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