         } else if (msr == MSR_MTRRfix64K_00000)
 898                 p[0] = data;
 899         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 900                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 901         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 902                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 903         else if (msr == MSR_IA32_CR_PAT)
 904                 vcpu->arch.pat = data;
 905         else {  /* Variable MTRRs */
 906                 int idx, is_mtrr_mask;
 907                 u64 *pt;
 908
 909                 idx = (msr - 0x200) / 2;
 910                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 911                 if (!is_mtrr_mask)
 912                         pt =
 913                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 914                 else
 915                         pt =
 916                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 917                 *pt = data;
 918         }
 919
 920         kvm_mmu_reset_context(vcpu);
 921         return 0;
 922 }
 923
 924 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 925 {
 926         u64 mcg_cap = vcpu->arch.mcg_cap;
 927         unsigned bank_num = mcg_cap & 0xff;
 928
 929         switch (msr) {
 930         case MSR_IA32_MCG_STATUS:
 931                 vcpu->arch.mcg_status = data;
 932                 break;
 933         case MSR_IA32_MCG_CTL:
 934                 if (!(mcg_cap & MCG_CTL_P))
 935                         return 1;
 936                 if (data != 0 && data != ~(u64)0)
 937                         return -1;
 938                 vcpu->arch.mcg_ctl = data;
 939                 break;
 940         default:
 941                 if (msr >= MSR_IA32_MC0_CTL &&
 942                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 943                         u32 offset = msr - MSR_IA32_MC0_CTL;
 944                         /* only 0 or all 1s can be written to IA32_MCi_CTL
 945                          * some Linux kernels though clear bit 10 in bank 4 to
 946                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
 947                          * this to avoid an uncatched #GP in the guest
 948                          */
 949                         if ((offset & 0x3) == 0 &&
 950                             data != 0 && (data | (1 << 10)) != ~(u64)0)
 951                                 return -1;
 952                         vcpu->arch.mce_banks[offset] = data;
 953                         break;
 954                 }
 955                 return 1;
 956         }
 957         return 0;
 958 }
 959
 960 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 961 {
 962         struct kvm *kvm = vcpu->kvm;
 963         int lm = is_long_mode(vcpu);
 964         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 965                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 966         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 967                 : kvm->arch.xen_hvm_config.blob_size_32;
 968         u32 page_num = data & ~PAGE_MASK;
 969         u64 page_addr = data & PAGE_MASK;
 970         u8 *page;
 971         int r;
 972
 973         r = -E2BIG;
 974         if (page_num >= blob_size)
 975                 goto out;
 976         r = -ENOMEM;
 977         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 978         if (!page)
 979                 goto out;
 980         r = -EFAULT;
 981         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 982                 goto out_free;
 983         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 984                 goto out_free;
 985         r = 0;
 986 out_free:
 987         kfree(page);
 988 out:
 989         return r;
 990 }
 991
 992 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
 993 {
 994         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
 995 }
 996
 997 static bool kvm_hv_msr_partition_wide(u32 msr)
 998 {
 999         bool r = false;
1000         switch (msr) {
1001         case HV_X64_MSR_GUEST_OS_ID:
1002         case HV_X64_MSR_HYPERCALL:
1003                 r = true;
1004                 break;
1005         }
1006
1007         return r;
1008 }
1009
1010 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1011 {
1012         struct kvm *kvm = vcpu->kvm;
1013
1014         switch (msr) {
1015         case HV_X64_MSR_GUEST_OS_ID:
1016                 kvm->arch.hv_guest_os_id = data;
1017                 /* setting guest os id to zero disables hypercall page */
1018                 if (!kvm->arch.hv_guest_os_id)
1019                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1020                 break;
1021         case HV_X64_MSR_HYPERCALL: {
1022                 u64 gfn;
1023                 unsigned long addr;
1024                 u8 instructions[4];
1025
1026                 /* if guest os id is not set hypercall should remain disabled */
1027                 if (!kvm->arch.hv_guest_os_id)
1028                         break;
1029                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1030                         kvm->arch.hv_hypercall = data;
1031                         break;
1032                 }
1033                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1034                 addr = gfn_to_hva(kvm, gfn);
1035                 if (kvm_is_error_hva(addr))
1036                         return 1;
1037                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1038                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1039                 if (copy_to_user((void __user *)addr, instructions, 4))
1040                         return 1;
1041                 kvm->arch.hv_hypercall = data;
1042                 break;
1043         }
1044         default:
1045                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1046                           "data 0x%llx\n", msr, data);
1047                 return 1;
1048         }
1049         return 0;
1050 }
1051
1052 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1053 {
1054         switch (msr) {
1055         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1056                 unsigned long addr;
1057
1058                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1059                         vcpu->arch.hv_vapic = data;
1060                         break;
1061                 }
1062                 addr = gfn_to_hva(vcpu->kvm, data >>
1063                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1064                 if (kvm_is_error_hva(addr))
1065                         return 1;
1066                 if (clear_user((void __user *)addr, PAGE_SIZE))
1067                         return 1;
1068                 vcpu->arch.hv_vapic = data;
1069                 break;
1070         }
1071         case HV_X64_MSR_EOI:
1072                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1073         case HV_X64_MSR_ICR:
1074                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1075         case HV_X64_MSR_TPR:
1076                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1077         default:
1078                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1079                           "data 0x%llx\n", msr, data);
1080                 return 1;
1081         }
1082
1083         return 0;
1084 }
1085
1086 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1087 {
1088         switch (msr) {
1089         case MSR_EFER:
1090                 set_efer(vcpu, data);
1091                 break;
1092         case MSR_K7_HWCR:
1093                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1094                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1095                 if (data != 0) {
1096                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1097                                 data);
1098                         return 1;
1099                 }
1100                 break;
1101         case MSR_FAM10H_MMIO_CONF_BASE:
1102                 if (data != 0) {
1103                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1104                                 "0x%llx\n", data);
1105                         return 1;
1106                 }
1107                 break;
1108         case MSR_AMD64_NB_CFG:
1109                 break;
1110         case MSR_IA32_DEBUGCTLMSR:
1111                 if (!data) {
1112                         /* We support the non-activated case already */
1113                         break;
1114                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1115                         /* Values other than LBR and BTF are vendor-specific,
1116                            thus reserved and should throw a #GP */
1117                         return 1;
1118                 }
1119                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1120                         __func__, data);
1121                 break;
1122         case MSR_IA32_UCODE_REV:
1123         case MSR_IA32_UCODE_WRITE:
1124         case MSR_VM_HSAVE_PA:
1125         case MSR_AMD64_PATCH_LOADER:
1126                 break;
1127         case 0x200 ... 0x2ff:
1128                 return set_msr_mtrr(vcpu, msr, data);
1129         case MSR_IA32_APICBASE:
1130                 kvm_set_apic_base(vcpu, data);
1131                 break;
1132         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1133                 return kvm_x2apic_msr_write(vcpu, msr, data);
1134         case MSR_IA32_MISC_ENABLE:
1135                 vcpu->arch.ia32_misc_enable_msr = data;
1136                 break;
1137         case MSR_KVM_WALL_CLOCK:
1138                 vcpu->kvm->arch.wall_clock = data;
1139                 kvm_write_wall_clock(vcpu->kvm, data);
1140                 break;
1141         case MSR_KVM_SYSTEM_TIME: {
1142                 if (vcpu->arch.time_page) {
1143                         kvm_release_page_dirty(vcpu->arch.time_page);
1144                         vcpu->arch.time_page = NULL;
1145                 }
1146
1147                 vcpu->arch.time = data;
1148
1149                 /* we verify if the enable bit is set... */
1150                 if (!(data & 1))
1151                         break;
1152
1153                 /* ...but clean it before doing the actual write */
1154                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1155
1156                 vcpu->arch.time_page =
1157                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1158
1159                 if (is_error_page(vcpu->arch.time_page)) {
1160                         kvm_release_page_clean(vcpu->arch.time_page);
1161                         vcpu->arch.time_page = NULL;
1162                 }
1163
1164                 kvm_request_guest_time_update(vcpu);
1165                 break;
1166         }
1167         case MSR_IA32_MCG_CTL:
1168         case MSR_IA32_MCG_STATUS:
1169         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1170                 return set_msr_mce(vcpu, msr, data);
1171
1172         /* Performance counters are not protected by a CPUID bit,
1173          * so we should check all of them in the generic path for the sake of
1174          * cross vendor migration.
1175          * Writing a zero into the event select MSRs disables them,
1176          * which we perfectly emulate ;-). Any other value should be at least
1177          * reported, some guests depend on them.
1178          */
1179         case MSR_P6_EVNTSEL0:
1180         case MSR_P6_EVNTSEL1:
1181         case MSR_K7_EVNTSEL0:
1182         case MSR_K7_EVNTSEL1:
1183         case MSR_K7_EVNTSEL2:
1184         case MSR_K7_EVNTSEL3:
1185                 if (data != 0)
1186                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1187                                 "0x%x data 0x%llx\n", msr, data);
1188                 break;
1189         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1190          * so we ignore writes to make it happy.
1191          */
1192         case MSR_P6_PERFCTR0:
1193         case MSR_P6_PERFCTR1:
1194         case MSR_K7_PERFCTR0:
1195         case MSR_K7_PERFCTR1:
1196         case MSR_K7_PERFCTR2:
1197         case MSR_K7_PERFCTR3:
1198                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1199                         "0x%x data 0x%llx\n", msr, data);
1200                 break;
1201         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1202                 if (kvm_hv_msr_partition_wide(msr)) {
1203                         int r;
1204                         mutex_lock(&vcpu->kvm->lock);
1205                         r = set_msr_hyperv_pw(vcpu, msr, data);
1206                         mutex_unlock(&vcpu->kvm->lock);
1207                         return r;
1208                 } else
1209                         return set_msr_hyperv(vcpu, msr, data);
1210                 break;
1211         default:
1212                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1213                         return xen_hvm_config(vcpu, data);
1214                 if (!ignore_msrs) {
1215                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1216                                 msr, data);
1217                         return 1;
1218                 } else {
1219                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1220                                 msr, data);
1221                         break;
1222                 }
1223         }
1224         return 0;
1225 }
1226 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1227
1228
1229 /*
1230  * Reads an msr value (of 'msr_index') into 'pdata'.
1231  * Returns 0 on success, non-0 otherwise.
1232  * Assumes vcpu_load() was already called.
1233  */
1234 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1235 {
1236         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1237 }
1238
1239 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1240 {
1241         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1242
1243         if (!msr_mtrr_valid(msr))
1244                 return 1;
1245
1246         if (msr == MSR_MTRRdefType)
1247                 *pdata = vcpu->arch.mtrr_state.def_type +
1248                          (vcpu->arch.mtrr_state.enabled << 10);
1249         else if (msr == MSR_MTRRfix64K_00000)
1250                 *pdata = p[0];
1251         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1252                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1253         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1254                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1255         else if (msr == MSR_IA32_CR_PAT)
1256                 *pdata = vcpu->arch.pat;
1257         else {  /* Variable MTRRs */
1258                 int idx, is_mtrr_mask;
1259                 u64 *pt;
1260
1261                 idx = (msr - 0x200) / 2;
1262                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1263                 if (!is_mtrr_mask)
1264                         pt =
1265                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1266                 else
1267                         pt =
1268                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1269                 *pdata = *pt;
1270         }
1271
1272         return 0;
1273 }
1274
1275 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1276 {
1277         u64 data;
1278         u64 mcg_cap = vcpu->arch.mcg_cap;
1279         unsigned bank_num = mcg_cap & 0xff;
1280
1281         switch (msr) {
1282         case MSR_IA32_P5_MC_ADDR:
1283         case MSR_IA32_P5_MC_TYPE:
1284                 data = 0;
1285                 break;
1286         case MSR_IA32_MCG_CAP:
1287                 data = vcpu->arch.mcg_cap;
1288                 break;
1289         case MSR_IA32_MCG_CTL:
1290                 if (!(mcg_cap & MCG_CTL_P))
1291                         return 1;
1292                 data = vcpu->arch.mcg_ctl;
1293                 break;
1294         case MSR_IA32_MCG_STATUS:
1295                 data = vcpu->arch.mcg_status;
1296                 break;
1297         default:
1298                 if (msr >= MSR_IA32_MC0_CTL &&
1299                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1300                         u32 offset = msr - MSR_IA32_MC0_CTL;
1301                         data = vcpu->arch.mce_banks[offset];
1302                         break;
1303                 }
1304                 return 1;
1305         }
1306         *pdata = data;
1307         return 0;
1308 }
1309
1310 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1311 {
1312         u64 data = 0;
1313         struct kvm *kvm = vcpu->kvm;
1314
1315         switch (msr) {
1316         case HV_X64_MSR_GUEST_OS_ID:
1317                 data = kvm->arch.hv_guest_os_id;
1318                 break;
1319         case HV_X64_MSR_HYPERCALL:
1320                 data = kvm->arch.hv_hypercall;
1321                 break;
1322         default:
1323                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1324                 return 1;
1325         }
1326
1327         *pdata = data;
1328         return 0;
1329 }
1330
1331 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1332 {
1333         u64 data = 0;
1334
1335         switch (msr) {
1336         case HV_X64_MSR_VP_INDEX: {
1337                 int r;
1338                 struct kvm_vcpu *v;
1339                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1340                         if (v == vcpu)
1341                                 data = r;
1342                 break;
1343         }
1344         case HV_X64_MSR_EOI:
1345                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1346         case HV_X64_MSR_ICR:
1347                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1348         case HV_X64_MSR_TPR:
1349                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1350         default:
1351                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1352                 return 1;
1353         }
1354         *pdata = data;
1355         return 0;
1356 }
1357
1358 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1359 {
1360         u64 data;
1361
1362         switch (msr) {
1363         case MSR_IA32_PLATFORM_ID:
1364         case MSR_IA32_UCODE_REV:
1365         case MSR_IA32_EBL_CR_POWERON:
1366         case MSR_IA32_DEBUGCTLMSR:
1367         case MSR_IA32_LASTBRANCHFROMIP:
1368         case MSR_IA32_LASTBRANCHTOIP:
1369         case MSR_IA32_LASTINTFROMIP:
1370         case MSR_IA32_LASTINTTOIP:
1371         case MSR_K8_SYSCFG:
1372         case MSR_K7_HWCR:
1373         case MSR_VM_HSAVE_PA:
1374         case MSR_P6_PERFCTR0:
1375         case MSR_P6_PERFCTR1:
1376         case MSR_P6_EVNTSEL0:
1377         case MSR_P6_EVNTSEL1:
1378         case MSR_K7_EVNTSEL0:
1379         case MSR_K7_PERFCTR0:
1380         case MSR_K8_INT_PENDING_MSG:
1381         case MSR_AMD64_NB_CFG:
1382         case MSR_FAM10H_MMIO_CONF_BASE:
1383                 data = 0;
1384                 break;
1385         case MSR_MTRRcap:
1386                 data = 0x500 | KVM_NR_VAR_MTRR;
1387                 break;
1388         case 0x200 ... 0x2ff:
1389                 return get_msr_mtrr(vcpu, msr, pdata);
1390         case 0xcd: /* fsb frequency */
1391                 data = 3;
1392                 break;
1393         case MSR_IA32_APICBASE:
1394                 data = kvm_get_apic_base(vcpu);
1395                 break;
1396         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1397                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1398                 break;
1399         case MSR_IA32_MISC_ENABLE:
1400                 data = vcpu->arch.ia32_misc_enable_msr;
1401                 break;
1402         case MSR_IA32_PERF_STATUS:
1403                 /* TSC increment by tick */
1404                 data = 1000ULL;
1405                 /* CPU multiplier */
1406                 data |= (((uint64_t)4ULL) << 40);
1407                 break;
1408         case MSR_EFER:
1409                 data = vcpu->arch.efer;
1410                 break;
1411         case MSR_KVM_WALL_CLOCK:
1412                 data = vcpu->kvm->arch.wall_clock;
1413                 break;
1414         case MSR_KVM_SYSTEM_TIME:
1415                 data = vcpu->arch.time;
1416                 break;
1417         case MSR_IA32_P5_MC_ADDR:
1418         case MSR_IA32_P5_MC_TYPE:
1419         case MSR_IA32_MCG_CAP:
1420         case MSR_IA32_MCG_CTL:
1421         case MSR_IA32_MCG_STATUS:
1422         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1423                 return get_msr_mce(vcpu, msr, pdata);
1424         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1425                 if (kvm_hv_msr_partition_wide(msr)) {
1426                         int r;
1427                         mutex_lock(&vcpu->kvm->lock);
1428                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1429                         mutex_unlock(&vcpu->kvm->lock);
1430                         return r;
1431                 } else
1432                         return get_msr_hyperv(vcpu, msr, pdata);
1433                 break;
1434         default:
1435                 if (!ignore_msrs) {
1436                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1437                         return 1;
1438                 } else {
1439                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1440                         data = 0;
1441                 }
1442                 break;
1443         }
1444         *pdata = data;
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1448
1449 /*
1450  * Read or write a bunch of msrs. All parameters are kernel addresses.
1451  *
1452  * @return number of msrs set successfully.
1453  */
1454 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1455                     struct kvm_msr_entry *entries,
1456                     int (*do_msr)(struct kvm_vcpu *vcpu,
1457                                   unsigned index, u64 *data))
1458 {
1459         int i, idx;
1460
1461         vcpu_load(vcpu);
1462
1463         idx = srcu_read_lock(&vcpu->kvm->srcu);
1464         for (i = 0; i < msrs->nmsrs; ++i)
1465                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1466                         break;
1467         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1468
1469         vcpu_put(vcpu);
1470
1471         return i;
1472 }
1473
1474 /*
1475  * Read or write a bunch of msrs. Parameters are user addresses.
1476  *
1477  * @return number of msrs set successfully.
1478  */
1479 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1480                   int (*do_msr)(struct kvm_vcpu *vcpu,
1481                                 unsigned index, u64 *data),
1482                   int writeback)
1483 {
1484         struct kvm_msrs msrs;
1485         struct kvm_msr_entry *entries;
1486         int r, n;
1487         unsigned size;
1488
1489         r = -EFAULT;
1490         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1491                 goto out;
1492
1493         r = -E2BIG;
1494         if (msrs.nmsrs >= MAX_IO_MSRS)
1495                 goto out;
1496
1497         r = -ENOMEM;
1498         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1499         entries = vmalloc(size);
1500         if (!entries)
1501                 goto out;
1502
1503         r = -EFAULT;
1504         if (copy_from_user(entries, user_msrs->entries, size))
1505                 goto out_free;
1506
1507         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1508         if (r < 0)
1509                 goto out_free;
1510
1511         r = -EFAULT;
1512         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1513                 goto out_free;
1514
1515         r = n;
1516
1517 out_free:
1518         vfree(entries);
1519 out:
1520         return r;
1521 }
1522
1523 int kvm_dev_ioctl_check_extension(long ext)
1524 {
1525         int r;
1526
1527         switch (ext) {
1528         case KVM_CAP_IRQCHIP:
1529         case KVM_CAP_HLT:
1530         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1531         case KVM_CAP_SET_TSS_ADDR:
1532         case KVM_CAP_EXT_CPUID:
1533         case KVM_CAP_CLOCKSOURCE:
1534         case KVM_CAP_PIT:
1535         case KVM_CAP_NOP_IO_DELAY:
1536         case KVM_CAP_MP_STATE:
1537         case KVM_CAP_SYNC_MMU:
1538         case KVM_CAP_REINJECT_CONTROL:
1539         case KVM_CAP_IRQ_INJECT_STATUS:
1540         case KVM_CAP_ASSIGN_DEV_IRQ:
1541         case KVM_CAP_IRQFD:
1542         case KVM_CAP_IOEVENTFD:
1543         case KVM_CAP_PIT2:
1544         case KVM_CAP_PIT_STATE2:
1545         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1546         case KVM_CAP_XEN_HVM:
1547         case KVM_CAP_ADJUST_CLOCK:
1548         case KVM_CAP_VCPU_EVENTS:
1549         case KVM_CAP_HYPERV:
1550         case KVM_CAP_HYPERV_VAPIC:
1551         case KVM_CAP_HYPERV_SPIN:
1552         case KVM_CAP_PCI_SEGMENT:
1553         case KVM_CAP_DEBUGREGS:
1554         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1555                 r = 1;
1556                 break;
1557         case KVM_CAP_COALESCED_MMIO:
1558                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1559                 break;
1560         case KVM_CAP_VAPIC:
1561                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1562                 break;
1563         case KVM_CAP_NR_VCPUS:
1564                 r = KVM_MAX_VCPUS;
1565                 break;
1566         case KVM_CAP_NR_MEMSLOTS:
1567                 r = KVM_MEMORY_SLOTS;
1568                 break;
1569         case KVM_CAP_PV_MMU:    /* obsolete */
1570                 r = 0;
1571                 break;
1572         case KVM_CAP_IOMMU:
1573                 r = iommu_found();
1574                 break;
1575         case KVM_CAP_MCE:
1576                 r = KVM_MAX_MCE_BANKS;
1577                 break;
1578         default:
1579                 r = 0;
1580                 break;
1581         }
1582         return r;
1583
1584 }
1585
1586 long kvm_arch_dev_ioctl(struct file *filp,
1587                         unsigned int ioctl, unsigned long arg)
1588 {
1589         void __user *argp = (void __user *)arg;
1590         long r;
1591
1592         switch (ioctl) {
1593         case KVM_GET_MSR_INDEX_LIST: {
1594                 struct kvm_msr_list __user *user_msr_list = argp;
1595                 struct kvm_msr_list msr_list;
1596                 unsigned n;
1597
1598                 r = -EFAULT;
1599                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1600                         goto out;
1601                 n = msr_list.nmsrs;
1602                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1603                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1604                         goto out;
1605                 r = -E2BIG;
1606                 if (n < msr_list.nmsrs)
1607                         goto out;
1608                 r = -EFAULT;
1609                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1610                                  num_msrs_to_save * sizeof(u32)))
1611                         goto out;
1612                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1613                                  &emulated_msrs,
1614                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1615                         goto out;
1616                 r = 0;
1617                 break;
1618         }
1619         case KVM_GET_SUPPORTED_CPUID: {
1620                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1621                 struct kvm_cpuid2 cpuid;
1622
1623                 r = -EFAULT;
1624                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1625                         goto out;
1626                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1627                                                       cpuid_arg->entries);
1628                 if (r)
1629                         goto out;
1630
1631                 r = -EFAULT;
1632                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1633                         goto out;
1634                 r = 0;
1635                 break;
1636         }
1637         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1638                 u64 mce_cap;
1639
1640                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1641                 r = -EFAULT;
1642                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1643                         goto out;
1644                 r = 0;
1645                 break;
1646         }
1647         default:
1648                 r = -EINVAL;
1649         }
1650 out:
1651         return r;
1652 }
1653
1654 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1655 {
1656         kvm_x86_ops->vcpu_load(vcpu, cpu);
1657         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1658                 unsigned long khz = cpufreq_quick_get(cpu);
1659                 if (!khz)
1660                         khz = tsc_khz;
1661                 per_cpu(cpu_tsc_khz, cpu) = khz;
1662         }
1663         kvm_request_guest_time_update(vcpu);
1664 }
1665
1666 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1667 {
1668         kvm_put_guest_fpu(vcpu);
1669         kvm_x86_ops->vcpu_put(vcpu);
1670 }
1671
1672 static int is_efer_nx(void)
1673 {
1674         unsigned long long efer = 0;
1675
1676         rdmsrl_safe(MSR_EFER, &efer);
1677         return efer & EFER_NX;
1678 }
1679
1680 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1681 {
1682         int i;
1683         struct kvm_cpuid_entry2 *e, *entry;
1684
1685         entry = NULL;
1686         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1687                 e = &vcpu->arch.cpuid_entries[i];
1688                 if (e->function == 0x80000001) {
1689                         entry = e;
1690                         break;
1691                 }
1692         }
1693         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1694                 entry->edx &= ~(1 << 20);
1695                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1696         }
1697 }
1698
1699 /* when an old userspace process fills a new kernel module */
1700 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1701                                     struct kvm_cpuid *cpuid,
1702                                     struct kvm_cpuid_entry __user *entries)
1703 {
1704         int r, i;
1705         struct kvm_cpuid_entry *cpuid_entries;
1706
1707         r = -E2BIG;
1708         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1709                 goto out;
1710         r = -ENOMEM;
1711         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1712         if (!cpuid_entries)
1713                 goto out;
1714         r = -EFAULT;
1715         if (copy_from_user(cpuid_entries, entries,
1716                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1717                 goto out_free;
1718         for (i = 0; i < cpuid->nent; i++) {
1719                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1720                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1721                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1722                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1723                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1724                 vcpu->arch.cpuid_entries[i].index = 0;
1725                 vcpu->arch.cpuid_entries[i].flags = 0;
1726                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1727                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1728                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1729         }
1730         vcpu->arch.cpuid_nent = cpuid->nent;
1731         cpuid_fix_nx_cap(vcpu);
1732         r = 0;
1733         kvm_apic_set_version(vcpu);
1734         kvm_x86_ops->cpuid_update(vcpu);
1735
1736 out_free:
1737         vfree(cpuid_entries);
1738 out:
1739         return r;
1740 }
1741
1742 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1743                                      struct kvm_cpuid2 *cpuid,
1744                                      struct kvm_cpuid_entry2 __user *entries)
1745 {
1746         int r;
1747
1748         r = -E2BIG;
1749         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1750                 goto out;
1751         r = -EFAULT;
1752         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1753                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1754                 goto out;
1755         vcpu->arch.cpuid_nent = cpuid->nent;
1756         kvm_apic_set_version(vcpu);
1757         kvm_x86_ops->cpuid_update(vcpu);
1758         return 0;
1759
1760 out:
1761         return r;
1762 }
1763
1764 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1765                                      struct kvm_cpuid2 *cpuid,
1766                                      struct kvm_cpuid_entry2 __user *entries)
1767 {
1768         int r;
1769
1770         r = -E2BIG;
1771         if (cpuid->nent < vcpu->arch.cpuid_nent)
1772                 goto out;
1773         r = -EFAULT;
1774         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1775                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1776                 goto out;
1777         return 0;
1778
1779 out:
1780         cpuid->nent = vcpu->arch.cpuid_nent;
1781         return r;
1782 }
1783
1784 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1785                            u32 index)
1786 {
1787         entry->function = function;
1788         entry->index = index;
1789         cpuid_count(entry->function, entry->index,
1790                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1791         entry->flags = 0;
1792 }
1793
1794 #define F(x) bit(X86_FEATURE_##x)
1795
1796 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1797                          u32 index, int *nent, int maxnent)
1798 {
1799         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1800 #ifdef CONFIG_X86_64
1801         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1802                                 ? F(GBPAGES) : 0;
1803         unsigned f_lm = F(LM);
1804 #else
1805         unsigned f_gbpages = 0;
1806         unsigned f_lm = 0;
1807 #endif
1808         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1809
1810         /* cpuid 1.edx */
1811         const u32 kvm_supported_word0_x86_features =
1812                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1813                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1814                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1815                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1816                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1817                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1818                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1819                 0 /* HTT, TM, Reserved, PBE */;
1820         /* cpuid 0x80000001.edx */
1821         const u32 kvm_supported_word1_x86_features =
1822                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1823                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1824                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1825                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1826                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1827                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1828                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1829                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1830         /* cpuid 1.ecx */
1831         const u32 kvm_supported_word4_x86_features =
1832                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1833                 0 /* DS-CPL, VMX, SMX, EST */ |
1834                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1835                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1836                 0 /* Reserved, DCA */ | F(XMM4_1) |
1837                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1838                 0 /* Reserved, XSAVE, OSXSAVE */;
1839         /* cpuid 0x80000001.ecx */
1840         const u32 kvm_supported_word6_x86_features =
1841                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1842                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1843                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1844                 0 /* SKINIT */ | 0 /* WDT */;
1845
1846         /* all calls to cpuid_count() should be made on the same cpu */
1847         get_cpu();
1848         do_cpuid_1_ent(entry, function, index);
1849         ++*nent;
1850
1851         switch (function) {
1852         case 0:
1853                 entry->eax = min(entry->eax, (u32)0xb);
1854                 break;
1855         case 1:
1856                 entry->edx &= kvm_supported_word0_x86_features;
1857                 entry->ecx &= kvm_supported_word4_x86_features;
1858                 /* we support x2apic emulation even if host does not support
1859                  * it since we emulate x2apic in software */
1860                 entry->ecx |= F(X2APIC);
1861                 break;
1862         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1863          * may return different values. This forces us to get_cpu() before
1864          * issuing the first command, and also to emulate this annoying behavior
1865          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1866         case 2: {
1867                 int t, times = entry->eax & 0xff;
1868
1869                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1870                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1871                 for (t = 1; t < times && *nent < maxnent; ++t) {
1872                         do_cpuid_1_ent(&entry[t], function, 0);
1873                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1874                         ++*nent;
1875                 }
1876                 break;
1877         }
1878         /* function 4 and 0xb have additional index. */
1879         case 4: {
1880                 int i, cache_type;
1881
1882                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1883                 /* read more entries until cache_type is zero */
1884                 for (i = 1; *nent < maxnent; ++i) {
1885                         cache_type = entry[i - 1].eax & 0x1f;
1886                         if (!cache_type)
1887                                 break;
1888                         do_cpuid_1_ent(&entry[i], function, i);
1889                         entry[i].flags |=
1890                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1891                         ++*nent;
1892                 }
1893                 break;
1894         }
1895         case 0xb: {
1896                 int i, level_type;
1897
1898                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1899                 /* read more entries until level_type is zero */
1900                 for (i = 1; *nent < maxnent; ++i) {
1901                         level_type = entry[i - 1].ecx & 0xff00;
1902                         if (!level_type)
1903                                 break;
1904                         do_cpuid_1_ent(&entry[i], function, i);
1905                         entry[i].flags |=
1906                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1907                         ++*nent;
1908                 }
1909                 break;
1910         }
1911         case 0x80000000:
1912                 entry->eax = min(entry->eax, 0x8000001a);
1913                 break;
1914         case 0x80000001:
1915                 entry->edx &= kvm_supported_word1_x86_features;
1916                 entry->ecx &= kvm_supported_word6_x86_features;
1917                 break;
1918         }
1919         put_cpu();
1920 }
1921
1922 #undef F
1923
1924 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1925                                      struct kvm_cpuid_entry2 __user *entries)
1926 {
1927         struct kvm_cpuid_entry2 *cpuid_entries;
1928         int limit, nent = 0, r = -E2BIG;
1929         u32 func;
1930
1931         if (cpuid->nent < 1)
1932                 goto out;
1933         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1934                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1935         r = -ENOMEM;
1936         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1937         if (!cpuid_entries)
1938                 goto out;
1939
1940         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1941         limit = cpuid_entries[0].eax;
1942         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1943                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1944                              &nent, cpuid->nent);
1945         r = -E2BIG;
1946         if (nent >= cpuid->nent)
1947                 goto out_free;
1948
1949         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1950         limit = cpuid_entries[nent - 1].eax;
1951         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1952                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1953                              &nent, cpuid->nent);
1954         r = -E2BIG;
1955         if (nent >= cpuid->nent)
1956                 goto out_free;
1957
1958         r = -EFAULT;
1959         if (copy_to_user(entries, cpuid_entries,
1960                          nent * sizeof(struct kvm_cpuid_entry2)))
1961                 goto out_free;
1962         cpuid->nent = nent;
1963         r = 0;
1964
1965 out_free:
1966         vfree(cpuid_entries);
1967 out:
1968         return r;
1969 }
1970
1971 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1972                                     struct kvm_lapic_state *s)
1973 {
1974         vcpu_load(vcpu);
1975         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1976         vcpu_put(vcpu);
1977
1978         return 0;
1979 }
1980
1981 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1982                                     struct kvm_lapic_state *s)
1983 {
1984         vcpu_load(vcpu);
1985         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1986         kvm_apic_post_state_restore(vcpu);
1987         update_cr8_intercept(vcpu);
1988         vcpu_put(vcpu);
1989
1990         return 0;
1991 }
1992
1993 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1994                                     struct kvm_interrupt *irq)
1995 {
1996         if (irq->irq < 0 || irq->irq >= 256)
1997                 return -EINVAL;
1998         if (irqchip_in_kernel(vcpu->kvm))
1999                 return -ENXIO;
2000         vcpu_load(vcpu);
2001
2002         kvm_queue_interrupt(vcpu, irq->irq, false);
2003
2004         vcpu_put(vcpu);
2005
2006         return 0;
2007 }
2008
2009 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2010 {
2011         vcpu_load(vcpu);
2012         kvm_inject_nmi(vcpu);
2013         vcpu_put(vcpu);
2014
2015         return 0;
2016 }
2017
2018 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2019                                            struct kvm_tpr_access_ctl *tac)
2020 {
2021         if (tac->flags)
2022                 return -EINVAL;
2023         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2024         return 0;
2025 }
2026
2027 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2028                                         u64 mcg_cap)
2029 {
2030         int r;
2031         unsigned bank_num = mcg_cap & 0xff, bank;
2032
2033         r = -EINVAL;
2034         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2035                 goto out;
2036         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2037                 goto out;
2038         r = 0;
2039         vcpu->arch.mcg_cap = mcg_cap;
2040         /* Init IA32_MCG_CTL to all 1s */
2041         if (mcg_cap & MCG_CTL_P)
2042                 vcpu->arch.mcg_ctl = ~(u64)0;
2043         /* Init IA32_MCi_CTL to all 1s */
2044         for (bank = 0; bank < bank_num; bank++)
2045                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2046 out:
2047         return r;
2048 }
2049
2050 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2051                                       struct kvm_x86_mce *mce)
2052 {
2053         u64 mcg_cap = vcpu->arch.mcg_cap;
2054         unsigned bank_num = mcg_cap & 0xff;
2055         u64 *banks = vcpu->arch.mce_banks;
2056
2057         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2058                 return -EINVAL;
2059         /*
2060          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2061          * reporting is disabled
2062          */
2063         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2064             vcpu->arch.mcg_ctl != ~(u64)0)
2065                 return 0;
2066         banks += 4 * mce->bank;
2067         /*
2068          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2069          * reporting is disabled for the bank
2070          */
2071         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2072                 return 0;
2073         if (mce->status & MCI_STATUS_UC) {
2074                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2075                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2076                         printk(KERN_DEBUG "kvm: set_mce: "
2077                                "injects mce exception while "
2078                                "previous one is in progress!\n");
2079                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2080                         return 0;
2081                 }
2082                 if (banks[1] & MCI_STATUS_VAL)
2083                         mce->status |= MCI_STATUS_OVER;
2084                 banks[2] = mce->addr;
2085                 banks[3] = mce->misc;
2086                 vcpu->arch.mcg_status = mce->mcg_status;
2087                 banks[1] = mce->status;
2088                 kvm_queue_exception(vcpu, MC_VECTOR);
2089         } else if (!(banks[1] & MCI_STATUS_VAL)
2090                    || !(banks[1] & MCI_STATUS_UC)) {
2091                 if (banks[1] & MCI_STATUS_VAL)
2092                         mce->status |= MCI_STATUS_OVER;
2093                 banks[2] = mce->addr;
2094                 banks[3] = mce->misc;
2095                 banks[1] = mce->status;
2096         } else
2097                 banks[1] |= MCI_STATUS_OVER;
2098         return 0;
2099 }
2100
2101 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2102                                                struct kvm_vcpu_events *events)
2103 {
2104         vcpu_load(vcpu);
2105
2106         events->exception.injected =
2107                 vcpu->arch.exception.pending &&
2108                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2109         events->exception.nr = vcpu->arch.exception.nr;
2110         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2111         events->exception.error_code = vcpu->arch.exception.error_code;
2112
2113         events->interrupt.injected =
2114                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2115         events->interrupt.nr = vcpu->arch.interrupt.nr;
2116         events->interrupt.soft = 0;
2117         events->interrupt.shadow =
2118                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2119                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2120
2121         events->nmi.injected = vcpu->arch.nmi_injected;
2122         events->nmi.pending = vcpu->arch.nmi_pending;
2123         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2124
2125         events->sipi_vector = vcpu->arch.sipi_vector;
2126
2127         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2128                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2129                          | KVM_VCPUEVENT_VALID_SHADOW);
2130
2131         vcpu_put(vcpu);
2132 }
2133
2134 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2135                                               struct kvm_vcpu_events *events)
2136 {
2137         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2138                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2139                               | KVM_VCPUEVENT_VALID_SHADOW))
2140                 return -EINVAL;
2141
2142         vcpu_load(vcpu);
2143
2144         vcpu->arch.exception.pending = events->exception.injected;
2145         vcpu->arch.exception.nr = events->exception.nr;
2146         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2147         vcpu->arch.exception.error_code = events->exception.error_code;
2148
2149         vcpu->arch.interrupt.pending = events->interrupt.injected;
2150         vcpu->arch.interrupt.nr = events->interrupt.nr;
2151         vcpu->arch.interrupt.soft = events->interrupt.soft;
2152         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2153                 kvm_pic_clear_isr_ack(vcpu->kvm);
2154         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2155                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2156                                                   events->interrupt.shadow);
2157
2158         vcpu->arch.nmi_injected = events->nmi.injected;
2159         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2160                 vcpu->arch.nmi_pending = events->nmi.pending;
2161         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2162
2163         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2164                 vcpu->arch.sipi_vector = events->sipi_vector;
2165
2166         vcpu_put(vcpu);
2167
2168         return 0;
2169 }
2170
2171 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2172                                              struct kvm_debugregs *dbgregs)
2173 {
2174         vcpu_load(vcpu);
2175
2176         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2177         dbgregs->dr6 = vcpu->arch.dr6;
2178         dbgregs->dr7 = vcpu->arch.dr7;
2179         dbgregs->flags = 0;
2180
2181         vcpu_put(vcpu);
2182 }
2183
2184 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2185                                             struct kvm_debugregs *dbgregs)
2186 {
2187         if (dbgregs->flags)
2188                 return -EINVAL;
2189
2190         vcpu_load(vcpu);
2191
2192         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2193         vcpu->arch.dr6 = dbgregs->dr6;
2194         vcpu->arch.dr7 = dbgregs->dr7;
2195
2196         vcpu_put(vcpu);
2197
2198         return 0;
2199 }
2200
2201 long kvm_arch_vcpu_ioctl(struct file *filp,
2202                          unsigned int ioctl, unsigned long arg)
2203 {
2204         struct kvm_vcpu *vcpu = filp->private_data;
2205         void __user *argp = (void __user *)arg;
2206         int r;
2207         struct kvm_lapic_state *lapic = NULL;
2208
2209         switch (ioctl) {
2210         case KVM_GET_LAPIC: {
2211                 r = -EINVAL;
2212                 if (!vcpu->arch.apic)
2213                         goto out;
2214                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2215
2216                 r = -ENOMEM;
2217                 if (!lapic)
2218                         goto out;
2219                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2220                 if (r)
2221                         goto out;
2222                 r = -EFAULT;
2223                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2224                         goto out;
2225                 r = 0;
2226                 break;
2227         }
2228         case KVM_SET_LAPIC: {
2229                 r = -EINVAL;
2230                 if (!vcpu->arch.apic)
2231                         goto out;
2232                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2233                 r = -ENOMEM;
2234                 if (!lapic)
2235                         goto out;
2236                 r = -EFAULT;
2237                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2238                         goto out;
2239                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2240                 if (r)
2241                         goto out;
2242                 r = 0;
2243                 break;
2244         }
2245         case KVM_INTERRUPT: {
2246                 struct kvm_interrupt irq;
2247
2248                 r = -EFAULT;
2249                 if (copy_from_user(&irq, argp, sizeof irq))
2250                         goto out;
2251                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2252                 if (r)
2253                         goto out;
2254                 r = 0;
2255                 break;
2256         }
2257         case KVM_NMI: {
2258                 r = kvm_vcpu_ioctl_nmi(vcpu);
2259                 if (r)
2260                         goto out;
2261                 r = 0;
2262                 break;
2263         }
2264         case KVM_SET_CPUID: {
2265                 struct kvm_cpuid __user *cpuid_arg = argp;
2266                 struct kvm_cpuid cpuid;
2267
2268                 r = -EFAULT;
2269                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2270                         goto out;
2271                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2272                 if (r)
2273                         goto out;
2274                 break;
2275         }
2276         case KVM_SET_CPUID2: {
2277                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2278                 struct kvm_cpuid2 cpuid;
2279
2280                 r = -EFAULT;
2281                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2282                         goto out;
2283                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2284                                               cpuid_arg->entries);
2285                 if (r)
2286                         goto out;
2287                 break;
2288         }
2289         case KVM_GET_CPUID2: {
2290                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2291                 struct kvm_cpuid2 cpuid;
2292
2293                 r = -EFAULT;
2294                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2295                         goto out;
2296                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2297                                               cpuid_arg->entries);
2298                 if (r)
2299                         goto out;
2300                 r = -EFAULT;
2301                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2302                         goto out;
2303                 r = 0;
2304                 break;
2305         }
2306         case KVM_GET_MSRS:
2307                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2308                 break;
2309         case KVM_SET_MSRS:
2310                 r = msr_io(vcpu, argp, do_set_msr, 0);
2311                 break;
2312         case KVM_TPR_ACCESS_REPORTING: {
2313                 struct kvm_tpr_access_ctl tac;
2314
2315                 r = -EFAULT;
2316                 if (copy_from_user(&tac, argp, sizeof tac))
2317                         goto out;
2318                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2319                 if (r)
2320                         goto out;
2321                 r = -EFAULT;
2322                 if (copy_to_user(argp, &tac, sizeof tac))
2323                         goto out;
2324                 r = 0;
2325                 break;
2326         };
2327         case KVM_SET_VAPIC_ADDR: {
2328                 struct kvm_vapic_addr va;
2329
2330                 r = -EINVAL;
2331                 if (!irqchip_in_kernel(vcpu->kvm))
2332                         goto out;
2333                 r = -EFAULT;
2334                 if (copy_from_user(&va, argp, sizeof va))
2335                         goto out;
2336                 r = 0;
2337                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2338                 break;
2339         }
2340         case KVM_X86_SETUP_MCE: {
2341                 u64 mcg_cap;
2342
2343                 r = -EFAULT;
2344                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2345                         goto out;
2346                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2347                 break;
2348         }
2349         case KVM_X86_SET_MCE: {
2350                 struct kvm_x86_mce mce;
2351
2352                 r = -EFAULT;
2353                 if (copy_from_user(&mce, argp, sizeof mce))
2354                         goto out;
2355                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2356                 break;
2357         }
2358         case KVM_GET_VCPU_EVENTS: {
2359                 struct kvm_vcpu_events events;
2360
2361                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2362
2363                 r = -EFAULT;
2364                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2365                         break;
2366                 r = 0;
2367                 break;
2368         }
2369         case KVM_SET_VCPU_EVENTS: {
2370                 struct kvm_vcpu_events events;
2371
2372                 r = -EFAULT;
2373                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2374                         break;
2375
2376                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2377                 break;
2378         }
2379         case KVM_GET_DEBUGREGS: {
2380                 struct kvm_debugregs dbgregs;
2381
2382                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2383
2384                 r = -EFAULT;
2385                 if (copy_to_user(argp, &dbgregs,
2386                                  sizeof(struct kvm_debugregs)))
2387                         break;
2388                 r = 0;
2389                 break;
2390         }
2391         case KVM_SET_DEBUGREGS: {
2392                 struct kvm_debugregs dbgregs;
2393
2394                 r = -EFAULT;
2395                 if (copy_from_user(&dbgregs, argp,
2396                                    sizeof(struct kvm_debugregs)))
2397                         break;
2398
2399                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2400                 break;
2401         }
2402         default:
2403                 r = -EINVAL;
2404         }
2405 out:
2406         kfree(lapic);
2407         return r;
2408 }
2409
2410 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2411 {
2412         int ret;
2413
2414         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2415                 return -1;
2416         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2417         return ret;
2418 }
2419
2420 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2421                                               u64 ident_addr)
2422 {
2423         kvm->arch.ept_identity_map_addr = ident_addr;
2424         return 0;
2425 }
2426
2427 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2428                                           u32 kvm_nr_mmu_pages)
2429 {
2430         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2431                 return -EINVAL;
2432
2433         mutex_lock(&kvm->slots_lock);
2434         spin_lock(&kvm->mmu_lock);
2435
2436         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2437         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2438
2439         spin_unlock(&kvm->mmu_lock);
2440         mutex_unlock(&kvm->slots_lock);
2441         return 0;
2442 }
2443
2444 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2445 {
2446         return kvm->arch.n_alloc_mmu_pages;
2447 }
2448
2449 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2450 {
2451         int i;
2452         struct kvm_mem_alias *alias;
2453         struct kvm_mem_aliases *aliases;
2454
2455         aliases = rcu_dereference(kvm->arch.aliases);
2456
2457         for (i = 0; i < aliases->naliases; ++i) {
2458                 alias = &aliases->aliases[i];
2459                 if (alias->flags & KVM_ALIAS_INVALID)
2460                         continue;
2461                 if (gfn >= alias->base_gfn
2462                     && gfn < alias->base_gfn + alias->npages)
2463                         return alias->target_gfn + gfn - alias->base_gfn;
2464         }
2465         return gfn;
2466 }
2467
2468 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2469 {
2470         int i;
2471         struct kvm_mem_alias *alias;
2472         struct kvm_mem_aliases *aliases;
2473
2474         aliases = rcu_dereference(kvm->arch.aliases);
2475
2476         for (i = 0; i < aliases->naliases; ++i) {
2477                 alias = &aliases->aliases[i];
2478                 if (gfn >= alias->base_gfn
2479                     && gfn < alias->base_gfn + alias->npages)
2480                         return alias->target_gfn + gfn - alias->base_gfn;
2481         }
2482         return gfn;
2483 }
2484
2485 /*
2486  * Set a new alias region.  Aliases map a portion of physical memory into
2487  * another portion.  This is useful for memory windows, for example the PC
2488  * VGA region.
2489  */
2490 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2491                                          struct kvm_memory_alias *alias)
2492 {
2493         int r, n;
2494         struct kvm_mem_alias *p;
2495         struct kvm_mem_aliases *aliases, *old_aliases;
2496
2497         r = -EINVAL;
2498         /* General sanity checks */
2499         if (alias->memory_size & (PAGE_SIZE - 1))
2500                 goto out;
2501         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2502                 goto out;
2503         if (alias->slot >= KVM_ALIAS_SLOTS)
2504                 goto out;
2505         if (alias->guest_phys_addr + alias->memory_size
2506             < alias->guest_phys_addr)
2507                 goto out;
2508         if (alias->target_phys_addr + alias->memory_size
2509             < alias->target_phys_addr)
2510                 goto out;
2511
2512         r = -ENOMEM;
2513         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2514         if (!aliases)
2515                 goto out;
2516
2517         mutex_lock(&kvm->slots_lock);
2518
2519         /* invalidate any gfn reference in case of deletion/shrinking */
2520         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2521         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2522         old_aliases = kvm->arch.aliases;
2523         rcu_assign_pointer(kvm->arch.aliases, aliases);
2524         synchronize_srcu_expedited(&kvm->srcu);
2525         kvm_mmu_zap_all(kvm);
2526         kfree(old_aliases);
2527
2528         r = -ENOMEM;
2529         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2530         if (!aliases)
2531                 goto out_unlock;
2532
2533         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2534
2535         p = &aliases->aliases[alias->slot];
2536         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2537         p->npages = alias->memory_size >> PAGE_SHIFT;
2538         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2539         p->flags &= ~(KVM_ALIAS_INVALID);
2540
2541         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2542                 if (aliases->aliases[n - 1].npages)
2543                         break;
2544         aliases->naliases = n;
2545
2546         old_aliases = kvm->arch.aliases;
2547         rcu_assign_pointer(kvm->arch.aliases, aliases);
2548         synchronize_srcu_expedited(&kvm->srcu);
2549         kfree(old_aliases);
2550         r = 0;
2551
2552 out_unlock:
2553         mutex_unlock(&kvm->slots_lock);
2554 out:
2555         return r;
2556 }
2557
2558 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2559 {
2560         int r;
2561
2562         r = 0;
2563         switch (chip->chip_id) {
2564         case KVM_IRQCHIP_PIC_MASTER:
2565                 memcpy(&chip->chip.pic,
2566                         &pic_irqchip(kvm)->pics[0],
2567                         sizeof(struct kvm_pic_state));
2568                 break;
2569         case KVM_IRQCHIP_PIC_SLAVE:
2570                 memcpy(&chip->chip.pic,
2571                         &pic_irqchip(kvm)->pics[1],
2572                         sizeof(struct kvm_pic_state));
2573                 break;
2574         case KVM_IRQCHIP_IOAPIC:
2575                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2576                 break;
2577         default:
2578                 r = -EINVAL;
2579                 break;
2580         }
2581         return r;
2582 }
2583
2584 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2585 {
2586         int r;
2587
2588         r = 0;
2589         switch (chip->chip_id) {
2590         case KVM_IRQCHIP_PIC_MASTER:
2591                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2592                 memcpy(&pic_irqchip(kvm)->pics[0],
2593                         &chip->chip.pic,
2594                         sizeof(struct kvm_pic_state));
2595                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2596                 break;
2597         case KVM_IRQCHIP_PIC_SLAVE:
2598                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2599                 memcpy(&pic_irqchip(kvm)->pics[1],
2600                         &chip->chip.pic,
2601                         sizeof(struct kvm_pic_state));
2602                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2603                 break;
2604         case KVM_IRQCHIP_IOAPIC:
2605                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2606                 break;
2607         default:
2608                 r = -EINVAL;
2609                 break;
2610         }
2611         kvm_pic_update_irq(pic_irqchip(kvm));
2612         return r;
2613 }
2614
2615 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2616 {
2617         int r = 0;
2618
2619         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2620         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2621         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2622         return r;
2623 }
2624
2625 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2626 {
2627         int r = 0;
2628
2629         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2630         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2631         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2632         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2633         return r;
2634 }
2635
2636 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2637 {
2638         int r = 0;
2639
2640         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2641         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2642                 sizeof(ps->channels));
2643         ps->flags = kvm->arch.vpit->pit_state.flags;
2644         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2645         return r;
2646 }
2647
2648 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2649 {
2650         int r = 0, start = 0;
2651         u32 prev_legacy, cur_legacy;
2652         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2653         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2654         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2655         if (!prev_legacy && cur_legacy)
2656                 start = 1;
2657         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2658                sizeof(kvm->arch.vpit->pit_state.channels));
2659         kvm->arch.vpit->pit_state.flags = ps->flags;
2660         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2661         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2662         return r;
2663 }
2664
2665 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2666                                  struct kvm_reinject_control *control)
2667 {
2668         if (!kvm->arch.vpit)
2669                 return -ENXIO;
2670         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2671         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2672         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2673         return 0;
2674 }
2675
2676 /*
2677  * Get (and clear) the dirty memory log for a memory slot.
2678  */
2679 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2680                                       struct kvm_dirty_log *log)
2681 {
2682         int r, i;
2683         struct kvm_memory_slot *memslot;
2684         unsigned long n;
2685         unsigned long is_dirty = 0;
2686         unsigned long *dirty_bitmap = NULL;
2687
2688         mutex_lock(&kvm->slots_lock);
2689
2690         r = -EINVAL;
2691         if (log->slot >= KVM_MEMORY_SLOTS)
2692                 goto out;
2693
2694         memslot = &kvm->memslots->memslots[log->slot];
2695         r = -ENOENT;
2696         if (!memslot->dirty_bitmap)
2697                 goto out;
2698
2699         n = kvm_dirty_bitmap_bytes(memslot);
2700
2701         r = -ENOMEM;
2702         dirty_bitmap = vmalloc(n);
2703         if (!dirty_bitmap)
2704                 goto out;
2705         memset(dirty_bitmap, 0, n);
2706
2707         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2708                 is_dirty = memslot->dirty_bitmap[i];
2709
2710         /* If nothing is dirty, don't bother messing with page tables. */
2711         if (is_dirty) {
2712                 struct kvm_memslots *slots, *old_slots;
2713
2714                 spin_lock(&kvm->mmu_lock);
2715                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2716                 spin_unlock(&kvm->mmu_lock);
2717
2718                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2719                 if (!slots)
2720                         goto out_free;
2721
2722                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2723                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2724
2725                 old_slots = kvm->memslots;
2726                 rcu_assign_pointer(kvm->memslots, slots);
2727                 synchronize_srcu_expedited(&kvm->srcu);
2728                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2729                 kfree(old_slots);
2730         }
2731
2732         r = 0;
2733         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2734                 r = -EFAULT;
2735 out_free:
2736         vfree(dirty_bitmap);
2737 out:
2738         mutex_unlock(&kvm->slots_lock);
2739         return r;
2740 }
2741
2742 long kvm_arch_vm_ioctl(struct file *filp,
2743                        unsigned int ioctl, unsigned long arg)
2744 {
2745         struct kvm *kvm = filp->private_data;
2746         void __user *argp = (void __user *)arg;
2747         int r = -ENOTTY;
2748         /*
2749          * This union makes it completely explicit to gcc-3.x
2750          * that these two variables' stack usage should be
2751          * combined, not added together.
2752          */
2753         union {
2754                 struct kvm_pit_state ps;
2755                 struct kvm_pit_state2 ps2;
2756                 struct kvm_memory_alias alias;
2757                 struct kvm_pit_config pit_config;
2758         } u;
2759
2760         switch (ioctl) {
2761         case KVM_SET_TSS_ADDR:
2762                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2763                 if (r < 0)
2764                         goto out;
2765                 break;
2766         case KVM_SET_IDENTITY_MAP_ADDR: {
2767                 u64 ident_addr;
2768
2769                 r = -EFAULT;
2770                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2771                         goto out;
2772                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2773                 if (r < 0)
2774                         goto out;
2775                 break;
2776         }
2777         case KVM_SET_MEMORY_REGION: {
2778                 struct kvm_memory_region kvm_mem;
2779                 struct kvm_userspace_memory_region kvm_userspace_mem;
2780
2781                 r = -EFAULT;
2782                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2783                         goto out;
2784                 kvm_userspace_mem.slot = kvm_mem.slot;
2785                 kvm_userspace_mem.flags = kvm_mem.flags;
2786                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2787                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2788                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2789                 if (r)
2790                         goto out;
2791                 break;
2792         }
2793         case KVM_SET_NR_MMU_PAGES:
2794                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2795                 if (r)
2796                         goto out;
2797                 break;
2798         case KVM_GET_NR_MMU_PAGES:
2799                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2800                 break;
2801         case KVM_SET_MEMORY_ALIAS:
2802                 r = -EFAULT;
2803                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2804                         goto out;
2805                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2806                 if (r)
2807                         goto out;
2808                 break;
2809         case KVM_CREATE_IRQCHIP: {
2810                 struct kvm_pic *vpic;
2811
2812                 mutex_lock(&kvm->lock);
2813                 r = -EEXIST;
2814                 if (kvm->arch.vpic)
2815                         goto create_irqchip_unlock;
2816                 r = -ENOMEM;
2817                 vpic = kvm_create_pic(kvm);
2818                 if (vpic) {
2819                         r = kvm_ioapic_init(kvm);
2820                         if (r) {
2821                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2822                                                           &vpic->dev);
2823                                 kfree(vpic);
2824                                 goto create_irqchip_unlock;
2825                         }
2826                 } else
2827                         goto create_irqchip_unlock;
2828                 smp_wmb();
2829                 kvm->arch.vpic = vpic;
2830                 smp_wmb();
2831                 r = kvm_setup_default_irq_routing(kvm);
2832                 if (r) {
2833                         mutex_lock(&kvm->irq_lock);
2834                         kvm_ioapic_destroy(kvm);
2835                         kvm_destroy_pic(kvm);
2836                         mutex_unlock(&kvm->irq_lock);
2837                 }
2838         create_irqchip_unlock:
2839                 mutex_unlock(&kvm->lock);
2840                 break;
2841         }
2842         case KVM_CREATE_PIT:
2843                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2844                 goto create_pit;
2845         case KVM_CREATE_PIT2:
2846                 r = -EFAULT;
2847                 if (copy_from_user(&u.pit_config, argp,
2848                                    sizeof(struct kvm_pit_config)))
2849                         goto out;
2850         create_pit:
2851                 mutex_lock(&kvm->slots_lock);
2852                 r = -EEXIST;
2853                 if (kvm->arch.vpit)
2854                         goto create_pit_unlock;
2855                 r = -ENOMEM;
2856                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2857                 if (kvm->arch.vpit)
2858                         r = 0;
2859         create_pit_unlock:
2860                 mutex_unlock(&kvm->slots_lock);
2861                 break;
2862         case KVM_IRQ_LINE_STATUS:
2863         case KVM_IRQ_LINE: {
2864                 struct kvm_irq_level irq_event;
2865
2866                 r = -EFAULT;
2867                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2868                         goto out;
2869                 if (irqchip_in_kernel(kvm)) {
2870                         __s32 status;
2871                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2872                                         irq_event.irq, irq_event.level);
2873                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2874                                 irq_event.status = status;
2875                                 if (copy_to_user(argp, &irq_event,
2876                                                         sizeof irq_event))
2877                                         goto out;
2878                         }
2879                         r = 0;
2880                 }
2881                 break;
2882         }
2883         case KVM_GET_IRQCHIP: {
2884                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2885                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2886
2887                 r = -ENOMEM;
2888                 if (!chip)
2889                         goto out;
2890                 r = -EFAULT;
2891                 if (copy_from_user(chip, argp, sizeof *chip))
2892                         goto get_irqchip_out;
2893                 r = -ENXIO;
2894                 if (!irqchip_in_kernel(kvm))
2895                         goto get_irqchip_out;
2896                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2897                 if (r)
2898                         goto get_irqchip_out;
2899                 r = -EFAULT;
2900                 if (copy_to_user(argp, chip, sizeof *chip))
2901                         goto get_irqchip_out;
2902                 r = 0;
2903         get_irqchip_out:
2904                 kfree(chip);
2905                 if (r)
2906                         goto out;
2907                 break;
2908         }
2909         case KVM_SET_IRQCHIP: {
2910                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2911                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2912
2913                 r = -ENOMEM;
2914                 if (!chip)
2915                         goto out;
2916                 r = -EFAULT;
2917                 if (copy_from_user(chip, argp, sizeof *chip))
2918                         goto set_irqchip_out;
2919                 r = -ENXIO;
2920                 if (!irqchip_in_kernel(kvm))
2921                         goto set_irqchip_out;
2922                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2923                 if (r)
2924                         goto set_irqchip_out;
2925                 r = 0;
2926         set_irqchip_out:
2927                 kfree(chip);
2928                 if (r)
2929                         goto out;
2930                 break;
2931         }
2932         case KVM_GET_PIT: {
2933                 r = -EFAULT;
2934                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2935                         goto out;
2936                 r = -ENXIO;
2937                 if (!kvm->arch.vpit)
2938                         goto out;
2939                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2940                 if (r)
2941                         goto out;
2942                 r = -EFAULT;
2943                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2944                         goto out;
2945                 r = 0;
2946                 break;
2947         }
2948         case KVM_SET_PIT: {
2949                 r = -EFAULT;
2950                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2951                         goto out;
2952                 r = -ENXIO;
2953                 if (!kvm->arch.vpit)
2954                         goto out;
2955                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2956                 if (r)
2957                         goto out;
2958                 r = 0;
2959                 break;
2960         }
2961         case KVM_GET_PIT2: {
2962                 r = -ENXIO;
2963                 if (!kvm->arch.vpit)
2964                         goto out;
2965                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2966                 if (r)
2967                         goto out;
2968                 r = -EFAULT;
2969                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2970                         goto out;
2971                 r = 0;
2972                 break;
2973         }
2974         case KVM_SET_PIT2: {
2975                 r = -EFAULT;
2976                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2977                         goto out;
2978                 r = -ENXIO;
2979                 if (!kvm->arch.vpit)
2980                         goto out;
2981                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2982                 if (r)
2983                         goto out;
2984                 r = 0;
2985                 break;
2986         }
2987         case KVM_REINJECT_CONTROL: {
2988                 struct kvm_reinject_control control;
2989                 r =  -EFAULT;
2990                 if (copy_from_user(&control, argp, sizeof(control)))
2991                         goto out;
2992                 r = kvm_vm_ioctl_reinject(kvm, &control);
2993                 if (r)
2994                         goto out;
2995                 r = 0;
2996                 break;
2997         }
2998         case KVM_XEN_HVM_CONFIG: {
2999                 r = -EFAULT;
3000                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3001                                    sizeof(struct kvm_xen_hvm_config)))
3002                         goto out;
3003                 r = -EINVAL;
3004                 if (kvm->arch.xen_hvm_config.flags)
3005                         goto out;
3006                 r = 0;
3007                 break;
3008         }
3009         case KVM_SET_CLOCK: {
3010                 struct timespec now;
3011                 struct kvm_clock_data user_ns;
3012                 u64 now_ns;
3013                 s64 delta;
3014
3015                 r = -EFAULT;
3016                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3017                         goto out;
3018
3019                 r = -EINVAL;
3020                 if (user_ns.flags)
3021                         goto out;
3022
3023                 r = 0;
3024                 ktime_get_ts(&now);
3025                 now_ns = timespec_to_ns(&now);
3026                 delta = user_ns.clock - now_ns;
3027                 kvm->arch.kvmclock_offset = delta;
3028                 break;
3029         }
3030         case KVM_GET_CLOCK: {
3031                 struct timespec now;
3032                 struct kvm_clock_data user_ns;
3033                 u64 now_ns;
3034
3035                 ktime_get_ts(&now);
3036                 now_ns = timespec_to_ns(&now);
3037                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3038                 user_ns.flags = 0;
3039
3040                 r = -EFAULT;
3041                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3042                         goto out;
3043                 r = 0;
3044                 break;
3045         }
3046
3047         default:
3048                 ;
3049         }
3050 out:
3051         return r;
3052 }
3053
3054 static void kvm_init_msr_list(void)
3055 {
3056         u32 dummy[2];
3057         unsigned i, j;
3058
3059         /* skip the first msrs in the list. KVM-specific */
3060         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3061                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3062                         continue;
3063                 if (j < i)
3064                         msrs_to_save[j] = msrs_to_save[i];
3065                 j++;
3066         }
3067         num_msrs_to_save = j;
3068 }
3069
3070 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3071                            const void *v)
3072 {
3073         if (vcpu->arch.apic &&
3074             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3075                 return 0;
3076
3077         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3078 }
3079
3080 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3081 {
3082         if (vcpu->arch.apic &&
3083             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3084                 return 0;
3085
3086         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3087 }
3088
3089 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3090 {
3091         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3092         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3093 }
3094
3095  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3096 {
3097         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3098         access |= PFERR_FETCH_MASK;
3099         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3100 }
3101
3102 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3103 {
3104         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3105         access |= PFERR_WRITE_MASK;
3106         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3107 }
3108
3109 /* uses this to access any guest's mapped memory without checking CPL */
3110 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3111 {
3112         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3113 }
3114
3115 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3116                                       struct kvm_vcpu *vcpu, u32 access,
3117                                       u32 *error)
3118 {
3119         void *data = val;
3120         int r = X86EMUL_CONTINUE;
3121
3122         while (bytes) {
3123                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3124                 unsigned offset = addr & (PAGE_SIZE-1);
3125                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3126                 int ret;
3127
3128                 if (gpa == UNMAPPED_GVA) {
3129                         r = X86EMUL_PROPAGATE_FAULT;
3130                         goto out;
3131                 }
3132                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3133                 if (ret < 0) {
3134                         r = X86EMUL_UNHANDLEABLE;
3135                         goto out;
3136                 }
3137
3138                 bytes -= toread;
3139                 data += toread;
3140                 addr += toread;
3141         }
3142 out:
3143         return r;
3144 }
3145
3146 /* used for instruction fetching */
3147 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3148                                 struct kvm_vcpu *vcpu, u32 *error)
3149 {
3150         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3151         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3152                                           access | PFERR_FETCH_MASK, error);
3153 }
3154
3155 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3156                                struct kvm_vcpu *vcpu, u32 *error)
3157 {
3158         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3159         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3160                                           error);
3161 }
3162
3163 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3164                                struct kvm_vcpu *vcpu, u32 *error)
3165 {
3166         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3167 }
3168
3169 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3170                                 struct kvm_vcpu *vcpu, u32 *error)
3171 {
3172         void *data = val;
3173         int r = X86EMUL_CONTINUE;
3174
3175         while (bytes) {
3176                 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3177                 unsigned offset = addr & (PAGE_SIZE-1);
3178                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3179                 int ret;
3180
3181                 if (gpa == UNMAPPED_GVA) {
3182                         r = X86EMUL_PROPAGATE_FAULT;
3183                         goto out;
3184                 }
3185                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3186                 if (ret < 0) {
3187                         r = X86EMUL_UNHANDLEABLE;
3188                         goto out;
3189                 }
3190
3191                 bytes -= towrite;
3192                 data += towrite;
3193                 addr += towrite;
3194         }
3195 out:
3196         return r;
3197 }
3198
3199
3200 static int emulator_read_emulated(unsigned long addr,
3201                                   void *val,
3202                                   unsigned int bytes,
3203                                   struct kvm_vcpu *vcpu)
3204 {
3205         gpa_t                 gpa;
3206         u32 error_code;
3207
3208         if (vcpu->mmio_read_completed) {
3209                 memcpy(val, vcpu->mmio_data, bytes);
3210                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3211                                vcpu->mmio_phys_addr, *(u64 *)val);
3212                 vcpu->mmio_read_completed = 0;
3213                 return X86EMUL_CONTINUE;
3214         }
3215
3216         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3217
3218         if (gpa == UNMAPPED_GVA) {
3219                 kvm_inject_page_fault(vcpu, addr, error_code);
3220                 return X86EMUL_PROPAGATE_FAULT;
3221         }
3222
3223         /* For APIC access vmexit */
3224         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3225                 goto mmio;
3226
3227         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3228                                 == X86EMUL_CONTINUE)
3229                 return X86EMUL_CONTINUE;
3230
3231 mmio:
3232         /*
3233          * Is this MMIO handled locally?
3234          */
3235         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3236                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3237                 return X86EMUL_CONTINUE;
3238         }
3239
3240         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3241
3242         vcpu->mmio_needed = 1;
3243         vcpu->mmio_phys_addr = gpa;
3244         vcpu->mmio_size = bytes;
3245         vcpu->mmio_is_write = 0;
3246
3247         return X86EMUL_UNHANDLEABLE;
3248 }
3249
3250 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3251                           const void *val, int bytes)
3252 {
3253         int ret;
3254
3255         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3256         if (ret < 0)
3257                 return 0;
3258         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3259         return 1;
3260 }
3261
3262 static int emulator_write_emulated_onepage(unsigned long addr,
3263                                            const void *val,
3264                                            unsigned int bytes,
3265                                            struct kvm_vcpu *vcpu)
3266 {
3267         gpa_t                 gpa;
3268         u32 error_code;
3269
3270         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3271
3272         if (gpa == UNMAPPED_GVA) {
3273                 kvm_inject_page_fault(vcpu, addr, error_code);
3274                 return X86EMUL_PROPAGATE_FAULT;
3275         }
3276
3277         /* For APIC access vmexit */
3278         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3279                 goto mmio;
3280
3281         if (emulator_write_phys(vcpu, gpa, val, bytes))
3282                 return X86EMUL_CONTINUE;
3283
3284 mmio:
3285         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3286         /*
3287          * Is this MMIO handled locally?
3288          */
3289         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3290                 return X86EMUL_CONTINUE;
3291
3292         vcpu->mmio_needed = 1;
3293         vcpu->mmio_phys_addr = gpa;
3294         vcpu->mmio_size = bytes;
3295         vcpu->mmio_is_write = 1;
3296         memcpy(vcpu->mmio_data, val, bytes);
3297
3298         return X86EMUL_CONTINUE;
3299 }
3300
3301 int emulator_write_emulated(unsigned long addr,
3302                                    const void *val,
3303                                    unsigned int bytes,
3304                                    struct kvm_vcpu *vcpu)
3305 {
3306         /* Crossing a page boundary? */
3307         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3308                 int rc, now;
3309
3310                 now = -addr & ~PAGE_MASK;
3311                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3312                 if (rc != X86EMUL_CONTINUE)
3313                         return rc;
3314                 addr += now;
3315                 val += now;
3316                 bytes -= now;
3317         }
3318         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3319 }
3320 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3321
3322 static int emulator_cmpxchg_emulated(unsigned long addr,
3323                                      const void *old,
3324                                      const void *new,
3325                                      unsigned int bytes,
3326                                      struct kvm_vcpu *vcpu)
3327 {
3328         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3329 #ifndef CONFIG_X86_64
3330         /* guests cmpxchg8b have to be emulated atomically */
3331         if (bytes == 8) {
3332                 gpa_t gpa;
3333                 struct page *page;
3334                 char *kaddr;
3335                 u64 val;
3336
3337                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3338
3339                 if (gpa == UNMAPPED_GVA ||
3340                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3341                         goto emul_write;
3342
3343                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3344                         goto emul_write;
3345
3346                 val = *(u64 *)new;
3347
3348                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3349
3350                 kaddr = kmap_atomic(page, KM_USER0);
3351                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3352                 kunmap_atomic(kaddr, KM_USER0);
3353                 kvm_release_page_dirty(page);
3354         }
3355 emul_write:
3356 #endif
3357
3358         return emulator_write_emulated(addr, new, bytes, vcpu);
3359 }
3360
3361 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3362 {
3363         return kvm_x86_ops->get_segment_base(vcpu, seg);
3364 }
3365
3366 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3367 {
3368         kvm_mmu_invlpg(vcpu, address);
3369         return X86EMUL_CONTINUE;
3370 }
3371
3372 int emulate_clts(struct kvm_vcpu *vcpu)
3373 {
3374         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3375         kvm_x86_ops->fpu_activate(vcpu);
3376         return X86EMUL_CONTINUE;
3377 }
3378
3379 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3380 {
3381         return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3382 }
3383
3384 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3385 {
3386         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3387
3388         return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3389 }
3390
3391 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3392 {
3393         u8 opcodes[4];
3394         unsigned long rip = kvm_rip_read(vcpu);
3395         unsigned long rip_linear;
3396
3397         if (!printk_ratelimit())
3398                 return;
3399
3400         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3401
3402         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3403
3404         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3405                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3406 }
3407 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3408
3409 static struct x86_emulate_ops emulate_ops = {
3410         .read_std            = kvm_read_guest_virt_system,
3411         .fetch               = kvm_fetch_guest_virt,
3412         .read_emulated       = emulator_read_emulated,
3413         .write_emulated      = emulator_write_emulated,
3414         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3415 };
3416
3417 static void cache_all_regs(struct kvm_vcpu *vcpu)
3418 {
3419         kvm_register_read(vcpu, VCPU_REGS_RAX);
3420         kvm_register_read(vcpu, VCPU_REGS_RSP);
3421         kvm_register_read(vcpu, VCPU_REGS_RIP);
3422         vcpu->arch.regs_dirty = ~0;
3423 }
3424
3425 int emulate_instruction(struct kvm_vcpu *vcpu,
3426                         unsigned long cr2,
3427                         u16 error_code,
3428                         int emulation_type)
3429 {
3430         int r, shadow_mask;
3431         struct decode_cache *c;
3432         struct kvm_run *run = vcpu->run;
3433
3434         kvm_clear_exception_queue(vcpu);
3435         vcpu->arch.mmio_fault_cr2 = cr2;
3436         /*
3437          * TODO: fix emulate.c to use guest_read/write_register
3438          * instead of direct ->regs accesses, can save hundred cycles
3439          * on Intel for instructions that don't read/change RSP, for
3440          * for example.
3441          */
3442         cache_all_regs(vcpu);
3443
3444         vcpu->mmio_is_write = 0;
3445         vcpu->arch.pio.string = 0;
3446
3447         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3448                 int cs_db, cs_l;
3449                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3450
3451                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3452                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3453                 vcpu->arch.emulate_ctxt.mode =
3454                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3455                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3456                         ? X86EMUL_MODE_VM86 : cs_l
3457                         ? X86EMUL_MODE_PROT64 : cs_db
3458                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3459
3460                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3461
3462                 /* Only allow emulation of specific instructions on #UD
3463                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3464                 c = &vcpu->arch.emulate_ctxt.decode;
3465                 if (emulation_type & EMULTYPE_TRAP_UD) {
3466                         if (!c->twobyte)
3467                                 return EMULATE_FAIL;
3468                         switch (c->b) {
3469                         case 0x01: /* VMMCALL */
3470                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3471                                         return EMULATE_FAIL;
3472                                 break;
3473                         case 0x34: /* sysenter */
3474                         case 0x35: /* sysexit */
3475                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3476                                         return EMULATE_FAIL;
3477                                 break;
3478                         case 0x05: /* syscall */
3479                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3480                                         return EMULATE_FAIL;
3481                                 break;
3482                         default:
3483                                 return EMULATE_FAIL;
3484                         }
3485
3486                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3487                                 return EMULATE_FAIL;
3488                 }
3489
3490                 ++vcpu->stat.insn_emulation;
3491                 if (r)  {
3492                         ++vcpu->stat.insn_emulation_fail;
3493                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3494                                 return EMULATE_DONE;
3495                         return EMULATE_FAIL;
3496                 }
3497         }
3498
3499         if (emulation_type & EMULTYPE_SKIP) {
3500                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3501                 return EMULATE_DONE;
3502         }
3503
3504         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3505         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3506
3507         if (r == 0)
3508                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3509
3510         if (vcpu->arch.pio.string)
3511                 return EMULATE_DO_MMIO;
3512
3513         if (r || vcpu->mmio_is_write) {
3514                 run->exit_reason = KVM_EXIT_MMIO;
3515                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3516                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3517                 run->mmio.len = vcpu->mmio_size;
3518                 run->mmio.is_write = vcpu->mmio_is_write;
3519         }
3520
3521         if (r) {
3522                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3523                         return EMULATE_DONE;
3524                 if (!vcpu->mmio_needed) {
3525                         kvm_report_emulation_failure(vcpu, "mmio");
3526                         return EMULATE_FAIL;
3527                 }
3528                 return EMULATE_DO_MMIO;
3529         }
3530
3531         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3532
3533         if (vcpu->mmio_is_write) {
3534                 vcpu->mmio_needed = 0;
3535                 return EMULATE_DO_MMIO;
3536         }
3537
3538         return EMULATE_DONE;
3539 }
3540 EXPORT_SYMBOL_GPL(emulate_instruction);
3541
3542 static int pio_copy_data(struct kvm_vcpu *vcpu)
3543 {
3544         void *p = vcpu->arch.pio_data;
3545         gva_t q = vcpu->arch.pio.guest_gva;
3546         unsigned bytes;
3547         int ret;
3548         u32 error_code;
3549
3550         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3551         if (vcpu->arch.pio.in)
3552                 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3553         else
3554                 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3555
3556         if (ret == X86EMUL_PROPAGATE_FAULT)
3557                 kvm_inject_page_fault(vcpu, q, error_code);
3558
3559         return ret;
3560 }
3561
3562 int complete_pio(struct kvm_vcpu *vcpu)
3563 {
3564         struct kvm_pio_request *io = &vcpu->arch.pio;
3565         long delta;
3566         int r;
3567         unsigned long val;
3568
3569         if (!io->string) {
3570                 if (io->in) {
3571                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3572                         memcpy(&val, vcpu->arch.pio_data, io->size);
3573                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3574                 }
3575         } else {
3576                 if (io->in) {
3577                         r = pio_copy_data(vcpu);
3578                         if (r)
3579                                 goto out;
3580                 }
3581
3582                 delta = 1;
3583                 if (io->rep) {
3584                         delta *= io->cur_count;
3585                         /*
3586                          * The size of the register should really depend on
3587                          * current address size.
3588                          */
3589                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3590                         val -= delta;
3591                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3592                 }
3593                 if (io->down)
3594                         delta = -delta;
3595                 delta *= io->size;
3596                 if (io->in) {
3597                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3598                         val += delta;
3599                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3600                 } else {
3601                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3602                         val += delta;
3603                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3604                 }
3605         }
3606 out:
3607         io->count -= io->cur_count;
3608         io->cur_count = 0;
3609
3610         return 0;
3611 }
3612
3613 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3614 {
3615         /* TODO: String I/O for in kernel device */
3616         int r;
3617
3618         if (vcpu->arch.pio.in)
3619                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3620                                     vcpu->arch.pio.size, pd);
3621         else
3622                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3623                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3624                                      pd);
3625         return r;
3626 }
3627
3628 static int pio_string_write(struct kvm_vcpu *vcpu)
3629 {
3630         struct kvm_pio_request *io = &vcpu->arch.pio;
3631         void *pd = vcpu->arch.pio_data;
3632         int i, r = 0;
3633
3634         for (i = 0; i < io->cur_count; i++) {
3635                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3636                                      io->port, io->size, pd)) {
3637                         r = -EOPNOTSUPP;
3638                         break;
3639                 }
3640                 pd += io->size;
3641         }
3642         return r;
3643 }
3644
3645 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3646 {
3647         unsigned long val;
3648
3649         trace_kvm_pio(!in, port, size, 1);
3650
3651         vcpu->run->exit_reason = KVM_EXIT_IO;
3652         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3653         vcpu->run->io.size = vcpu->arch.pio.size = size;
3654         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3655         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3656         vcpu->run->io.port = vcpu->arch.pio.port = port;
3657         vcpu->arch.pio.in = in;
3658         vcpu->arch.pio.string = 0;
3659         vcpu->arch.pio.down = 0;
3660         vcpu->arch.pio.rep = 0;
3661
3662         if (!vcpu->arch.pio.in) {
3663                 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3664                 memcpy(vcpu->arch.pio_data, &val, 4);
3665         }
3666
3667         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3668                 complete_pio(vcpu);
3669                 return 1;
3670         }
3671         return 0;
3672 }
3673 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3674
3675 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3676                   int size, unsigned long count, int down,
3677                   gva_t address, int rep, unsigned port)
3678 {
3679         unsigned now, in_page;
3680         int ret = 0;
3681
3682         trace_kvm_pio(!in, port, size, count);
3683
3684         vcpu->run->exit_reason = KVM_EXIT_IO;
3685         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3686         vcpu->run->io.size = vcpu->arch.pio.size = size;
3687         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3688         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3689         vcpu->run->io.port = vcpu->arch.pio.port = port;
3690         vcpu->arch.pio.in = in;
3691         vcpu->arch.pio.string = 1;
3692         vcpu->arch.pio.down = down;
3693         vcpu->arch.pio.rep = rep;
3694
3695         if (!count) {
3696                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3697                 return 1;
3698         }
3699
3700         if (!down)
3701                 in_page = PAGE_SIZE - offset_in_page(address);
3702         else
3703                 in_page = offset_in_page(address) + size;
3704         now = min(count, (unsigned long)in_page / size);
3705         if (!now)
3706                 now = 1;
3707         if (down) {
3708                 /*
3709                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3710                  */
3711                 pr_unimpl(vcpu, "guest string pio down\n");
3712                 kvm_inject_gp(vcpu, 0);
3713                 return 1;
3714         }
3715         vcpu->run->io.count = now;
3716         vcpu->arch.pio.cur_count = now;
3717
3718         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3719                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3720
3721         vcpu->arch.pio.guest_gva = address;
3722
3723         if (!vcpu->arch.pio.in) {
3724                 /* string PIO write */
3725                 ret = pio_copy_data(vcpu);
3726                 if (ret == X86EMUL_PROPAGATE_FAULT)
3727                         return 1;
3728                 if (ret == 0 && !pio_string_write(vcpu)) {
3729                         complete_pio(vcpu);
3730                         if (vcpu->arch.pio.count == 0)
3731                                 ret = 1;
3732                 }
3733         }
3734         /* no string PIO read support yet */
3735
3736         return ret;
3737 }
3738 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3739
3740 static void bounce_off(void *info)
3741 {
3742         /* nothing */
3743 }
3744
3745 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3746                                      void *data)
3747 {
3748         struct cpufreq_freqs *freq = data;
3749         struct kvm *kvm;
3750         struct kvm_vcpu *vcpu;
3751         int i, send_ipi = 0;
3752
3753         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3754                 return 0;
3755         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3756                 return 0;
3757         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3758
3759         spin_lock(&kvm_lock);
3760         list_for_each_entry(kvm, &vm_list, vm_list) {
3761                 kvm_for_each_vcpu(i, vcpu, kvm) {
3762                         if (vcpu->cpu != freq->cpu)
3763                                 continue;
3764                         if (!kvm_request_guest_time_update(vcpu))
3765                                 continue;
3766                         if (vcpu->cpu != smp_processor_id())
3767                                 send_ipi++;
3768                 }
3769         }
3770         spin_unlock(&kvm_lock);
3771
3772         if (freq->old < freq->new && send_ipi) {
3773                 /*
3774                  * We upscale the frequency.  Must make the guest
3775                  * doesn't see old kvmclock values while running with
3776                  * the new frequency, otherwise we risk the guest sees
3777                  * time go backwards.
3778                  *
3779                  * In case we update the frequency for another cpu
3780                  * (which might be in guest context) send an interrupt
3781                  * to kick the cpu out of guest context.  Next time
3782                  * guest context is entered kvmclock will be updated,
3783                  * so the guest will not see stale values.
3784                  */
3785                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3786         }
3787         return 0;
3788 }
3789
3790 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3791         .notifier_call  = kvmclock_cpufreq_notifier
3792 };
3793
3794 static void kvm_timer_init(void)
3795 {
3796         int cpu;
3797
3798         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3799                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3800                                           CPUFREQ_TRANSITION_NOTIFIER);
3801                 for_each_online_cpu(cpu) {
3802                         unsigned long khz = cpufreq_get(cpu);
3803                         if (!khz)
3804                                 khz = tsc_khz;
3805                         per_cpu(cpu_tsc_khz, cpu) = khz;
3806                 }
3807         } else {
3808                 for_each_possible_cpu(cpu)
3809                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3810         }
3811 }
3812
3813 int kvm_arch_init(void *opaque)
3814 {
3815         int r;
3816         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3817
3818         if (kvm_x86_ops) {
3819                 printk(KERN_ERR "kvm: already loaded the other module\n");
3820                 r = -EEXIST;
3821                 goto out;
3822         }
3823
3824         if (!ops->cpu_has_kvm_support()) {
3825                 printk(KERN_ERR "kvm: no hardware support\n");
3826                 r = -EOPNOTSUPP;
3827                 goto out;
3828         }
3829         if (ops->disabled_by_bios()) {
3830                 printk(KERN_ERR "kvm: disabled by bios\n");
3831                 r = -EOPNOTSUPP;
3832                 goto out;
3833         }
3834
3835         r = kvm_mmu_module_init();
3836         if (r)
3837                 goto out;
3838
3839         kvm_init_msr_list();
3840
3841         kvm_x86_ops = ops;
3842         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3843         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3844         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3845                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3846
3847         kvm_timer_init();
3848
3849         return 0;
3850
3851 out:
3852         return r;
3853 }
3854
3855 void kvm_arch_exit(void)
3856 {
3857         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3858                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3859                                             CPUFREQ_TRANSITION_NOTIFIER);
3860         kvm_x86_ops = NULL;
3861         kvm_mmu_module_exit();
3862 }
3863
3864 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3865 {
3866         ++vcpu->stat.halt_exits;
3867         if (irqchip_in_kernel(vcpu->kvm)) {
3868                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3869                 return 1;
3870         } else {
3871                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3872                 return 0;
3873         }
3874 }
3875 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3876
3877 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3878                            unsigned long a1)
3879 {
3880         if (is_long_mode(vcpu))
3881                 return a0;
3882         else
3883                 return a0 | ((gpa_t)a1 << 32);
3884 }
3885
3886 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3887 {
3888         u64 param, ingpa, outgpa, ret;
3889         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3890         bool fast, longmode;
3891         int cs_db, cs_l;
3892
3893         /*
3894          * hypercall generates UD from non zero cpl and real mode
3895          * per HYPER-V spec
3896          */
3897         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3898                 kvm_queue_exception(vcpu, UD_VECTOR);
3899                 return 0;
3900         }
3901
3902         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3903         longmode = is_long_mode(vcpu) && cs_l == 1;
3904
3905         if (!longmode) {
3906                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3907                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3908                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3909                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3910                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3911                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3912         }
3913 #ifdef CONFIG_X86_64
3914         else {
3915                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3916                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3917                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3918         }
3919 #endif
3920
3921         code = param & 0xffff;
3922         fast = (param >> 16) & 0x1;
3923         rep_cnt = (param >> 32) & 0xfff;
3924         rep_idx = (param >> 48) & 0xfff;
3925
3926         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3927
3928         switch (code) {
3929         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3930                 kvm_vcpu_on_spin(vcpu);
3931                 break;
3932         default:
3933                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3934                 break;
3935         }
3936
3937         ret = res | (((u64)rep_done & 0xfff) << 32);
3938         if (longmode) {
3939                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3940         } else {
3941                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3942                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3943         }
3944
3945         return 1;
3946 }
3947
3948 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3949 {
3950         unsigned long nr, a0, a1, a2, a3, ret;
3951         int r = 1;
3952
3953         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3954                 return kvm_hv_hypercall(vcpu);
3955
3956         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3957         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3958         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3959         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3960         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3961
3962         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3963
3964         if (!is_long_mode(vcpu)) {
3965                 nr &= 0xFFFFFFFF;
3966                 a0 &= 0xFFFFFFFF;
3967                 a1 &= 0xFFFFFFFF;
3968                 a2 &= 0xFFFFFFFF;
3969                 a3 &= 0xFFFFFFFF;
3970         }
3971
3972         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3973                 ret = -KVM_EPERM;
3974                 goto out;
3975         }
3976
3977         switch (nr) {
3978         case KVM_HC_VAPIC_POLL_IRQ:
3979                 ret = 0;
3980                 break;
3981         case KVM_HC_MMU_OP:
3982                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3983                 break;
3984         default:
3985                 ret = -KVM_ENOSYS;
3986                 break;
3987         }
3988 out:
3989         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3990         ++vcpu->stat.hypercalls;
3991         return r;
3992 }
3993 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3994
3995 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3996 {
3997         char instruction[3];
3998         unsigned long rip = kvm_rip_read(vcpu);
3999
4000         /*
4001          * Blow out the MMU to ensure that no other VCPU has an active mapping
4002          * to ensure that the updated hypercall appears atomically across all
4003          * VCPUs.
4004          */
4005         kvm_mmu_zap_all(vcpu->kvm);
4006
4007         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4008
4009         return emulator_write_emulated(rip, instruction, 3, vcpu);
4010 }
4011
4012 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4013 {
4014         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4015 }
4016
4017 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4018 {
4019         struct desc_ptr dt = { limit, base };
4020
4021         kvm_x86_ops->set_gdt(vcpu, &dt);
4022 }
4023
4024 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4025 {
4026         struct desc_ptr dt = { limit, base };
4027
4028         kvm_x86_ops->set_idt(vcpu, &dt);
4029 }
4030
4031 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
4032                    unsigned long *rflags)
4033 {
4034         kvm_lmsw(vcpu, msw);
4035         *rflags = kvm_get_rflags(vcpu);
4036 }
4037
4038 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
4039 {
4040         unsigned long value;
4041
4042         switch (cr) {
4043         case 0:
4044                 value = kvm_read_cr0(vcpu);
4045                 break;
4046         case 2:
4047                 value = vcpu->arch.cr2;
4048                 break;
4049         case 3:
4050                 value = vcpu->arch.cr3;
4051                 break;
4052         case 4:
4053                 value = kvm_read_cr4(vcpu);
4054                 break;
4055         case 8:
4056                 value = kvm_get_cr8(vcpu);
4057                 break;
4058         default:
4059                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4060                 return 0;
4061         }
4062
4063         return value;
4064 }
4065
4066 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
4067                      unsigned long *rflags)
4068 {
4069         switch (cr) {
4070         case 0:
4071                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4072                 *rflags = kvm_get_rflags(vcpu);
4073                 break;
4074         case 2:
4075                 vcpu->arch.cr2 = val;
4076                 break;
4077         case 3:
4078                 kvm_set_cr3(vcpu, val);
4079                 break;
4080         case 4:
4081                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4082                 break;
4083         case 8:
4084                 kvm_set_cr8(vcpu, val & 0xfUL);
4085                 break;
4086         default:
4087                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4088         }
4089 }
4090
4091 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4092 {
4093         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4094         int j, nent = vcpu->arch.cpuid_nent;
4095
4096         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4097         /* when no next entry is found, the current entry[i] is reselected */
4098         for (j = i + 1; ; j = (j + 1) % nent) {
4099                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4100                 if (ej->function == e->function) {
4101                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4102                         return j;
4103                 }
4104         }
4105         return 0; /* silence gcc, even though control never reaches here */
4106 }
4107
4108 /* find an entry with matching function, matching index (if needed), and that
4109  * should be read next (if it's stateful) */
4110 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4111         u32 function, u32 index)
4112 {
4113         if (e->function != function)
4114                 return 0;
4115         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4116                 return 0;
4117         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4118             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4119                 return 0;
4120         return 1;
4121 }
4122
4123 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4124                                               u32 function, u32 index)
4125 {
4126         int i;
4127         struct kvm_cpuid_entry2 *best = NULL;
4128
4129         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4130                 struct kvm_cpuid_entry2 *e;
4131
4132                 e = &vcpu->arch.cpuid_entries[i];
4133                 if (is_matching_cpuid_entry(e, function, index)) {
4134                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4135                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4136                         best = e;
4137                         break;
4138                 }
4139                 /*
4140                  * Both basic or both extended?
4141                  */
4142                 if (((e->function ^ function) & 0x80000000) == 0)
4143                         if (!best || e->function > best->function)
4144                                 best = e;
4145         }
4146         return best;
4147 }
4148 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4149
4150 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4151 {
4152         struct kvm_cpuid_entry2 *best;
4153
4154         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4155         if (best)
4156                 return best->eax & 0xff;
4157         return 36;
4158 }
4159
4160 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4161 {
4162         u32 function, index;
4163         struct kvm_cpuid_entry2 *best;
4164
4165         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4166         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4167         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4168         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4169         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4170         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4171         best = kvm_find_cpuid_entry(vcpu, function, index);
4172         if (best) {
4173                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4174                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4175                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4176                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4177         }
4178         kvm_x86_ops->skip_emulated_instruction(vcpu);
4179         trace_kvm_cpuid(function,
4180                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4181                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4182                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4183                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4184 }
4185 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4186
4187 /*
4188  * Check if userspace requested an interrupt window, and that the
4189  * interrupt window is open.
4190  *
4191  * No need to exit to userspace if we already have an interrupt queued.
4192  */
4193 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4194 {
4195         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4196                 vcpu->run->request_interrupt_window &&
4197                 kvm_arch_interrupt_allowed(vcpu));
4198 }
4199
4200 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4201 {
4202         struct kvm_run *kvm_run = vcpu->run;
4203
4204         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4205         kvm_run->cr8 = kvm_get_cr8(vcpu);
4206         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4207         if (irqchip_in_kernel(vcpu->kvm))
4208                 kvm_run->ready_for_interrupt_injection = 1;
4209         else
4210                 kvm_run->ready_for_interrupt_injection =
4211                         kvm_arch_interrupt_allowed(vcpu) &&
4212                         !kvm_cpu_has_interrupt(vcpu) &&
4213                         !kvm_event_needs_reinjection(vcpu);
4214 }
4215
4216 static void vapic_enter(struct kvm_vcpu *vcpu)
4217 {
4218         struct kvm_lapic *apic = vcpu->arch.apic;
4219         struct page *page;
4220
4221         if (!apic || !apic->vapic_addr)
4222                 return;
4223
4224         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4225
4226         vcpu->arch.apic->vapic_page = page;
4227 }
4228
4229 static void vapic_exit(struct kvm_vcpu *vcpu)
4230 {
4231         struct kvm_lapic *apic = vcpu->arch.apic;
4232         int idx;
4233
4234         if (!apic || !apic->vapic_addr)
4235                 return;
4236
4237         idx = srcu_read_lock(&vcpu->kvm->srcu);
4238         kvm_release_page_dirty(apic->vapic_page);
4239         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4240         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4241 }
4242
4243 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4244 {
4245         int max_irr, tpr;
4246
4247         if (!kvm_x86_ops->update_cr8_intercept)
4248                 return;
4249
4250         if (!vcpu->arch.apic)
4251                 return;
4252
4253         if (!vcpu->arch.apic->vapic_addr)
4254                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4255         else
4256                 max_irr = -1;
4257
4258         if (max_irr != -1)
4259                 max_irr >>= 4;
4260
4261         tpr = kvm_lapic_get_cr8(vcpu);
4262
4263         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4264 }
4265
4266 static void inject_pending_event(struct kvm_vcpu *vcpu)
4267 {
4268         /* try to reinject previous events if any */
4269         if (vcpu->arch.exception.pending) {
4270                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4271                                           vcpu->arch.exception.has_error_code,
4272                                           vcpu->arch.exception.error_code);
4273                 return;
4274         }
4275
4276         if (vcpu->arch.nmi_injected) {
4277                 kvm_x86_ops->set_nmi(vcpu);
4278                 return;
4279         }
4280
4281         if (vcpu->arch.interrupt.pending) {
4282                 kvm_x86_ops->set_irq(vcpu);
4283                 return;
4284         }
4285
4286         /* try to inject new event if pending */
4287         if (vcpu->arch.nmi_pending) {
4288                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4289                         vcpu->arch.nmi_pending = false;
4290                         vcpu->arch.nmi_injected = true;
4291                         kvm_x86_ops->set_nmi(vcpu);
4292                 }
4293         } else if (kvm_cpu_has_interrupt(vcpu)) {
4294                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4295                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4296                                             false);
4297                         kvm_x86_ops->set_irq(vcpu);
4298                 }
4299         }
4300 }
4301
4302 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4303 {
4304         int r;
4305         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4306                 vcpu->run->request_interrupt_window;
4307
4308         if (vcpu->requests)
4309                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4310                         kvm_mmu_unload(vcpu);
4311
4312         r = kvm_mmu_reload(vcpu);
4313         if (unlikely(r))
4314                 goto out;
4315
4316         if (vcpu->requests) {
4317                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4318                         __kvm_migrate_timers(vcpu);
4319                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4320                         kvm_write_guest_time(vcpu);
4321                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4322                         kvm_mmu_sync_roots(vcpu);
4323                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4324                         kvm_x86_ops->tlb_flush(vcpu);
4325                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4326                                        &vcpu->requests)) {
4327                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4328                         r = 0;
4329                         goto out;
4330                 }
4331                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4332                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4333                         r = 0;
4334                         goto out;
4335                 }
4336                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4337                         vcpu->fpu_active = 0;
4338                         kvm_x86_ops->fpu_deactivate(vcpu);
4339                 }
4340         }
4341
4342         preempt_disable();
4343
4344         kvm_x86_ops->prepare_guest_switch(vcpu);
4345         if (vcpu->fpu_active)
4346                 kvm_load_guest_fpu(vcpu);
4347
4348         local_irq_disable();
4349
4350         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4351         smp_mb__after_clear_bit();
4352
4353         if (vcpu->requests || need_resched() || signal_pending(current)) {
4354                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4355                 local_irq_enable();
4356                 preempt_enable();
4357                 r = 1;
4358                 goto out;
4359         }
4360
4361         inject_pending_event(vcpu);
4362
4363         /* enable NMI/IRQ window open exits if needed */
4364         if (vcpu->arch.nmi_pending)
4365                 kvm_x86_ops->enable_nmi_window(vcpu);
4366         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4367                 kvm_x86_ops->enable_irq_window(vcpu);
4368
4369         if (kvm_lapic_enabled(vcpu)) {
4370                 update_cr8_intercept(vcpu);
4371                 kvm_lapic_sync_to_vapic(vcpu);
4372         }
4373
4374         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4375
4376         kvm_guest_enter();
4377
4378         if (unlikely(vcpu->arch.switch_db_regs)) {
4379                 set_debugreg(0, 7);
4380                 set_debugreg(vcpu->arch.eff_db[0], 0);
4381                 set_debugreg(vcpu->arch.eff_db[1], 1);
4382                 set_debugreg(vcpu->arch.eff_db[2], 2);
4383                 set_debugreg(vcpu->arch.eff_db[3], 3);
4384         }
4385
4386         trace_kvm_entry(vcpu->vcpu_id);
4387         kvm_x86_ops->run(vcpu);
4388
4389         /*
4390          * If the guest has used debug registers, at least dr7
4391          * will be disabled while returning to the host.
4392          * If we don't have active breakpoints in the host, we don't
4393          * care about the messed up debug address registers. But if
4394          * we have some of them active, restore the old state.
4395          */
4396         if (hw_breakpoint_active())
4397                 hw_breakpoint_restore();
4398
4399         set_bit(KVM_REQ_KICK, &vcpu->requests);
4400         local_irq_enable();
4401
4402         ++vcpu->stat.exits;
4403
4404         /*
4405          * We must have an instruction between local_irq_enable() and
4406          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4407          * the interrupt shadow.  The stat.exits increment will do nicely.
4408          * But we need to prevent reordering, hence this barrier():
4409          */
4410         barrier();
4411
4412         kvm_guest_exit();
4413
4414         preempt_enable();
4415
4416         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4417
4418         /*
4419          * Profile KVM exit RIPs:
4420          */
4421         if (unlikely(prof_on == KVM_PROFILING)) {
4422                 unsigned long rip = kvm_rip_read(vcpu);
4423                 profile_hit(KVM_PROFILING, (void *)rip);
4424         }
4425
4426
4427         kvm_lapic_sync_from_vapic(vcpu);
4428
4429         r = kvm_x86_ops->handle_exit(vcpu);
4430 out:
4431         return r;
4432 }
4433
4434
4435 static int __vcpu_run(struct kvm_vcpu *vcpu)
4436 {
4437         int r;
4438         struct kvm *kvm = vcpu->kvm;
4439
4440         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4441                 pr_debug("vcpu %d received sipi with vector # %x\n",
4442                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4443                 kvm_lapic_reset(vcpu);
4444                 r = kvm_arch_vcpu_reset(vcpu);
4445                 if (r)
4446                         return r;
4447                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4448         }
4449
4450         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4451         vapic_enter(vcpu);
4452
4453         r = 1;
4454         while (r > 0) {
4455                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4456                         r = vcpu_enter_guest(vcpu);
4457                 else {
4458                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4459                         kvm_vcpu_block(vcpu);
4460                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4461                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4462                         {
4463                                 switch(vcpu->arch.mp_state) {
4464                                 case KVM_MP_STATE_HALTED:
4465                                         vcpu->arch.mp_state =
4466                                                 KVM_MP_STATE_RUNNABLE;
4467                                 case KVM_MP_STATE_RUNNABLE:
4468                                         break;
4469                                 case KVM_MP_STATE_SIPI_RECEIVED:
4470                                 default:
4471                                         r = -EINTR;
4472                                         break;
4473                                 }
4474                         }
4475                 }
4476
4477                 if (r <= 0)
4478                         break;
4479
4480                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4481                 if (kvm_cpu_has_pending_timer(vcpu))
4482                         kvm_inject_pending_timer_irqs(vcpu);
4483
4484                 if (dm_request_for_irq_injection(vcpu)) {
4485                         r = -EINTR;
4486                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4487                         ++vcpu->stat.request_irq_exits;
4488                 }
4489                 if (signal_pending(current)) {
4490                         r = -EINTR;
4491                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4492                         ++vcpu->stat.signal_exits;
4493                 }
4494                 if (need_resched()) {
4495                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4496                         kvm_resched(vcpu);
4497                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4498                 }
4499         }
4500
4501         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4502         post_kvm_run_save(vcpu);
4503
4504         vapic_exit(vcpu);
4505
4506         return r;
4507 }
4508
4509 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4510 {
4511         int r;
4512         sigset_t sigsaved;
4513
4514         vcpu_load(vcpu);
4515
4516         if (vcpu->sigset_active)
4517                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4518
4519         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4520                 kvm_vcpu_block(vcpu);
4521                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4522                 r = -EAGAIN;
4523                 goto out;
4524         }
4525
4526         /* re-sync apic's tpr */
4527         if (!irqchip_in_kernel(vcpu->kvm))
4528                 kvm_set_cr8(vcpu, kvm_run->cr8);
4529
4530         if (vcpu->arch.pio.cur_count) {
4531                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4532                 r = complete_pio(vcpu);
4533                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4534                 if (r)
4535                         goto out;
4536         }
4537         if (vcpu->mmio_needed) {
4538                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4539                 vcpu->mmio_read_completed = 1;
4540                 vcpu->mmio_needed = 0;
4541
4542                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4543                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4544                                         EMULTYPE_NO_DECODE);
4545                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4546                 if (r == EMULATE_DO_MMIO) {
4547                         /*
4548                          * Read-modify-write.  Back to userspace.
4549                          */
4550                         r = 0;
4551                         goto out;
4552                 }
4553         }
4554         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4555                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4556                                      kvm_run->hypercall.ret);
4557
4558         r = __vcpu_run(vcpu);
4559
4560 out:
4561         if (vcpu->sigset_active)
4562                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4563
4564         vcpu_put(vcpu);
4565         return r;
4566 }
4567
4568 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4569 {
4570         vcpu_load(vcpu);
4571
4572         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4573         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4574         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4575         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4576         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4577         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4578         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4579         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4580 #ifdef CONFIG_X86_64
4581         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4582         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4583         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4584         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4585         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4586         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4587         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4588         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4589 #endif
4590
4591         regs->rip = kvm_rip_read(vcpu);
4592         regs->rflags = kvm_get_rflags(vcpu);
4593
4594         vcpu_put(vcpu);
4595
4596         return 0;
4597 }
4598
4599 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4600 {
4601         vcpu_load(vcpu);
4602
4603         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4604         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4605         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4606         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4607         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4608         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4609         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4610         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4611 #ifdef CONFIG_X86_64
4612         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4613         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4614         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4615         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4616         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4617         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4618         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4619         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4620 #endif
4621
4622         kvm_rip_write(vcpu, regs->rip);
4623         kvm_set_rflags(vcpu, regs->rflags);
4624
4625         vcpu->arch.exception.pending = false;
4626
4627         vcpu_put(vcpu);
4628
4629         return 0;
4630 }
4631
4632 void kvm_get_segment(struct kvm_vcpu *vcpu,
4633                      struct kvm_segment *var, int seg)
4634 {
4635         kvm_x86_ops->get_segment(vcpu, var, seg);
4636 }
4637
4638 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4639 {
4640         struct kvm_segment cs;
4641
4642         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4643         *db = cs.db;
4644         *l = cs.l;
4645 }
4646 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4647
4648 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4649                                   struct kvm_sregs *sregs)
4650 {
4651         struct desc_ptr dt;
4652
4653         vcpu_load(vcpu);
4654
4655         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4656         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4657         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4658         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4659         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4660         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4661
4662         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4663         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4664
4665         kvm_x86_ops->get_idt(vcpu, &dt);
4666         sregs->idt.limit = dt.size;
4667         sregs->idt.base = dt.address;
4668         kvm_x86_ops->get_gdt(vcpu, &dt);
4669         sregs->gdt.limit = dt.size;
4670         sregs->gdt.base = dt.address;
4671
4672         sregs->cr0 = kvm_read_cr0(vcpu);
4673         sregs->cr2 = vcpu->arch.cr2;
4674         sregs->cr3 = vcpu->arch.cr3;
4675         sregs->cr4 = kvm_read_cr4(vcpu);
4676         sregs->cr8 = kvm_get_cr8(vcpu);
4677         sregs->efer = vcpu->arch.efer;
4678         sregs->apic_base = kvm_get_apic_base(vcpu);
4679
4680         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4681
4682         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4683                 set_bit(vcpu->arch.interrupt.nr,
4684                         (unsigned long *)sregs->interrupt_bitmap);
4685
4686         vcpu_put(vcpu);
4687
4688         return 0;
4689 }
4690
4691 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4692                                     struct kvm_mp_state *mp_state)
4693 {
4694         vcpu_load(vcpu);
4695         mp_state->mp_state = vcpu->arch.mp_state;
4696         vcpu_put(vcpu);
4697         return 0;
4698 }
4699
4700 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4701                                     struct kvm_mp_state *mp_state)
4702 {
4703         vcpu_load(vcpu);
4704         vcpu->arch.mp_state = mp_state->mp_state;
4705         vcpu_put(vcpu);
4706         return 0;
4707 }
4708
4709 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4710                         struct kvm_segment *var, int seg)
4711 {
4712         kvm_x86_ops->set_segment(vcpu, var, seg);
4713 }
4714
4715 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4716                                    struct kvm_segment *kvm_desct)
4717 {
4718         kvm_desct->base = get_desc_base(seg_desc);
4719         kvm_desct->limit = get_desc_limit(seg_desc);
4720         if (seg_desc->g) {
4721                 kvm_desct->limit <<= 12;
4722                 kvm_desct->limit |= 0xfff;
4723         }
4724         kvm_desct->selector = selector;
4725         kvm_desct->type = seg_desc->type;
4726         kvm_desct->present = seg_desc->p;
4727         kvm_desct->dpl = seg_desc->dpl;
4728         kvm_desct->db = seg_desc->d;
4729         kvm_desct->s = seg_desc->s;
4730         kvm_desct->l = seg_desc->l;
4731         kvm_desct->g = seg_desc->g;
4732         kvm_desct->avl = seg_desc->avl;
4733         if (!selector)
4734                 kvm_desct->unusable = 1;
4735         else
4736                 kvm_desct->unusable = 0;
4737         kvm_desct->padding = 0;
4738 }
4739
4740 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4741                                           u16 selector,
4742                                           struct desc_ptr *dtable)
4743 {
4744         if (selector & 1 << 2) {
4745                 struct kvm_segment kvm_seg;
4746
4747                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4748
4749                 if (kvm_seg.unusable)
4750                         dtable->size = 0;
4751                 else
4752                         dtable->size = kvm_seg.limit;
4753                 dtable->address = kvm_seg.base;
4754         }
4755         else
4756                 kvm_x86_ops->get_gdt(vcpu, dtable);
4757 }
4758
4759 /* allowed just for 8 bytes segments */
4760 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4761                                          struct desc_struct *seg_desc)
4762 {
4763         struct desc_ptr dtable;
4764         u16 index = selector >> 3;
4765         int ret;
4766         u32 err;
4767         gva_t addr;
4768
4769         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4770
4771         if (dtable.size < index * 8 + 7) {
4772                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4773                 return X86EMUL_PROPAGATE_FAULT;
4774         }
4775         addr = dtable.base + index * 8;
4776         ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4777                                          vcpu,  &err);
4778         if (ret == X86EMUL_PROPAGATE_FAULT)
4779                 kvm_inject_page_fault(vcpu, addr, err);
4780
4781        return ret;
4782 }
4783
4784 /* allowed just for 8 bytes segments */
4785 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4786                                          struct desc_struct *seg_desc)
4787 {
4788         struct desc_ptr dtable;
4789         u16 index = selector >> 3;
4790
4791         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4792
4793         if (dtable.size < index * 8 + 7)
4794                 return 1;
4795         return kvm_write_guest_virt(dtable.address + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4796 }
4797
4798 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4799                                struct desc_struct *seg_desc)
4800 {
4801         u32 base_addr = get_desc_base(seg_desc);
4802
4803         return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4804 }
4805
4806 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4807                              struct desc_struct *seg_desc)
4808 {
4809         u32 base_addr = get_desc_base(seg_desc);
4810
4811         return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4812 }
4813
4814 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4815 {
4816         struct kvm_segment kvm_seg;
4817
4818         kvm_get_segment(vcpu, &kvm_seg, seg);
4819         return kvm_seg.selector;
4820 }
4821
4822 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4823 {
4824         struct kvm_segment segvar = {
4825                 .base = selector << 4,
4826                 .limit = 0xffff,
4827                 .selector = selector,
4828                 .type = 3,
4829                 .present = 1,
4830                 .dpl = 3,
4831                 .db = 0,
4832                 .s = 1,
4833                 .l = 0,
4834                 .g = 0,
4835                 .avl = 0,
4836                 .unusable = 0,
4837         };
4838         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4839         return X86EMUL_CONTINUE;
4840 }
4841
4842 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4843 {
4844         return (seg != VCPU_SREG_LDTR) &&
4845                 (seg != VCPU_SREG_TR) &&
4846                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4847 }
4848
4849 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4850 {
4851         struct kvm_segment kvm_seg;
4852         struct desc_struct seg_desc;
4853         u8 dpl, rpl, cpl;
4854         unsigned err_vec = GP_VECTOR;
4855         u32 err_code = 0;
4856         bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4857         int ret;
4858
4859         if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4860                 return kvm_load_realmode_segment(vcpu, selector, seg);
4861
4862         /* NULL selector is not valid for TR, CS and SS */
4863         if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4864             && null_selector)
4865                 goto exception;
4866
4867         /* TR should be in GDT only */
4868         if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4869                 goto exception;
4870
4871         ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4872         if (ret)
4873                 return ret;
4874
4875         seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4876
4877         if (null_selector) { /* for NULL selector skip all following checks */
4878                 kvm_seg.unusable = 1;
4879                 goto load;
4880         }
4881
4882         err_code = selector & 0xfffc;
4883         err_vec = GP_VECTOR;
4884
4885         /* can't load system descriptor into segment selecor */
4886         if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4887                 goto exception;
4888
4889         if (!kvm_seg.present) {
4890                 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4891                 goto exception;
4892         }
4893
4894         rpl = selector & 3;
4895         dpl = kvm_seg.dpl;
4896         cpl = kvm_x86_ops->get_cpl(vcpu);
4897
4898         switch (seg) {
4899         case VCPU_SREG_SS:
4900                 /*
4901                  * segment is not a writable data segment or segment
4902                  * selector's RPL != CPL or segment selector's RPL != CPL
4903                  */
4904                 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4905                         goto exception;
4906                 break;
4907         case VCPU_SREG_CS:
4908                 if (!(kvm_seg.type & 8))
4909                         goto exception;
4910
4911                 if (kvm_seg.type & 4) {
4912                         /* conforming */
4913                         if (dpl > cpl)
4914                                 goto exception;
4915                 } else {
4916                         /* nonconforming */
4917                         if (rpl > cpl || dpl != cpl)
4918                                 goto exception;
4919                 }
4920                 /* CS(RPL) <- CPL */
4921                 selector = (selector & 0xfffc) | cpl;
4922             break;
4923         case VCPU_SREG_TR:
4924                 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4925                         goto exception;
4926                 break;
4927         case VCPU_SREG_LDTR:
4928                 if (kvm_seg.s || kvm_seg.type != 2)
4929                         goto exception;
4930                 break;
4931         default: /*  DS, ES, FS, or GS */
4932                 /*
4933                  * segment is not a data or readable code segment or
4934                  * ((segment is a data or nonconforming code segment)
4935                  * and (both RPL and CPL > DPL))
4936                  */
4937                 if ((kvm_seg.type & 0xa) == 0x8 ||
4938                     (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4939                         goto exception;
4940                 break;
4941         }
4942
4943         if (!kvm_seg.unusable && kvm_seg.s) {
4944                 /* mark segment as accessed */
4945                 kvm_seg.type |= 1;
4946                 seg_desc.type |= 1;
4947                 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4948         }
4949 load:
4950         kvm_set_segment(vcpu, &kvm_seg, seg);
4951         return X86EMUL_CONTINUE;
4952 exception:
4953         kvm_queue_exception_e(vcpu, err_vec, err_code);
4954         return X86EMUL_PROPAGATE_FAULT;
4955 }
4956
4957 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4958                                 struct tss_segment_32 *tss)
4959 {
4960         tss->cr3 = vcpu->arch.cr3;
4961         tss->eip = kvm_rip_read(vcpu);
4962         tss->eflags = kvm_get_rflags(vcpu);
4963         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4964         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4965         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4966         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4967         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4968         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4969         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4970         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4971         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4972         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4973         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4974         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4975         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4976         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4977         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4978 }
4979
4980 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4981 {
4982         struct kvm_segment kvm_seg;
4983         kvm_get_segment(vcpu, &kvm_seg, seg);
4984         kvm_seg.selector = sel;
4985         kvm_set_segment(vcpu, &kvm_seg, seg);
4986 }
4987
4988 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4989                                   struct tss_segment_32 *tss)
4990 {
4991         kvm_set_cr3(vcpu, tss->cr3);
4992
4993         kvm_rip_write(vcpu, tss->eip);
4994         kvm_set_rflags(vcpu, tss->eflags | 2);
4995
4996         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4997         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4998         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4999         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
5000         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
5001         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
5002         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
5003         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
5004
5005         /*
5006          * SDM says that segment selectors are loaded before segment
5007          * descriptors
5008          */
5009         kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
5010         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5011         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5012         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5013         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5014         kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
5015         kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
5016
5017         /*
5018          * Now load segment descriptors. If fault happenes at this stage
5019          * it is handled in a context of new task
5020          */
5021         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
5022                 return 1;
5023
5024         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5025                 return 1;
5026
5027         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5028                 return 1;
5029
5030         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5031                 return 1;
5032
5033         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5034                 return 1;
5035
5036         if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
5037                 return 1;
5038
5039         if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
5040                 return 1;
5041         return 0;
5042 }
5043
5044 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
5045                                 struct tss_segment_16 *tss)
5046 {
5047         tss->ip = kvm_rip_read(vcpu);
5048         tss->flag = kvm_get_rflags(vcpu);
5049         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5050         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5051         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5052         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5053         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5054         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5055         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
5056         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
5057
5058         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
5059         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
5060         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
5061         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
5062         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
5063 }
5064
5065 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
5066                                  struct tss_segment_16 *tss)
5067 {
5068         kvm_rip_write(vcpu, tss->ip);
5069         kvm_set_rflags(vcpu, tss->flag | 2);
5070         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5071         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5072         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5073         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5074         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5075         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5076         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5077         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5078
5079         /*
5080          * SDM says that segment selectors are loaded before segment
5081          * descriptors
5082          */
5083         kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5084         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5085         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5086         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5087         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5088
5089         /*
5090          * Now load segment descriptors. If fault happenes at this stage
5091          * it is handled in a context of new task
5092          */
5093         if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5094                 return 1;
5095
5096         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5097                 return 1;
5098
5099         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5100                 return 1;
5101
5102         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5103                 return 1;
5104
5105         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5106                 return 1;
5107         return 0;
5108 }
5109
5110 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5111                               u16 old_tss_sel, u32 old_tss_base,
5112                               struct desc_struct *nseg_desc)
5113 {
5114         struct tss_segment_16 tss_segment_16;
5115         int ret = 0;
5116
5117         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5118                            sizeof tss_segment_16))
5119                 goto out;
5120
5121         save_state_to_tss16(vcpu, &tss_segment_16);
5122
5123         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5124                             sizeof tss_segment_16))
5125                 goto out;
5126
5127         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5128                            &tss_segment_16, sizeof tss_segment_16))
5129                 goto out;
5130
5131         if (old_tss_sel != 0xffff) {
5132                 tss_segment_16.prev_task_link = old_tss_sel;
5133
5134                 if (kvm_write_guest(vcpu->kvm,
5135                                     get_tss_base_addr_write(vcpu, nseg_desc),
5136                                     &tss_segment_16.prev_task_link,
5137                                     sizeof tss_segment_16.prev_task_link))
5138                         goto out;
5139         }
5140
5141         if (load_state_from_tss16(vcpu, &tss_segment_16))
5142                 goto out;
5143
5144         ret = 1;
5145 out:
5146         return ret;
5147 }
5148
5149 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5150                        u16 old_tss_sel, u32 old_tss_base,
5151                        struct desc_struct *nseg_desc)
5152 {
5153         struct tss_segment_32 tss_segment_32;
5154         int ret = 0;
5155
5156         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5157                            sizeof tss_segment_32))
5158                 goto out;
5159
5160         save_state_to_tss32(vcpu, &tss_segment_32);
5161
5162         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5163                             sizeof tss_segment_32))
5164                 goto out;
5165
5166         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5167                            &tss_segment_32, sizeof tss_segment_32))
5168                 goto out;
5169
5170         if (old_tss_sel != 0xffff) {
5171                 tss_segment_32.prev_task_link = old_tss_sel;
5172
5173                 if (kvm_write_guest(vcpu->kvm,
5174                                     get_tss_base_addr_write(vcpu, nseg_desc),
5175                                     &tss_segment_32.prev_task_link,
5176                                     sizeof tss_segment_32.prev_task_link))
5177                         goto out;
5178         }
5179
5180         if (load_state_from_tss32(vcpu, &tss_segment_32))
5181                 goto out;
5182
5183         ret = 1;
5184 out:
5185         return ret;
5186 }
5187
5188 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5189 {
5190         struct kvm_segment tr_seg;
5191         struct desc_struct cseg_desc;
5192         struct desc_struct nseg_desc;
5193         int ret = 0;
5194         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5195         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5196         u32 desc_limit;
5197
5198         old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5199
5200         /* FIXME: Handle errors. Failure to read either TSS or their
5201          * descriptors should generate a pagefault.
5202          */
5203         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5204                 goto out;
5205
5206         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5207                 goto out;
5208
5209         if (reason != TASK_SWITCH_IRET) {
5210                 int cpl;
5211
5212                 cpl = kvm_x86_ops->get_cpl(vcpu);
5213                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5214                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5215                         return 1;
5216                 }
5217         }
5218
5219         desc_limit = get_desc_limit(&nseg_desc);
5220         if (!nseg_desc.p ||
5221             ((desc_limit < 0x67 && (nseg_desc.type & 8)) ||
5222              desc_limit < 0x2b)) {
5223                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5224                 return 1;
5225         }
5226
5227         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5228                 cseg_desc.type &= ~(1 << 1); //clear the B flag
5229                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5230         }
5231
5232         if (reason == TASK_SWITCH_IRET) {
5233                 u32 eflags = kvm_get_rflags(vcpu);
5234                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5235         }
5236
5237         /* set back link to prev task only if NT bit is set in eflags
5238            note that old_tss_sel is not used afetr this point */
5239         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5240                 old_tss_sel = 0xffff;
5241
5242         if (nseg_desc.type & 8)
5243                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5244                                          old_tss_base, &nseg_desc);
5245         else
5246                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5247                                          old_tss_base, &nseg_desc);
5248
5249         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5250                 u32 eflags = kvm_get_rflags(vcpu);
5251                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5252         }
5253
5254         if (reason != TASK_SWITCH_IRET) {
5255                 nseg_desc.type |= (1 << 1);
5256                 save_guest_segment_descriptor(vcpu, tss_selector,
5257                                               &nseg_desc);
5258         }
5259
5260         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5261         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5262         tr_seg.type = 11;
5263         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5264 out:
5265         return ret;
5266 }
5267 EXPORT_SYMBOL_GPL(kvm_task_switch);
5268
5269 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5270                                   struct kvm_sregs *sregs)
5271 {
5272         int mmu_reset_needed = 0;
5273         int pending_vec, max_bits;
5274         struct desc_ptr dt;
5275
5276         vcpu_load(vcpu);
5277
5278         dt.size = sregs->idt.limit;
5279         dt.address = sregs->idt.base;
5280         kvm_x86_ops->set_idt(vcpu, &dt);
5281         dt.size = sregs->gdt.limit;
5282         dt.address = sregs->gdt.base;
5283         kvm_x86_ops->set_gdt(vcpu, &dt);
5284
5285         vcpu->arch.cr2 = sregs->cr2;
5286         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5287         vcpu->arch.cr3 = sregs->cr3;
5288
5289         kvm_set_cr8(vcpu, sregs->cr8);
5290
5291         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5292         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5293         kvm_set_apic_base(vcpu, sregs->apic_base);
5294
5295         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5296         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5297         vcpu->arch.cr0 = sregs->cr0;
5298
5299         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5300         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5301         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5302                 load_pdptrs(vcpu, vcpu->arch.cr3);
5303                 mmu_reset_needed = 1;
5304         }
5305
5306         if (mmu_reset_needed)
5307                 kvm_mmu_reset_context(vcpu);
5308
5309         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5310         pending_vec = find_first_bit(
5311                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5312         if (pending_vec < max_bits) {
5313                 kvm_queue_interrupt(vcpu, pending_vec, false);
5314                 pr_debug("Set back pending irq %d\n", pending_vec);
5315                 if (irqchip_in_kernel(vcpu->kvm))
5316                         kvm_pic_clear_isr_ack(vcpu->kvm);
5317         }
5318
5319         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5320         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5321         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5322         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5323         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5324         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5325
5326         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5327         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5328
5329         update_cr8_intercept(vcpu);
5330
5331         /* Older userspace won't unhalt the vcpu on reset. */
5332         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5333             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5334             !is_protmode(vcpu))
5335                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5336
5337         vcpu_put(vcpu);
5338
5339         return 0;
5340 }
5341
5342 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5343                                         struct kvm_guest_debug *dbg)
5344 {
5345         unsigned long rflags;
5346         int i, r;
5347
5348         vcpu_load(vcpu);
5349
5350         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5351                 r = -EBUSY;
5352                 if (vcpu->arch.exception.pending)
5353                         goto unlock_out;
5354                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5355                         kvm_queue_exception(vcpu, DB_VECTOR);
5356                 else
5357                         kvm_queue_exception(vcpu, BP_VECTOR);
5358         }
5359
5360         /*
5361          * Read rflags as long as potentially injected trace flags are still
5362          * filtered out.
5363          */
5364         rflags = kvm_get_rflags(vcpu);
5365
5366         vcpu->guest_debug = dbg->control;
5367         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5368                 vcpu->guest_debug = 0;
5369
5370         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5371                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5372                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5373                 vcpu->arch.switch_db_regs =
5374                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5375         } else {
5376                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5377                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5378                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5379         }
5380
5381         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5382                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5383                         get_segment_base(vcpu, VCPU_SREG_CS);
5384
5385         /*
5386          * Trigger an rflags update that will inject or remove the trace
5387          * flags.
5388          */
5389         kvm_set_rflags(vcpu, rflags);
5390
5391         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5392
5393         r = 0;
5394
5395 unlock_out:
5396         vcpu_put(vcpu);
5397
5398         return r;
5399 }
5400
5401 /*
5402  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5403  * we have asm/x86/processor.h
5404  */
5405 struct fxsave {
5406         u16     cwd;
5407         u16     swd;
5408         u16     twd;
5409         u16     fop;
5410         u64     rip;
5411         u64     rdp;
5412         u32     mxcsr;
5413         u32     mxcsr_mask;
5414         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5415 #ifdef CONFIG_X86_64
5416         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5417 #else
5418         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5419 #endif
5420 };
5421
5422 /*
5423  * Translate a guest virtual address to a guest physical address.
5424  */
5425 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5426                                     struct kvm_translation *tr)
5427 {
5428         unsigned long vaddr = tr->linear_address;
5429         gpa_t gpa;
5430         int idx;
5431
5432         vcpu_load(vcpu);
5433         idx = srcu_read_lock(&vcpu->kvm->srcu);
5434         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5435         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5436         tr->physical_address = gpa;
5437         tr->valid = gpa != UNMAPPED_GVA;
5438         tr->writeable = 1;
5439         tr->usermode = 0;
5440         vcpu_put(vcpu);
5441
5442         return 0;
5443 }
5444
5445 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5446 {
5447         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5448
5449         vcpu_load(vcpu);
5450
5451         memcpy(fpu->fpr, fxsave->st_space, 128);
5452         fpu->fcw = fxsave->cwd;
5453         fpu->fsw = fxsave->swd;
5454         fpu->ftwx = fxsave->twd;
5455         fpu->last_opcode = fxsave->fop;
5456         fpu->last_ip = fxsave->rip;
5457         fpu->last_dp = fxsave->rdp;
5458         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5459
5460         vcpu_put(vcpu);
5461
5462         return 0;
5463 }
5464
5465 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5466 {
5467         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5468
5469         vcpu_load(vcpu);
5470
5471         memcpy(fxsave->st_space, fpu->fpr, 128);
5472         fxsave->cwd = fpu->fcw;
5473         fxsave->swd = fpu->fsw;
5474         fxsave->twd = fpu->ftwx;
5475         fxsave->fop = fpu->last_opcode;
5476         fxsave->rip = fpu->last_ip;
5477         fxsave->rdp = fpu->last_dp;
5478         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5479
5480         vcpu_put(vcpu);
5481
5482         return 0;
5483 }
5484
5485 void fx_init(struct kvm_vcpu *vcpu)
5486 {
5487         unsigned after_mxcsr_mask;
5488
5489         /*
5490          * Touch the fpu the first time in non atomic context as if
5491          * this is the first fpu instruction the exception handler
5492          * will fire before the instruction returns and it'll have to
5493          * allocate ram with GFP_KERNEL.
5494          */
5495         if (!used_math())
5496                 kvm_fx_save(&vcpu->arch.host_fx_image);
5497
5498         /* Initialize guest FPU by resetting ours and saving into guest's */
5499         preempt_disable();
5500         kvm_fx_save(&vcpu->arch.host_fx_image);
5501         kvm_fx_finit();
5502         kvm_fx_save(&vcpu->arch.guest_fx_image);
5503         kvm_fx_restore(&vcpu->arch.host_fx_image);
5504         preempt_enable();
5505
5506         vcpu->arch.cr0 |= X86_CR0_ET;
5507         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5508         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5509         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5510                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5511 }
5512 EXPORT_SYMBOL_GPL(fx_init);
5513
5514 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5515 {
5516         if (vcpu->guest_fpu_loaded)
5517                 return;
5518
5519         vcpu->guest_fpu_loaded = 1;
5520         kvm_fx_save(&vcpu->arch.host_fx_image);
5521         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5522         trace_kvm_fpu(1);
5523 }
5524
5525 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5526 {
5527         if (!vcpu->guest_fpu_loaded)
5528                 return;
5529
5530         vcpu->guest_fpu_loaded = 0;
5531         kvm_fx_save(&vcpu->arch.guest_fx_image);
5532         kvm_fx_restore(&vcpu->arch.host_fx_image);
5533         ++vcpu->stat.fpu_reload;
5534         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5535         trace_kvm_fpu(0);
5536 }
5537
5538 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5539 {
5540         if (vcpu->arch.time_page) {
5541                 kvm_release_page_dirty(vcpu->arch.time_page);
5542                 vcpu->arch.time_page = NULL;
5543         }
5544
5545         kvm_x86_ops->vcpu_free(vcpu);
5546 }
5547
5548 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5549                                                 unsigned int id)
5550 {
5551         return kvm_x86_ops->vcpu_create(kvm, id);
5552 }
5553
5554 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5555 {
5556         int r;
5557
5558         /* We do fxsave: this must be aligned. */
5559         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5560
5561         vcpu->arch.mtrr_state.have_fixed = 1;
5562         vcpu_load(vcpu);
5563         r = kvm_arch_vcpu_reset(vcpu);
5564         if (r == 0)
5565                 r = kvm_mmu_setup(vcpu);
5566         vcpu_put(vcpu);
5567         if (r < 0)
5568                 goto free_vcpu;
5569
5570         return 0;
5571 free_vcpu:
5572         kvm_x86_ops->vcpu_free(vcpu);
5573         return r;
5574 }
5575
5576 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5577 {
5578         vcpu_load(vcpu);
5579         kvm_mmu_unload(vcpu);
5580         vcpu_put(vcpu);
5581
5582         kvm_x86_ops->vcpu_free(vcpu);
5583 }
5584
5585 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5586 {
5587         vcpu->arch.nmi_pending = false;
5588         vcpu->arch.nmi_injected = false;
5589
5590         vcpu->arch.switch_db_regs = 0;
5591         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5592         vcpu->arch.dr6 = DR6_FIXED_1;
5593         vcpu->arch.dr7 = DR7_FIXED_1;
5594
5595         return kvm_x86_ops->vcpu_reset(vcpu);
5596 }
5597
5598 int kvm_arch_hardware_enable(void *garbage)
5599 {
5600         /*
5601          * Since this may be called from a hotplug notifcation,
5602          * we can't get the CPU frequency directly.
5603          */
5604         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5605                 int cpu = raw_smp_processor_id();
5606                 per_cpu(cpu_tsc_khz, cpu) = 0;
5607         }
5608
5609         kvm_shared_msr_cpu_online();
5610
5611         return kvm_x86_ops->hardware_enable(garbage);
5612 }
5613
5614 void kvm_arch_hardware_disable(void *garbage)
5615 {
5616         kvm_x86_ops->hardware_disable(garbage);
5617         drop_user_return_notifiers(garbage);
5618 }
5619
5620 int kvm_arch_hardware_setup(void)
5621 {
5622         return kvm_x86_ops->hardware_setup();
5623 }
5624
5625 void kvm_arch_hardware_unsetup(void)
5626 {
5627         kvm_x86_ops->hardware_unsetup();
5628 }
5629
5630 void kvm_arch_check_processor_compat(void *rtn)
5631 {
5632         kvm_x86_ops->check_processor_compatibility(rtn);
5633 }
5634
5635 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5636 {
5637         struct page *page;
5638         struct kvm *kvm;
5639         int r;
5640
5641         BUG_ON(vcpu->kvm == NULL);
5642         kvm = vcpu->kvm;
5643
5644         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5645         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5646                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5647         else
5648                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5649
5650         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5651         if (!page) {
5652                 r = -ENOMEM;
5653                 goto fail;
5654         }
5655         vcpu->arch.pio_data = page_address(page);
5656
5657         r = kvm_mmu_create(vcpu);
5658         if (r < 0)
5659                 goto fail_free_pio_data;
5660
5661         if (irqchip_in_kernel(kvm)) {
5662                 r = kvm_create_lapic(vcpu);
5663                 if (r < 0)
5664                         goto fail_mmu_destroy;
5665         }
5666
5667         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5668                                        GFP_KERNEL);
5669         if (!vcpu->arch.mce_banks) {
5670                 r = -ENOMEM;
5671                 goto fail_free_lapic;
5672         }
5673         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5674
5675         return 0;
5676 fail_free_lapic:
5677         kvm_free_lapic(vcpu);
5678 fail_mmu_destroy:
5679         kvm_mmu_destroy(vcpu);
5680 fail_free_pio_data:
5681         free_page((unsigned long)vcpu->arch.pio_data);
5682 fail:
5683         return r;
5684 }
5685
5686 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5687 {
5688         int idx;
5689
5690         kfree(vcpu->arch.mce_banks);
5691         kvm_free_lapic(vcpu);
5692         idx = srcu_read_lock(&vcpu->kvm->srcu);
5693         kvm_mmu_destroy(vcpu);
5694         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5695         free_page((unsigned long)vcpu->arch.pio_data);
5696 }
5697
5698 struct  kvm *kvm_arch_create_vm(void)
5699 {
5700         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5701
5702         if (!kvm)
5703                 return ERR_PTR(-ENOMEM);
5704
5705         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5706         if (!kvm->arch.aliases) {
5707                 kfree(kvm);
5708                 return ERR_PTR(-ENOMEM);
5709         }
5710
5711         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5712         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5713
5714         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5715         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5716
5717         rdtscll(kvm->arch.vm_init_tsc);
5718
5719         return kvm;
5720 }
5721
5722 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5723 {
5724         vcpu_load(vcpu);
5725         kvm_mmu_unload(vcpu);
5726         vcpu_put(vcpu);
5727 }
5728
5729 static void kvm_free_vcpus(struct kvm *kvm)
5730 {
5731         unsigned int i;
5732         struct kvm_vcpu *vcpu;
5733
5734         /*
5735          * Unpin any mmu pages first.
5736          */
5737         kvm_for_each_vcpu(i, vcpu, kvm)
5738                 kvm_unload_vcpu_mmu(vcpu);
5739         kvm_for_each_vcpu(i, vcpu, kvm)
5740                 kvm_arch_vcpu_free(vcpu);
5741
5742         mutex_lock(&kvm->lock);
5743         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5744                 kvm->vcpus[i] = NULL;
5745
5746         atomic_set(&kvm->online_vcpus, 0);
5747         mutex_unlock(&kvm->lock);
5748 }
5749
5750 void kvm_arch_sync_events(struct kvm *kvm)
5751 {
5752         kvm_free_all_assigned_devices(kvm);
5753 }
5754
5755 void kvm_arch_destroy_vm(struct kvm *kvm)
5756 {
5757         kvm_iommu_unmap_guest(kvm);
5758         kvm_free_pit(kvm);
5759         kfree(kvm->arch.vpic);
5760         kfree(kvm->arch.vioapic);
5761         kvm_free_vcpus(kvm);
5762         kvm_free_physmem(kvm);
5763         if (kvm->arch.apic_access_page)
5764                 put_page(kvm->arch.apic_access_page);
5765         if (kvm->arch.ept_identity_pagetable)
5766                 put_page(kvm->arch.ept_identity_pagetable);
5767         cleanup_srcu_struct(&kvm->srcu);
5768         kfree(kvm->arch.aliases);
5769         kfree(kvm);
5770 }
5771
5772 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5773                                 struct kvm_memory_slot *memslot,
5774                                 struct kvm_memory_slot old,
5775                                 struct kvm_userspace_memory_region *mem,
5776                                 int user_alloc)
5777 {
5778         int npages = memslot->npages;
5779
5780         /*To keep backward compatibility with older userspace,
5781          *x86 needs to hanlde !user_alloc case.
5782          */
5783         if (!user_alloc) {
5784                 if (npages && !old.rmap) {
5785                         unsigned long userspace_addr;
5786
5787                         down_write(&current->mm->mmap_sem);
5788                         userspace_addr = do_mmap(NULL, 0,
5789                                                  npages * PAGE_SIZE,
5790                                                  PROT_READ | PROT_WRITE,
5791                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5792                                                  0);
5793                         up_write(&current->mm->mmap_sem);
5794
5795                         if (IS_ERR((void *)userspace_addr))
5796                                 return PTR_ERR((void *)userspace_addr);
5797
5798                         memslot->userspace_addr = userspace_addr;
5799                 }
5800         }
5801
5802
5803         return 0;
5804 }
5805
5806 void kvm_arch_commit_memory_region(struct kvm *kvm,
5807                                 struct kvm_userspace_memory_region *mem,
5808                                 struct kvm_memory_slot old,
5809                                 int user_alloc)
5810 {
5811
5812         int npages = mem->memory_size >> PAGE_SHIFT;
5813
5814         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5815                 int ret;
5816
5817                 down_write(&current->mm->mmap_sem);
5818                 ret = do_munmap(current->mm, old.userspace_addr,
5819                                 old.npages * PAGE_SIZE);
5820                 up_write(&current->mm->mmap_sem);
5821                 if (ret < 0)
5822                         printk(KERN_WARNING
5823                                "kvm_vm_ioctl_set_memory_region: "
5824                                "failed to munmap memory\n");
5825         }
5826
5827         spin_lock(&kvm->mmu_lock);
5828         if (!kvm->arch.n_requested_mmu_pages) {
5829                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5830                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5831         }
5832
5833         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5834         spin_unlock(&kvm->mmu_lock);
5835 }
5836
5837 void kvm_arch_flush_shadow(struct kvm *kvm)
5838 {
5839         kvm_mmu_zap_all(kvm);
5840         kvm_reload_remote_mmus(kvm);
5841 }
5842
5843 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5844 {
5845         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5846                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5847                 || vcpu->arch.nmi_pending ||
5848                 (kvm_arch_interrupt_allowed(vcpu) &&
5849                  kvm_cpu_has_interrupt(vcpu));
5850 }
5851
5852 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5853 {
5854         int me;
5855         int cpu = vcpu->cpu;
5856
5857         if (waitqueue_active(&vcpu->wq)) {
5858                 wake_up_interruptible(&vcpu->wq);
5859                 ++vcpu->stat.halt_wakeup;
5860         }
5861
5862         me = get_cpu();
5863         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5864                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5865                         smp_send_reschedule(cpu);
5866         put_cpu();
5867 }
5868
5869 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5870 {
5871         return kvm_x86_ops->interrupt_allowed(vcpu);
5872 }
5873
5874 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5875 {
5876         unsigned long current_rip = kvm_rip_read(vcpu) +
5877                 get_segment_base(vcpu, VCPU_SREG_CS);
5878
5879         return current_rip == linear_rip;
5880 }
5881 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5882
5883 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5884 {
5885         unsigned long rflags;
5886
5887         rflags = kvm_x86_ops->get_rflags(vcpu);
5888         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5889                 rflags &= ~X86_EFLAGS_TF;
5890         return rflags;
5891 }
5892 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5893
5894 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5895 {
5896         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5897             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5898                 rflags |= X86_EFLAGS_TF;
5899         kvm_x86_ops->set_rflags(vcpu, rflags);
5900 }
5901 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5902
5903 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5904 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5905 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5906 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5907 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5908 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5909 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5910 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5911 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5912 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5913 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5914 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);