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 descriptor_table 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.base;
 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_X86_ROBUST_SINGLESTEP:
1552                 r = 1;
1553                 break;
1554         case KVM_CAP_COALESCED_MMIO:
1555                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1556                 break;
1557         case KVM_CAP_VAPIC:
1558                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1559                 break;
1560         case KVM_CAP_NR_VCPUS:
1561                 r = KVM_MAX_VCPUS;
1562                 break;
1563         case KVM_CAP_NR_MEMSLOTS:
1564                 r = KVM_MEMORY_SLOTS;
1565                 break;
1566         case KVM_CAP_PV_MMU:    /* obsolete */
1567                 r = 0;
1568                 break;
1569         case KVM_CAP_IOMMU:
1570                 r = iommu_found();
1571                 break;
1572         case KVM_CAP_MCE:
1573                 r = KVM_MAX_MCE_BANKS;
1574                 break;
1575         default:
1576                 r = 0;
1577                 break;
1578         }
1579         return r;
1580
1581 }
1582
1583 long kvm_arch_dev_ioctl(struct file *filp,
1584                         unsigned int ioctl, unsigned long arg)
1585 {
1586         void __user *argp = (void __user *)arg;
1587         long r;
1588
1589         switch (ioctl) {
1590         case KVM_GET_MSR_INDEX_LIST: {
1591                 struct kvm_msr_list __user *user_msr_list = argp;
1592                 struct kvm_msr_list msr_list;
1593                 unsigned n;
1594
1595                 r = -EFAULT;
1596                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1597                         goto out;
1598                 n = msr_list.nmsrs;
1599                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1600                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1601                         goto out;
1602                 r = -E2BIG;
1603                 if (n < msr_list.nmsrs)
1604                         goto out;
1605                 r = -EFAULT;
1606                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1607                                  num_msrs_to_save * sizeof(u32)))
1608                         goto out;
1609                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1610                                  &emulated_msrs,
1611                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1612                         goto out;
1613                 r = 0;
1614                 break;
1615         }
1616         case KVM_GET_SUPPORTED_CPUID: {
1617                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1618                 struct kvm_cpuid2 cpuid;
1619
1620                 r = -EFAULT;
1621                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1622                         goto out;
1623                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1624                                                       cpuid_arg->entries);
1625                 if (r)
1626                         goto out;
1627
1628                 r = -EFAULT;
1629                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1630                         goto out;
1631                 r = 0;
1632                 break;
1633         }
1634         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1635                 u64 mce_cap;
1636
1637                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1638                 r = -EFAULT;
1639                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1640                         goto out;
1641                 r = 0;
1642                 break;
1643         }
1644         default:
1645                 r = -EINVAL;
1646         }
1647 out:
1648         return r;
1649 }
1650
1651 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1652 {
1653         kvm_x86_ops->vcpu_load(vcpu, cpu);
1654         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1655                 unsigned long khz = cpufreq_quick_get(cpu);
1656                 if (!khz)
1657                         khz = tsc_khz;
1658                 per_cpu(cpu_tsc_khz, cpu) = khz;
1659         }
1660         kvm_request_guest_time_update(vcpu);
1661 }
1662
1663 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1664 {
1665         kvm_put_guest_fpu(vcpu);
1666         kvm_x86_ops->vcpu_put(vcpu);
1667 }
1668
1669 static int is_efer_nx(void)
1670 {
1671         unsigned long long efer = 0;
1672
1673         rdmsrl_safe(MSR_EFER, &efer);
1674         return efer & EFER_NX;
1675 }
1676
1677 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1678 {
1679         int i;
1680         struct kvm_cpuid_entry2 *e, *entry;
1681
1682         entry = NULL;
1683         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1684                 e = &vcpu->arch.cpuid_entries[i];
1685                 if (e->function == 0x80000001) {
1686                         entry = e;
1687                         break;
1688                 }
1689         }
1690         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1691                 entry->edx &= ~(1 << 20);
1692                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1693         }
1694 }
1695
1696 /* when an old userspace process fills a new kernel module */
1697 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1698                                     struct kvm_cpuid *cpuid,
1699                                     struct kvm_cpuid_entry __user *entries)
1700 {
1701         int r, i;
1702         struct kvm_cpuid_entry *cpuid_entries;
1703
1704         r = -E2BIG;
1705         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1706                 goto out;
1707         r = -ENOMEM;
1708         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1709         if (!cpuid_entries)
1710                 goto out;
1711         r = -EFAULT;
1712         if (copy_from_user(cpuid_entries, entries,
1713                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1714                 goto out_free;
1715         for (i = 0; i < cpuid->nent; i++) {
1716                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1717                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1718                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1719                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1720                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1721                 vcpu->arch.cpuid_entries[i].index = 0;
1722                 vcpu->arch.cpuid_entries[i].flags = 0;
1723                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1724                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1725                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1726         }
1727         vcpu->arch.cpuid_nent = cpuid->nent;
1728         cpuid_fix_nx_cap(vcpu);
1729         r = 0;
1730         kvm_apic_set_version(vcpu);
1731         kvm_x86_ops->cpuid_update(vcpu);
1732
1733 out_free:
1734         vfree(cpuid_entries);
1735 out:
1736         return r;
1737 }
1738
1739 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1740                                      struct kvm_cpuid2 *cpuid,
1741                                      struct kvm_cpuid_entry2 __user *entries)
1742 {
1743         int r;
1744
1745         r = -E2BIG;
1746         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1747                 goto out;
1748         r = -EFAULT;
1749         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1750                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1751                 goto out;
1752         vcpu->arch.cpuid_nent = cpuid->nent;
1753         kvm_apic_set_version(vcpu);
1754         kvm_x86_ops->cpuid_update(vcpu);
1755         return 0;
1756
1757 out:
1758         return r;
1759 }
1760
1761 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1762                                      struct kvm_cpuid2 *cpuid,
1763                                      struct kvm_cpuid_entry2 __user *entries)
1764 {
1765         int r;
1766
1767         r = -E2BIG;
1768         if (cpuid->nent < vcpu->arch.cpuid_nent)
1769                 goto out;
1770         r = -EFAULT;
1771         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1772                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1773                 goto out;
1774         return 0;
1775
1776 out:
1777         cpuid->nent = vcpu->arch.cpuid_nent;
1778         return r;
1779 }
1780
1781 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1782                            u32 index)
1783 {
1784         entry->function = function;
1785         entry->index = index;
1786         cpuid_count(entry->function, entry->index,
1787                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1788         entry->flags = 0;
1789 }
1790
1791 #define F(x) bit(X86_FEATURE_##x)
1792
1793 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1794                          u32 index, int *nent, int maxnent)
1795 {
1796         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1797 #ifdef CONFIG_X86_64
1798         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1799                                 ? F(GBPAGES) : 0;
1800         unsigned f_lm = F(LM);
1801 #else
1802         unsigned f_gbpages = 0;
1803         unsigned f_lm = 0;
1804 #endif
1805         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1806
1807         /* cpuid 1.edx */
1808         const u32 kvm_supported_word0_x86_features =
1809                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1810                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1811                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1812                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1813                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1814                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1815                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1816                 0 /* HTT, TM, Reserved, PBE */;
1817         /* cpuid 0x80000001.edx */
1818         const u32 kvm_supported_word1_x86_features =
1819                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1820                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1821                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1822                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1823                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1824                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1825                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1826                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1827         /* cpuid 1.ecx */
1828         const u32 kvm_supported_word4_x86_features =
1829                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1830                 0 /* DS-CPL, VMX, SMX, EST */ |
1831                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1832                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1833                 0 /* Reserved, DCA */ | F(XMM4_1) |
1834                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1835                 0 /* Reserved, XSAVE, OSXSAVE */;
1836         /* cpuid 0x80000001.ecx */
1837         const u32 kvm_supported_word6_x86_features =
1838                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1839                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1840                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1841                 0 /* SKINIT */ | 0 /* WDT */;
1842
1843         /* all calls to cpuid_count() should be made on the same cpu */
1844         get_cpu();
1845         do_cpuid_1_ent(entry, function, index);
1846         ++*nent;
1847
1848         switch (function) {
1849         case 0:
1850                 entry->eax = min(entry->eax, (u32)0xb);
1851                 break;
1852         case 1:
1853                 entry->edx &= kvm_supported_word0_x86_features;
1854                 entry->ecx &= kvm_supported_word4_x86_features;
1855                 /* we support x2apic emulation even if host does not support
1856                  * it since we emulate x2apic in software */
1857                 entry->ecx |= F(X2APIC);
1858                 break;
1859         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1860          * may return different values. This forces us to get_cpu() before
1861          * issuing the first command, and also to emulate this annoying behavior
1862          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1863         case 2: {
1864                 int t, times = entry->eax & 0xff;
1865
1866                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1867                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1868                 for (t = 1; t < times && *nent < maxnent; ++t) {
1869                         do_cpuid_1_ent(&entry[t], function, 0);
1870                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1871                         ++*nent;
1872                 }
1873                 break;
1874         }
1875         /* function 4 and 0xb have additional index. */
1876         case 4: {
1877                 int i, cache_type;
1878
1879                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1880                 /* read more entries until cache_type is zero */
1881                 for (i = 1; *nent < maxnent; ++i) {
1882                         cache_type = entry[i - 1].eax & 0x1f;
1883                         if (!cache_type)
1884                                 break;
1885                         do_cpuid_1_ent(&entry[i], function, i);
1886                         entry[i].flags |=
1887                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1888                         ++*nent;
1889                 }
1890                 break;
1891         }
1892         case 0xb: {
1893                 int i, level_type;
1894
1895                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1896                 /* read more entries until level_type is zero */
1897                 for (i = 1; *nent < maxnent; ++i) {
1898                         level_type = entry[i - 1].ecx & 0xff00;
1899                         if (!level_type)
1900                                 break;
1901                         do_cpuid_1_ent(&entry[i], function, i);
1902                         entry[i].flags |=
1903                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1904                         ++*nent;
1905                 }
1906                 break;
1907         }
1908         case 0x80000000:
1909                 entry->eax = min(entry->eax, 0x8000001a);
1910                 break;
1911         case 0x80000001:
1912                 entry->edx &= kvm_supported_word1_x86_features;
1913                 entry->ecx &= kvm_supported_word6_x86_features;
1914                 break;
1915         }
1916         put_cpu();
1917 }
1918
1919 #undef F
1920
1921 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1922                                      struct kvm_cpuid_entry2 __user *entries)
1923 {
1924         struct kvm_cpuid_entry2 *cpuid_entries;
1925         int limit, nent = 0, r = -E2BIG;
1926         u32 func;
1927
1928         if (cpuid->nent < 1)
1929                 goto out;
1930         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1931                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1932         r = -ENOMEM;
1933         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1934         if (!cpuid_entries)
1935                 goto out;
1936
1937         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1938         limit = cpuid_entries[0].eax;
1939         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1940                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1941                              &nent, cpuid->nent);
1942         r = -E2BIG;
1943         if (nent >= cpuid->nent)
1944                 goto out_free;
1945
1946         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1947         limit = cpuid_entries[nent - 1].eax;
1948         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1949                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1950                              &nent, cpuid->nent);
1951         r = -E2BIG;
1952         if (nent >= cpuid->nent)
1953                 goto out_free;
1954
1955         r = -EFAULT;
1956         if (copy_to_user(entries, cpuid_entries,
1957                          nent * sizeof(struct kvm_cpuid_entry2)))
1958                 goto out_free;
1959         cpuid->nent = nent;
1960         r = 0;
1961
1962 out_free:
1963         vfree(cpuid_entries);
1964 out:
1965         return r;
1966 }
1967
1968 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1969                                     struct kvm_lapic_state *s)
1970 {
1971         vcpu_load(vcpu);
1972         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1973         vcpu_put(vcpu);
1974
1975         return 0;
1976 }
1977
1978 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1979                                     struct kvm_lapic_state *s)
1980 {
1981         vcpu_load(vcpu);
1982         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1983         kvm_apic_post_state_restore(vcpu);
1984         update_cr8_intercept(vcpu);
1985         vcpu_put(vcpu);
1986
1987         return 0;
1988 }
1989
1990 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1991                                     struct kvm_interrupt *irq)
1992 {
1993         if (irq->irq < 0 || irq->irq >= 256)
1994                 return -EINVAL;
1995         if (irqchip_in_kernel(vcpu->kvm))
1996                 return -ENXIO;
1997         vcpu_load(vcpu);
1998
1999         kvm_queue_interrupt(vcpu, irq->irq, false);
2000
2001         vcpu_put(vcpu);
2002
2003         return 0;
2004 }
2005
2006 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2007 {
2008         vcpu_load(vcpu);
2009         kvm_inject_nmi(vcpu);
2010         vcpu_put(vcpu);
2011
2012         return 0;
2013 }
2014
2015 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2016                                            struct kvm_tpr_access_ctl *tac)
2017 {
2018         if (tac->flags)
2019                 return -EINVAL;
2020         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2021         return 0;
2022 }
2023
2024 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2025                                         u64 mcg_cap)
2026 {
2027         int r;
2028         unsigned bank_num = mcg_cap & 0xff, bank;
2029
2030         r = -EINVAL;
2031         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2032                 goto out;
2033         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2034                 goto out;
2035         r = 0;
2036         vcpu->arch.mcg_cap = mcg_cap;
2037         /* Init IA32_MCG_CTL to all 1s */
2038         if (mcg_cap & MCG_CTL_P)
2039                 vcpu->arch.mcg_ctl = ~(u64)0;
2040         /* Init IA32_MCi_CTL to all 1s */
2041         for (bank = 0; bank < bank_num; bank++)
2042                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2043 out:
2044         return r;
2045 }
2046
2047 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2048                                       struct kvm_x86_mce *mce)
2049 {
2050         u64 mcg_cap = vcpu->arch.mcg_cap;
2051         unsigned bank_num = mcg_cap & 0xff;
2052         u64 *banks = vcpu->arch.mce_banks;
2053
2054         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2055                 return -EINVAL;
2056         /*
2057          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2058          * reporting is disabled
2059          */
2060         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2061             vcpu->arch.mcg_ctl != ~(u64)0)
2062                 return 0;
2063         banks += 4 * mce->bank;
2064         /*
2065          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2066          * reporting is disabled for the bank
2067          */
2068         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2069                 return 0;
2070         if (mce->status & MCI_STATUS_UC) {
2071                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2072                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2073                         printk(KERN_DEBUG "kvm: set_mce: "
2074                                "injects mce exception while "
2075                                "previous one is in progress!\n");
2076                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2077                         return 0;
2078                 }
2079                 if (banks[1] & MCI_STATUS_VAL)
2080                         mce->status |= MCI_STATUS_OVER;
2081                 banks[2] = mce->addr;
2082                 banks[3] = mce->misc;
2083                 vcpu->arch.mcg_status = mce->mcg_status;
2084                 banks[1] = mce->status;
2085                 kvm_queue_exception(vcpu, MC_VECTOR);
2086         } else if (!(banks[1] & MCI_STATUS_VAL)
2087                    || !(banks[1] & MCI_STATUS_UC)) {
2088                 if (banks[1] & MCI_STATUS_VAL)
2089                         mce->status |= MCI_STATUS_OVER;
2090                 banks[2] = mce->addr;
2091                 banks[3] = mce->misc;
2092                 banks[1] = mce->status;
2093         } else
2094                 banks[1] |= MCI_STATUS_OVER;
2095         return 0;
2096 }
2097
2098 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2099                                                struct kvm_vcpu_events *events)
2100 {
2101         vcpu_load(vcpu);
2102
2103         events->exception.injected = vcpu->arch.exception.pending;
2104         events->exception.nr = vcpu->arch.exception.nr;
2105         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2106         events->exception.error_code = vcpu->arch.exception.error_code;
2107
2108         events->interrupt.injected = vcpu->arch.interrupt.pending;
2109         events->interrupt.nr = vcpu->arch.interrupt.nr;
2110         events->interrupt.soft = vcpu->arch.interrupt.soft;
2111
2112         events->nmi.injected = vcpu->arch.nmi_injected;
2113         events->nmi.pending = vcpu->arch.nmi_pending;
2114         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2115
2116         events->sipi_vector = vcpu->arch.sipi_vector;
2117
2118         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2119                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2120
2121         vcpu_put(vcpu);
2122 }
2123
2124 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2125                                               struct kvm_vcpu_events *events)
2126 {
2127         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2128                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2129                 return -EINVAL;
2130
2131         vcpu_load(vcpu);
2132
2133         vcpu->arch.exception.pending = events->exception.injected;
2134         vcpu->arch.exception.nr = events->exception.nr;
2135         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2136         vcpu->arch.exception.error_code = events->exception.error_code;
2137
2138         vcpu->arch.interrupt.pending = events->interrupt.injected;
2139         vcpu->arch.interrupt.nr = events->interrupt.nr;
2140         vcpu->arch.interrupt.soft = events->interrupt.soft;
2141         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2142                 kvm_pic_clear_isr_ack(vcpu->kvm);
2143
2144         vcpu->arch.nmi_injected = events->nmi.injected;
2145         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2146                 vcpu->arch.nmi_pending = events->nmi.pending;
2147         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2148
2149         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2150                 vcpu->arch.sipi_vector = events->sipi_vector;
2151
2152         vcpu_put(vcpu);
2153
2154         return 0;
2155 }
2156
2157 long kvm_arch_vcpu_ioctl(struct file *filp,
2158                          unsigned int ioctl, unsigned long arg)
2159 {
2160         struct kvm_vcpu *vcpu = filp->private_data;
2161         void __user *argp = (void __user *)arg;
2162         int r;
2163         struct kvm_lapic_state *lapic = NULL;
2164
2165         switch (ioctl) {
2166         case KVM_GET_LAPIC: {
2167                 r = -EINVAL;
2168                 if (!vcpu->arch.apic)
2169                         goto out;
2170                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2171
2172                 r = -ENOMEM;
2173                 if (!lapic)
2174                         goto out;
2175                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2176                 if (r)
2177                         goto out;
2178                 r = -EFAULT;
2179                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2180                         goto out;
2181                 r = 0;
2182                 break;
2183         }
2184         case KVM_SET_LAPIC: {
2185                 r = -EINVAL;
2186                 if (!vcpu->arch.apic)
2187                         goto out;
2188                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2189                 r = -ENOMEM;
2190                 if (!lapic)
2191                         goto out;
2192                 r = -EFAULT;
2193                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2194                         goto out;
2195                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2196                 if (r)
2197                         goto out;
2198                 r = 0;
2199                 break;
2200         }
2201         case KVM_INTERRUPT: {
2202                 struct kvm_interrupt irq;
2203
2204                 r = -EFAULT;
2205                 if (copy_from_user(&irq, argp, sizeof irq))
2206                         goto out;
2207                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2208                 if (r)
2209                         goto out;
2210                 r = 0;
2211                 break;
2212         }
2213         case KVM_NMI: {
2214                 r = kvm_vcpu_ioctl_nmi(vcpu);
2215                 if (r)
2216                         goto out;
2217                 r = 0;
2218                 break;
2219         }
2220         case KVM_SET_CPUID: {
2221                 struct kvm_cpuid __user *cpuid_arg = argp;
2222                 struct kvm_cpuid cpuid;
2223
2224                 r = -EFAULT;
2225                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2226                         goto out;
2227                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2228                 if (r)
2229                         goto out;
2230                 break;
2231         }
2232         case KVM_SET_CPUID2: {
2233                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2234                 struct kvm_cpuid2 cpuid;
2235
2236                 r = -EFAULT;
2237                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2238                         goto out;
2239                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2240                                               cpuid_arg->entries);
2241                 if (r)
2242                         goto out;
2243                 break;
2244         }
2245         case KVM_GET_CPUID2: {
2246                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2247                 struct kvm_cpuid2 cpuid;
2248
2249                 r = -EFAULT;
2250                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2251                         goto out;
2252                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2253                                               cpuid_arg->entries);
2254                 if (r)
2255                         goto out;
2256                 r = -EFAULT;
2257                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2258                         goto out;
2259                 r = 0;
2260                 break;
2261         }
2262         case KVM_GET_MSRS:
2263                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2264                 break;
2265         case KVM_SET_MSRS:
2266                 r = msr_io(vcpu, argp, do_set_msr, 0);
2267                 break;
2268         case KVM_TPR_ACCESS_REPORTING: {
2269                 struct kvm_tpr_access_ctl tac;
2270
2271                 r = -EFAULT;
2272                 if (copy_from_user(&tac, argp, sizeof tac))
2273                         goto out;
2274                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2275                 if (r)
2276                         goto out;
2277                 r = -EFAULT;
2278                 if (copy_to_user(argp, &tac, sizeof tac))
2279                         goto out;
2280                 r = 0;
2281                 break;
2282         };
2283         case KVM_SET_VAPIC_ADDR: {
2284                 struct kvm_vapic_addr va;
2285
2286                 r = -EINVAL;
2287                 if (!irqchip_in_kernel(vcpu->kvm))
2288                         goto out;
2289                 r = -EFAULT;
2290                 if (copy_from_user(&va, argp, sizeof va))
2291                         goto out;
2292                 r = 0;
2293                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2294                 break;
2295         }
2296         case KVM_X86_SETUP_MCE: {
2297                 u64 mcg_cap;
2298
2299                 r = -EFAULT;
2300                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2301                         goto out;
2302                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2303                 break;
2304         }
2305         case KVM_X86_SET_MCE: {
2306                 struct kvm_x86_mce mce;
2307
2308                 r = -EFAULT;
2309                 if (copy_from_user(&mce, argp, sizeof mce))
2310                         goto out;
2311                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2312                 break;
2313         }
2314         case KVM_GET_VCPU_EVENTS: {
2315                 struct kvm_vcpu_events events;
2316
2317                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2318
2319                 r = -EFAULT;
2320                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2321                         break;
2322                 r = 0;
2323                 break;
2324         }
2325         case KVM_SET_VCPU_EVENTS: {
2326                 struct kvm_vcpu_events events;
2327
2328                 r = -EFAULT;
2329                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2330                         break;
2331
2332                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2333                 break;
2334         }
2335         default:
2336                 r = -EINVAL;
2337         }
2338 out:
2339         kfree(lapic);
2340         return r;
2341 }
2342
2343 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2344 {
2345         int ret;
2346
2347         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2348                 return -1;
2349         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2350         return ret;
2351 }
2352
2353 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2354                                               u64 ident_addr)
2355 {
2356         kvm->arch.ept_identity_map_addr = ident_addr;
2357         return 0;
2358 }
2359
2360 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2361                                           u32 kvm_nr_mmu_pages)
2362 {
2363         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2364                 return -EINVAL;
2365
2366         mutex_lock(&kvm->slots_lock);
2367         spin_lock(&kvm->mmu_lock);
2368
2369         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2370         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2371
2372         spin_unlock(&kvm->mmu_lock);
2373         mutex_unlock(&kvm->slots_lock);
2374         return 0;
2375 }
2376
2377 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2378 {
2379         return kvm->arch.n_alloc_mmu_pages;
2380 }
2381
2382 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2383 {
2384         int i;
2385         struct kvm_mem_alias *alias;
2386         struct kvm_mem_aliases *aliases;
2387
2388         aliases = rcu_dereference(kvm->arch.aliases);
2389
2390         for (i = 0; i < aliases->naliases; ++i) {
2391                 alias = &aliases->aliases[i];
2392                 if (alias->flags & KVM_ALIAS_INVALID)
2393                         continue;
2394                 if (gfn >= alias->base_gfn
2395                     && gfn < alias->base_gfn + alias->npages)
2396                         return alias->target_gfn + gfn - alias->base_gfn;
2397         }
2398         return gfn;
2399 }
2400
2401 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2402 {
2403         int i;
2404         struct kvm_mem_alias *alias;
2405         struct kvm_mem_aliases *aliases;
2406
2407         aliases = rcu_dereference(kvm->arch.aliases);
2408
2409         for (i = 0; i < aliases->naliases; ++i) {
2410                 alias = &aliases->aliases[i];
2411                 if (gfn >= alias->base_gfn
2412                     && gfn < alias->base_gfn + alias->npages)
2413                         return alias->target_gfn + gfn - alias->base_gfn;
2414         }
2415         return gfn;
2416 }
2417
2418 /*
2419  * Set a new alias region.  Aliases map a portion of physical memory into
2420  * another portion.  This is useful for memory windows, for example the PC
2421  * VGA region.
2422  */
2423 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2424                                          struct kvm_memory_alias *alias)
2425 {
2426         int r, n;
2427         struct kvm_mem_alias *p;
2428         struct kvm_mem_aliases *aliases, *old_aliases;
2429
2430         r = -EINVAL;
2431         /* General sanity checks */
2432         if (alias->memory_size & (PAGE_SIZE - 1))
2433                 goto out;
2434         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2435                 goto out;
2436         if (alias->slot >= KVM_ALIAS_SLOTS)
2437                 goto out;
2438         if (alias->guest_phys_addr + alias->memory_size
2439             < alias->guest_phys_addr)
2440                 goto out;
2441         if (alias->target_phys_addr + alias->memory_size
2442             < alias->target_phys_addr)
2443                 goto out;
2444
2445         r = -ENOMEM;
2446         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2447         if (!aliases)
2448                 goto out;
2449
2450         mutex_lock(&kvm->slots_lock);
2451
2452         /* invalidate any gfn reference in case of deletion/shrinking */
2453         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2454         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2455         old_aliases = kvm->arch.aliases;
2456         rcu_assign_pointer(kvm->arch.aliases, aliases);
2457         synchronize_srcu_expedited(&kvm->srcu);
2458         kvm_mmu_zap_all(kvm);
2459         kfree(old_aliases);
2460
2461         r = -ENOMEM;
2462         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2463         if (!aliases)
2464                 goto out_unlock;
2465
2466         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2467
2468         p = &aliases->aliases[alias->slot];
2469         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2470         p->npages = alias->memory_size >> PAGE_SHIFT;
2471         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2472         p->flags &= ~(KVM_ALIAS_INVALID);
2473
2474         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2475                 if (aliases->aliases[n - 1].npages)
2476                         break;
2477         aliases->naliases = n;
2478
2479         old_aliases = kvm->arch.aliases;
2480         rcu_assign_pointer(kvm->arch.aliases, aliases);
2481         synchronize_srcu_expedited(&kvm->srcu);
2482         kfree(old_aliases);
2483         r = 0;
2484
2485 out_unlock:
2486         mutex_unlock(&kvm->slots_lock);
2487 out:
2488         return r;
2489 }
2490
2491 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2492 {
2493         int r;
2494
2495         r = 0;
2496         switch (chip->chip_id) {
2497         case KVM_IRQCHIP_PIC_MASTER:
2498                 memcpy(&chip->chip.pic,
2499                         &pic_irqchip(kvm)->pics[0],
2500                         sizeof(struct kvm_pic_state));
2501                 break;
2502         case KVM_IRQCHIP_PIC_SLAVE:
2503                 memcpy(&chip->chip.pic,
2504                         &pic_irqchip(kvm)->pics[1],
2505                         sizeof(struct kvm_pic_state));
2506                 break;
2507         case KVM_IRQCHIP_IOAPIC:
2508                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2509                 break;
2510         default:
2511                 r = -EINVAL;
2512                 break;
2513         }
2514         return r;
2515 }
2516
2517 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2518 {
2519         int r;
2520
2521         r = 0;
2522         switch (chip->chip_id) {
2523         case KVM_IRQCHIP_PIC_MASTER:
2524                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2525                 memcpy(&pic_irqchip(kvm)->pics[0],
2526                         &chip->chip.pic,
2527                         sizeof(struct kvm_pic_state));
2528                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2529                 break;
2530         case KVM_IRQCHIP_PIC_SLAVE:
2531                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2532                 memcpy(&pic_irqchip(kvm)->pics[1],
2533                         &chip->chip.pic,
2534                         sizeof(struct kvm_pic_state));
2535                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2536                 break;
2537         case KVM_IRQCHIP_IOAPIC:
2538                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2539                 break;
2540         default:
2541                 r = -EINVAL;
2542                 break;
2543         }
2544         kvm_pic_update_irq(pic_irqchip(kvm));
2545         return r;
2546 }
2547
2548 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2549 {
2550         int r = 0;
2551
2552         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2553         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2554         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2555         return r;
2556 }
2557
2558 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2559 {
2560         int r = 0;
2561
2562         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2563         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2564         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2565         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2566         return r;
2567 }
2568
2569 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2570 {
2571         int r = 0;
2572
2573         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2574         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2575                 sizeof(ps->channels));
2576         ps->flags = kvm->arch.vpit->pit_state.flags;
2577         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2578         return r;
2579 }
2580
2581 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2582 {
2583         int r = 0, start = 0;
2584         u32 prev_legacy, cur_legacy;
2585         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2586         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2587         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2588         if (!prev_legacy && cur_legacy)
2589                 start = 1;
2590         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2591                sizeof(kvm->arch.vpit->pit_state.channels));
2592         kvm->arch.vpit->pit_state.flags = ps->flags;
2593         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2594         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2595         return r;
2596 }
2597
2598 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2599                                  struct kvm_reinject_control *control)
2600 {
2601         if (!kvm->arch.vpit)
2602                 return -ENXIO;
2603         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2604         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2605         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2606         return 0;
2607 }
2608
2609 /*
2610  * Get (and clear) the dirty memory log for a memory slot.
2611  */
2612 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2613                                       struct kvm_dirty_log *log)
2614 {
2615         int r, n, i;
2616         struct kvm_memory_slot *memslot;
2617         unsigned long is_dirty = 0;
2618         unsigned long *dirty_bitmap = NULL;
2619
2620         mutex_lock(&kvm->slots_lock);
2621
2622         r = -EINVAL;
2623         if (log->slot >= KVM_MEMORY_SLOTS)
2624                 goto out;
2625
2626         memslot = &kvm->memslots->memslots[log->slot];
2627         r = -ENOENT;
2628         if (!memslot->dirty_bitmap)
2629                 goto out;
2630
2631         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2632
2633         r = -ENOMEM;
2634         dirty_bitmap = vmalloc(n);
2635         if (!dirty_bitmap)
2636                 goto out;
2637         memset(dirty_bitmap, 0, n);
2638
2639         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2640                 is_dirty = memslot->dirty_bitmap[i];
2641
2642         /* If nothing is dirty, don't bother messing with page tables. */
2643         if (is_dirty) {
2644                 struct kvm_memslots *slots, *old_slots;
2645
2646                 spin_lock(&kvm->mmu_lock);
2647                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2648                 spin_unlock(&kvm->mmu_lock);
2649
2650                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2651                 if (!slots)
2652                         goto out_free;
2653
2654                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2655                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2656
2657                 old_slots = kvm->memslots;
2658                 rcu_assign_pointer(kvm->memslots, slots);
2659                 synchronize_srcu_expedited(&kvm->srcu);
2660                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2661                 kfree(old_slots);
2662         }
2663
2664         r = 0;
2665         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2666                 r = -EFAULT;
2667 out_free:
2668         vfree(dirty_bitmap);
2669 out:
2670         mutex_unlock(&kvm->slots_lock);
2671         return r;
2672 }
2673
2674 long kvm_arch_vm_ioctl(struct file *filp,
2675                        unsigned int ioctl, unsigned long arg)
2676 {
2677         struct kvm *kvm = filp->private_data;
2678         void __user *argp = (void __user *)arg;
2679         int r = -ENOTTY;
2680         /*
2681          * This union makes it completely explicit to gcc-3.x
2682          * that these two variables' stack usage should be
2683          * combined, not added together.
2684          */
2685         union {
2686                 struct kvm_pit_state ps;
2687                 struct kvm_pit_state2 ps2;
2688                 struct kvm_memory_alias alias;
2689                 struct kvm_pit_config pit_config;
2690         } u;
2691
2692         switch (ioctl) {
2693         case KVM_SET_TSS_ADDR:
2694                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2695                 if (r < 0)
2696                         goto out;
2697                 break;
2698         case KVM_SET_IDENTITY_MAP_ADDR: {
2699                 u64 ident_addr;
2700
2701                 r = -EFAULT;
2702                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2703                         goto out;
2704                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2705                 if (r < 0)
2706                         goto out;
2707                 break;
2708         }
2709         case KVM_SET_MEMORY_REGION: {
2710                 struct kvm_memory_region kvm_mem;
2711                 struct kvm_userspace_memory_region kvm_userspace_mem;
2712
2713                 r = -EFAULT;
2714                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2715                         goto out;
2716                 kvm_userspace_mem.slot = kvm_mem.slot;
2717                 kvm_userspace_mem.flags = kvm_mem.flags;
2718                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2719                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2720                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2721                 if (r)
2722                         goto out;
2723                 break;
2724         }
2725         case KVM_SET_NR_MMU_PAGES:
2726                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2727                 if (r)
2728                         goto out;
2729                 break;
2730         case KVM_GET_NR_MMU_PAGES:
2731                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2732                 break;
2733         case KVM_SET_MEMORY_ALIAS:
2734                 r = -EFAULT;
2735                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2736                         goto out;
2737                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2738                 if (r)
2739                         goto out;
2740                 break;
2741         case KVM_CREATE_IRQCHIP: {
2742                 struct kvm_pic *vpic;
2743
2744                 mutex_lock(&kvm->lock);
2745                 r = -EEXIST;
2746                 if (kvm->arch.vpic)
2747                         goto create_irqchip_unlock;
2748                 r = -ENOMEM;
2749                 vpic = kvm_create_pic(kvm);
2750                 if (vpic) {
2751                         r = kvm_ioapic_init(kvm);
2752                         if (r) {
2753                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2754                                                           &vpic->dev);
2755                                 kfree(vpic);
2756                                 goto create_irqchip_unlock;
2757                         }
2758                 } else
2759                         goto create_irqchip_unlock;
2760                 smp_wmb();
2761                 kvm->arch.vpic = vpic;
2762                 smp_wmb();
2763                 r = kvm_setup_default_irq_routing(kvm);
2764                 if (r) {
2765                         mutex_lock(&kvm->irq_lock);
2766                         kvm_ioapic_destroy(kvm);
2767                         kvm_destroy_pic(kvm);
2768                         mutex_unlock(&kvm->irq_lock);
2769                 }
2770         create_irqchip_unlock:
2771                 mutex_unlock(&kvm->lock);
2772                 break;
2773         }
2774         case KVM_CREATE_PIT:
2775                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2776                 goto create_pit;
2777         case KVM_CREATE_PIT2:
2778                 r = -EFAULT;
2779                 if (copy_from_user(&u.pit_config, argp,
2780                                    sizeof(struct kvm_pit_config)))
2781                         goto out;
2782         create_pit:
2783                 mutex_lock(&kvm->slots_lock);
2784                 r = -EEXIST;
2785                 if (kvm->arch.vpit)
2786                         goto create_pit_unlock;
2787                 r = -ENOMEM;
2788                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2789                 if (kvm->arch.vpit)
2790                         r = 0;
2791         create_pit_unlock:
2792                 mutex_unlock(&kvm->slots_lock);
2793                 break;
2794         case KVM_IRQ_LINE_STATUS:
2795         case KVM_IRQ_LINE: {
2796                 struct kvm_irq_level irq_event;
2797
2798                 r = -EFAULT;
2799                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2800                         goto out;
2801                 if (irqchip_in_kernel(kvm)) {
2802                         __s32 status;
2803                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2804                                         irq_event.irq, irq_event.level);
2805                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2806                                 irq_event.status = status;
2807                                 if (copy_to_user(argp, &irq_event,
2808                                                         sizeof irq_event))
2809                                         goto out;
2810                         }
2811                         r = 0;
2812                 }
2813                 break;
2814         }
2815         case KVM_GET_IRQCHIP: {
2816                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2817                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2818
2819                 r = -ENOMEM;
2820                 if (!chip)
2821                         goto out;
2822                 r = -EFAULT;
2823                 if (copy_from_user(chip, argp, sizeof *chip))
2824                         goto get_irqchip_out;
2825                 r = -ENXIO;
2826                 if (!irqchip_in_kernel(kvm))
2827                         goto get_irqchip_out;
2828                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2829                 if (r)
2830                         goto get_irqchip_out;
2831                 r = -EFAULT;
2832                 if (copy_to_user(argp, chip, sizeof *chip))
2833                         goto get_irqchip_out;
2834                 r = 0;
2835         get_irqchip_out:
2836                 kfree(chip);
2837                 if (r)
2838                         goto out;
2839                 break;
2840         }
2841         case KVM_SET_IRQCHIP: {
2842                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2843                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2844
2845                 r = -ENOMEM;
2846                 if (!chip)
2847                         goto out;
2848                 r = -EFAULT;
2849                 if (copy_from_user(chip, argp, sizeof *chip))
2850                         goto set_irqchip_out;
2851                 r = -ENXIO;
2852                 if (!irqchip_in_kernel(kvm))
2853                         goto set_irqchip_out;
2854                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2855                 if (r)
2856                         goto set_irqchip_out;
2857                 r = 0;
2858         set_irqchip_out:
2859                 kfree(chip);
2860                 if (r)
2861                         goto out;
2862                 break;
2863         }
2864         case KVM_GET_PIT: {
2865                 r = -EFAULT;
2866                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2867                         goto out;
2868                 r = -ENXIO;
2869                 if (!kvm->arch.vpit)
2870                         goto out;
2871                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2872                 if (r)
2873                         goto out;
2874                 r = -EFAULT;
2875                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2876                         goto out;
2877                 r = 0;
2878                 break;
2879         }
2880         case KVM_SET_PIT: {
2881                 r = -EFAULT;
2882                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2883                         goto out;
2884                 r = -ENXIO;
2885                 if (!kvm->arch.vpit)
2886                         goto out;
2887                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2888                 if (r)
2889                         goto out;
2890                 r = 0;
2891                 break;
2892         }
2893         case KVM_GET_PIT2: {
2894                 r = -ENXIO;
2895                 if (!kvm->arch.vpit)
2896                         goto out;
2897                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2898                 if (r)
2899                         goto out;
2900                 r = -EFAULT;
2901                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2902                         goto out;
2903                 r = 0;
2904                 break;
2905         }
2906         case KVM_SET_PIT2: {
2907                 r = -EFAULT;
2908                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2909                         goto out;
2910                 r = -ENXIO;
2911                 if (!kvm->arch.vpit)
2912                         goto out;
2913                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2914                 if (r)
2915                         goto out;
2916                 r = 0;
2917                 break;
2918         }
2919         case KVM_REINJECT_CONTROL: {
2920                 struct kvm_reinject_control control;
2921                 r =  -EFAULT;
2922                 if (copy_from_user(&control, argp, sizeof(control)))
2923                         goto out;
2924                 r = kvm_vm_ioctl_reinject(kvm, &control);
2925                 if (r)
2926                         goto out;
2927                 r = 0;
2928                 break;
2929         }
2930         case KVM_XEN_HVM_CONFIG: {
2931                 r = -EFAULT;
2932                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2933                                    sizeof(struct kvm_xen_hvm_config)))
2934                         goto out;
2935                 r = -EINVAL;
2936                 if (kvm->arch.xen_hvm_config.flags)
2937                         goto out;
2938                 r = 0;
2939                 break;
2940         }
2941         case KVM_SET_CLOCK: {
2942                 struct timespec now;
2943                 struct kvm_clock_data user_ns;
2944                 u64 now_ns;
2945                 s64 delta;
2946
2947                 r = -EFAULT;
2948                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2949                         goto out;
2950
2951                 r = -EINVAL;
2952                 if (user_ns.flags)
2953                         goto out;
2954
2955                 r = 0;
2956                 ktime_get_ts(&now);
2957                 now_ns = timespec_to_ns(&now);
2958                 delta = user_ns.clock - now_ns;
2959                 kvm->arch.kvmclock_offset = delta;
2960                 break;
2961         }
2962         case KVM_GET_CLOCK: {
2963                 struct timespec now;
2964                 struct kvm_clock_data user_ns;
2965                 u64 now_ns;
2966
2967                 ktime_get_ts(&now);
2968                 now_ns = timespec_to_ns(&now);
2969                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2970                 user_ns.flags = 0;
2971
2972                 r = -EFAULT;
2973                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2974                         goto out;
2975                 r = 0;
2976                 break;
2977         }
2978
2979         default:
2980                 ;
2981         }
2982 out:
2983         return r;
2984 }
2985
2986 static void kvm_init_msr_list(void)
2987 {
2988         u32 dummy[2];
2989         unsigned i, j;
2990
2991         /* skip the first msrs in the list. KVM-specific */
2992         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2993                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2994                         continue;
2995                 if (j < i)
2996                         msrs_to_save[j] = msrs_to_save[i];
2997                 j++;
2998         }
2999         num_msrs_to_save = j;
3000 }
3001
3002 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3003                            const void *v)
3004 {
3005         if (vcpu->arch.apic &&
3006             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3007                 return 0;
3008
3009         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3010 }
3011
3012 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3013 {
3014         if (vcpu->arch.apic &&
3015             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3016                 return 0;
3017
3018         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3019 }
3020
3021 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3022 {
3023         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3024         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3025 }
3026
3027  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3028 {
3029         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3030         access |= PFERR_FETCH_MASK;
3031         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3032 }
3033
3034 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3035 {
3036         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3037         access |= PFERR_WRITE_MASK;
3038         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3039 }
3040
3041 /* uses this to access any guest's mapped memory without checking CPL */
3042 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3043 {
3044         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3045 }
3046
3047 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3048                                       struct kvm_vcpu *vcpu, u32 access,
3049                                       u32 *error)
3050 {
3051         void *data = val;
3052         int r = X86EMUL_CONTINUE;
3053
3054         while (bytes) {
3055                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3056                 unsigned offset = addr & (PAGE_SIZE-1);
3057                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3058                 int ret;
3059
3060                 if (gpa == UNMAPPED_GVA) {
3061                         r = X86EMUL_PROPAGATE_FAULT;
3062                         goto out;
3063                 }
3064                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3065                 if (ret < 0) {
3066                         r = X86EMUL_UNHANDLEABLE;
3067                         goto out;
3068                 }
3069
3070                 bytes -= toread;
3071                 data += toread;
3072                 addr += toread;
3073         }
3074 out:
3075         return r;
3076 }
3077
3078 /* used for instruction fetching */
3079 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3080                                 struct kvm_vcpu *vcpu, u32 *error)
3081 {
3082         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3083         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3084                                           access | PFERR_FETCH_MASK, error);
3085 }
3086
3087 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3088                                struct kvm_vcpu *vcpu, u32 *error)
3089 {
3090         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3091         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3092                                           error);
3093 }
3094
3095 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3096                                struct kvm_vcpu *vcpu, u32 *error)
3097 {
3098         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3099 }
3100
3101 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3102                                 struct kvm_vcpu *vcpu, u32 *error)
3103 {
3104         void *data = val;
3105         int r = X86EMUL_CONTINUE;
3106
3107         while (bytes) {
3108                 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3109                 unsigned offset = addr & (PAGE_SIZE-1);
3110                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3111                 int ret;
3112
3113                 if (gpa == UNMAPPED_GVA) {
3114                         r = X86EMUL_PROPAGATE_FAULT;
3115                         goto out;
3116                 }
3117                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3118                 if (ret < 0) {
3119                         r = X86EMUL_UNHANDLEABLE;
3120                         goto out;
3121                 }
3122
3123                 bytes -= towrite;
3124                 data += towrite;
3125                 addr += towrite;
3126         }
3127 out:
3128         return r;
3129 }
3130
3131
3132 static int emulator_read_emulated(unsigned long addr,
3133                                   void *val,
3134                                   unsigned int bytes,
3135                                   struct kvm_vcpu *vcpu)
3136 {
3137         gpa_t                 gpa;
3138         u32 error_code;
3139
3140         if (vcpu->mmio_read_completed) {
3141                 memcpy(val, vcpu->mmio_data, bytes);
3142                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3143                                vcpu->mmio_phys_addr, *(u64 *)val);
3144                 vcpu->mmio_read_completed = 0;
3145                 return X86EMUL_CONTINUE;
3146         }
3147
3148         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3149
3150         if (gpa == UNMAPPED_GVA) {
3151                 kvm_inject_page_fault(vcpu, addr, error_code);
3152                 return X86EMUL_PROPAGATE_FAULT;
3153         }
3154
3155         /* For APIC access vmexit */
3156         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3157                 goto mmio;
3158
3159         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3160                                 == X86EMUL_CONTINUE)
3161                 return X86EMUL_CONTINUE;
3162
3163 mmio:
3164         /*
3165          * Is this MMIO handled locally?
3166          */
3167         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3168                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3169                 return X86EMUL_CONTINUE;
3170         }
3171
3172         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3173
3174         vcpu->mmio_needed = 1;
3175         vcpu->mmio_phys_addr = gpa;
3176         vcpu->mmio_size = bytes;
3177         vcpu->mmio_is_write = 0;
3178
3179         return X86EMUL_UNHANDLEABLE;
3180 }
3181
3182 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3183                           const void *val, int bytes)
3184 {
3185         int ret;
3186
3187         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3188         if (ret < 0)
3189                 return 0;
3190         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3191         return 1;
3192 }
3193
3194 static int emulator_write_emulated_onepage(unsigned long addr,
3195                                            const void *val,
3196                                            unsigned int bytes,
3197                                            struct kvm_vcpu *vcpu)
3198 {
3199         gpa_t                 gpa;
3200         u32 error_code;
3201
3202         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3203
3204         if (gpa == UNMAPPED_GVA) {
3205                 kvm_inject_page_fault(vcpu, addr, error_code);
3206                 return X86EMUL_PROPAGATE_FAULT;
3207         }
3208
3209         /* For APIC access vmexit */
3210         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3211                 goto mmio;
3212
3213         if (emulator_write_phys(vcpu, gpa, val, bytes))
3214                 return X86EMUL_CONTINUE;
3215
3216 mmio:
3217         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3218         /*
3219          * Is this MMIO handled locally?
3220          */
3221         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3222                 return X86EMUL_CONTINUE;
3223
3224         vcpu->mmio_needed = 1;
3225         vcpu->mmio_phys_addr = gpa;
3226         vcpu->mmio_size = bytes;
3227         vcpu->mmio_is_write = 1;
3228         memcpy(vcpu->mmio_data, val, bytes);
3229
3230         return X86EMUL_CONTINUE;
3231 }
3232
3233 int emulator_write_emulated(unsigned long addr,
3234                                    const void *val,
3235                                    unsigned int bytes,
3236                                    struct kvm_vcpu *vcpu)
3237 {
3238         /* Crossing a page boundary? */
3239         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3240                 int rc, now;
3241
3242                 now = -addr & ~PAGE_MASK;
3243                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3244                 if (rc != X86EMUL_CONTINUE)
3245                         return rc;
3246                 addr += now;
3247                 val += now;
3248                 bytes -= now;
3249         }
3250         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3251 }
3252 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3253
3254 static int emulator_cmpxchg_emulated(unsigned long addr,
3255                                      const void *old,
3256                                      const void *new,
3257                                      unsigned int bytes,
3258                                      struct kvm_vcpu *vcpu)
3259 {
3260         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3261 #ifndef CONFIG_X86_64
3262         /* guests cmpxchg8b have to be emulated atomically */
3263         if (bytes == 8) {
3264                 gpa_t gpa;
3265                 struct page *page;
3266                 char *kaddr;
3267                 u64 val;
3268
3269                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3270
3271                 if (gpa == UNMAPPED_GVA ||
3272                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3273                         goto emul_write;
3274
3275                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3276                         goto emul_write;
3277
3278                 val = *(u64 *)new;
3279
3280                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3281
3282                 kaddr = kmap_atomic(page, KM_USER0);
3283                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3284                 kunmap_atomic(kaddr, KM_USER0);
3285                 kvm_release_page_dirty(page);
3286         }
3287 emul_write:
3288 #endif
3289
3290         return emulator_write_emulated(addr, new, bytes, vcpu);
3291 }
3292
3293 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3294 {
3295         return kvm_x86_ops->get_segment_base(vcpu, seg);
3296 }
3297
3298 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3299 {
3300         kvm_mmu_invlpg(vcpu, address);
3301         return X86EMUL_CONTINUE;
3302 }
3303
3304 int emulate_clts(struct kvm_vcpu *vcpu)
3305 {
3306         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3307         kvm_x86_ops->fpu_activate(vcpu);
3308         return X86EMUL_CONTINUE;
3309 }
3310
3311 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3312 {
3313         return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3314 }
3315
3316 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3317 {
3318         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3319
3320         return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3321 }
3322
3323 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3324 {
3325         u8 opcodes[4];
3326         unsigned long rip = kvm_rip_read(vcpu);
3327         unsigned long rip_linear;
3328
3329         if (!printk_ratelimit())
3330                 return;
3331
3332         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3333
3334         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3335
3336         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3337                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3338 }
3339 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3340
3341 static struct x86_emulate_ops emulate_ops = {
3342         .read_std            = kvm_read_guest_virt_system,
3343         .fetch               = kvm_fetch_guest_virt,
3344         .read_emulated       = emulator_read_emulated,
3345         .write_emulated      = emulator_write_emulated,
3346         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3347 };
3348
3349 static void cache_all_regs(struct kvm_vcpu *vcpu)
3350 {
3351         kvm_register_read(vcpu, VCPU_REGS_RAX);
3352         kvm_register_read(vcpu, VCPU_REGS_RSP);
3353         kvm_register_read(vcpu, VCPU_REGS_RIP);
3354         vcpu->arch.regs_dirty = ~0;
3355 }
3356
3357 int emulate_instruction(struct kvm_vcpu *vcpu,
3358                         unsigned long cr2,
3359                         u16 error_code,
3360                         int emulation_type)
3361 {
3362         int r, shadow_mask;
3363         struct decode_cache *c;
3364         struct kvm_run *run = vcpu->run;
3365
3366         kvm_clear_exception_queue(vcpu);
3367         vcpu->arch.mmio_fault_cr2 = cr2;
3368         /*
3369          * TODO: fix emulate.c to use guest_read/write_register
3370          * instead of direct ->regs accesses, can save hundred cycles
3371          * on Intel for instructions that don't read/change RSP, for
3372          * for example.
3373          */
3374         cache_all_regs(vcpu);
3375
3376         vcpu->mmio_is_write = 0;
3377         vcpu->arch.pio.string = 0;
3378
3379         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3380                 int cs_db, cs_l;
3381                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3382
3383                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3384                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3385                 vcpu->arch.emulate_ctxt.mode =
3386                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3387                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3388                         ? X86EMUL_MODE_VM86 : cs_l
3389                         ? X86EMUL_MODE_PROT64 : cs_db
3390                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3391
3392                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3393
3394                 /* Only allow emulation of specific instructions on #UD
3395                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3396                 c = &vcpu->arch.emulate_ctxt.decode;
3397                 if (emulation_type & EMULTYPE_TRAP_UD) {
3398                         if (!c->twobyte)
3399                                 return EMULATE_FAIL;
3400                         switch (c->b) {
3401                         case 0x01: /* VMMCALL */
3402                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3403                                         return EMULATE_FAIL;
3404                                 break;
3405                         case 0x34: /* sysenter */
3406                         case 0x35: /* sysexit */
3407                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3408                                         return EMULATE_FAIL;
3409                                 break;
3410                         case 0x05: /* syscall */
3411                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3412                                         return EMULATE_FAIL;
3413                                 break;
3414                         default:
3415                                 return EMULATE_FAIL;
3416                         }
3417
3418                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3419                                 return EMULATE_FAIL;
3420                 }
3421
3422                 ++vcpu->stat.insn_emulation;
3423                 if (r)  {
3424                         ++vcpu->stat.insn_emulation_fail;
3425                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3426                                 return EMULATE_DONE;
3427                         return EMULATE_FAIL;
3428                 }
3429         }
3430
3431         if (emulation_type & EMULTYPE_SKIP) {
3432                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3433                 return EMULATE_DONE;
3434         }
3435
3436         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3437         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3438
3439         if (r == 0)
3440                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3441
3442         if (vcpu->arch.pio.string)
3443                 return EMULATE_DO_MMIO;
3444
3445         if ((r || vcpu->mmio_is_write) && run) {
3446                 run->exit_reason = KVM_EXIT_MMIO;
3447                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3448                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3449                 run->mmio.len = vcpu->mmio_size;
3450                 run->mmio.is_write = vcpu->mmio_is_write;
3451         }
3452
3453         if (r) {
3454                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3455                         return EMULATE_DONE;
3456                 if (!vcpu->mmio_needed) {
3457                         kvm_report_emulation_failure(vcpu, "mmio");
3458                         return EMULATE_FAIL;
3459                 }
3460                 return EMULATE_DO_MMIO;
3461         }
3462
3463         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3464
3465         if (vcpu->mmio_is_write) {
3466                 vcpu->mmio_needed = 0;
3467                 return EMULATE_DO_MMIO;
3468         }
3469
3470         return EMULATE_DONE;
3471 }
3472 EXPORT_SYMBOL_GPL(emulate_instruction);
3473
3474 static int pio_copy_data(struct kvm_vcpu *vcpu)
3475 {
3476         void *p = vcpu->arch.pio_data;
3477         gva_t q = vcpu->arch.pio.guest_gva;
3478         unsigned bytes;
3479         int ret;
3480         u32 error_code;
3481
3482         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3483         if (vcpu->arch.pio.in)
3484                 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3485         else
3486                 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3487
3488         if (ret == X86EMUL_PROPAGATE_FAULT)
3489                 kvm_inject_page_fault(vcpu, q, error_code);
3490
3491         return ret;
3492 }
3493
3494 int complete_pio(struct kvm_vcpu *vcpu)
3495 {
3496         struct kvm_pio_request *io = &vcpu->arch.pio;
3497         long delta;
3498         int r;
3499         unsigned long val;
3500
3501         if (!io->string) {
3502                 if (io->in) {
3503                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3504                         memcpy(&val, vcpu->arch.pio_data, io->size);
3505                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3506                 }
3507         } else {
3508                 if (io->in) {
3509                         r = pio_copy_data(vcpu);
3510                         if (r)
3511                                 goto out;
3512                 }
3513
3514                 delta = 1;
3515                 if (io->rep) {
3516                         delta *= io->cur_count;
3517                         /*
3518                          * The size of the register should really depend on
3519                          * current address size.
3520                          */
3521                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3522                         val -= delta;
3523                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3524                 }
3525                 if (io->down)
3526                         delta = -delta;
3527                 delta *= io->size;
3528                 if (io->in) {
3529                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3530                         val += delta;
3531                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3532                 } else {
3533                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3534                         val += delta;
3535                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3536                 }
3537         }
3538 out:
3539         io->count -= io->cur_count;
3540         io->cur_count = 0;
3541
3542         return 0;
3543 }
3544
3545 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3546 {
3547         /* TODO: String I/O for in kernel device */
3548         int r;
3549
3550         if (vcpu->arch.pio.in)
3551                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3552                                     vcpu->arch.pio.size, pd);
3553         else
3554                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3555                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3556                                      pd);
3557         return r;
3558 }
3559
3560 static int pio_string_write(struct kvm_vcpu *vcpu)
3561 {
3562         struct kvm_pio_request *io = &vcpu->arch.pio;
3563         void *pd = vcpu->arch.pio_data;
3564         int i, r = 0;
3565
3566         for (i = 0; i < io->cur_count; i++) {
3567                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3568                                      io->port, io->size, pd)) {
3569                         r = -EOPNOTSUPP;
3570                         break;
3571                 }
3572                 pd += io->size;
3573         }
3574         return r;
3575 }
3576
3577 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3578 {
3579         unsigned long val;
3580
3581         trace_kvm_pio(!in, port, size, 1);
3582
3583         vcpu->run->exit_reason = KVM_EXIT_IO;
3584         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3585         vcpu->run->io.size = vcpu->arch.pio.size = size;
3586         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3587         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3588         vcpu->run->io.port = vcpu->arch.pio.port = port;
3589         vcpu->arch.pio.in = in;
3590         vcpu->arch.pio.string = 0;
3591         vcpu->arch.pio.down = 0;
3592         vcpu->arch.pio.rep = 0;
3593
3594         if (!vcpu->arch.pio.in) {
3595                 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3596                 memcpy(vcpu->arch.pio_data, &val, 4);
3597         }
3598
3599         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3600                 complete_pio(vcpu);
3601                 return 1;
3602         }
3603         return 0;
3604 }
3605 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3606
3607 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3608                   int size, unsigned long count, int down,
3609                   gva_t address, int rep, unsigned port)
3610 {
3611         unsigned now, in_page;
3612         int ret = 0;
3613
3614         trace_kvm_pio(!in, port, size, count);
3615
3616         vcpu->run->exit_reason = KVM_EXIT_IO;
3617         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3618         vcpu->run->io.size = vcpu->arch.pio.size = size;
3619         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3620         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3621         vcpu->run->io.port = vcpu->arch.pio.port = port;
3622         vcpu->arch.pio.in = in;
3623         vcpu->arch.pio.string = 1;
3624         vcpu->arch.pio.down = down;
3625         vcpu->arch.pio.rep = rep;
3626
3627         if (!count) {
3628                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3629                 return 1;
3630         }
3631
3632         if (!down)
3633                 in_page = PAGE_SIZE - offset_in_page(address);
3634         else
3635                 in_page = offset_in_page(address) + size;
3636         now = min(count, (unsigned long)in_page / size);
3637         if (!now)
3638                 now = 1;
3639         if (down) {
3640                 /*
3641                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3642                  */
3643                 pr_unimpl(vcpu, "guest string pio down\n");
3644                 kvm_inject_gp(vcpu, 0);
3645                 return 1;
3646         }
3647         vcpu->run->io.count = now;
3648         vcpu->arch.pio.cur_count = now;
3649
3650         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3651                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3652
3653         vcpu->arch.pio.guest_gva = address;
3654
3655         if (!vcpu->arch.pio.in) {
3656                 /* string PIO write */
3657                 ret = pio_copy_data(vcpu);
3658                 if (ret == X86EMUL_PROPAGATE_FAULT)
3659                         return 1;
3660                 if (ret == 0 && !pio_string_write(vcpu)) {
3661                         complete_pio(vcpu);
3662                         if (vcpu->arch.pio.count == 0)
3663                                 ret = 1;
3664                 }
3665         }
3666         /* no string PIO read support yet */
3667
3668         return ret;
3669 }
3670 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3671
3672 static void bounce_off(void *info)
3673 {
3674         /* nothing */
3675 }
3676
3677 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3678                                      void *data)
3679 {
3680         struct cpufreq_freqs *freq = data;
3681         struct kvm *kvm;
3682         struct kvm_vcpu *vcpu;
3683         int i, send_ipi = 0;
3684
3685         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3686                 return 0;
3687         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3688                 return 0;
3689         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3690
3691         spin_lock(&kvm_lock);
3692         list_for_each_entry(kvm, &vm_list, vm_list) {
3693                 kvm_for_each_vcpu(i, vcpu, kvm) {
3694                         if (vcpu->cpu != freq->cpu)
3695                                 continue;
3696                         if (!kvm_request_guest_time_update(vcpu))
3697                                 continue;
3698                         if (vcpu->cpu != smp_processor_id())
3699                                 send_ipi++;
3700                 }
3701         }
3702         spin_unlock(&kvm_lock);
3703
3704         if (freq->old < freq->new && send_ipi) {
3705                 /*
3706                  * We upscale the frequency.  Must make the guest
3707                  * doesn't see old kvmclock values while running with
3708                  * the new frequency, otherwise we risk the guest sees
3709                  * time go backwards.
3710                  *
3711                  * In case we update the frequency for another cpu
3712                  * (which might be in guest context) send an interrupt
3713                  * to kick the cpu out of guest context.  Next time
3714                  * guest context is entered kvmclock will be updated,
3715                  * so the guest will not see stale values.
3716                  */
3717                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3718         }
3719         return 0;
3720 }
3721
3722 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3723         .notifier_call  = kvmclock_cpufreq_notifier
3724 };
3725
3726 static void kvm_timer_init(void)
3727 {
3728         int cpu;
3729
3730         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3731                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3732                                           CPUFREQ_TRANSITION_NOTIFIER);
3733                 for_each_online_cpu(cpu) {
3734                         unsigned long khz = cpufreq_get(cpu);
3735                         if (!khz)
3736                                 khz = tsc_khz;
3737                         per_cpu(cpu_tsc_khz, cpu) = khz;
3738                 }
3739         } else {
3740                 for_each_possible_cpu(cpu)
3741                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3742         }
3743 }
3744
3745 int kvm_arch_init(void *opaque)
3746 {
3747         int r;
3748         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3749
3750         if (kvm_x86_ops) {
3751                 printk(KERN_ERR "kvm: already loaded the other module\n");
3752                 r = -EEXIST;
3753                 goto out;
3754         }
3755
3756         if (!ops->cpu_has_kvm_support()) {
3757                 printk(KERN_ERR "kvm: no hardware support\n");
3758                 r = -EOPNOTSUPP;
3759                 goto out;
3760         }
3761         if (ops->disabled_by_bios()) {
3762                 printk(KERN_ERR "kvm: disabled by bios\n");
3763                 r = -EOPNOTSUPP;
3764                 goto out;
3765         }
3766
3767         r = kvm_mmu_module_init();
3768         if (r)
3769                 goto out;
3770
3771         kvm_init_msr_list();
3772
3773         kvm_x86_ops = ops;
3774         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3775         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3776         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3777                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3778
3779         kvm_timer_init();
3780
3781         return 0;
3782
3783 out:
3784         return r;
3785 }
3786
3787 void kvm_arch_exit(void)
3788 {
3789         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3790                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3791                                             CPUFREQ_TRANSITION_NOTIFIER);
3792         kvm_x86_ops = NULL;
3793         kvm_mmu_module_exit();
3794 }
3795
3796 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3797 {
3798         ++vcpu->stat.halt_exits;
3799         if (irqchip_in_kernel(vcpu->kvm)) {
3800                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3801                 return 1;
3802         } else {
3803                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3804                 return 0;
3805         }
3806 }
3807 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3808
3809 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3810                            unsigned long a1)
3811 {
3812         if (is_long_mode(vcpu))
3813                 return a0;
3814         else
3815                 return a0 | ((gpa_t)a1 << 32);
3816 }
3817
3818 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3819 {
3820         u64 param, ingpa, outgpa, ret;
3821         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3822         bool fast, longmode;
3823         int cs_db, cs_l;
3824
3825         /*
3826          * hypercall generates UD from non zero cpl and real mode
3827          * per HYPER-V spec
3828          */
3829         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3830                 kvm_queue_exception(vcpu, UD_VECTOR);
3831                 return 0;
3832         }
3833
3834         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3835         longmode = is_long_mode(vcpu) && cs_l == 1;
3836
3837         if (!longmode) {
3838                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3839                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3840                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3841                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3842                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3843                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3844         }
3845 #ifdef CONFIG_X86_64
3846         else {
3847                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3848                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3849                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3850         }
3851 #endif
3852
3853         code = param & 0xffff;
3854         fast = (param >> 16) & 0x1;
3855         rep_cnt = (param >> 32) & 0xfff;
3856         rep_idx = (param >> 48) & 0xfff;
3857
3858         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3859
3860         switch (code) {
3861         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3862                 kvm_vcpu_on_spin(vcpu);
3863                 break;
3864         default:
3865                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3866                 break;
3867         }
3868
3869         ret = res | (((u64)rep_done & 0xfff) << 32);
3870         if (longmode) {
3871                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3872         } else {
3873                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3874                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3875         }
3876
3877         return 1;
3878 }
3879
3880 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3881 {
3882         unsigned long nr, a0, a1, a2, a3, ret;
3883         int r = 1;
3884
3885         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3886                 return kvm_hv_hypercall(vcpu);
3887
3888         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3889         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3890         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3891         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3892         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3893
3894         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3895
3896         if (!is_long_mode(vcpu)) {
3897                 nr &= 0xFFFFFFFF;
3898                 a0 &= 0xFFFFFFFF;
3899                 a1 &= 0xFFFFFFFF;
3900                 a2 &= 0xFFFFFFFF;
3901                 a3 &= 0xFFFFFFFF;
3902         }
3903
3904         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3905                 ret = -KVM_EPERM;
3906                 goto out;
3907         }
3908
3909         switch (nr) {
3910         case KVM_HC_VAPIC_POLL_IRQ:
3911                 ret = 0;
3912                 break;
3913         case KVM_HC_MMU_OP:
3914                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3915                 break;
3916         default:
3917                 ret = -KVM_ENOSYS;
3918                 break;
3919         }
3920 out:
3921         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3922         ++vcpu->stat.hypercalls;
3923         return r;
3924 }
3925 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3926
3927 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3928 {
3929         char instruction[3];
3930         unsigned long rip = kvm_rip_read(vcpu);
3931
3932         /*
3933          * Blow out the MMU to ensure that no other VCPU has an active mapping
3934          * to ensure that the updated hypercall appears atomically across all
3935          * VCPUs.
3936          */
3937         kvm_mmu_zap_all(vcpu->kvm);
3938
3939         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3940
3941         return emulator_write_emulated(rip, instruction, 3, vcpu);
3942 }
3943
3944 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3945 {
3946         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3947 }
3948
3949 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3950 {
3951         struct descriptor_table dt = { limit, base };
3952
3953         kvm_x86_ops->set_gdt(vcpu, &dt);
3954 }
3955
3956 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3957 {
3958         struct descriptor_table dt = { limit, base };
3959
3960         kvm_x86_ops->set_idt(vcpu, &dt);
3961 }
3962
3963 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3964                    unsigned long *rflags)
3965 {
3966         kvm_lmsw(vcpu, msw);
3967         *rflags = kvm_get_rflags(vcpu);
3968 }
3969
3970 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3971 {
3972         unsigned long value;
3973
3974         switch (cr) {
3975         case 0:
3976                 value = kvm_read_cr0(vcpu);
3977                 break;
3978         case 2:
3979                 value = vcpu->arch.cr2;
3980                 break;
3981         case 3:
3982                 value = vcpu->arch.cr3;
3983                 break;
3984         case 4:
3985                 value = kvm_read_cr4(vcpu);
3986                 break;
3987         case 8:
3988                 value = kvm_get_cr8(vcpu);
3989                 break;
3990         default:
3991                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3992                 return 0;
3993         }
3994
3995         return value;
3996 }
3997
3998 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3999                      unsigned long *rflags)
4000 {
4001         switch (cr) {
4002         case 0:
4003                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4004                 *rflags = kvm_get_rflags(vcpu);
4005                 break;
4006         case 2:
4007                 vcpu->arch.cr2 = val;
4008                 break;
4009         case 3:
4010                 kvm_set_cr3(vcpu, val);
4011                 break;
4012         case 4:
4013                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4014                 break;
4015         case 8:
4016                 kvm_set_cr8(vcpu, val & 0xfUL);
4017                 break;
4018         default:
4019                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4020         }
4021 }
4022
4023 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4024 {
4025         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4026         int j, nent = vcpu->arch.cpuid_nent;
4027
4028         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4029         /* when no next entry is found, the current entry[i] is reselected */
4030         for (j = i + 1; ; j = (j + 1) % nent) {
4031                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4032                 if (ej->function == e->function) {
4033                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4034                         return j;
4035                 }
4036         }
4037         return 0; /* silence gcc, even though control never reaches here */
4038 }
4039
4040 /* find an entry with matching function, matching index (if needed), and that
4041  * should be read next (if it's stateful) */
4042 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4043         u32 function, u32 index)
4044 {
4045         if (e->function != function)
4046                 return 0;
4047         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4048                 return 0;
4049         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4050             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4051                 return 0;
4052         return 1;
4053 }
4054
4055 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4056                                               u32 function, u32 index)
4057 {
4058         int i;
4059         struct kvm_cpuid_entry2 *best = NULL;
4060
4061         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4062                 struct kvm_cpuid_entry2 *e;
4063
4064                 e = &vcpu->arch.cpuid_entries[i];
4065                 if (is_matching_cpuid_entry(e, function, index)) {
4066                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4067                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4068                         best = e;
4069                         break;
4070                 }
4071                 /*
4072                  * Both basic or both extended?
4073                  */
4074                 if (((e->function ^ function) & 0x80000000) == 0)
4075                         if (!best || e->function > best->function)
4076                                 best = e;
4077         }
4078         return best;
4079 }
4080 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4081
4082 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4083 {
4084         struct kvm_cpuid_entry2 *best;
4085
4086         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4087         if (best)
4088                 return best->eax & 0xff;
4089         return 36;
4090 }
4091
4092 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4093 {
4094         u32 function, index;
4095         struct kvm_cpuid_entry2 *best;
4096
4097         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4098         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4099         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4100         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4101         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4102         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4103         best = kvm_find_cpuid_entry(vcpu, function, index);
4104         if (best) {
4105                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4106                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4107                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4108                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4109         }
4110         kvm_x86_ops->skip_emulated_instruction(vcpu);
4111         trace_kvm_cpuid(function,
4112                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4113                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4114                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4115                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4116 }
4117 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4118
4119 /*
4120  * Check if userspace requested an interrupt window, and that the
4121  * interrupt window is open.
4122  *
4123  * No need to exit to userspace if we already have an interrupt queued.
4124  */
4125 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4126 {
4127         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4128                 vcpu->run->request_interrupt_window &&
4129                 kvm_arch_interrupt_allowed(vcpu));
4130 }
4131
4132 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4133 {
4134         struct kvm_run *kvm_run = vcpu->run;
4135
4136         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4137         kvm_run->cr8 = kvm_get_cr8(vcpu);
4138         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4139         if (irqchip_in_kernel(vcpu->kvm))
4140                 kvm_run->ready_for_interrupt_injection = 1;
4141         else
4142                 kvm_run->ready_for_interrupt_injection =
4143                         kvm_arch_interrupt_allowed(vcpu) &&
4144                         !kvm_cpu_has_interrupt(vcpu) &&
4145                         !kvm_event_needs_reinjection(vcpu);
4146 }
4147
4148 static void vapic_enter(struct kvm_vcpu *vcpu)
4149 {
4150         struct kvm_lapic *apic = vcpu->arch.apic;
4151         struct page *page;
4152
4153         if (!apic || !apic->vapic_addr)
4154                 return;
4155
4156         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4157
4158         vcpu->arch.apic->vapic_page = page;
4159 }
4160
4161 static void vapic_exit(struct kvm_vcpu *vcpu)
4162 {
4163         struct kvm_lapic *apic = vcpu->arch.apic;
4164         int idx;
4165
4166         if (!apic || !apic->vapic_addr)
4167                 return;
4168
4169         idx = srcu_read_lock(&vcpu->kvm->srcu);
4170         kvm_release_page_dirty(apic->vapic_page);
4171         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4172         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4173 }
4174
4175 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4176 {
4177         int max_irr, tpr;
4178
4179         if (!kvm_x86_ops->update_cr8_intercept)
4180                 return;
4181
4182         if (!vcpu->arch.apic)
4183                 return;
4184
4185         if (!vcpu->arch.apic->vapic_addr)
4186                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4187         else
4188                 max_irr = -1;
4189
4190         if (max_irr != -1)
4191                 max_irr >>= 4;
4192
4193         tpr = kvm_lapic_get_cr8(vcpu);
4194
4195         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4196 }
4197
4198 static void inject_pending_event(struct kvm_vcpu *vcpu)
4199 {
4200         /* try to reinject previous events if any */
4201         if (vcpu->arch.exception.pending) {
4202                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4203                                           vcpu->arch.exception.has_error_code,
4204                                           vcpu->arch.exception.error_code);
4205                 return;
4206         }
4207
4208         if (vcpu->arch.nmi_injected) {
4209                 kvm_x86_ops->set_nmi(vcpu);
4210                 return;
4211         }
4212
4213         if (vcpu->arch.interrupt.pending) {
4214                 kvm_x86_ops->set_irq(vcpu);
4215                 return;
4216         }
4217
4218         /* try to inject new event if pending */
4219         if (vcpu->arch.nmi_pending) {
4220                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4221                         vcpu->arch.nmi_pending = false;
4222                         vcpu->arch.nmi_injected = true;
4223                         kvm_x86_ops->set_nmi(vcpu);
4224                 }
4225         } else if (kvm_cpu_has_interrupt(vcpu)) {
4226                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4227                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4228                                             false);
4229                         kvm_x86_ops->set_irq(vcpu);
4230                 }
4231         }
4232 }
4233
4234 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4235 {
4236         int r;
4237         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4238                 vcpu->run->request_interrupt_window;
4239
4240         if (vcpu->requests)
4241                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4242                         kvm_mmu_unload(vcpu);
4243
4244         r = kvm_mmu_reload(vcpu);
4245         if (unlikely(r))
4246                 goto out;
4247
4248         if (vcpu->requests) {
4249                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4250                         __kvm_migrate_timers(vcpu);
4251                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4252                         kvm_write_guest_time(vcpu);
4253                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4254                         kvm_mmu_sync_roots(vcpu);
4255                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4256                         kvm_x86_ops->tlb_flush(vcpu);
4257                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4258                                        &vcpu->requests)) {
4259                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4260                         r = 0;
4261                         goto out;
4262                 }
4263                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4264                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4265                         r = 0;
4266                         goto out;
4267                 }
4268                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4269                         vcpu->fpu_active = 0;
4270                         kvm_x86_ops->fpu_deactivate(vcpu);
4271                 }
4272         }
4273
4274         preempt_disable();
4275
4276         kvm_x86_ops->prepare_guest_switch(vcpu);
4277         if (vcpu->fpu_active)
4278                 kvm_load_guest_fpu(vcpu);
4279
4280         local_irq_disable();
4281
4282         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4283         smp_mb__after_clear_bit();
4284
4285         if (vcpu->requests || need_resched() || signal_pending(current)) {
4286                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4287                 local_irq_enable();
4288                 preempt_enable();
4289                 r = 1;
4290                 goto out;
4291         }
4292
4293         inject_pending_event(vcpu);
4294
4295         /* enable NMI/IRQ window open exits if needed */
4296         if (vcpu->arch.nmi_pending)
4297                 kvm_x86_ops->enable_nmi_window(vcpu);
4298         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4299                 kvm_x86_ops->enable_irq_window(vcpu);
4300
4301         if (kvm_lapic_enabled(vcpu)) {
4302                 update_cr8_intercept(vcpu);
4303                 kvm_lapic_sync_to_vapic(vcpu);
4304         }
4305
4306         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4307
4308         kvm_guest_enter();
4309
4310         if (unlikely(vcpu->arch.switch_db_regs)) {
4311                 set_debugreg(0, 7);
4312                 set_debugreg(vcpu->arch.eff_db[0], 0);
4313                 set_debugreg(vcpu->arch.eff_db[1], 1);
4314                 set_debugreg(vcpu->arch.eff_db[2], 2);
4315                 set_debugreg(vcpu->arch.eff_db[3], 3);
4316         }
4317
4318         trace_kvm_entry(vcpu->vcpu_id);
4319         kvm_x86_ops->run(vcpu);
4320
4321         /*
4322          * If the guest has used debug registers, at least dr7
4323          * will be disabled while returning to the host.
4324          * If we don't have active breakpoints in the host, we don't
4325          * care about the messed up debug address registers. But if
4326          * we have some of them active, restore the old state.
4327          */
4328         if (hw_breakpoint_active())
4329                 hw_breakpoint_restore();
4330
4331         set_bit(KVM_REQ_KICK, &vcpu->requests);
4332         local_irq_enable();
4333
4334         ++vcpu->stat.exits;
4335
4336         /*
4337          * We must have an instruction between local_irq_enable() and
4338          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4339          * the interrupt shadow.  The stat.exits increment will do nicely.
4340          * But we need to prevent reordering, hence this barrier():
4341          */
4342         barrier();
4343
4344         kvm_guest_exit();
4345
4346         preempt_enable();
4347
4348         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4349
4350         /*
4351          * Profile KVM exit RIPs:
4352          */
4353         if (unlikely(prof_on == KVM_PROFILING)) {
4354                 unsigned long rip = kvm_rip_read(vcpu);
4355                 profile_hit(KVM_PROFILING, (void *)rip);
4356         }
4357
4358
4359         kvm_lapic_sync_from_vapic(vcpu);
4360
4361         r = kvm_x86_ops->handle_exit(vcpu);
4362 out:
4363         return r;
4364 }
4365
4366
4367 static int __vcpu_run(struct kvm_vcpu *vcpu)
4368 {
4369         int r;
4370         struct kvm *kvm = vcpu->kvm;
4371
4372         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4373                 pr_debug("vcpu %d received sipi with vector # %x\n",
4374                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4375                 kvm_lapic_reset(vcpu);
4376                 r = kvm_arch_vcpu_reset(vcpu);
4377                 if (r)
4378                         return r;
4379                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4380         }
4381
4382         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4383         vapic_enter(vcpu);
4384
4385         r = 1;
4386         while (r > 0) {
4387                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4388                         r = vcpu_enter_guest(vcpu);
4389                 else {
4390                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4391                         kvm_vcpu_block(vcpu);
4392                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4393                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4394                         {
4395                                 switch(vcpu->arch.mp_state) {
4396                                 case KVM_MP_STATE_HALTED:
4397                                         vcpu->arch.mp_state =
4398                                                 KVM_MP_STATE_RUNNABLE;
4399                                 case KVM_MP_STATE_RUNNABLE:
4400                                         break;
4401                                 case KVM_MP_STATE_SIPI_RECEIVED:
4402                                 default:
4403                                         r = -EINTR;
4404                                         break;
4405                                 }
4406                         }
4407                 }
4408
4409                 if (r <= 0)
4410                         break;
4411
4412                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4413                 if (kvm_cpu_has_pending_timer(vcpu))
4414                         kvm_inject_pending_timer_irqs(vcpu);
4415
4416                 if (dm_request_for_irq_injection(vcpu)) {
4417                         r = -EINTR;
4418                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4419                         ++vcpu->stat.request_irq_exits;
4420                 }
4421                 if (signal_pending(current)) {
4422                         r = -EINTR;
4423                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4424                         ++vcpu->stat.signal_exits;
4425                 }
4426                 if (need_resched()) {
4427                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4428                         kvm_resched(vcpu);
4429                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4430                 }
4431         }
4432
4433         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4434         post_kvm_run_save(vcpu);
4435
4436         vapic_exit(vcpu);
4437
4438         return r;
4439 }
4440
4441 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4442 {
4443         int r;
4444         sigset_t sigsaved;
4445
4446         vcpu_load(vcpu);
4447
4448         if (vcpu->sigset_active)
4449                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4450
4451         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4452                 kvm_vcpu_block(vcpu);
4453                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4454                 r = -EAGAIN;
4455                 goto out;
4456         }
4457
4458         /* re-sync apic's tpr */
4459         if (!irqchip_in_kernel(vcpu->kvm))
4460                 kvm_set_cr8(vcpu, kvm_run->cr8);
4461
4462         if (vcpu->arch.pio.cur_count) {
4463                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4464                 r = complete_pio(vcpu);
4465                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4466                 if (r)
4467                         goto out;
4468         }
4469         if (vcpu->mmio_needed) {
4470                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4471                 vcpu->mmio_read_completed = 1;
4472                 vcpu->mmio_needed = 0;
4473
4474                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4475                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4476                                         EMULTYPE_NO_DECODE);
4477                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4478                 if (r == EMULATE_DO_MMIO) {
4479                         /*
4480                          * Read-modify-write.  Back to userspace.
4481                          */
4482                         r = 0;
4483                         goto out;
4484                 }
4485         }
4486         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4487                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4488                                      kvm_run->hypercall.ret);
4489
4490         r = __vcpu_run(vcpu);
4491
4492 out:
4493         if (vcpu->sigset_active)
4494                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4495
4496         vcpu_put(vcpu);
4497         return r;
4498 }
4499
4500 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4501 {
4502         vcpu_load(vcpu);
4503
4504         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4505         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4506         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4507         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4508         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4509         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4510         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4511         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4512 #ifdef CONFIG_X86_64
4513         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4514         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4515         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4516         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4517         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4518         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4519         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4520         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4521 #endif
4522
4523         regs->rip = kvm_rip_read(vcpu);
4524         regs->rflags = kvm_get_rflags(vcpu);
4525
4526         vcpu_put(vcpu);
4527
4528         return 0;
4529 }
4530
4531 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4532 {
4533         vcpu_load(vcpu);
4534
4535         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4536         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4537         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4538         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4539         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4540         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4541         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4542         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4543 #ifdef CONFIG_X86_64
4544         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4545         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4546         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4547         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4548         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4549         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4550         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4551         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4552 #endif
4553
4554         kvm_rip_write(vcpu, regs->rip);
4555         kvm_set_rflags(vcpu, regs->rflags);
4556
4557         vcpu->arch.exception.pending = false;
4558
4559         vcpu_put(vcpu);
4560
4561         return 0;
4562 }
4563
4564 void kvm_get_segment(struct kvm_vcpu *vcpu,
4565                      struct kvm_segment *var, int seg)
4566 {
4567         kvm_x86_ops->get_segment(vcpu, var, seg);
4568 }
4569
4570 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4571 {
4572         struct kvm_segment cs;
4573
4574         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4575         *db = cs.db;
4576         *l = cs.l;
4577 }
4578 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4579
4580 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4581                                   struct kvm_sregs *sregs)
4582 {
4583         struct descriptor_table dt;
4584
4585         vcpu_load(vcpu);
4586
4587         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4588         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4589         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4590         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4591         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4592         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4593
4594         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4595         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4596
4597         kvm_x86_ops->get_idt(vcpu, &dt);
4598         sregs->idt.limit = dt.limit;
4599         sregs->idt.base = dt.base;
4600         kvm_x86_ops->get_gdt(vcpu, &dt);
4601         sregs->gdt.limit = dt.limit;
4602         sregs->gdt.base = dt.base;
4603
4604         sregs->cr0 = kvm_read_cr0(vcpu);
4605         sregs->cr2 = vcpu->arch.cr2;
4606         sregs->cr3 = vcpu->arch.cr3;
4607         sregs->cr4 = kvm_read_cr4(vcpu);
4608         sregs->cr8 = kvm_get_cr8(vcpu);
4609         sregs->efer = vcpu->arch.efer;
4610         sregs->apic_base = kvm_get_apic_base(vcpu);
4611
4612         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4613
4614         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4615                 set_bit(vcpu->arch.interrupt.nr,
4616                         (unsigned long *)sregs->interrupt_bitmap);
4617
4618         vcpu_put(vcpu);
4619
4620         return 0;
4621 }
4622
4623 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4624                                     struct kvm_mp_state *mp_state)
4625 {
4626         vcpu_load(vcpu);
4627         mp_state->mp_state = vcpu->arch.mp_state;
4628         vcpu_put(vcpu);
4629         return 0;
4630 }
4631
4632 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4633                                     struct kvm_mp_state *mp_state)
4634 {
4635         vcpu_load(vcpu);
4636         vcpu->arch.mp_state = mp_state->mp_state;
4637         vcpu_put(vcpu);
4638         return 0;
4639 }
4640
4641 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4642                         struct kvm_segment *var, int seg)
4643 {
4644         kvm_x86_ops->set_segment(vcpu, var, seg);
4645 }
4646
4647 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4648                                    struct kvm_segment *kvm_desct)
4649 {
4650         kvm_desct->base = get_desc_base(seg_desc);
4651         kvm_desct->limit = get_desc_limit(seg_desc);
4652         if (seg_desc->g) {
4653                 kvm_desct->limit <<= 12;
4654                 kvm_desct->limit |= 0xfff;
4655         }
4656         kvm_desct->selector = selector;
4657         kvm_desct->type = seg_desc->type;
4658         kvm_desct->present = seg_desc->p;
4659         kvm_desct->dpl = seg_desc->dpl;
4660         kvm_desct->db = seg_desc->d;
4661         kvm_desct->s = seg_desc->s;
4662         kvm_desct->l = seg_desc->l;
4663         kvm_desct->g = seg_desc->g;
4664         kvm_desct->avl = seg_desc->avl;
4665         if (!selector)
4666                 kvm_desct->unusable = 1;
4667         else
4668                 kvm_desct->unusable = 0;
4669         kvm_desct->padding = 0;
4670 }
4671
4672 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4673                                           u16 selector,
4674                                           struct descriptor_table *dtable)
4675 {
4676         if (selector & 1 << 2) {
4677                 struct kvm_segment kvm_seg;
4678
4679                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4680
4681                 if (kvm_seg.unusable)
4682                         dtable->limit = 0;
4683                 else
4684                         dtable->limit = kvm_seg.limit;
4685                 dtable->base = kvm_seg.base;
4686         }
4687         else
4688                 kvm_x86_ops->get_gdt(vcpu, dtable);
4689 }
4690
4691 /* allowed just for 8 bytes segments */
4692 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4693                                          struct desc_struct *seg_desc)
4694 {
4695         struct descriptor_table dtable;
4696         u16 index = selector >> 3;
4697         int ret;
4698         u32 err;
4699         gva_t addr;
4700
4701         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4702
4703         if (dtable.limit < index * 8 + 7) {
4704                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4705                 return X86EMUL_PROPAGATE_FAULT;
4706         }
4707         addr = dtable.base + index * 8;
4708         ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4709                                          vcpu,  &err);
4710         if (ret == X86EMUL_PROPAGATE_FAULT)
4711                 kvm_inject_page_fault(vcpu, addr, err);
4712
4713        return ret;
4714 }
4715
4716 /* allowed just for 8 bytes segments */
4717 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4718                                          struct desc_struct *seg_desc)
4719 {
4720         struct descriptor_table dtable;
4721         u16 index = selector >> 3;
4722
4723         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4724
4725         if (dtable.limit < index * 8 + 7)
4726                 return 1;
4727         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4728 }
4729
4730 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4731                                struct desc_struct *seg_desc)
4732 {
4733         u32 base_addr = get_desc_base(seg_desc);
4734
4735         return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4736 }
4737
4738 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4739                              struct desc_struct *seg_desc)
4740 {
4741         u32 base_addr = get_desc_base(seg_desc);
4742
4743         return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4744 }
4745
4746 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4747 {
4748         struct kvm_segment kvm_seg;
4749
4750         kvm_get_segment(vcpu, &kvm_seg, seg);
4751         return kvm_seg.selector;
4752 }
4753
4754 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4755 {
4756         struct kvm_segment segvar = {
4757                 .base = selector << 4,
4758                 .limit = 0xffff,
4759                 .selector = selector,
4760                 .type = 3,
4761                 .present = 1,
4762                 .dpl = 3,
4763                 .db = 0,
4764                 .s = 1,
4765                 .l = 0,
4766                 .g = 0,
4767                 .avl = 0,
4768                 .unusable = 0,
4769         };
4770         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4771         return X86EMUL_CONTINUE;
4772 }
4773
4774 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4775 {
4776         return (seg != VCPU_SREG_LDTR) &&
4777                 (seg != VCPU_SREG_TR) &&
4778                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4779 }
4780
4781 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4782 {
4783         struct kvm_segment kvm_seg;
4784         struct desc_struct seg_desc;
4785         u8 dpl, rpl, cpl;
4786         unsigned err_vec = GP_VECTOR;
4787         u32 err_code = 0;
4788         bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4789         int ret;
4790
4791         if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4792                 return kvm_load_realmode_segment(vcpu, selector, seg);
4793
4794         /* NULL selector is not valid for TR, CS and SS */
4795         if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4796             && null_selector)
4797                 goto exception;
4798
4799         /* TR should be in GDT only */
4800         if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4801                 goto exception;
4802
4803         ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4804         if (ret)
4805                 return ret;
4806
4807         seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4808
4809         if (null_selector) { /* for NULL selector skip all following checks */
4810                 kvm_seg.unusable = 1;
4811                 goto load;
4812         }
4813
4814         err_code = selector & 0xfffc;
4815         err_vec = GP_VECTOR;
4816
4817         /* can't load system descriptor into segment selecor */
4818         if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4819                 goto exception;
4820
4821         if (!kvm_seg.present) {
4822                 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4823                 goto exception;
4824         }
4825
4826         rpl = selector & 3;
4827         dpl = kvm_seg.dpl;
4828         cpl = kvm_x86_ops->get_cpl(vcpu);
4829
4830         switch (seg) {
4831         case VCPU_SREG_SS:
4832                 /*
4833                  * segment is not a writable data segment or segment
4834                  * selector's RPL != CPL or segment selector's RPL != CPL
4835                  */
4836                 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4837                         goto exception;
4838                 break;
4839         case VCPU_SREG_CS:
4840                 if (!(kvm_seg.type & 8))
4841                         goto exception;
4842
4843                 if (kvm_seg.type & 4) {
4844                         /* conforming */
4845                         if (dpl > cpl)
4846                                 goto exception;
4847                 } else {
4848                         /* nonconforming */
4849                         if (rpl > cpl || dpl != cpl)
4850                                 goto exception;
4851                 }
4852                 /* CS(RPL) <- CPL */
4853                 selector = (selector & 0xfffc) | cpl;
4854             break;
4855         case VCPU_SREG_TR:
4856                 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4857                         goto exception;
4858                 break;
4859         case VCPU_SREG_LDTR:
4860                 if (kvm_seg.s || kvm_seg.type != 2)
4861                         goto exception;
4862                 break;
4863         default: /*  DS, ES, FS, or GS */
4864                 /*
4865                  * segment is not a data or readable code segment or
4866                  * ((segment is a data or nonconforming code segment)
4867                  * and (both RPL and CPL > DPL))
4868                  */
4869                 if ((kvm_seg.type & 0xa) == 0x8 ||
4870                     (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4871                         goto exception;
4872                 break;
4873         }
4874
4875         if (!kvm_seg.unusable && kvm_seg.s) {
4876                 /* mark segment as accessed */
4877                 kvm_seg.type |= 1;
4878                 seg_desc.type |= 1;
4879                 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4880         }
4881 load:
4882         kvm_set_segment(vcpu, &kvm_seg, seg);
4883         return X86EMUL_CONTINUE;
4884 exception:
4885         kvm_queue_exception_e(vcpu, err_vec, err_code);
4886         return X86EMUL_PROPAGATE_FAULT;
4887 }
4888
4889 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4890                                 struct tss_segment_32 *tss)
4891 {
4892         tss->cr3 = vcpu->arch.cr3;
4893         tss->eip = kvm_rip_read(vcpu);
4894         tss->eflags = kvm_get_rflags(vcpu);
4895         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4896         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4897         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4898         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4899         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4900         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4901         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4902         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4903         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4904         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4905         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4906         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4907         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4908         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4909         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4910 }
4911
4912 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4913 {
4914         struct kvm_segment kvm_seg;
4915         kvm_get_segment(vcpu, &kvm_seg, seg);
4916         kvm_seg.selector = sel;
4917         kvm_set_segment(vcpu, &kvm_seg, seg);
4918 }
4919
4920 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4921                                   struct tss_segment_32 *tss)
4922 {
4923         kvm_set_cr3(vcpu, tss->cr3);
4924
4925         kvm_rip_write(vcpu, tss->eip);
4926         kvm_set_rflags(vcpu, tss->eflags | 2);
4927
4928         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4929         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4930         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4931         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4932         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4933         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4934         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4935         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4936
4937         /*
4938          * SDM says that segment selectors are loaded before segment
4939          * descriptors
4940          */
4941         kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
4942         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
4943         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
4944         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
4945         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
4946         kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
4947         kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
4948
4949         /*
4950          * Now load segment descriptors. If fault happenes at this stage
4951          * it is handled in a context of new task
4952          */
4953         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
4954                 return 1;
4955
4956         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
4957                 return 1;
4958
4959         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
4960                 return 1;
4961
4962         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
4963                 return 1;
4964
4965         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
4966                 return 1;
4967
4968         if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
4969                 return 1;
4970
4971         if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
4972                 return 1;
4973         return 0;
4974 }
4975
4976 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4977                                 struct tss_segment_16 *tss)
4978 {
4979         tss->ip = kvm_rip_read(vcpu);
4980         tss->flag = kvm_get_rflags(vcpu);
4981         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4982         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4983         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4984         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4985         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4986         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4987         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4988         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4989
4990         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4991         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4992         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4993         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4994         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4995 }
4996
4997 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4998                                  struct tss_segment_16 *tss)
4999 {
5000         kvm_rip_write(vcpu, tss->ip);
5001         kvm_set_rflags(vcpu, tss->flag | 2);
5002         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5003         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5004         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5005         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5006         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5007         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5008         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5009         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5010
5011         /*
5012          * SDM says that segment selectors are loaded before segment
5013          * descriptors
5014          */
5015         kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5016         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5017         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5018         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5019         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5020
5021         /*
5022          * Now load segment descriptors. If fault happenes at this stage
5023          * it is handled in a context of new task
5024          */
5025         if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5026                 return 1;
5027
5028         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5029                 return 1;
5030
5031         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5032                 return 1;
5033
5034         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5035                 return 1;
5036
5037         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5038                 return 1;
5039         return 0;
5040 }
5041
5042 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5043                               u16 old_tss_sel, u32 old_tss_base,
5044                               struct desc_struct *nseg_desc)
5045 {
5046         struct tss_segment_16 tss_segment_16;
5047         int ret = 0;
5048
5049         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5050                            sizeof tss_segment_16))
5051                 goto out;
5052
5053         save_state_to_tss16(vcpu, &tss_segment_16);
5054
5055         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5056                             sizeof tss_segment_16))
5057                 goto out;
5058
5059         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5060                            &tss_segment_16, sizeof tss_segment_16))
5061                 goto out;
5062
5063         if (old_tss_sel != 0xffff) {
5064                 tss_segment_16.prev_task_link = old_tss_sel;
5065
5066                 if (kvm_write_guest(vcpu->kvm,
5067                                     get_tss_base_addr_write(vcpu, nseg_desc),
5068                                     &tss_segment_16.prev_task_link,
5069                                     sizeof tss_segment_16.prev_task_link))
5070                         goto out;
5071         }
5072
5073         if (load_state_from_tss16(vcpu, &tss_segment_16))
5074                 goto out;
5075
5076         ret = 1;
5077 out:
5078         return ret;
5079 }
5080
5081 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5082                        u16 old_tss_sel, u32 old_tss_base,
5083                        struct desc_struct *nseg_desc)
5084 {
5085         struct tss_segment_32 tss_segment_32;
5086         int ret = 0;
5087
5088         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5089                            sizeof tss_segment_32))
5090                 goto out;
5091
5092         save_state_to_tss32(vcpu, &tss_segment_32);
5093
5094         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5095                             sizeof tss_segment_32))
5096                 goto out;
5097
5098         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5099                            &tss_segment_32, sizeof tss_segment_32))
5100                 goto out;
5101
5102         if (old_tss_sel != 0xffff) {
5103                 tss_segment_32.prev_task_link = old_tss_sel;
5104
5105                 if (kvm_write_guest(vcpu->kvm,
5106                                     get_tss_base_addr_write(vcpu, nseg_desc),
5107                                     &tss_segment_32.prev_task_link,
5108                                     sizeof tss_segment_32.prev_task_link))
5109                         goto out;
5110         }
5111
5112         if (load_state_from_tss32(vcpu, &tss_segment_32))
5113                 goto out;
5114
5115         ret = 1;
5116 out:
5117         return ret;
5118 }
5119
5120 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5121 {
5122         struct kvm_segment tr_seg;
5123         struct desc_struct cseg_desc;
5124         struct desc_struct nseg_desc;
5125         int ret = 0;
5126         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5127         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5128
5129         old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5130
5131         /* FIXME: Handle errors. Failure to read either TSS or their
5132          * descriptors should generate a pagefault.
5133          */
5134         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5135                 goto out;
5136
5137         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5138                 goto out;
5139
5140         if (reason != TASK_SWITCH_IRET) {
5141                 int cpl;
5142
5143                 cpl = kvm_x86_ops->get_cpl(vcpu);
5144                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5145                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5146                         return 1;
5147                 }
5148         }
5149
5150         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
5151                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5152                 return 1;
5153         }
5154
5155         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5156                 cseg_desc.type &= ~(1 << 1); //clear the B flag
5157                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5158         }
5159
5160         if (reason == TASK_SWITCH_IRET) {
5161                 u32 eflags = kvm_get_rflags(vcpu);
5162                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5163         }
5164
5165         /* set back link to prev task only if NT bit is set in eflags
5166            note that old_tss_sel is not used afetr this point */
5167         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5168                 old_tss_sel = 0xffff;
5169
5170         if (nseg_desc.type & 8)
5171                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5172                                          old_tss_base, &nseg_desc);
5173         else
5174                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5175                                          old_tss_base, &nseg_desc);
5176
5177         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5178                 u32 eflags = kvm_get_rflags(vcpu);
5179                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5180         }
5181
5182         if (reason != TASK_SWITCH_IRET) {
5183                 nseg_desc.type |= (1 << 1);
5184                 save_guest_segment_descriptor(vcpu, tss_selector,
5185                                               &nseg_desc);
5186         }
5187
5188         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5189         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5190         tr_seg.type = 11;
5191         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5192 out:
5193         return ret;
5194 }
5195 EXPORT_SYMBOL_GPL(kvm_task_switch);
5196
5197 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5198                                   struct kvm_sregs *sregs)
5199 {
5200         int mmu_reset_needed = 0;
5201         int pending_vec, max_bits;
5202         struct descriptor_table dt;
5203
5204         vcpu_load(vcpu);
5205
5206         dt.limit = sregs->idt.limit;
5207         dt.base = sregs->idt.base;
5208         kvm_x86_ops->set_idt(vcpu, &dt);
5209         dt.limit = sregs->gdt.limit;
5210         dt.base = sregs->gdt.base;
5211         kvm_x86_ops->set_gdt(vcpu, &dt);
5212
5213         vcpu->arch.cr2 = sregs->cr2;
5214         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5215         vcpu->arch.cr3 = sregs->cr3;
5216
5217         kvm_set_cr8(vcpu, sregs->cr8);
5218
5219         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5220         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5221         kvm_set_apic_base(vcpu, sregs->apic_base);
5222
5223         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5224         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5225         vcpu->arch.cr0 = sregs->cr0;
5226
5227         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5228         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5229         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5230                 load_pdptrs(vcpu, vcpu->arch.cr3);
5231                 mmu_reset_needed = 1;
5232         }
5233
5234         if (mmu_reset_needed)
5235                 kvm_mmu_reset_context(vcpu);
5236
5237         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5238         pending_vec = find_first_bit(
5239                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5240         if (pending_vec < max_bits) {
5241                 kvm_queue_interrupt(vcpu, pending_vec, false);
5242                 pr_debug("Set back pending irq %d\n", pending_vec);
5243                 if (irqchip_in_kernel(vcpu->kvm))
5244                         kvm_pic_clear_isr_ack(vcpu->kvm);
5245         }
5246
5247         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5248         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5249         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5250         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5251         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5252         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5253
5254         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5255         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5256
5257         update_cr8_intercept(vcpu);
5258
5259         /* Older userspace won't unhalt the vcpu on reset. */
5260         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5261             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5262             !is_protmode(vcpu))
5263                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5264
5265         vcpu_put(vcpu);
5266
5267         return 0;
5268 }
5269
5270 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5271                                         struct kvm_guest_debug *dbg)
5272 {
5273         unsigned long rflags;
5274         int i, r;
5275
5276         vcpu_load(vcpu);
5277
5278         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5279                 r = -EBUSY;
5280                 if (vcpu->arch.exception.pending)
5281                         goto unlock_out;
5282                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5283                         kvm_queue_exception(vcpu, DB_VECTOR);
5284                 else
5285                         kvm_queue_exception(vcpu, BP_VECTOR);
5286         }
5287
5288         /*
5289          * Read rflags as long as potentially injected trace flags are still
5290          * filtered out.
5291          */
5292         rflags = kvm_get_rflags(vcpu);
5293
5294         vcpu->guest_debug = dbg->control;
5295         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5296                 vcpu->guest_debug = 0;
5297
5298         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5299                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5300                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5301                 vcpu->arch.switch_db_regs =
5302                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5303         } else {
5304                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5305                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5306                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5307         }
5308
5309         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5310                 vcpu->arch.singlestep_cs =
5311                         get_segment_selector(vcpu, VCPU_SREG_CS);
5312                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5313         }
5314
5315         /*
5316          * Trigger an rflags update that will inject or remove the trace
5317          * flags.
5318          */
5319         kvm_set_rflags(vcpu, rflags);
5320
5321         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5322
5323         r = 0;
5324
5325 unlock_out:
5326         vcpu_put(vcpu);
5327
5328         return r;
5329 }
5330
5331 /*
5332  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5333  * we have asm/x86/processor.h
5334  */
5335 struct fxsave {
5336         u16     cwd;
5337         u16     swd;
5338         u16     twd;
5339         u16     fop;
5340         u64     rip;
5341         u64     rdp;
5342         u32     mxcsr;
5343         u32     mxcsr_mask;
5344         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5345 #ifdef CONFIG_X86_64
5346         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5347 #else
5348         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5349 #endif
5350 };
5351
5352 /*
5353  * Translate a guest virtual address to a guest physical address.
5354  */
5355 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5356                                     struct kvm_translation *tr)
5357 {
5358         unsigned long vaddr = tr->linear_address;
5359         gpa_t gpa;
5360         int idx;
5361
5362         vcpu_load(vcpu);
5363         idx = srcu_read_lock(&vcpu->kvm->srcu);
5364         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5365         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5366         tr->physical_address = gpa;
5367         tr->valid = gpa != UNMAPPED_GVA;
5368         tr->writeable = 1;
5369         tr->usermode = 0;
5370         vcpu_put(vcpu);
5371
5372         return 0;
5373 }
5374
5375 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5376 {
5377         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5378
5379         vcpu_load(vcpu);
5380
5381         memcpy(fpu->fpr, fxsave->st_space, 128);
5382         fpu->fcw = fxsave->cwd;
5383         fpu->fsw = fxsave->swd;
5384         fpu->ftwx = fxsave->twd;
5385         fpu->last_opcode = fxsave->fop;
5386         fpu->last_ip = fxsave->rip;
5387         fpu->last_dp = fxsave->rdp;
5388         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5389
5390         vcpu_put(vcpu);
5391
5392         return 0;
5393 }
5394
5395 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5396 {
5397         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5398
5399         vcpu_load(vcpu);
5400
5401         memcpy(fxsave->st_space, fpu->fpr, 128);
5402         fxsave->cwd = fpu->fcw;
5403         fxsave->swd = fpu->fsw;
5404         fxsave->twd = fpu->ftwx;
5405         fxsave->fop = fpu->last_opcode;
5406         fxsave->rip = fpu->last_ip;
5407         fxsave->rdp = fpu->last_dp;
5408         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5409
5410         vcpu_put(vcpu);
5411
5412         return 0;
5413 }
5414
5415 void fx_init(struct kvm_vcpu *vcpu)
5416 {
5417         unsigned after_mxcsr_mask;
5418
5419         /*
5420          * Touch the fpu the first time in non atomic context as if
5421          * this is the first fpu instruction the exception handler
5422          * will fire before the instruction returns and it'll have to
5423          * allocate ram with GFP_KERNEL.
5424          */
5425         if (!used_math())
5426                 kvm_fx_save(&vcpu->arch.host_fx_image);
5427
5428         /* Initialize guest FPU by resetting ours and saving into guest's */
5429         preempt_disable();
5430         kvm_fx_save(&vcpu->arch.host_fx_image);
5431         kvm_fx_finit();
5432         kvm_fx_save(&vcpu->arch.guest_fx_image);
5433         kvm_fx_restore(&vcpu->arch.host_fx_image);
5434         preempt_enable();
5435
5436         vcpu->arch.cr0 |= X86_CR0_ET;
5437         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5438         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5439         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5440                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5441 }
5442 EXPORT_SYMBOL_GPL(fx_init);
5443
5444 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5445 {
5446         if (vcpu->guest_fpu_loaded)
5447                 return;
5448
5449         vcpu->guest_fpu_loaded = 1;
5450         kvm_fx_save(&vcpu->arch.host_fx_image);
5451         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5452         trace_kvm_fpu(1);
5453 }
5454
5455 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5456 {
5457         if (!vcpu->guest_fpu_loaded)
5458                 return;
5459
5460         vcpu->guest_fpu_loaded = 0;
5461         kvm_fx_save(&vcpu->arch.guest_fx_image);
5462         kvm_fx_restore(&vcpu->arch.host_fx_image);
5463         ++vcpu->stat.fpu_reload;
5464         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5465         trace_kvm_fpu(0);
5466 }
5467
5468 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5469 {
5470         if (vcpu->arch.time_page) {
5471                 kvm_release_page_dirty(vcpu->arch.time_page);
5472                 vcpu->arch.time_page = NULL;
5473         }
5474
5475         kvm_x86_ops->vcpu_free(vcpu);
5476 }
5477
5478 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5479                                                 unsigned int id)
5480 {
5481         return kvm_x86_ops->vcpu_create(kvm, id);
5482 }
5483
5484 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5485 {
5486         int r;
5487
5488         /* We do fxsave: this must be aligned. */
5489         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5490
5491         vcpu->arch.mtrr_state.have_fixed = 1;
5492         vcpu_load(vcpu);
5493         r = kvm_arch_vcpu_reset(vcpu);
5494         if (r == 0)
5495                 r = kvm_mmu_setup(vcpu);
5496         vcpu_put(vcpu);
5497         if (r < 0)
5498                 goto free_vcpu;
5499
5500         return 0;
5501 free_vcpu:
5502         kvm_x86_ops->vcpu_free(vcpu);
5503         return r;
5504 }
5505
5506 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5507 {
5508         vcpu_load(vcpu);
5509         kvm_mmu_unload(vcpu);
5510         vcpu_put(vcpu);
5511
5512         kvm_x86_ops->vcpu_free(vcpu);
5513 }
5514
5515 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5516 {
5517         vcpu->arch.nmi_pending = false;
5518         vcpu->arch.nmi_injected = false;
5519
5520         vcpu->arch.switch_db_regs = 0;
5521         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5522         vcpu->arch.dr6 = DR6_FIXED_1;
5523         vcpu->arch.dr7 = DR7_FIXED_1;
5524
5525         return kvm_x86_ops->vcpu_reset(vcpu);
5526 }
5527
5528 int kvm_arch_hardware_enable(void *garbage)
5529 {
5530         /*
5531          * Since this may be called from a hotplug notifcation,
5532          * we can't get the CPU frequency directly.
5533          */
5534         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5535                 int cpu = raw_smp_processor_id();
5536                 per_cpu(cpu_tsc_khz, cpu) = 0;
5537         }
5538
5539         kvm_shared_msr_cpu_online();
5540
5541         return kvm_x86_ops->hardware_enable(garbage);
5542 }
5543
5544 void kvm_arch_hardware_disable(void *garbage)
5545 {
5546         kvm_x86_ops->hardware_disable(garbage);
5547         drop_user_return_notifiers(garbage);
5548 }
5549
5550 int kvm_arch_hardware_setup(void)
5551 {
5552         return kvm_x86_ops->hardware_setup();
5553 }
5554
5555 void kvm_arch_hardware_unsetup(void)
5556 {
5557         kvm_x86_ops->hardware_unsetup();
5558 }
5559
5560 void kvm_arch_check_processor_compat(void *rtn)
5561 {
5562         kvm_x86_ops->check_processor_compatibility(rtn);
5563 }
5564
5565 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5566 {
5567         struct page *page;
5568         struct kvm *kvm;
5569         int r;
5570
5571         BUG_ON(vcpu->kvm == NULL);
5572         kvm = vcpu->kvm;
5573
5574         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5575         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5576                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5577         else
5578                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5579
5580         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5581         if (!page) {
5582                 r = -ENOMEM;
5583                 goto fail;
5584         }
5585         vcpu->arch.pio_data = page_address(page);
5586
5587         r = kvm_mmu_create(vcpu);
5588         if (r < 0)
5589                 goto fail_free_pio_data;
5590
5591         if (irqchip_in_kernel(kvm)) {
5592                 r = kvm_create_lapic(vcpu);
5593                 if (r < 0)
5594                         goto fail_mmu_destroy;
5595         }
5596
5597         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5598                                        GFP_KERNEL);
5599         if (!vcpu->arch.mce_banks) {
5600                 r = -ENOMEM;
5601                 goto fail_free_lapic;
5602         }
5603         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5604
5605         return 0;
5606 fail_free_lapic:
5607         kvm_free_lapic(vcpu);
5608 fail_mmu_destroy:
5609         kvm_mmu_destroy(vcpu);
5610 fail_free_pio_data:
5611         free_page((unsigned long)vcpu->arch.pio_data);
5612 fail:
5613         return r;
5614 }
5615
5616 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5617 {
5618         int idx;
5619
5620         kfree(vcpu->arch.mce_banks);
5621         kvm_free_lapic(vcpu);
5622         idx = srcu_read_lock(&vcpu->kvm->srcu);
5623         kvm_mmu_destroy(vcpu);
5624         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5625         free_page((unsigned long)vcpu->arch.pio_data);
5626 }
5627
5628 struct  kvm *kvm_arch_create_vm(void)
5629 {
5630         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5631
5632         if (!kvm)
5633                 return ERR_PTR(-ENOMEM);
5634
5635         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5636         if (!kvm->arch.aliases) {
5637                 kfree(kvm);
5638                 return ERR_PTR(-ENOMEM);
5639         }
5640
5641         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5642         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5643
5644         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5645         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5646
5647         rdtscll(kvm->arch.vm_init_tsc);
5648
5649         return kvm;
5650 }
5651
5652 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5653 {
5654         vcpu_load(vcpu);
5655         kvm_mmu_unload(vcpu);
5656         vcpu_put(vcpu);
5657 }
5658
5659 static void kvm_free_vcpus(struct kvm *kvm)
5660 {
5661         unsigned int i;
5662         struct kvm_vcpu *vcpu;
5663
5664         /*
5665          * Unpin any mmu pages first.
5666          */
5667         kvm_for_each_vcpu(i, vcpu, kvm)
5668                 kvm_unload_vcpu_mmu(vcpu);
5669         kvm_for_each_vcpu(i, vcpu, kvm)
5670                 kvm_arch_vcpu_free(vcpu);
5671
5672         mutex_lock(&kvm->lock);
5673         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5674                 kvm->vcpus[i] = NULL;
5675
5676         atomic_set(&kvm->online_vcpus, 0);
5677         mutex_unlock(&kvm->lock);
5678 }
5679
5680 void kvm_arch_sync_events(struct kvm *kvm)
5681 {
5682         kvm_free_all_assigned_devices(kvm);
5683 }
5684
5685 void kvm_arch_destroy_vm(struct kvm *kvm)
5686 {
5687         kvm_iommu_unmap_guest(kvm);
5688         kvm_free_pit(kvm);
5689         kfree(kvm->arch.vpic);
5690         kfree(kvm->arch.vioapic);
5691         kvm_free_vcpus(kvm);
5692         kvm_free_physmem(kvm);
5693         if (kvm->arch.apic_access_page)
5694                 put_page(kvm->arch.apic_access_page);
5695         if (kvm->arch.ept_identity_pagetable)
5696                 put_page(kvm->arch.ept_identity_pagetable);
5697         cleanup_srcu_struct(&kvm->srcu);
5698         kfree(kvm->arch.aliases);
5699         kfree(kvm);
5700 }
5701
5702 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5703                                 struct kvm_memory_slot *memslot,
5704                                 struct kvm_memory_slot old,
5705                                 struct kvm_userspace_memory_region *mem,
5706                                 int user_alloc)
5707 {
5708         int npages = memslot->npages;
5709
5710         /*To keep backward compatibility with older userspace,
5711          *x86 needs to hanlde !user_alloc case.
5712          */
5713         if (!user_alloc) {
5714                 if (npages && !old.rmap) {
5715                         unsigned long userspace_addr;
5716
5717                         down_write(&current->mm->mmap_sem);
5718                         userspace_addr = do_mmap(NULL, 0,
5719                                                  npages * PAGE_SIZE,
5720                                                  PROT_READ | PROT_WRITE,
5721                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5722                                                  0);
5723                         up_write(&current->mm->mmap_sem);
5724
5725                         if (IS_ERR((void *)userspace_addr))
5726                                 return PTR_ERR((void *)userspace_addr);
5727
5728                         memslot->userspace_addr = userspace_addr;
5729                 }
5730         }
5731
5732
5733         return 0;
5734 }
5735
5736 void kvm_arch_commit_memory_region(struct kvm *kvm,
5737                                 struct kvm_userspace_memory_region *mem,
5738                                 struct kvm_memory_slot old,
5739                                 int user_alloc)
5740 {
5741
5742         int npages = mem->memory_size >> PAGE_SHIFT;
5743
5744         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5745                 int ret;
5746
5747                 down_write(&current->mm->mmap_sem);
5748                 ret = do_munmap(current->mm, old.userspace_addr,
5749                                 old.npages * PAGE_SIZE);
5750                 up_write(&current->mm->mmap_sem);
5751                 if (ret < 0)
5752                         printk(KERN_WARNING
5753                                "kvm_vm_ioctl_set_memory_region: "
5754                                "failed to munmap memory\n");
5755         }
5756
5757         spin_lock(&kvm->mmu_lock);
5758         if (!kvm->arch.n_requested_mmu_pages) {
5759                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5760                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5761         }
5762
5763         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5764         spin_unlock(&kvm->mmu_lock);
5765 }
5766
5767 void kvm_arch_flush_shadow(struct kvm *kvm)
5768 {
5769         kvm_mmu_zap_all(kvm);
5770         kvm_reload_remote_mmus(kvm);
5771 }
5772
5773 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5774 {
5775         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5776                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5777                 || vcpu->arch.nmi_pending ||
5778                 (kvm_arch_interrupt_allowed(vcpu) &&
5779                  kvm_cpu_has_interrupt(vcpu));
5780 }
5781
5782 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5783 {
5784         int me;
5785         int cpu = vcpu->cpu;
5786
5787         if (waitqueue_active(&vcpu->wq)) {
5788                 wake_up_interruptible(&vcpu->wq);
5789                 ++vcpu->stat.halt_wakeup;
5790         }
5791
5792         me = get_cpu();
5793         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5794                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5795                         smp_send_reschedule(cpu);
5796         put_cpu();
5797 }
5798
5799 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5800 {
5801         return kvm_x86_ops->interrupt_allowed(vcpu);
5802 }
5803
5804 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5805 {
5806         unsigned long rflags;
5807
5808         rflags = kvm_x86_ops->get_rflags(vcpu);
5809         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5810                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5811         return rflags;
5812 }
5813 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5814
5815 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5816 {
5817         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5818             vcpu->arch.singlestep_cs ==
5819                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5820             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5821                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5822         kvm_x86_ops->set_rflags(vcpu, rflags);
5823 }
5824 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5825
5826 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5827 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5828 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5829 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5830 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5831 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5832 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5833 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5834 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5835 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5836 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);