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 <trace/events/kvm.h>
  43 #undef TRACE_INCLUDE_FILE
  44 #define CREATE_TRACE_POINTS
  45 #include "trace.h"
  46
  47 #include <asm/debugreg.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/msr.h>
  50 #include <asm/desc.h>
  51 #include <asm/mtrr.h>
  52 #include <asm/mce.h>
  53
  54 #define MAX_IO_MSRS 256
  55 #define CR0_RESERVED_BITS                                               \
  56         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  57                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  58                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  59 #define CR4_RESERVED_BITS                                               \
  60         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  61                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  62                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  63                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  64
  65 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  66
  67 #define KVM_MAX_MCE_BANKS 32
  68 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  69
  70 /* EFER defaults:
  71  * - enable syscall per default because its emulated by KVM
  72  * - enable LME and LMA per default on 64 bit KVM
  73  */
  74 #ifdef CONFIG_X86_64
  75 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  76 #else
  77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  78 #endif
  79
  80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  82
  83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  85                                     struct kvm_cpuid_entry2 __user *entries);
  86
  87 struct kvm_x86_ops *kvm_x86_ops;
  88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  89
  90 int ignore_msrs = 0;
  91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  92
  93 #define KVM_NR_SHARED_MSRS 16
  94
  95 struct kvm_shared_msrs_global {
  96         int nr;
  97         u32 msrs[KVM_NR_SHARED_MSRS];
  98 };
  99
 100 struct kvm_shared_msrs {
 101         struct user_return_notifier urn;
 102         bool registered;
 103         struct kvm_shared_msr_values {
 104                 u64 host;
 105                 u64 curr;
 106         } values[KVM_NR_SHARED_MSRS];
 107 };
 108
 109 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 110 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 111
 112 struct kvm_stats_debugfs_item debugfs_entries[] = {
 113         { "pf_fixed", VCPU_STAT(pf_fixed) },
 114         { "pf_guest", VCPU_STAT(pf_guest) },
 115         { "tlb_flush", VCPU_STAT(tlb_flush) },
 116         { "invlpg", VCPU_STAT(invlpg) },
 117         { "exits", VCPU_STAT(exits) },
 118         { "io_exits", VCPU_STAT(io_exits) },
 119         { "mmio_exits", VCPU_STAT(mmio_exits) },
 120         { "signal_exits", VCPU_STAT(signal_exits) },
 121         { "irq_window", VCPU_STAT(irq_window_exits) },
 122         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 123         { "halt_exits", VCPU_STAT(halt_exits) },
 124         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 125         { "hypercalls", VCPU_STAT(hypercalls) },
 126         { "request_irq", VCPU_STAT(request_irq_exits) },
 127         { "irq_exits", VCPU_STAT(irq_exits) },
 128         { "host_state_reload", VCPU_STAT(host_state_reload) },
 129         { "efer_reload", VCPU_STAT(efer_reload) },
 130         { "fpu_reload", VCPU_STAT(fpu_reload) },
 131         { "insn_emulation", VCPU_STAT(insn_emulation) },
 132         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 133         { "irq_injections", VCPU_STAT(irq_injections) },
 134         { "nmi_injections", VCPU_STAT(nmi_injections) },
 135         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 136         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 137         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 138         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 139         { "mmu_flooded", VM_STAT(mmu_flooded) },
 140         { "mmu_recycled", VM_STAT(mmu_recycled) },
 141         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 142         { "mmu_unsync", VM_STAT(mmu_unsync) },
 143         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 144         { "largepages", VM_STAT(lpages) },
 145         { NULL }
 146 };
 147
 148 static void kvm_on_user_return(struct user_return_notifier *urn)
 149 {
 150         unsigned slot;
 151         struct kvm_shared_msrs *locals
 152                 = container_of(urn, struct kvm_shared_msrs, urn);
 153         struct kvm_shared_msr_values *values;
 154
 155         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 156                 values = &locals->values[slot];
 157                 if (values->host != values->curr) {
 158                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 159                         values->curr = values->host;
 160                 }
 161         }
 162         locals->registered = false;
 163         user_return_notifier_unregister(urn);
 164 }
 165
 166 static void shared_msr_update(unsigned slot, u32 msr)
 167 {
 168         struct kvm_shared_msrs *smsr;
 169         u64 value;
 170
 171         smsr = &__get_cpu_var(shared_msrs);
 172         /* only read, and nobody should modify it at this time,
 173          * so don't need lock */
 174         if (slot >= shared_msrs_global.nr) {
 175                 printk(KERN_ERR "kvm: invalid MSR slot!");
 176                 return;
 177         }
 178         rdmsrl_safe(msr, &value);
 179         smsr->values[slot].host = value;
 180         smsr->values[slot].curr = value;
 181 }
 182
 183 void kvm_define_shared_msr(unsigned slot, u32 msr)
 184 {
 185         if (slot >= shared_msrs_global.nr)
 186                 shared_msrs_global.nr = slot + 1;
 187         shared_msrs_global.msrs[slot] = msr;
 188         /* we need ensured the shared_msr_global have been updated */
 189         smp_wmb();
 190 }
 191 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 192
 193 static void kvm_shared_msr_cpu_online(void)
 194 {
 195         unsigned i;
 196
 197         for (i = 0; i < shared_msrs_global.nr; ++i)
 198                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 199 }
 200
 201 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 202 {
 203         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 204
 205         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 206                 return;
 207         smsr->values[slot].curr = value;
 208         wrmsrl(shared_msrs_global.msrs[slot], value);
 209         if (!smsr->registered) {
 210                 smsr->urn.on_user_return = kvm_on_user_return;
 211                 user_return_notifier_register(&smsr->urn);
 212                 smsr->registered = true;
 213         }
 214 }
 215 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 216
 217 static void drop_user_return_notifiers(void *ignore)
 218 {
 219         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 220
 221         if (smsr->registered)
 222                 kvm_on_user_return(&smsr->urn);
 223 }
 224
 225 unsigned long segment_base(u16 selector)
 226 {
 227         struct descriptor_table gdt;
 228         struct desc_struct *d;
 229         unsigned long table_base;
 230         unsigned long v;
 231
 232         if (selector == 0)
 233                 return 0;
 234
 235         kvm_get_gdt(&gdt);
 236         table_base = gdt.base;
 237
 238         if (selector & 4) {           /* from ldt */
 239                 u16 ldt_selector = kvm_read_ldt();
 240
 241                 table_base = segment_base(ldt_selector);
 242         }
 243         d = (struct desc_struct *)(table_base + (selector & ~7));
 244         v = get_desc_base(d);
 245 #ifdef CONFIG_X86_64
 246         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 247                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 248 #endif
 249         return v;
 250 }
 251 EXPORT_SYMBOL_GPL(segment_base);
 252
 253 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 254 {
 255         if (irqchip_in_kernel(vcpu->kvm))
 256                 return vcpu->arch.apic_base;
 257         else
 258                 return vcpu->arch.apic_base;
 259 }
 260 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 261
 262 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 263 {
 264         /* TODO: reserve bits check */
 265         if (irqchip_in_kernel(vcpu->kvm))
 266                 kvm_lapic_set_base(vcpu, data);
 267         else
 268                 vcpu->arch.apic_base = data;
 269 }
 270 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 271
 272 #define EXCPT_BENIGN            0
 273 #define EXCPT_CONTRIBUTORY      1
 274 #define EXCPT_PF                2
 275
 276 static int exception_class(int vector)
 277 {
 278         switch (vector) {
 279         case PF_VECTOR:
 280                 return EXCPT_PF;
 281         case DE_VECTOR:
 282         case TS_VECTOR:
 283         case NP_VECTOR:
 284         case SS_VECTOR:
 285         case GP_VECTOR:
 286                 return EXCPT_CONTRIBUTORY;
 287         default:
 288                 break;
 289         }
 290         return EXCPT_BENIGN;
 291 }
 292
 293 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 294                 unsigned nr, bool has_error, u32 error_code)
 295 {
 296         u32 prev_nr;
 297         int class1, class2;
 298
 299         if (!vcpu->arch.exception.pending) {
 300         queue:
 301                 vcpu->arch.exception.pending = true;
 302                 vcpu->arch.exception.has_error_code = has_error;
 303                 vcpu->arch.exception.nr = nr;
 304                 vcpu->arch.exception.error_code = error_code;
 305                 return;
 306         }
 307
 308         /* to check exception */
 309         prev_nr = vcpu->arch.exception.nr;
 310         if (prev_nr == DF_VECTOR) {
 311                 /* triple fault -> shutdown */
 312                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 313                 return;
 314         }
 315         class1 = exception_class(prev_nr);
 316         class2 = exception_class(nr);
 317         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 318                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 319                 /* generate double fault per SDM Table 5-5 */
 320                 vcpu->arch.exception.pending = true;
 321                 vcpu->arch.exception.has_error_code = true;
 322                 vcpu->arch.exception.nr = DF_VECTOR;
 323                 vcpu->arch.exception.error_code = 0;
 324         } else
 325                 /* replace previous exception with a new one in a hope
 326                    that instruction re-execution will regenerate lost
 327                    exception */
 328                 goto queue;
 329 }
 330
 331 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 332 {
 333         kvm_multiple_exception(vcpu, nr, false, 0);
 334 }
 335 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 336
 337 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 338                            u32 error_code)
 339 {
 340         ++vcpu->stat.pf_guest;
 341         vcpu->arch.cr2 = addr;
 342         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 343 }
 344
 345 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 346 {
 347         vcpu->arch.nmi_pending = 1;
 348 }
 349 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 350
 351 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 352 {
 353         kvm_multiple_exception(vcpu, nr, true, error_code);
 354 }
 355 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 356
 357 /*
 358  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 359  * a #GP and return false.
 360  */
 361 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 362 {
 363         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 364                 return true;
 365         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 366         return false;
 367 }
 368 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 369
 370 /*
 371  * Load the pae pdptrs.  Return true is they are all valid.
 372  */
 373 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 374 {
 375         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 376         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 377         int i;
 378         int ret;
 379         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 380
 381         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 382                                   offset * sizeof(u64), sizeof(pdpte));
 383         if (ret < 0) {
 384                 ret = 0;
 385                 goto out;
 386         }
 387         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 388                 if (is_present_gpte(pdpte[i]) &&
 389                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 390                         ret = 0;
 391                         goto out;
 392                 }
 393         }
 394         ret = 1;
 395
 396         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 397         __set_bit(VCPU_EXREG_PDPTR,
 398                   (unsigned long *)&vcpu->arch.regs_avail);
 399         __set_bit(VCPU_EXREG_PDPTR,
 400                   (unsigned long *)&vcpu->arch.regs_dirty);
 401 out:
 402
 403         return ret;
 404 }
 405 EXPORT_SYMBOL_GPL(load_pdptrs);
 406
 407 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 408 {
 409         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 410         bool changed = true;
 411         int r;
 412
 413         if (is_long_mode(vcpu) || !is_pae(vcpu))
 414                 return false;
 415
 416         if (!test_bit(VCPU_EXREG_PDPTR,
 417                       (unsigned long *)&vcpu->arch.regs_avail))
 418                 return true;
 419
 420         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 421         if (r < 0)
 422                 goto out;
 423         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 424 out:
 425
 426         return changed;
 427 }
 428
 429 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 430 {
 431         if (cr0 & CR0_RESERVED_BITS) {
 432                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 433                        cr0, vcpu->arch.cr0);
 434                 kvm_inject_gp(vcpu, 0);
 435                 return;
 436         }
 437
 438         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 439                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 440                 kvm_inject_gp(vcpu, 0);
 441                 return;
 442         }
 443
 444         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 445                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 446                        "and a clear PE flag\n");
 447                 kvm_inject_gp(vcpu, 0);
 448                 return;
 449         }
 450
 451         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 452 #ifdef CONFIG_X86_64
 453                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 454                         int cs_db, cs_l;
 455
 456                         if (!is_pae(vcpu)) {
 457                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 458                                        "in long mode while PAE is disabled\n");
 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                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 465                                        "in long mode while CS.L == 1\n");
 466                                 kvm_inject_gp(vcpu, 0);
 467                                 return;
 468
 469                         }
 470                 } else
 471 #endif
 472                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 473                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 474                                "reserved bits\n");
 475                         kvm_inject_gp(vcpu, 0);
 476                         return;
 477                 }
 478
 479         }
 480
 481         kvm_x86_ops->set_cr0(vcpu, cr0);
 482         vcpu->arch.cr0 = cr0;
 483
 484         kvm_mmu_reset_context(vcpu);
 485         return;
 486 }
 487 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 488
 489 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 490 {
 491         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 492 }
 493 EXPORT_SYMBOL_GPL(kvm_lmsw);
 494
 495 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 496 {
 497         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 498         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 499
 500         if (cr4 & CR4_RESERVED_BITS) {
 501                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 502                 kvm_inject_gp(vcpu, 0);
 503                 return;
 504         }
 505
 506         if (is_long_mode(vcpu)) {
 507                 if (!(cr4 & X86_CR4_PAE)) {
 508                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 509                                "in long mode\n");
 510                         kvm_inject_gp(vcpu, 0);
 511                         return;
 512                 }
 513         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 514                    && ((cr4 ^ old_cr4) & pdptr_bits)
 515                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 516                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 517                 kvm_inject_gp(vcpu, 0);
 518                 return;
 519         }
 520
 521         if (cr4 & X86_CR4_VMXE) {
 522                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 523                 kvm_inject_gp(vcpu, 0);
 524                 return;
 525         }
 526         kvm_x86_ops->set_cr4(vcpu, cr4);
 527         vcpu->arch.cr4 = cr4;
 528         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 529         kvm_mmu_reset_context(vcpu);
 530 }
 531 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 532
 533 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 534 {
 535         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 536                 kvm_mmu_sync_roots(vcpu);
 537                 kvm_mmu_flush_tlb(vcpu);
 538                 return;
 539         }
 540
 541         if (is_long_mode(vcpu)) {
 542                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 543                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 544                         kvm_inject_gp(vcpu, 0);
 545                         return;
 546                 }
 547         } else {
 548                 if (is_pae(vcpu)) {
 549                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 550                                 printk(KERN_DEBUG
 551                                        "set_cr3: #GP, reserved bits\n");
 552                                 kvm_inject_gp(vcpu, 0);
 553                                 return;
 554                         }
 555                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 556                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 557                                        "reserved bits\n");
 558                                 kvm_inject_gp(vcpu, 0);
 559                                 return;
 560                         }
 561                 }
 562                 /*
 563                  * We don't check reserved bits in nonpae mode, because
 564                  * this isn't enforced, and VMware depends on this.
 565                  */
 566         }
 567
 568         /*
 569          * Does the new cr3 value map to physical memory? (Note, we
 570          * catch an invalid cr3 even in real-mode, because it would
 571          * cause trouble later on when we turn on paging anyway.)
 572          *
 573          * A real CPU would silently accept an invalid cr3 and would
 574          * attempt to use it - with largely undefined (and often hard
 575          * to debug) behavior on the guest side.
 576          */
 577         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 578                 kvm_inject_gp(vcpu, 0);
 579         else {
 580                 vcpu->arch.cr3 = cr3;
 581                 vcpu->arch.mmu.new_cr3(vcpu);
 582         }
 583 }
 584 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 585
 586 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 587 {
 588         if (cr8 & CR8_RESERVED_BITS) {
 589                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 590                 kvm_inject_gp(vcpu, 0);
 591                 return;
 592         }
 593         if (irqchip_in_kernel(vcpu->kvm))
 594                 kvm_lapic_set_tpr(vcpu, cr8);
 595         else
 596                 vcpu->arch.cr8 = cr8;
 597 }
 598 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 599
 600 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 601 {
 602         if (irqchip_in_kernel(vcpu->kvm))
 603                 return kvm_lapic_get_cr8(vcpu);
 604         else
 605                 return vcpu->arch.cr8;
 606 }
 607 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 608
 609 static inline u32 bit(int bitno)
 610 {
 611         return 1 << (bitno & 31);
 612 }
 613
 614 /*
 615  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 616  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 617  *
 618  * This list is modified at module load time to reflect the
 619  * capabilities of the host cpu. This capabilities test skips MSRs that are
 620  * kvm-specific. Those are put in the beginning of the list.
 621  */
 622
 623 #define KVM_SAVE_MSRS_BEGIN     2
 624 static u32 msrs_to_save[] = {
 625         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 626         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 627         MSR_K6_STAR,
 628 #ifdef CONFIG_X86_64
 629         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 630 #endif
 631         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 632 };
 633
 634 static unsigned num_msrs_to_save;
 635
 636 static u32 emulated_msrs[] = {
 637         MSR_IA32_MISC_ENABLE,
 638 };
 639
 640 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 641 {
 642         if (efer & efer_reserved_bits) {
 643                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 644                        efer);
 645                 kvm_inject_gp(vcpu, 0);
 646                 return;
 647         }
 648
 649         if (is_paging(vcpu)
 650             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 651                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 652                 kvm_inject_gp(vcpu, 0);
 653                 return;
 654         }
 655
 656         if (efer & EFER_FFXSR) {
 657                 struct kvm_cpuid_entry2 *feat;
 658
 659                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 660                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 661                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 662                         kvm_inject_gp(vcpu, 0);
 663                         return;
 664                 }
 665         }
 666
 667         if (efer & EFER_SVME) {
 668                 struct kvm_cpuid_entry2 *feat;
 669
 670                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 671                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 672                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 673                         kvm_inject_gp(vcpu, 0);
 674                         return;
 675                 }
 676         }
 677
 678         kvm_x86_ops->set_efer(vcpu, efer);
 679
 680         efer &= ~EFER_LMA;
 681         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 682
 683         vcpu->arch.shadow_efer = efer;
 684
 685         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 686         kvm_mmu_reset_context(vcpu);
 687 }
 688
 689 void kvm_enable_efer_bits(u64 mask)
 690 {
 691        efer_reserved_bits &= ~mask;
 692 }
 693 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 694
 695
 696 /*
 697  * Writes msr value into into the appropriate "register".
 698  * Returns 0 on success, non-0 otherwise.
 699  * Assumes vcpu_load() was already called.
 700  */
 701 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 702 {
 703         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 704 }
 705
 706 /*
 707  * Adapt set_msr() to msr_io()'s calling convention
 708  */
 709 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 710 {
 711         return kvm_set_msr(vcpu, index, *data);
 712 }
 713
 714 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 715 {
 716         static int version;
 717         struct pvclock_wall_clock wc;
 718         struct timespec boot;
 719
 720         if (!wall_clock)
 721                 return;
 722
 723         version++;
 724
 725         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 726
 727         /*
 728          * The guest calculates current wall clock time by adding
 729          * system time (updated by kvm_write_guest_time below) to the
 730          * wall clock specified here.  guest system time equals host
 731          * system time for us, thus we must fill in host boot time here.
 732          */
 733         getboottime(&boot);
 734
 735         wc.sec = boot.tv_sec;
 736         wc.nsec = boot.tv_nsec;
 737         wc.version = version;
 738
 739         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 740
 741         version++;
 742         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 743 }
 744
 745 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 746 {
 747         uint32_t quotient, remainder;
 748
 749         /* Don't try to replace with do_div(), this one calculates
 750          * "(dividend << 32) / divisor" */
 751         __asm__ ( "divl %4"
 752                   : "=a" (quotient), "=d" (remainder)
 753                   : "0" (0), "1" (dividend), "r" (divisor) );
 754         return quotient;
 755 }
 756
 757 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 758 {
 759         uint64_t nsecs = 1000000000LL;
 760         int32_t  shift = 0;
 761         uint64_t tps64;
 762         uint32_t tps32;
 763
 764         tps64 = tsc_khz * 1000LL;
 765         while (tps64 > nsecs*2) {
 766                 tps64 >>= 1;
 767                 shift--;
 768         }
 769
 770         tps32 = (uint32_t)tps64;
 771         while (tps32 <= (uint32_t)nsecs) {
 772                 tps32 <<= 1;
 773                 shift++;
 774         }
 775
 776         hv_clock->tsc_shift = shift;
 777         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 778
 779         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 780                  __func__, tsc_khz, hv_clock->tsc_shift,
 781                  hv_clock->tsc_to_system_mul);
 782 }
 783
 784 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 785
 786 static void kvm_write_guest_time(struct kvm_vcpu *v)
 787 {
 788         struct timespec ts;
 789         unsigned long flags;
 790         struct kvm_vcpu_arch *vcpu = &v->arch;
 791         void *shared_kaddr;
 792         unsigned long this_tsc_khz;
 793
 794         if ((!vcpu->time_page))
 795                 return;
 796
 797         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 798         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 799                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 800                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 801         }
 802         put_cpu_var(cpu_tsc_khz);
 803
 804         /* Keep irq disabled to prevent changes to the clock */
 805         local_irq_save(flags);
 806         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 807         ktime_get_ts(&ts);
 808         monotonic_to_bootbased(&ts);
 809         local_irq_restore(flags);
 810
 811         /* With all the info we got, fill in the values */
 812
 813         vcpu->hv_clock.system_time = ts.tv_nsec +
 814                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 815
 816         /*
 817          * The interface expects us to write an even number signaling that the
 818          * update is finished. Since the guest won't see the intermediate
 819          * state, we just increase by 2 at the end.
 820          */
 821         vcpu->hv_clock.version += 2;
 822
 823         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 824
 825         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 826                sizeof(vcpu->hv_clock));
 827
 828         kunmap_atomic(shared_kaddr, KM_USER0);
 829
 830         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 831 }
 832
 833 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 834 {
 835         struct kvm_vcpu_arch *vcpu = &v->arch;
 836
 837         if (!vcpu->time_page)
 838                 return 0;
 839         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 840         return 1;
 841 }
 842
 843 static bool msr_mtrr_valid(unsigned msr)
 844 {
 845         switch (msr) {
 846         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 847         case MSR_MTRRfix64K_00000:
 848         case MSR_MTRRfix16K_80000:
 849         case MSR_MTRRfix16K_A0000:
 850         case MSR_MTRRfix4K_C0000:
 851         case MSR_MTRRfix4K_C8000:
 852         case MSR_MTRRfix4K_D0000:
 853         case MSR_MTRRfix4K_D8000:
 854         case MSR_MTRRfix4K_E0000:
 855         case MSR_MTRRfix4K_E8000:
 856         case MSR_MTRRfix4K_F0000:
 857         case MSR_MTRRfix4K_F8000:
 858         case MSR_MTRRdefType:
 859         case MSR_IA32_CR_PAT:
 860                 return true;
 861         case 0x2f8:
 862                 return true;
 863         }
 864         return false;
 865 }
 866
 867 static bool valid_pat_type(unsigned t)
 868 {
 869         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 870 }
 871
 872 static bool valid_mtrr_type(unsigned t)
 873 {
 874         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 875 }
 876
 877 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 878 {
 879         int i;
 880
 881         if (!msr_mtrr_valid(msr))
 882                 return false;
 883
 884         if (msr == MSR_IA32_CR_PAT) {
 885                 for (i = 0; i < 8; i++)
 886                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 887                                 return false;
 888                 return true;
 889         } else if (msr == MSR_MTRRdefType) {
 890                 if (data & ~0xcff)
 891                         return false;
 892                 return valid_mtrr_type(data & 0xff);
 893         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 894                 for (i = 0; i < 8 ; i++)
 895                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 896                                 return false;
 897                 return true;
 898         }
 899
 900         /* variable MTRRs */
 901         return valid_mtrr_type(data & 0xff);
 902 }
 903
 904 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 905 {
 906         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 907
 908         if (!mtrr_valid(vcpu, msr, data))
 909                 return 1;
 910
 911         if (msr == MSR_MTRRdefType) {
 912                 vcpu->arch.mtrr_state.def_type = data;
 913                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 914         } else if (msr == MSR_MTRRfix64K_00000)
 915                 p[0] = data;
 916         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 917                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 918         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 919                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 920         else if (msr == MSR_IA32_CR_PAT)
 921                 vcpu->arch.pat = data;
 922         else {  /* Variable MTRRs */
 923                 int idx, is_mtrr_mask;
 924                 u64 *pt;
 925
 926                 idx = (msr - 0x200) / 2;
 927                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 928                 if (!is_mtrr_mask)
 929                         pt =
 930                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 931                 else
 932                         pt =
 933                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 934                 *pt = data;
 935         }
 936
 937         kvm_mmu_reset_context(vcpu);
 938         return 0;
 939 }
 940
 941 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 942 {
 943         u64 mcg_cap = vcpu->arch.mcg_cap;
 944         unsigned bank_num = mcg_cap & 0xff;
 945
 946         switch (msr) {
 947         case MSR_IA32_MCG_STATUS:
 948                 vcpu->arch.mcg_status = data;
 949                 break;
 950         case MSR_IA32_MCG_CTL:
 951                 if (!(mcg_cap & MCG_CTL_P))
 952                         return 1;
 953                 if (data != 0 && data != ~(u64)0)
 954                         return -1;
 955                 vcpu->arch.mcg_ctl = data;
 956                 break;
 957         default:
 958                 if (msr >= MSR_IA32_MC0_CTL &&
 959                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 960                         u32 offset = msr - MSR_IA32_MC0_CTL;
 961                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 962                         if ((offset & 0x3) == 0 &&
 963                             data != 0 && data != ~(u64)0)
 964                                 return -1;
 965                         vcpu->arch.mce_banks[offset] = data;
 966                         break;
 967                 }
 968                 return 1;
 969         }
 970         return 0;
 971 }
 972
 973 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 974 {
 975         struct kvm *kvm = vcpu->kvm;
 976         int lm = is_long_mode(vcpu);
 977         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 978                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 979         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 980                 : kvm->arch.xen_hvm_config.blob_size_32;
 981         u32 page_num = data & ~PAGE_MASK;
 982         u64 page_addr = data & PAGE_MASK;
 983         u8 *page;
 984         int r;
 985
 986         r = -E2BIG;
 987         if (page_num >= blob_size)
 988                 goto out;
 989         r = -ENOMEM;
 990         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 991         if (!page)
 992                 goto out;
 993         r = -EFAULT;
 994         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 995                 goto out_free;
 996         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 997                 goto out_free;
 998         r = 0;
 999 out_free:
1000         kfree(page);
1001 out:
1002         return r;
1003 }
1004
1005 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1006 {
1007         switch (msr) {
1008         case MSR_EFER:
1009                 set_efer(vcpu, data);
1010                 break;
1011         case MSR_K7_HWCR:
1012                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1013                 if (data != 0) {
1014                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1015                                 data);
1016                         return 1;
1017                 }
1018                 break;
1019         case MSR_FAM10H_MMIO_CONF_BASE:
1020                 if (data != 0) {
1021                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1022                                 "0x%llx\n", data);
1023                         return 1;
1024                 }
1025                 break;
1026         case MSR_AMD64_NB_CFG:
1027                 break;
1028         case MSR_IA32_DEBUGCTLMSR:
1029                 if (!data) {
1030                         /* We support the non-activated case already */
1031                         break;
1032                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1033                         /* Values other than LBR and BTF are vendor-specific,
1034                            thus reserved and should throw a #GP */
1035                         return 1;
1036                 }
1037                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1038                         __func__, data);
1039                 break;
1040         case MSR_IA32_UCODE_REV:
1041         case MSR_IA32_UCODE_WRITE:
1042         case MSR_VM_HSAVE_PA:
1043         case MSR_AMD64_PATCH_LOADER:
1044                 break;
1045         case 0x200 ... 0x2ff:
1046                 return set_msr_mtrr(vcpu, msr, data);
1047         case MSR_IA32_APICBASE:
1048                 kvm_set_apic_base(vcpu, data);
1049                 break;
1050         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1051                 return kvm_x2apic_msr_write(vcpu, msr, data);
1052         case MSR_IA32_MISC_ENABLE:
1053                 vcpu->arch.ia32_misc_enable_msr = data;
1054                 break;
1055         case MSR_KVM_WALL_CLOCK:
1056                 vcpu->kvm->arch.wall_clock = data;
1057                 kvm_write_wall_clock(vcpu->kvm, data);
1058                 break;
1059         case MSR_KVM_SYSTEM_TIME: {
1060                 if (vcpu->arch.time_page) {
1061                         kvm_release_page_dirty(vcpu->arch.time_page);
1062                         vcpu->arch.time_page = NULL;
1063                 }
1064
1065                 vcpu->arch.time = data;
1066
1067                 /* we verify if the enable bit is set... */
1068                 if (!(data & 1))
1069                         break;
1070
1071                 /* ...but clean it before doing the actual write */
1072                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1073
1074                 vcpu->arch.time_page =
1075                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1076
1077                 if (is_error_page(vcpu->arch.time_page)) {
1078                         kvm_release_page_clean(vcpu->arch.time_page);
1079                         vcpu->arch.time_page = NULL;
1080                 }
1081
1082                 kvm_request_guest_time_update(vcpu);
1083                 break;
1084         }
1085         case MSR_IA32_MCG_CTL:
1086         case MSR_IA32_MCG_STATUS:
1087         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1088                 return set_msr_mce(vcpu, msr, data);
1089
1090         /* Performance counters are not protected by a CPUID bit,
1091          * so we should check all of them in the generic path for the sake of
1092          * cross vendor migration.
1093          * Writing a zero into the event select MSRs disables them,
1094          * which we perfectly emulate ;-). Any other value should be at least
1095          * reported, some guests depend on them.
1096          */
1097         case MSR_P6_EVNTSEL0:
1098         case MSR_P6_EVNTSEL1:
1099         case MSR_K7_EVNTSEL0:
1100         case MSR_K7_EVNTSEL1:
1101         case MSR_K7_EVNTSEL2:
1102         case MSR_K7_EVNTSEL3:
1103                 if (data != 0)
1104                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1105                                 "0x%x data 0x%llx\n", msr, data);
1106                 break;
1107         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1108          * so we ignore writes to make it happy.
1109          */
1110         case MSR_P6_PERFCTR0:
1111         case MSR_P6_PERFCTR1:
1112         case MSR_K7_PERFCTR0:
1113         case MSR_K7_PERFCTR1:
1114         case MSR_K7_PERFCTR2:
1115         case MSR_K7_PERFCTR3:
1116                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1117                         "0x%x data 0x%llx\n", msr, data);
1118                 break;
1119         default:
1120                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1121                         return xen_hvm_config(vcpu, data);
1122                 if (!ignore_msrs) {
1123                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1124                                 msr, data);
1125                         return 1;
1126                 } else {
1127                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1128                                 msr, data);
1129                         break;
1130                 }
1131         }
1132         return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1135
1136
1137 /*
1138  * Reads an msr value (of 'msr_index') into 'pdata'.
1139  * Returns 0 on success, non-0 otherwise.
1140  * Assumes vcpu_load() was already called.
1141  */
1142 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1143 {
1144         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1145 }
1146
1147 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1148 {
1149         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1150
1151         if (!msr_mtrr_valid(msr))
1152                 return 1;
1153
1154         if (msr == MSR_MTRRdefType)
1155                 *pdata = vcpu->arch.mtrr_state.def_type +
1156                          (vcpu->arch.mtrr_state.enabled << 10);
1157         else if (msr == MSR_MTRRfix64K_00000)
1158                 *pdata = p[0];
1159         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1160                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1161         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1162                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1163         else if (msr == MSR_IA32_CR_PAT)
1164                 *pdata = vcpu->arch.pat;
1165         else {  /* Variable MTRRs */
1166                 int idx, is_mtrr_mask;
1167                 u64 *pt;
1168
1169                 idx = (msr - 0x200) / 2;
1170                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1171                 if (!is_mtrr_mask)
1172                         pt =
1173                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1174                 else
1175                         pt =
1176                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1177                 *pdata = *pt;
1178         }
1179
1180         return 0;
1181 }
1182
1183 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1184 {
1185         u64 data;
1186         u64 mcg_cap = vcpu->arch.mcg_cap;
1187         unsigned bank_num = mcg_cap & 0xff;
1188
1189         switch (msr) {
1190         case MSR_IA32_P5_MC_ADDR:
1191         case MSR_IA32_P5_MC_TYPE:
1192                 data = 0;
1193                 break;
1194         case MSR_IA32_MCG_CAP:
1195                 data = vcpu->arch.mcg_cap;
1196                 break;
1197         case MSR_IA32_MCG_CTL:
1198                 if (!(mcg_cap & MCG_CTL_P))
1199                         return 1;
1200                 data = vcpu->arch.mcg_ctl;
1201                 break;
1202         case MSR_IA32_MCG_STATUS:
1203                 data = vcpu->arch.mcg_status;
1204                 break;
1205         default:
1206                 if (msr >= MSR_IA32_MC0_CTL &&
1207                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1208                         u32 offset = msr - MSR_IA32_MC0_CTL;
1209                         data = vcpu->arch.mce_banks[offset];
1210                         break;
1211                 }
1212                 return 1;
1213         }
1214         *pdata = data;
1215         return 0;
1216 }
1217
1218 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1219 {
1220         u64 data;
1221
1222         switch (msr) {
1223         case MSR_IA32_PLATFORM_ID:
1224         case MSR_IA32_UCODE_REV:
1225         case MSR_IA32_EBL_CR_POWERON:
1226         case MSR_IA32_DEBUGCTLMSR:
1227         case MSR_IA32_LASTBRANCHFROMIP:
1228         case MSR_IA32_LASTBRANCHTOIP:
1229         case MSR_IA32_LASTINTFROMIP:
1230         case MSR_IA32_LASTINTTOIP:
1231         case MSR_K8_SYSCFG:
1232         case MSR_K7_HWCR:
1233         case MSR_VM_HSAVE_PA:
1234         case MSR_P6_PERFCTR0:
1235         case MSR_P6_PERFCTR1:
1236         case MSR_P6_EVNTSEL0:
1237         case MSR_P6_EVNTSEL1:
1238         case MSR_K7_EVNTSEL0:
1239         case MSR_K7_PERFCTR0:
1240         case MSR_K8_INT_PENDING_MSG:
1241         case MSR_AMD64_NB_CFG:
1242         case MSR_FAM10H_MMIO_CONF_BASE:
1243                 data = 0;
1244                 break;
1245         case MSR_MTRRcap:
1246                 data = 0x500 | KVM_NR_VAR_MTRR;
1247                 break;
1248         case 0x200 ... 0x2ff:
1249                 return get_msr_mtrr(vcpu, msr, pdata);
1250         case 0xcd: /* fsb frequency */
1251                 data = 3;
1252                 break;
1253         case MSR_IA32_APICBASE:
1254                 data = kvm_get_apic_base(vcpu);
1255                 break;
1256         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1257                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1258                 break;
1259         case MSR_IA32_MISC_ENABLE:
1260                 data = vcpu->arch.ia32_misc_enable_msr;
1261                 break;
1262         case MSR_IA32_PERF_STATUS:
1263                 /* TSC increment by tick */
1264                 data = 1000ULL;
1265                 /* CPU multiplier */
1266                 data |= (((uint64_t)4ULL) << 40);
1267                 break;
1268         case MSR_EFER:
1269                 data = vcpu->arch.shadow_efer;
1270                 break;
1271         case MSR_KVM_WALL_CLOCK:
1272                 data = vcpu->kvm->arch.wall_clock;
1273                 break;
1274         case MSR_KVM_SYSTEM_TIME:
1275                 data = vcpu->arch.time;
1276                 break;
1277         case MSR_IA32_P5_MC_ADDR:
1278         case MSR_IA32_P5_MC_TYPE:
1279         case MSR_IA32_MCG_CAP:
1280         case MSR_IA32_MCG_CTL:
1281         case MSR_IA32_MCG_STATUS:
1282         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1283                 return get_msr_mce(vcpu, msr, pdata);
1284         default:
1285                 if (!ignore_msrs) {
1286                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1287                         return 1;
1288                 } else {
1289                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1290                         data = 0;
1291                 }
1292                 break;
1293         }
1294         *pdata = data;
1295         return 0;
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1298
1299 /*
1300  * Read or write a bunch of msrs. All parameters are kernel addresses.
1301  *
1302  * @return number of msrs set successfully.
1303  */
1304 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1305                     struct kvm_msr_entry *entries,
1306                     int (*do_msr)(struct kvm_vcpu *vcpu,
1307                                   unsigned index, u64 *data))
1308 {
1309         int i, idx;
1310
1311         vcpu_load(vcpu);
1312
1313         idx = srcu_read_lock(&vcpu->kvm->srcu);
1314         for (i = 0; i < msrs->nmsrs; ++i)
1315                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1316                         break;
1317         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1318
1319         vcpu_put(vcpu);
1320
1321         return i;
1322 }
1323
1324 /*
1325  * Read or write a bunch of msrs. Parameters are user addresses.
1326  *
1327  * @return number of msrs set successfully.
1328  */
1329 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1330                   int (*do_msr)(struct kvm_vcpu *vcpu,
1331                                 unsigned index, u64 *data),
1332                   int writeback)
1333 {
1334         struct kvm_msrs msrs;
1335         struct kvm_msr_entry *entries;
1336         int r, n;
1337         unsigned size;
1338
1339         r = -EFAULT;
1340         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1341                 goto out;
1342
1343         r = -E2BIG;
1344         if (msrs.nmsrs >= MAX_IO_MSRS)
1345                 goto out;
1346
1347         r = -ENOMEM;
1348         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1349         entries = vmalloc(size);
1350         if (!entries)
1351                 goto out;
1352
1353         r = -EFAULT;
1354         if (copy_from_user(entries, user_msrs->entries, size))
1355                 goto out_free;
1356
1357         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1358         if (r < 0)
1359                 goto out_free;
1360
1361         r = -EFAULT;
1362         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1363                 goto out_free;
1364
1365         r = n;
1366
1367 out_free:
1368         vfree(entries);
1369 out:
1370         return r;
1371 }
1372
1373 int kvm_dev_ioctl_check_extension(long ext)
1374 {
1375         int r;
1376
1377         switch (ext) {
1378         case KVM_CAP_IRQCHIP:
1379         case KVM_CAP_HLT:
1380         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1381         case KVM_CAP_SET_TSS_ADDR:
1382         case KVM_CAP_EXT_CPUID:
1383         case KVM_CAP_CLOCKSOURCE:
1384         case KVM_CAP_PIT:
1385         case KVM_CAP_NOP_IO_DELAY:
1386         case KVM_CAP_MP_STATE:
1387         case KVM_CAP_SYNC_MMU:
1388         case KVM_CAP_REINJECT_CONTROL:
1389         case KVM_CAP_IRQ_INJECT_STATUS:
1390         case KVM_CAP_ASSIGN_DEV_IRQ:
1391         case KVM_CAP_IRQFD:
1392         case KVM_CAP_IOEVENTFD:
1393         case KVM_CAP_PIT2:
1394         case KVM_CAP_PIT_STATE2:
1395         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1396         case KVM_CAP_XEN_HVM:
1397         case KVM_CAP_ADJUST_CLOCK:
1398         case KVM_CAP_VCPU_EVENTS:
1399                 r = 1;
1400                 break;
1401         case KVM_CAP_COALESCED_MMIO:
1402                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1403                 break;
1404         case KVM_CAP_VAPIC:
1405                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1406                 break;
1407         case KVM_CAP_NR_VCPUS:
1408                 r = KVM_MAX_VCPUS;
1409                 break;
1410         case KVM_CAP_NR_MEMSLOTS:
1411                 r = KVM_MEMORY_SLOTS;
1412                 break;
1413         case KVM_CAP_PV_MMU:    /* obsolete */
1414                 r = 0;
1415                 break;
1416         case KVM_CAP_IOMMU:
1417                 r = iommu_found();
1418                 break;
1419         case KVM_CAP_MCE:
1420                 r = KVM_MAX_MCE_BANKS;
1421                 break;
1422         default:
1423                 r = 0;
1424                 break;
1425         }
1426         return r;
1427
1428 }
1429
1430 long kvm_arch_dev_ioctl(struct file *filp,
1431                         unsigned int ioctl, unsigned long arg)
1432 {
1433         void __user *argp = (void __user *)arg;
1434         long r;
1435
1436         switch (ioctl) {
1437         case KVM_GET_MSR_INDEX_LIST: {
1438                 struct kvm_msr_list __user *user_msr_list = argp;
1439                 struct kvm_msr_list msr_list;
1440                 unsigned n;
1441
1442                 r = -EFAULT;
1443                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1444                         goto out;
1445                 n = msr_list.nmsrs;
1446                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1447                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1448                         goto out;
1449                 r = -E2BIG;
1450                 if (n < msr_list.nmsrs)
1451                         goto out;
1452                 r = -EFAULT;
1453                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1454                                  num_msrs_to_save * sizeof(u32)))
1455                         goto out;
1456                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1457                                  &emulated_msrs,
1458                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1459                         goto out;
1460                 r = 0;
1461                 break;
1462         }
1463         case KVM_GET_SUPPORTED_CPUID: {
1464                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1465                 struct kvm_cpuid2 cpuid;
1466
1467                 r = -EFAULT;
1468                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1469                         goto out;
1470                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1471                                                       cpuid_arg->entries);
1472                 if (r)
1473                         goto out;
1474
1475                 r = -EFAULT;
1476                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1477                         goto out;
1478                 r = 0;
1479                 break;
1480         }
1481         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1482                 u64 mce_cap;
1483
1484                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1485                 r = -EFAULT;
1486                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1487                         goto out;
1488                 r = 0;
1489                 break;
1490         }
1491         default:
1492                 r = -EINVAL;
1493         }
1494 out:
1495         return r;
1496 }
1497
1498 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1499 {
1500         kvm_x86_ops->vcpu_load(vcpu, cpu);
1501         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1502                 unsigned long khz = cpufreq_quick_get(cpu);
1503                 if (!khz)
1504                         khz = tsc_khz;
1505                 per_cpu(cpu_tsc_khz, cpu) = khz;
1506         }
1507         kvm_request_guest_time_update(vcpu);
1508 }
1509
1510 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1511 {
1512         kvm_x86_ops->vcpu_put(vcpu);
1513         kvm_put_guest_fpu(vcpu);
1514 }
1515
1516 static int is_efer_nx(void)
1517 {
1518         unsigned long long efer = 0;
1519
1520         rdmsrl_safe(MSR_EFER, &efer);
1521         return efer & EFER_NX;
1522 }
1523
1524 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1525 {
1526         int i;
1527         struct kvm_cpuid_entry2 *e, *entry;
1528
1529         entry = NULL;
1530         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1531                 e = &vcpu->arch.cpuid_entries[i];
1532                 if (e->function == 0x80000001) {
1533                         entry = e;
1534                         break;
1535                 }
1536         }
1537         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1538                 entry->edx &= ~(1 << 20);
1539                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1540         }
1541 }
1542
1543 /* when an old userspace process fills a new kernel module */
1544 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1545                                     struct kvm_cpuid *cpuid,
1546                                     struct kvm_cpuid_entry __user *entries)
1547 {
1548         int r, i;
1549         struct kvm_cpuid_entry *cpuid_entries;
1550
1551         r = -E2BIG;
1552         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1553                 goto out;
1554         r = -ENOMEM;
1555         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1556         if (!cpuid_entries)
1557                 goto out;
1558         r = -EFAULT;
1559         if (copy_from_user(cpuid_entries, entries,
1560                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1561                 goto out_free;
1562         for (i = 0; i < cpuid->nent; i++) {
1563                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1564                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1565                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1566                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1567                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1568                 vcpu->arch.cpuid_entries[i].index = 0;
1569                 vcpu->arch.cpuid_entries[i].flags = 0;
1570                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1571                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1572                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1573         }
1574         vcpu->arch.cpuid_nent = cpuid->nent;
1575         cpuid_fix_nx_cap(vcpu);
1576         r = 0;
1577         kvm_apic_set_version(vcpu);
1578         kvm_x86_ops->cpuid_update(vcpu);
1579
1580 out_free:
1581         vfree(cpuid_entries);
1582 out:
1583         return r;
1584 }
1585
1586 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1587                                      struct kvm_cpuid2 *cpuid,
1588                                      struct kvm_cpuid_entry2 __user *entries)
1589 {
1590         int r;
1591
1592         r = -E2BIG;
1593         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1594                 goto out;
1595         r = -EFAULT;
1596         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1597                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1598                 goto out;
1599         vcpu->arch.cpuid_nent = cpuid->nent;
1600         kvm_apic_set_version(vcpu);
1601         kvm_x86_ops->cpuid_update(vcpu);
1602         return 0;
1603
1604 out:
1605         return r;
1606 }
1607
1608 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1609                                      struct kvm_cpuid2 *cpuid,
1610                                      struct kvm_cpuid_entry2 __user *entries)
1611 {
1612         int r;
1613
1614         r = -E2BIG;
1615         if (cpuid->nent < vcpu->arch.cpuid_nent)
1616                 goto out;
1617         r = -EFAULT;
1618         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1619                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1620                 goto out;
1621         return 0;
1622
1623 out:
1624         cpuid->nent = vcpu->arch.cpuid_nent;
1625         return r;
1626 }
1627
1628 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1629                            u32 index)
1630 {
1631         entry->function = function;
1632         entry->index = index;
1633         cpuid_count(entry->function, entry->index,
1634                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1635         entry->flags = 0;
1636 }
1637
1638 #define F(x) bit(X86_FEATURE_##x)
1639
1640 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1641                          u32 index, int *nent, int maxnent)
1642 {
1643         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1644         unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1645 #ifdef CONFIG_X86_64
1646         unsigned f_lm = F(LM);
1647 #else
1648         unsigned f_lm = 0;
1649 #endif
1650         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1651
1652         /* cpuid 1.edx */
1653         const u32 kvm_supported_word0_x86_features =
1654                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1655                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1656                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1657                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1658                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1659                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1660                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1661                 0 /* HTT, TM, Reserved, PBE */;
1662         /* cpuid 0x80000001.edx */
1663         const u32 kvm_supported_word1_x86_features =
1664                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1665                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1666                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1667                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1668                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1669                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1670                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1671                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1672         /* cpuid 1.ecx */
1673         const u32 kvm_supported_word4_x86_features =
1674                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1675                 0 /* DS-CPL, VMX, SMX, EST */ |
1676                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1677                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1678                 0 /* Reserved, DCA */ | F(XMM4_1) |
1679                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1680                 0 /* Reserved, XSAVE, OSXSAVE */;
1681         /* cpuid 0x80000001.ecx */
1682         const u32 kvm_supported_word6_x86_features =
1683                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1684                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1685                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1686                 0 /* SKINIT */ | 0 /* WDT */;
1687
1688         /* all calls to cpuid_count() should be made on the same cpu */
1689         get_cpu();
1690         do_cpuid_1_ent(entry, function, index);
1691         ++*nent;
1692
1693         switch (function) {
1694         case 0:
1695                 entry->eax = min(entry->eax, (u32)0xb);
1696                 break;
1697         case 1:
1698                 entry->edx &= kvm_supported_word0_x86_features;
1699                 entry->ecx &= kvm_supported_word4_x86_features;
1700                 /* we support x2apic emulation even if host does not support
1701                  * it since we emulate x2apic in software */
1702                 entry->ecx |= F(X2APIC);
1703                 break;
1704         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1705          * may return different values. This forces us to get_cpu() before
1706          * issuing the first command, and also to emulate this annoying behavior
1707          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1708         case 2: {
1709                 int t, times = entry->eax & 0xff;
1710
1711                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1712                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1713                 for (t = 1; t < times && *nent < maxnent; ++t) {
1714                         do_cpuid_1_ent(&entry[t], function, 0);
1715                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1716                         ++*nent;
1717                 }
1718                 break;
1719         }
1720         /* function 4 and 0xb have additional index. */
1721         case 4: {
1722                 int i, cache_type;
1723
1724                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1725                 /* read more entries until cache_type is zero */
1726                 for (i = 1; *nent < maxnent; ++i) {
1727                         cache_type = entry[i - 1].eax & 0x1f;
1728                         if (!cache_type)
1729                                 break;
1730                         do_cpuid_1_ent(&entry[i], function, i);
1731                         entry[i].flags |=
1732                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1733                         ++*nent;
1734                 }
1735                 break;
1736         }
1737         case 0xb: {
1738                 int i, level_type;
1739
1740                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1741                 /* read more entries until level_type is zero */
1742                 for (i = 1; *nent < maxnent; ++i) {
1743                         level_type = entry[i - 1].ecx & 0xff00;
1744                         if (!level_type)
1745                                 break;
1746                         do_cpuid_1_ent(&entry[i], function, i);
1747                         entry[i].flags |=
1748                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1749                         ++*nent;
1750                 }
1751                 break;
1752         }
1753         case 0x80000000:
1754                 entry->eax = min(entry->eax, 0x8000001a);
1755                 break;
1756         case 0x80000001:
1757                 entry->edx &= kvm_supported_word1_x86_features;
1758                 entry->ecx &= kvm_supported_word6_x86_features;
1759                 break;
1760         }
1761         put_cpu();
1762 }
1763
1764 #undef F
1765
1766 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1767                                      struct kvm_cpuid_entry2 __user *entries)
1768 {
1769         struct kvm_cpuid_entry2 *cpuid_entries;
1770         int limit, nent = 0, r = -E2BIG;
1771         u32 func;
1772
1773         if (cpuid->nent < 1)
1774                 goto out;
1775         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1776                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1777         r = -ENOMEM;
1778         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1779         if (!cpuid_entries)
1780                 goto out;
1781
1782         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1783         limit = cpuid_entries[0].eax;
1784         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1785                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1786                              &nent, cpuid->nent);
1787         r = -E2BIG;
1788         if (nent >= cpuid->nent)
1789                 goto out_free;
1790
1791         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1792         limit = cpuid_entries[nent - 1].eax;
1793         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1794                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1795                              &nent, cpuid->nent);
1796         r = -E2BIG;
1797         if (nent >= cpuid->nent)
1798                 goto out_free;
1799
1800         r = -EFAULT;
1801         if (copy_to_user(entries, cpuid_entries,
1802                          nent * sizeof(struct kvm_cpuid_entry2)))
1803                 goto out_free;
1804         cpuid->nent = nent;
1805         r = 0;
1806
1807 out_free:
1808         vfree(cpuid_entries);
1809 out:
1810         return r;
1811 }
1812
1813 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1814                                     struct kvm_lapic_state *s)
1815 {
1816         vcpu_load(vcpu);
1817         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1818         vcpu_put(vcpu);
1819
1820         return 0;
1821 }
1822
1823 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1824                                     struct kvm_lapic_state *s)
1825 {
1826         vcpu_load(vcpu);
1827         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1828         kvm_apic_post_state_restore(vcpu);
1829         update_cr8_intercept(vcpu);
1830         vcpu_put(vcpu);
1831
1832         return 0;
1833 }
1834
1835 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1836                                     struct kvm_interrupt *irq)
1837 {
1838         if (irq->irq < 0 || irq->irq >= 256)
1839                 return -EINVAL;
1840         if (irqchip_in_kernel(vcpu->kvm))
1841                 return -ENXIO;
1842         vcpu_load(vcpu);
1843
1844         kvm_queue_interrupt(vcpu, irq->irq, false);
1845
1846         vcpu_put(vcpu);
1847
1848         return 0;
1849 }
1850
1851 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1852 {
1853         vcpu_load(vcpu);
1854         kvm_inject_nmi(vcpu);
1855         vcpu_put(vcpu);
1856
1857         return 0;
1858 }
1859
1860 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1861                                            struct kvm_tpr_access_ctl *tac)
1862 {
1863         if (tac->flags)
1864                 return -EINVAL;
1865         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1866         return 0;
1867 }
1868
1869 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1870                                         u64 mcg_cap)
1871 {
1872         int r;
1873         unsigned bank_num = mcg_cap & 0xff, bank;
1874
1875         r = -EINVAL;
1876         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1877                 goto out;
1878         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1879                 goto out;
1880         r = 0;
1881         vcpu->arch.mcg_cap = mcg_cap;
1882         /* Init IA32_MCG_CTL to all 1s */
1883         if (mcg_cap & MCG_CTL_P)
1884                 vcpu->arch.mcg_ctl = ~(u64)0;
1885         /* Init IA32_MCi_CTL to all 1s */
1886         for (bank = 0; bank < bank_num; bank++)
1887                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1888 out:
1889         return r;
1890 }
1891
1892 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1893                                       struct kvm_x86_mce *mce)
1894 {
1895         u64 mcg_cap = vcpu->arch.mcg_cap;
1896         unsigned bank_num = mcg_cap & 0xff;
1897         u64 *banks = vcpu->arch.mce_banks;
1898
1899         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1900                 return -EINVAL;
1901         /*
1902          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1903          * reporting is disabled
1904          */
1905         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1906             vcpu->arch.mcg_ctl != ~(u64)0)
1907                 return 0;
1908         banks += 4 * mce->bank;
1909         /*
1910          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1911          * reporting is disabled for the bank
1912          */
1913         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1914                 return 0;
1915         if (mce->status & MCI_STATUS_UC) {
1916                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1917                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
1918                         printk(KERN_DEBUG "kvm: set_mce: "
1919                                "injects mce exception while "
1920                                "previous one is in progress!\n");
1921                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1922                         return 0;
1923                 }
1924                 if (banks[1] & MCI_STATUS_VAL)
1925                         mce->status |= MCI_STATUS_OVER;
1926                 banks[2] = mce->addr;
1927                 banks[3] = mce->misc;
1928                 vcpu->arch.mcg_status = mce->mcg_status;
1929                 banks[1] = mce->status;
1930                 kvm_queue_exception(vcpu, MC_VECTOR);
1931         } else if (!(banks[1] & MCI_STATUS_VAL)
1932                    || !(banks[1] & MCI_STATUS_UC)) {
1933                 if (banks[1] & MCI_STATUS_VAL)
1934                         mce->status |= MCI_STATUS_OVER;
1935                 banks[2] = mce->addr;
1936                 banks[3] = mce->misc;
1937                 banks[1] = mce->status;
1938         } else
1939                 banks[1] |= MCI_STATUS_OVER;
1940         return 0;
1941 }
1942
1943 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
1944                                                struct kvm_vcpu_events *events)
1945 {
1946         vcpu_load(vcpu);
1947
1948         events->exception.injected = vcpu->arch.exception.pending;
1949         events->exception.nr = vcpu->arch.exception.nr;
1950         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
1951         events->exception.error_code = vcpu->arch.exception.error_code;
1952
1953         events->interrupt.injected = vcpu->arch.interrupt.pending;
1954         events->interrupt.nr = vcpu->arch.interrupt.nr;
1955         events->interrupt.soft = vcpu->arch.interrupt.soft;
1956
1957         events->nmi.injected = vcpu->arch.nmi_injected;
1958         events->nmi.pending = vcpu->arch.nmi_pending;
1959         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
1960
1961         events->sipi_vector = vcpu->arch.sipi_vector;
1962
1963         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
1964                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
1965
1966         vcpu_put(vcpu);
1967 }
1968
1969 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
1970                                               struct kvm_vcpu_events *events)
1971 {
1972         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
1973                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
1974                 return -EINVAL;
1975
1976         vcpu_load(vcpu);
1977
1978         vcpu->arch.exception.pending = events->exception.injected;
1979         vcpu->arch.exception.nr = events->exception.nr;
1980         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
1981         vcpu->arch.exception.error_code = events->exception.error_code;
1982
1983         vcpu->arch.interrupt.pending = events->interrupt.injected;
1984         vcpu->arch.interrupt.nr = events->interrupt.nr;
1985         vcpu->arch.interrupt.soft = events->interrupt.soft;
1986         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
1987                 kvm_pic_clear_isr_ack(vcpu->kvm);
1988
1989         vcpu->arch.nmi_injected = events->nmi.injected;
1990         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
1991                 vcpu->arch.nmi_pending = events->nmi.pending;
1992         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
1993
1994         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
1995                 vcpu->arch.sipi_vector = events->sipi_vector;
1996
1997         vcpu_put(vcpu);
1998
1999         return 0;
2000 }
2001
2002 long kvm_arch_vcpu_ioctl(struct file *filp,
2003                          unsigned int ioctl, unsigned long arg)
2004 {
2005         struct kvm_vcpu *vcpu = filp->private_data;
2006         void __user *argp = (void __user *)arg;
2007         int r;
2008         struct kvm_lapic_state *lapic = NULL;
2009
2010         switch (ioctl) {
2011         case KVM_GET_LAPIC: {
2012                 r = -EINVAL;
2013                 if (!vcpu->arch.apic)
2014                         goto out;
2015                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2016
2017                 r = -ENOMEM;
2018                 if (!lapic)
2019                         goto out;
2020                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2021                 if (r)
2022                         goto out;
2023                 r = -EFAULT;
2024                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2025                         goto out;
2026                 r = 0;
2027                 break;
2028         }
2029         case KVM_SET_LAPIC: {
2030                 r = -EINVAL;
2031                 if (!vcpu->arch.apic)
2032                         goto out;
2033                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2034                 r = -ENOMEM;
2035                 if (!lapic)
2036                         goto out;
2037                 r = -EFAULT;
2038                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2039                         goto out;
2040                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2041                 if (r)
2042                         goto out;
2043                 r = 0;
2044                 break;
2045         }
2046         case KVM_INTERRUPT: {
2047                 struct kvm_interrupt irq;
2048
2049                 r = -EFAULT;
2050                 if (copy_from_user(&irq, argp, sizeof irq))
2051                         goto out;
2052                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2053                 if (r)
2054                         goto out;
2055                 r = 0;
2056                 break;
2057         }
2058         case KVM_NMI: {
2059                 r = kvm_vcpu_ioctl_nmi(vcpu);
2060                 if (r)
2061                         goto out;
2062                 r = 0;
2063                 break;
2064         }
2065         case KVM_SET_CPUID: {
2066                 struct kvm_cpuid __user *cpuid_arg = argp;
2067                 struct kvm_cpuid cpuid;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2071                         goto out;
2072                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2073                 if (r)
2074                         goto out;
2075                 break;
2076         }
2077         case KVM_SET_CPUID2: {
2078                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2079                 struct kvm_cpuid2 cpuid;
2080
2081                 r = -EFAULT;
2082                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2083                         goto out;
2084                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2085                                               cpuid_arg->entries);
2086                 if (r)
2087                         goto out;
2088                 break;
2089         }
2090         case KVM_GET_CPUID2: {
2091                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2092                 struct kvm_cpuid2 cpuid;
2093
2094                 r = -EFAULT;
2095                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2096                         goto out;
2097                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2098                                               cpuid_arg->entries);
2099                 if (r)
2100                         goto out;
2101                 r = -EFAULT;
2102                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2103                         goto out;
2104                 r = 0;
2105                 break;
2106         }
2107         case KVM_GET_MSRS:
2108                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2109                 break;
2110         case KVM_SET_MSRS:
2111                 r = msr_io(vcpu, argp, do_set_msr, 0);
2112                 break;
2113         case KVM_TPR_ACCESS_REPORTING: {
2114                 struct kvm_tpr_access_ctl tac;
2115
2116                 r = -EFAULT;
2117                 if (copy_from_user(&tac, argp, sizeof tac))
2118                         goto out;
2119                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2120                 if (r)
2121                         goto out;
2122                 r = -EFAULT;
2123                 if (copy_to_user(argp, &tac, sizeof tac))
2124                         goto out;
2125                 r = 0;
2126                 break;
2127         };
2128         case KVM_SET_VAPIC_ADDR: {
2129                 struct kvm_vapic_addr va;
2130
2131                 r = -EINVAL;
2132                 if (!irqchip_in_kernel(vcpu->kvm))
2133                         goto out;
2134                 r = -EFAULT;
2135                 if (copy_from_user(&va, argp, sizeof va))
2136                         goto out;
2137                 r = 0;
2138                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2139                 break;
2140         }
2141         case KVM_X86_SETUP_MCE: {
2142                 u64 mcg_cap;
2143
2144                 r = -EFAULT;
2145                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2146                         goto out;
2147                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2148                 break;
2149         }
2150         case KVM_X86_SET_MCE: {
2151                 struct kvm_x86_mce mce;
2152
2153                 r = -EFAULT;
2154                 if (copy_from_user(&mce, argp, sizeof mce))
2155                         goto out;
2156                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2157                 break;
2158         }
2159         case KVM_GET_VCPU_EVENTS: {
2160                 struct kvm_vcpu_events events;
2161
2162                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2163
2164                 r = -EFAULT;
2165                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2166                         break;
2167                 r = 0;
2168                 break;
2169         }
2170         case KVM_SET_VCPU_EVENTS: {
2171                 struct kvm_vcpu_events events;
2172
2173                 r = -EFAULT;
2174                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2175                         break;
2176
2177                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2178                 break;
2179         }
2180         default:
2181                 r = -EINVAL;
2182         }
2183 out:
2184         kfree(lapic);
2185         return r;
2186 }
2187
2188 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2189 {
2190         int ret;
2191
2192         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2193                 return -1;
2194         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2195         return ret;
2196 }
2197
2198 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2199                                               u64 ident_addr)
2200 {
2201         kvm->arch.ept_identity_map_addr = ident_addr;
2202         return 0;
2203 }
2204
2205 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2206                                           u32 kvm_nr_mmu_pages)
2207 {
2208         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2209                 return -EINVAL;
2210
2211         mutex_lock(&kvm->slots_lock);
2212         spin_lock(&kvm->mmu_lock);
2213
2214         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2215         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2216
2217         spin_unlock(&kvm->mmu_lock);
2218         mutex_unlock(&kvm->slots_lock);
2219         return 0;
2220 }
2221
2222 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2223 {
2224         return kvm->arch.n_alloc_mmu_pages;
2225 }
2226
2227 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2228 {
2229         int i;
2230         struct kvm_mem_alias *alias;
2231         struct kvm_mem_aliases *aliases;
2232
2233         aliases = rcu_dereference(kvm->arch.aliases);
2234
2235         for (i = 0; i < aliases->naliases; ++i) {
2236                 alias = &aliases->aliases[i];
2237                 if (alias->flags & KVM_ALIAS_INVALID)
2238                         continue;
2239                 if (gfn >= alias->base_gfn
2240                     && gfn < alias->base_gfn + alias->npages)
2241                         return alias->target_gfn + gfn - alias->base_gfn;
2242         }
2243         return gfn;
2244 }
2245
2246 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2247 {
2248         int i;
2249         struct kvm_mem_alias *alias;
2250         struct kvm_mem_aliases *aliases;
2251
2252         aliases = rcu_dereference(kvm->arch.aliases);
2253
2254         for (i = 0; i < aliases->naliases; ++i) {
2255                 alias = &aliases->aliases[i];
2256                 if (gfn >= alias->base_gfn
2257                     && gfn < alias->base_gfn + alias->npages)
2258                         return alias->target_gfn + gfn - alias->base_gfn;
2259         }
2260         return gfn;
2261 }
2262
2263 /*
2264  * Set a new alias region.  Aliases map a portion of physical memory into
2265  * another portion.  This is useful for memory windows, for example the PC
2266  * VGA region.
2267  */
2268 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2269                                          struct kvm_memory_alias *alias)
2270 {
2271         int r, n;
2272         struct kvm_mem_alias *p;
2273         struct kvm_mem_aliases *aliases, *old_aliases;
2274
2275         r = -EINVAL;
2276         /* General sanity checks */
2277         if (alias->memory_size & (PAGE_SIZE - 1))
2278                 goto out;
2279         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2280                 goto out;
2281         if (alias->slot >= KVM_ALIAS_SLOTS)
2282                 goto out;
2283         if (alias->guest_phys_addr + alias->memory_size
2284             < alias->guest_phys_addr)
2285                 goto out;
2286         if (alias->target_phys_addr + alias->memory_size
2287             < alias->target_phys_addr)
2288                 goto out;
2289
2290         r = -ENOMEM;
2291         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2292         if (!aliases)
2293                 goto out;
2294
2295         mutex_lock(&kvm->slots_lock);
2296
2297         /* invalidate any gfn reference in case of deletion/shrinking */
2298         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2299         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2300         old_aliases = kvm->arch.aliases;
2301         rcu_assign_pointer(kvm->arch.aliases, aliases);
2302         synchronize_srcu_expedited(&kvm->srcu);
2303         kvm_mmu_zap_all(kvm);
2304         kfree(old_aliases);
2305
2306         r = -ENOMEM;
2307         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2308         if (!aliases)
2309                 goto out_unlock;
2310
2311         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2312
2313         p = &aliases->aliases[alias->slot];
2314         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2315         p->npages = alias->memory_size >> PAGE_SHIFT;
2316         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2317         p->flags &= ~(KVM_ALIAS_INVALID);
2318
2319         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2320                 if (aliases->aliases[n - 1].npages)
2321                         break;
2322         aliases->naliases = n;
2323
2324         old_aliases = kvm->arch.aliases;
2325         rcu_assign_pointer(kvm->arch.aliases, aliases);
2326         synchronize_srcu_expedited(&kvm->srcu);
2327         kfree(old_aliases);
2328         r = 0;
2329
2330 out_unlock:
2331         mutex_unlock(&kvm->slots_lock);
2332 out:
2333         return r;
2334 }
2335
2336 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2337 {
2338         int r;
2339
2340         r = 0;
2341         switch (chip->chip_id) {
2342         case KVM_IRQCHIP_PIC_MASTER:
2343                 memcpy(&chip->chip.pic,
2344                         &pic_irqchip(kvm)->pics[0],
2345                         sizeof(struct kvm_pic_state));
2346                 break;
2347         case KVM_IRQCHIP_PIC_SLAVE:
2348                 memcpy(&chip->chip.pic,
2349                         &pic_irqchip(kvm)->pics[1],
2350                         sizeof(struct kvm_pic_state));
2351                 break;
2352         case KVM_IRQCHIP_IOAPIC:
2353                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2354                 break;
2355         default:
2356                 r = -EINVAL;
2357                 break;
2358         }
2359         return r;
2360 }
2361
2362 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2363 {
2364         int r;
2365
2366         r = 0;
2367         switch (chip->chip_id) {
2368         case KVM_IRQCHIP_PIC_MASTER:
2369                 spin_lock(&pic_irqchip(kvm)->lock);
2370                 memcpy(&pic_irqchip(kvm)->pics[0],
2371                         &chip->chip.pic,
2372                         sizeof(struct kvm_pic_state));
2373                 spin_unlock(&pic_irqchip(kvm)->lock);
2374                 break;
2375         case KVM_IRQCHIP_PIC_SLAVE:
2376                 spin_lock(&pic_irqchip(kvm)->lock);
2377                 memcpy(&pic_irqchip(kvm)->pics[1],
2378                         &chip->chip.pic,
2379                         sizeof(struct kvm_pic_state));
2380                 spin_unlock(&pic_irqchip(kvm)->lock);
2381                 break;
2382         case KVM_IRQCHIP_IOAPIC:
2383                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2384                 break;
2385         default:
2386                 r = -EINVAL;
2387                 break;
2388         }
2389         kvm_pic_update_irq(pic_irqchip(kvm));
2390         return r;
2391 }
2392
2393 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2394 {
2395         int r = 0;
2396
2397         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2398         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2399         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2400         return r;
2401 }
2402
2403 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2404 {
2405         int r = 0;
2406
2407         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2408         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2409         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2410         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2411         return r;
2412 }
2413
2414 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2415 {
2416         int r = 0;
2417
2418         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2419         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2420                 sizeof(ps->channels));
2421         ps->flags = kvm->arch.vpit->pit_state.flags;
2422         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2423         return r;
2424 }
2425
2426 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2427 {
2428         int r = 0, start = 0;
2429         u32 prev_legacy, cur_legacy;
2430         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2431         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2432         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2433         if (!prev_legacy && cur_legacy)
2434                 start = 1;
2435         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2436                sizeof(kvm->arch.vpit->pit_state.channels));
2437         kvm->arch.vpit->pit_state.flags = ps->flags;
2438         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2439         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2440         return r;
2441 }
2442
2443 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2444                                  struct kvm_reinject_control *control)
2445 {
2446         if (!kvm->arch.vpit)
2447                 return -ENXIO;
2448         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2449         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2450         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2451         return 0;
2452 }
2453
2454 /*
2455  * Get (and clear) the dirty memory log for a memory slot.
2456  */
2457 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2458                                       struct kvm_dirty_log *log)
2459 {
2460         int r, n, i;
2461         struct kvm_memory_slot *memslot;
2462         unsigned long is_dirty = 0;
2463         unsigned long *dirty_bitmap = NULL;
2464
2465         mutex_lock(&kvm->slots_lock);
2466
2467         r = -EINVAL;
2468         if (log->slot >= KVM_MEMORY_SLOTS)
2469                 goto out;
2470
2471         memslot = &kvm->memslots->memslots[log->slot];
2472         r = -ENOENT;
2473         if (!memslot->dirty_bitmap)
2474                 goto out;
2475
2476         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2477
2478         r = -ENOMEM;
2479         dirty_bitmap = vmalloc(n);
2480         if (!dirty_bitmap)
2481                 goto out;
2482         memset(dirty_bitmap, 0, n);
2483
2484         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2485                 is_dirty = memslot->dirty_bitmap[i];
2486
2487         /* If nothing is dirty, don't bother messing with page tables. */
2488         if (is_dirty) {
2489                 struct kvm_memslots *slots, *old_slots;
2490
2491                 spin_lock(&kvm->mmu_lock);
2492                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2493                 spin_unlock(&kvm->mmu_lock);
2494
2495                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2496                 if (!slots)
2497                         goto out_free;
2498
2499                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2500                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2501
2502                 old_slots = kvm->memslots;
2503                 rcu_assign_pointer(kvm->memslots, slots);
2504                 synchronize_srcu_expedited(&kvm->srcu);
2505                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2506                 kfree(old_slots);
2507         }
2508
2509         r = 0;
2510         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2511                 r = -EFAULT;
2512 out_free:
2513         vfree(dirty_bitmap);
2514 out:
2515         mutex_unlock(&kvm->slots_lock);
2516         return r;
2517 }
2518
2519 long kvm_arch_vm_ioctl(struct file *filp,
2520                        unsigned int ioctl, unsigned long arg)
2521 {
2522         struct kvm *kvm = filp->private_data;
2523         void __user *argp = (void __user *)arg;
2524         int r = -ENOTTY;
2525         /*
2526          * This union makes it completely explicit to gcc-3.x
2527          * that these two variables' stack usage should be
2528          * combined, not added together.
2529          */
2530         union {
2531                 struct kvm_pit_state ps;
2532                 struct kvm_pit_state2 ps2;
2533                 struct kvm_memory_alias alias;
2534                 struct kvm_pit_config pit_config;
2535         } u;
2536
2537         switch (ioctl) {
2538         case KVM_SET_TSS_ADDR:
2539                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2540                 if (r < 0)
2541                         goto out;
2542                 break;
2543         case KVM_SET_IDENTITY_MAP_ADDR: {
2544                 u64 ident_addr;
2545
2546                 r = -EFAULT;
2547                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2548                         goto out;
2549                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2550                 if (r < 0)
2551                         goto out;
2552                 break;
2553         }
2554         case KVM_SET_MEMORY_REGION: {
2555                 struct kvm_memory_region kvm_mem;
2556                 struct kvm_userspace_memory_region kvm_userspace_mem;
2557
2558                 r = -EFAULT;
2559                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2560                         goto out;
2561                 kvm_userspace_mem.slot = kvm_mem.slot;
2562                 kvm_userspace_mem.flags = kvm_mem.flags;
2563                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2564                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2565                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2566                 if (r)
2567                         goto out;
2568                 break;
2569         }
2570         case KVM_SET_NR_MMU_PAGES:
2571                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2572                 if (r)
2573                         goto out;
2574                 break;
2575         case KVM_GET_NR_MMU_PAGES:
2576                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2577                 break;
2578         case KVM_SET_MEMORY_ALIAS:
2579                 r = -EFAULT;
2580                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2581                         goto out;
2582                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2583                 if (r)
2584                         goto out;
2585                 break;
2586         case KVM_CREATE_IRQCHIP: {
2587                 struct kvm_pic *vpic;
2588
2589                 mutex_lock(&kvm->lock);
2590                 r = -EEXIST;
2591                 if (kvm->arch.vpic)
2592                         goto create_irqchip_unlock;
2593                 r = -ENOMEM;
2594                 vpic = kvm_create_pic(kvm);
2595                 if (vpic) {
2596                         r = kvm_ioapic_init(kvm);
2597                         if (r) {
2598                                 kfree(vpic);
2599                                 goto create_irqchip_unlock;
2600                         }
2601                 } else
2602                         goto create_irqchip_unlock;
2603                 smp_wmb();
2604                 kvm->arch.vpic = vpic;
2605                 smp_wmb();
2606                 r = kvm_setup_default_irq_routing(kvm);
2607                 if (r) {
2608                         mutex_lock(&kvm->irq_lock);
2609                         kfree(kvm->arch.vpic);
2610                         kfree(kvm->arch.vioapic);
2611                         kvm->arch.vpic = NULL;
2612                         kvm->arch.vioapic = NULL;
2613                         mutex_unlock(&kvm->irq_lock);
2614                 }
2615         create_irqchip_unlock:
2616                 mutex_unlock(&kvm->lock);
2617                 break;
2618         }
2619         case KVM_CREATE_PIT:
2620                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2621                 goto create_pit;
2622         case KVM_CREATE_PIT2:
2623                 r = -EFAULT;
2624                 if (copy_from_user(&u.pit_config, argp,
2625                                    sizeof(struct kvm_pit_config)))
2626                         goto out;
2627         create_pit:
2628                 mutex_lock(&kvm->slots_lock);
2629                 r = -EEXIST;
2630                 if (kvm->arch.vpit)
2631                         goto create_pit_unlock;
2632                 r = -ENOMEM;
2633                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2634                 if (kvm->arch.vpit)
2635                         r = 0;
2636         create_pit_unlock:
2637                 mutex_unlock(&kvm->slots_lock);
2638                 break;
2639         case KVM_IRQ_LINE_STATUS:
2640         case KVM_IRQ_LINE: {
2641                 struct kvm_irq_level irq_event;
2642
2643                 r = -EFAULT;
2644                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2645                         goto out;
2646                 if (irqchip_in_kernel(kvm)) {
2647                         __s32 status;
2648                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2649                                         irq_event.irq, irq_event.level);
2650                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2651                                 irq_event.status = status;
2652                                 if (copy_to_user(argp, &irq_event,
2653                                                         sizeof irq_event))
2654                                         goto out;
2655                         }
2656                         r = 0;
2657                 }
2658                 break;
2659         }
2660         case KVM_GET_IRQCHIP: {
2661                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2662                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2663
2664                 r = -ENOMEM;
2665                 if (!chip)
2666                         goto out;
2667                 r = -EFAULT;
2668                 if (copy_from_user(chip, argp, sizeof *chip))
2669                         goto get_irqchip_out;
2670                 r = -ENXIO;
2671                 if (!irqchip_in_kernel(kvm))
2672                         goto get_irqchip_out;
2673                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2674                 if (r)
2675                         goto get_irqchip_out;
2676                 r = -EFAULT;
2677                 if (copy_to_user(argp, chip, sizeof *chip))
2678                         goto get_irqchip_out;
2679                 r = 0;
2680         get_irqchip_out:
2681                 kfree(chip);
2682                 if (r)
2683                         goto out;
2684                 break;
2685         }
2686         case KVM_SET_IRQCHIP: {
2687                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2688                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2689
2690                 r = -ENOMEM;
2691                 if (!chip)
2692                         goto out;
2693                 r = -EFAULT;
2694                 if (copy_from_user(chip, argp, sizeof *chip))
2695                         goto set_irqchip_out;
2696                 r = -ENXIO;
2697                 if (!irqchip_in_kernel(kvm))
2698                         goto set_irqchip_out;
2699                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2700                 if (r)
2701                         goto set_irqchip_out;
2702                 r = 0;
2703         set_irqchip_out:
2704                 kfree(chip);
2705                 if (r)
2706                         goto out;
2707                 break;
2708         }
2709         case KVM_GET_PIT: {
2710                 r = -EFAULT;
2711                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2712                         goto out;
2713                 r = -ENXIO;
2714                 if (!kvm->arch.vpit)
2715                         goto out;
2716                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2717                 if (r)
2718                         goto out;
2719                 r = -EFAULT;
2720                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2721                         goto out;
2722                 r = 0;
2723                 break;
2724         }
2725         case KVM_SET_PIT: {
2726                 r = -EFAULT;
2727                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2728                         goto out;
2729                 r = -ENXIO;
2730                 if (!kvm->arch.vpit)
2731                         goto out;
2732                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2733                 if (r)
2734                         goto out;
2735                 r = 0;
2736                 break;
2737         }
2738         case KVM_GET_PIT2: {
2739                 r = -ENXIO;
2740                 if (!kvm->arch.vpit)
2741                         goto out;
2742                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2743                 if (r)
2744                         goto out;
2745                 r = -EFAULT;
2746                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2747                         goto out;
2748                 r = 0;
2749                 break;
2750         }
2751         case KVM_SET_PIT2: {
2752                 r = -EFAULT;
2753                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2754                         goto out;
2755                 r = -ENXIO;
2756                 if (!kvm->arch.vpit)
2757                         goto out;
2758                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2759                 if (r)
2760                         goto out;
2761                 r = 0;
2762                 break;
2763         }
2764         case KVM_REINJECT_CONTROL: {
2765                 struct kvm_reinject_control control;
2766                 r =  -EFAULT;
2767                 if (copy_from_user(&control, argp, sizeof(control)))
2768                         goto out;
2769                 r = kvm_vm_ioctl_reinject(kvm, &control);
2770                 if (r)
2771                         goto out;
2772                 r = 0;
2773                 break;
2774         }
2775         case KVM_XEN_HVM_CONFIG: {
2776                 r = -EFAULT;
2777                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2778                                    sizeof(struct kvm_xen_hvm_config)))
2779                         goto out;
2780                 r = -EINVAL;
2781                 if (kvm->arch.xen_hvm_config.flags)
2782                         goto out;
2783                 r = 0;
2784                 break;
2785         }
2786         case KVM_SET_CLOCK: {
2787                 struct timespec now;
2788                 struct kvm_clock_data user_ns;
2789                 u64 now_ns;
2790                 s64 delta;
2791
2792                 r = -EFAULT;
2793                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2794                         goto out;
2795
2796                 r = -EINVAL;
2797                 if (user_ns.flags)
2798                         goto out;
2799
2800                 r = 0;
2801                 ktime_get_ts(&now);
2802                 now_ns = timespec_to_ns(&now);
2803                 delta = user_ns.clock - now_ns;
2804                 kvm->arch.kvmclock_offset = delta;
2805                 break;
2806         }
2807         case KVM_GET_CLOCK: {
2808                 struct timespec now;
2809                 struct kvm_clock_data user_ns;
2810                 u64 now_ns;
2811
2812                 ktime_get_ts(&now);
2813                 now_ns = timespec_to_ns(&now);
2814                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2815                 user_ns.flags = 0;
2816
2817                 r = -EFAULT;
2818                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2819                         goto out;
2820                 r = 0;
2821                 break;
2822         }
2823
2824         default:
2825                 ;
2826         }
2827 out:
2828         return r;
2829 }
2830
2831 static void kvm_init_msr_list(void)
2832 {
2833         u32 dummy[2];
2834         unsigned i, j;
2835
2836         /* skip the first msrs in the list. KVM-specific */
2837         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2838                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2839                         continue;
2840                 if (j < i)
2841                         msrs_to_save[j] = msrs_to_save[i];
2842                 j++;
2843         }
2844         num_msrs_to_save = j;
2845 }
2846
2847 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2848                            const void *v)
2849 {
2850         if (vcpu->arch.apic &&
2851             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2852                 return 0;
2853
2854         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
2855 }
2856
2857 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2858 {
2859         if (vcpu->arch.apic &&
2860             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2861                 return 0;
2862
2863         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
2864 }
2865
2866 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2867                                struct kvm_vcpu *vcpu)
2868 {
2869         void *data = val;
2870         int r = X86EMUL_CONTINUE;
2871
2872         while (bytes) {
2873                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2874                 unsigned offset = addr & (PAGE_SIZE-1);
2875                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2876                 int ret;
2877
2878                 if (gpa == UNMAPPED_GVA) {
2879                         r = X86EMUL_PROPAGATE_FAULT;
2880                         goto out;
2881                 }
2882                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2883                 if (ret < 0) {
2884                         r = X86EMUL_UNHANDLEABLE;
2885                         goto out;
2886                 }
2887
2888                 bytes -= toread;
2889                 data += toread;
2890                 addr += toread;
2891         }
2892 out:
2893         return r;
2894 }
2895
2896 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2897                                 struct kvm_vcpu *vcpu)
2898 {
2899         void *data = val;
2900         int r = X86EMUL_CONTINUE;
2901
2902         while (bytes) {
2903                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2904                 unsigned offset = addr & (PAGE_SIZE-1);
2905                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2906                 int ret;
2907
2908                 if (gpa == UNMAPPED_GVA) {
2909                         r = X86EMUL_PROPAGATE_FAULT;
2910                         goto out;
2911                 }
2912                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2913                 if (ret < 0) {
2914                         r = X86EMUL_UNHANDLEABLE;
2915                         goto out;
2916                 }
2917
2918                 bytes -= towrite;
2919                 data += towrite;
2920                 addr += towrite;
2921         }
2922 out:
2923         return r;
2924 }
2925
2926
2927 static int emulator_read_emulated(unsigned long addr,
2928                                   void *val,
2929                                   unsigned int bytes,
2930                                   struct kvm_vcpu *vcpu)
2931 {
2932         gpa_t                 gpa;
2933
2934         if (vcpu->mmio_read_completed) {
2935                 memcpy(val, vcpu->mmio_data, bytes);
2936                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2937                                vcpu->mmio_phys_addr, *(u64 *)val);
2938                 vcpu->mmio_read_completed = 0;
2939                 return X86EMUL_CONTINUE;
2940         }
2941
2942         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2943
2944         /* For APIC access vmexit */
2945         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2946                 goto mmio;
2947
2948         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2949                                 == X86EMUL_CONTINUE)
2950                 return X86EMUL_CONTINUE;
2951         if (gpa == UNMAPPED_GVA)
2952                 return X86EMUL_PROPAGATE_FAULT;
2953
2954 mmio:
2955         /*
2956          * Is this MMIO handled locally?
2957          */
2958         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2959                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2960                 return X86EMUL_CONTINUE;
2961         }
2962
2963         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2964
2965         vcpu->mmio_needed = 1;
2966         vcpu->mmio_phys_addr = gpa;
2967         vcpu->mmio_size = bytes;
2968         vcpu->mmio_is_write = 0;
2969
2970         return X86EMUL_UNHANDLEABLE;
2971 }
2972
2973 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2974                           const void *val, int bytes)
2975 {
2976         int ret;
2977
2978         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2979         if (ret < 0)
2980                 return 0;
2981         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2982         return 1;
2983 }
2984
2985 static int emulator_write_emulated_onepage(unsigned long addr,
2986                                            const void *val,
2987                                            unsigned int bytes,
2988                                            struct kvm_vcpu *vcpu)
2989 {
2990         gpa_t                 gpa;
2991
2992         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2993
2994         if (gpa == UNMAPPED_GVA) {
2995                 kvm_inject_page_fault(vcpu, addr, 2);
2996                 return X86EMUL_PROPAGATE_FAULT;
2997         }
2998
2999         /* For APIC access vmexit */
3000         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3001                 goto mmio;
3002
3003         if (emulator_write_phys(vcpu, gpa, val, bytes))
3004                 return X86EMUL_CONTINUE;
3005
3006 mmio:
3007         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3008         /*
3009          * Is this MMIO handled locally?
3010          */
3011         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3012                 return X86EMUL_CONTINUE;
3013
3014         vcpu->mmio_needed = 1;
3015         vcpu->mmio_phys_addr = gpa;
3016         vcpu->mmio_size = bytes;
3017         vcpu->mmio_is_write = 1;
3018         memcpy(vcpu->mmio_data, val, bytes);
3019
3020         return X86EMUL_CONTINUE;
3021 }
3022
3023 int emulator_write_emulated(unsigned long addr,
3024                                    const void *val,
3025                                    unsigned int bytes,
3026                                    struct kvm_vcpu *vcpu)
3027 {
3028         /* Crossing a page boundary? */
3029         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3030                 int rc, now;
3031
3032                 now = -addr & ~PAGE_MASK;
3033                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3034                 if (rc != X86EMUL_CONTINUE)
3035                         return rc;
3036                 addr += now;
3037                 val += now;
3038                 bytes -= now;
3039         }
3040         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3041 }
3042 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3043
3044 static int emulator_cmpxchg_emulated(unsigned long addr,
3045                                      const void *old,
3046                                      const void *new,
3047                                      unsigned int bytes,
3048                                      struct kvm_vcpu *vcpu)
3049 {
3050         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3051 #ifndef CONFIG_X86_64
3052         /* guests cmpxchg8b have to be emulated atomically */
3053         if (bytes == 8) {
3054                 gpa_t gpa;
3055                 struct page *page;
3056                 char *kaddr;
3057                 u64 val;
3058
3059                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3060
3061                 if (gpa == UNMAPPED_GVA ||
3062                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3063                         goto emul_write;
3064
3065                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3066                         goto emul_write;
3067
3068                 val = *(u64 *)new;
3069
3070                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3071
3072                 kaddr = kmap_atomic(page, KM_USER0);
3073                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3074                 kunmap_atomic(kaddr, KM_USER0);
3075                 kvm_release_page_dirty(page);
3076         }
3077 emul_write:
3078 #endif
3079
3080         return emulator_write_emulated(addr, new, bytes, vcpu);
3081 }
3082
3083 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3084 {
3085         return kvm_x86_ops->get_segment_base(vcpu, seg);
3086 }
3087
3088 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3089 {
3090         kvm_mmu_invlpg(vcpu, address);
3091         return X86EMUL_CONTINUE;
3092 }
3093
3094 int emulate_clts(struct kvm_vcpu *vcpu)
3095 {
3096         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
3097         return X86EMUL_CONTINUE;
3098 }
3099
3100 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3101 {
3102         struct kvm_vcpu *vcpu = ctxt->vcpu;
3103
3104         switch (dr) {
3105         case 0 ... 3:
3106                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3107                 return X86EMUL_CONTINUE;
3108         default:
3109                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3110                 return X86EMUL_UNHANDLEABLE;
3111         }
3112 }
3113
3114 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3115 {
3116         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3117         int exception;
3118
3119         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3120         if (exception) {
3121                 /* FIXME: better handling */
3122                 return X86EMUL_UNHANDLEABLE;
3123         }
3124         return X86EMUL_CONTINUE;
3125 }
3126
3127 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3128 {
3129         u8 opcodes[4];
3130         unsigned long rip = kvm_rip_read(vcpu);
3131         unsigned long rip_linear;
3132
3133         if (!printk_ratelimit())
3134                 return;
3135
3136         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3137
3138         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3139
3140         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3141                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3142 }
3143 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3144
3145 static struct x86_emulate_ops emulate_ops = {
3146         .read_std            = kvm_read_guest_virt,
3147         .read_emulated       = emulator_read_emulated,
3148         .write_emulated      = emulator_write_emulated,
3149         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3150 };
3151
3152 static void cache_all_regs(struct kvm_vcpu *vcpu)
3153 {
3154         kvm_register_read(vcpu, VCPU_REGS_RAX);
3155         kvm_register_read(vcpu, VCPU_REGS_RSP);
3156         kvm_register_read(vcpu, VCPU_REGS_RIP);
3157         vcpu->arch.regs_dirty = ~0;
3158 }
3159
3160 int emulate_instruction(struct kvm_vcpu *vcpu,
3161                         unsigned long cr2,
3162                         u16 error_code,
3163                         int emulation_type)
3164 {
3165         int r, shadow_mask;
3166         struct decode_cache *c;
3167         struct kvm_run *run = vcpu->run;
3168
3169         kvm_clear_exception_queue(vcpu);
3170         vcpu->arch.mmio_fault_cr2 = cr2;
3171         /*
3172          * TODO: fix emulate.c to use guest_read/write_register
3173          * instead of direct ->regs accesses, can save hundred cycles
3174          * on Intel for instructions that don't read/change RSP, for
3175          * for example.
3176          */
3177         cache_all_regs(vcpu);
3178
3179         vcpu->mmio_is_write = 0;
3180         vcpu->arch.pio.string = 0;
3181
3182         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3183                 int cs_db, cs_l;
3184                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3185
3186                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3187                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3188                 vcpu->arch.emulate_ctxt.mode =
3189                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3190                         ? X86EMUL_MODE_REAL : cs_l
3191                         ? X86EMUL_MODE_PROT64 : cs_db
3192                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3193
3194                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3195
3196                 /* Only allow emulation of specific instructions on #UD
3197                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3198                 c = &vcpu->arch.emulate_ctxt.decode;
3199                 if (emulation_type & EMULTYPE_TRAP_UD) {
3200                         if (!c->twobyte)
3201                                 return EMULATE_FAIL;
3202                         switch (c->b) {
3203                         case 0x01: /* VMMCALL */
3204                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3205                                         return EMULATE_FAIL;
3206                                 break;
3207                         case 0x34: /* sysenter */
3208                         case 0x35: /* sysexit */
3209                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3210                                         return EMULATE_FAIL;
3211                                 break;
3212                         case 0x05: /* syscall */
3213                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3214                                         return EMULATE_FAIL;
3215                                 break;
3216                         default:
3217                                 return EMULATE_FAIL;
3218                         }
3219
3220                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3221                                 return EMULATE_FAIL;
3222                 }
3223
3224                 ++vcpu->stat.insn_emulation;
3225                 if (r)  {
3226                         ++vcpu->stat.insn_emulation_fail;
3227                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3228                                 return EMULATE_DONE;
3229                         return EMULATE_FAIL;
3230                 }
3231         }
3232
3233         if (emulation_type & EMULTYPE_SKIP) {
3234                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3235                 return EMULATE_DONE;
3236         }
3237
3238         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3239         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3240
3241         if (r == 0)
3242                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3243
3244         if (vcpu->arch.pio.string)
3245                 return EMULATE_DO_MMIO;
3246
3247         if ((r || vcpu->mmio_is_write) && run) {
3248                 run->exit_reason = KVM_EXIT_MMIO;
3249                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3250                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3251                 run->mmio.len = vcpu->mmio_size;
3252                 run->mmio.is_write = vcpu->mmio_is_write;
3253         }
3254
3255         if (r) {
3256                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3257                         return EMULATE_DONE;
3258                 if (!vcpu->mmio_needed) {
3259                         kvm_report_emulation_failure(vcpu, "mmio");
3260                         return EMULATE_FAIL;
3261                 }
3262                 return EMULATE_DO_MMIO;
3263         }
3264
3265         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3266
3267         if (vcpu->mmio_is_write) {
3268                 vcpu->mmio_needed = 0;
3269                 return EMULATE_DO_MMIO;
3270         }
3271
3272         return EMULATE_DONE;
3273 }
3274 EXPORT_SYMBOL_GPL(emulate_instruction);
3275
3276 static int pio_copy_data(struct kvm_vcpu *vcpu)
3277 {
3278         void *p = vcpu->arch.pio_data;
3279         gva_t q = vcpu->arch.pio.guest_gva;
3280         unsigned bytes;
3281         int ret;
3282
3283         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3284         if (vcpu->arch.pio.in)
3285                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3286         else
3287                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3288         return ret;
3289 }
3290
3291 int complete_pio(struct kvm_vcpu *vcpu)
3292 {
3293         struct kvm_pio_request *io = &vcpu->arch.pio;
3294         long delta;
3295         int r;
3296         unsigned long val;
3297
3298         if (!io->string) {
3299                 if (io->in) {
3300                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3301                         memcpy(&val, vcpu->arch.pio_data, io->size);
3302                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3303                 }
3304         } else {
3305                 if (io->in) {
3306                         r = pio_copy_data(vcpu);
3307                         if (r)
3308                                 return r;
3309                 }
3310
3311                 delta = 1;
3312                 if (io->rep) {
3313                         delta *= io->cur_count;
3314                         /*
3315                          * The size of the register should really depend on
3316                          * current address size.
3317                          */
3318                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3319                         val -= delta;
3320                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3321                 }
3322                 if (io->down)
3323                         delta = -delta;
3324                 delta *= io->size;
3325                 if (io->in) {
3326                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3327                         val += delta;
3328                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3329                 } else {
3330                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3331                         val += delta;
3332                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3333                 }
3334         }
3335
3336         io->count -= io->cur_count;
3337         io->cur_count = 0;
3338
3339         return 0;
3340 }
3341
3342 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3343 {
3344         /* TODO: String I/O for in kernel device */
3345         int r;
3346
3347         if (vcpu->arch.pio.in)
3348                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3349                                     vcpu->arch.pio.size, pd);
3350         else
3351                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3352                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3353                                      pd);
3354         return r;
3355 }
3356
3357 static int pio_string_write(struct kvm_vcpu *vcpu)
3358 {
3359         struct kvm_pio_request *io = &vcpu->arch.pio;
3360         void *pd = vcpu->arch.pio_data;
3361         int i, r = 0;
3362
3363         for (i = 0; i < io->cur_count; i++) {
3364                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3365                                      io->port, io->size, pd)) {
3366                         r = -EOPNOTSUPP;
3367                         break;
3368                 }
3369                 pd += io->size;
3370         }
3371         return r;
3372 }
3373
3374 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3375 {
3376         unsigned long val;
3377
3378         vcpu->run->exit_reason = KVM_EXIT_IO;
3379         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3380         vcpu->run->io.size = vcpu->arch.pio.size = size;
3381         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3382         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3383         vcpu->run->io.port = vcpu->arch.pio.port = port;
3384         vcpu->arch.pio.in = in;
3385         vcpu->arch.pio.string = 0;
3386         vcpu->arch.pio.down = 0;
3387         vcpu->arch.pio.rep = 0;
3388
3389         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3390                       size, 1);
3391
3392         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3393         memcpy(vcpu->arch.pio_data, &val, 4);
3394
3395         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3396                 complete_pio(vcpu);
3397                 return 1;
3398         }
3399         return 0;
3400 }
3401 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3402
3403 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3404                   int size, unsigned long count, int down,
3405                   gva_t address, int rep, unsigned port)
3406 {
3407         unsigned now, in_page;
3408         int ret = 0;
3409
3410         vcpu->run->exit_reason = KVM_EXIT_IO;
3411         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3412         vcpu->run->io.size = vcpu->arch.pio.size = size;
3413         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3414         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3415         vcpu->run->io.port = vcpu->arch.pio.port = port;
3416         vcpu->arch.pio.in = in;
3417         vcpu->arch.pio.string = 1;
3418         vcpu->arch.pio.down = down;
3419         vcpu->arch.pio.rep = rep;
3420
3421         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3422                       size, count);
3423
3424         if (!count) {
3425                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3426                 return 1;
3427         }
3428
3429         if (!down)
3430                 in_page = PAGE_SIZE - offset_in_page(address);
3431         else
3432                 in_page = offset_in_page(address) + size;
3433         now = min(count, (unsigned long)in_page / size);
3434         if (!now)
3435                 now = 1;
3436         if (down) {
3437                 /*
3438                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3439                  */
3440                 pr_unimpl(vcpu, "guest string pio down\n");
3441                 kvm_inject_gp(vcpu, 0);
3442                 return 1;
3443         }
3444         vcpu->run->io.count = now;
3445         vcpu->arch.pio.cur_count = now;
3446
3447         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3448                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3449
3450         vcpu->arch.pio.guest_gva = address;
3451
3452         if (!vcpu->arch.pio.in) {
3453                 /* string PIO write */
3454                 ret = pio_copy_data(vcpu);
3455                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3456                         kvm_inject_gp(vcpu, 0);
3457                         return 1;
3458                 }
3459                 if (ret == 0 && !pio_string_write(vcpu)) {
3460                         complete_pio(vcpu);
3461                         if (vcpu->arch.pio.count == 0)
3462                                 ret = 1;
3463                 }
3464         }
3465         /* no string PIO read support yet */
3466
3467         return ret;
3468 }
3469 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3470
3471 static void bounce_off(void *info)
3472 {
3473         /* nothing */
3474 }
3475
3476 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3477                                      void *data)
3478 {
3479         struct cpufreq_freqs *freq = data;
3480         struct kvm *kvm;
3481         struct kvm_vcpu *vcpu;
3482         int i, send_ipi = 0;
3483
3484         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3485                 return 0;
3486         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3487                 return 0;
3488         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3489
3490         spin_lock(&kvm_lock);
3491         list_for_each_entry(kvm, &vm_list, vm_list) {
3492                 kvm_for_each_vcpu(i, vcpu, kvm) {
3493                         if (vcpu->cpu != freq->cpu)
3494                                 continue;
3495                         if (!kvm_request_guest_time_update(vcpu))
3496                                 continue;
3497                         if (vcpu->cpu != smp_processor_id())
3498                                 send_ipi++;
3499                 }
3500         }
3501         spin_unlock(&kvm_lock);
3502
3503         if (freq->old < freq->new && send_ipi) {
3504                 /*
3505                  * We upscale the frequency.  Must make the guest
3506                  * doesn't see old kvmclock values while running with
3507                  * the new frequency, otherwise we risk the guest sees
3508                  * time go backwards.
3509                  *
3510                  * In case we update the frequency for another cpu
3511                  * (which might be in guest context) send an interrupt
3512                  * to kick the cpu out of guest context.  Next time
3513                  * guest context is entered kvmclock will be updated,
3514                  * so the guest will not see stale values.
3515                  */
3516                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3517         }
3518         return 0;
3519 }
3520
3521 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3522         .notifier_call  = kvmclock_cpufreq_notifier
3523 };
3524
3525 static void kvm_timer_init(void)
3526 {
3527         int cpu;
3528
3529         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3530                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3531                                           CPUFREQ_TRANSITION_NOTIFIER);
3532                 for_each_online_cpu(cpu) {
3533                         unsigned long khz = cpufreq_get(cpu);
3534                         if (!khz)
3535                                 khz = tsc_khz;
3536                         per_cpu(cpu_tsc_khz, cpu) = khz;
3537                 }
3538         } else {
3539                 for_each_possible_cpu(cpu)
3540                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3541         }
3542 }
3543
3544 int kvm_arch_init(void *opaque)
3545 {
3546         int r;
3547         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3548
3549         if (kvm_x86_ops) {
3550                 printk(KERN_ERR "kvm: already loaded the other module\n");
3551                 r = -EEXIST;
3552                 goto out;
3553         }
3554
3555         if (!ops->cpu_has_kvm_support()) {
3556                 printk(KERN_ERR "kvm: no hardware support\n");
3557                 r = -EOPNOTSUPP;
3558                 goto out;
3559         }
3560         if (ops->disabled_by_bios()) {
3561                 printk(KERN_ERR "kvm: disabled by bios\n");
3562                 r = -EOPNOTSUPP;
3563                 goto out;
3564         }
3565
3566         r = kvm_mmu_module_init();
3567         if (r)
3568                 goto out;
3569
3570         kvm_init_msr_list();
3571
3572         kvm_x86_ops = ops;
3573         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3574         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3575         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3576                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3577
3578         kvm_timer_init();
3579
3580         return 0;
3581
3582 out:
3583         return r;
3584 }
3585
3586 void kvm_arch_exit(void)
3587 {
3588         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3589                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3590                                             CPUFREQ_TRANSITION_NOTIFIER);
3591         kvm_x86_ops = NULL;
3592         kvm_mmu_module_exit();
3593 }
3594
3595 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3596 {
3597         ++vcpu->stat.halt_exits;
3598         if (irqchip_in_kernel(vcpu->kvm)) {
3599                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3600                 return 1;
3601         } else {
3602                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3603                 return 0;
3604         }
3605 }
3606 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3607
3608 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3609                            unsigned long a1)
3610 {
3611         if (is_long_mode(vcpu))
3612                 return a0;
3613         else
3614                 return a0 | ((gpa_t)a1 << 32);
3615 }
3616
3617 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3618 {
3619         unsigned long nr, a0, a1, a2, a3, ret;
3620         int r = 1;
3621
3622         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3623         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3624         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3625         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3626         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3627
3628         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3629
3630         if (!is_long_mode(vcpu)) {
3631                 nr &= 0xFFFFFFFF;
3632                 a0 &= 0xFFFFFFFF;
3633                 a1 &= 0xFFFFFFFF;
3634                 a2 &= 0xFFFFFFFF;
3635                 a3 &= 0xFFFFFFFF;
3636         }
3637
3638         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3639                 ret = -KVM_EPERM;
3640                 goto out;
3641         }
3642
3643         switch (nr) {
3644         case KVM_HC_VAPIC_POLL_IRQ:
3645                 ret = 0;
3646                 break;
3647         case KVM_HC_MMU_OP:
3648                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3649                 break;
3650         default:
3651                 ret = -KVM_ENOSYS;
3652                 break;
3653         }
3654 out:
3655         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3656         ++vcpu->stat.hypercalls;
3657         return r;
3658 }
3659 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3660
3661 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3662 {
3663         char instruction[3];
3664         int ret = 0;
3665         unsigned long rip = kvm_rip_read(vcpu);
3666
3667
3668         /*
3669          * Blow out the MMU to ensure that no other VCPU has an active mapping
3670          * to ensure that the updated hypercall appears atomically across all
3671          * VCPUs.
3672          */
3673         kvm_mmu_zap_all(vcpu->kvm);
3674
3675         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3676         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3677             != X86EMUL_CONTINUE)
3678                 ret = -EFAULT;
3679
3680         return ret;
3681 }
3682
3683 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3684 {
3685         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3686 }
3687
3688 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3689 {
3690         struct descriptor_table dt = { limit, base };
3691
3692         kvm_x86_ops->set_gdt(vcpu, &dt);
3693 }
3694
3695 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3696 {
3697         struct descriptor_table dt = { limit, base };
3698
3699         kvm_x86_ops->set_idt(vcpu, &dt);
3700 }
3701
3702 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3703                    unsigned long *rflags)
3704 {
3705         kvm_lmsw(vcpu, msw);
3706         *rflags = kvm_get_rflags(vcpu);
3707 }
3708
3709 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3710 {
3711         unsigned long value;
3712
3713         switch (cr) {
3714         case 0:
3715                 value = vcpu->arch.cr0;
3716                 break;
3717         case 2:
3718                 value = vcpu->arch.cr2;
3719                 break;
3720         case 3:
3721                 value = vcpu->arch.cr3;
3722                 break;
3723         case 4:
3724                 value = kvm_read_cr4(vcpu);
3725                 break;
3726         case 8:
3727                 value = kvm_get_cr8(vcpu);
3728                 break;
3729         default:
3730                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3731                 return 0;
3732         }
3733
3734         return value;
3735 }
3736
3737 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3738                      unsigned long *rflags)
3739 {
3740         switch (cr) {
3741         case 0:
3742                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3743                 *rflags = kvm_get_rflags(vcpu);
3744                 break;
3745         case 2:
3746                 vcpu->arch.cr2 = val;
3747                 break;
3748         case 3:
3749                 kvm_set_cr3(vcpu, val);
3750                 break;
3751         case 4:
3752                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3753                 break;
3754         case 8:
3755                 kvm_set_cr8(vcpu, val & 0xfUL);
3756                 break;
3757         default:
3758                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3759         }
3760 }
3761
3762 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3763 {
3764         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3765         int j, nent = vcpu->arch.cpuid_nent;
3766
3767         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3768         /* when no next entry is found, the current entry[i] is reselected */
3769         for (j = i + 1; ; j = (j + 1) % nent) {
3770                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3771                 if (ej->function == e->function) {
3772                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3773                         return j;
3774                 }
3775         }
3776         return 0; /* silence gcc, even though control never reaches here */
3777 }
3778
3779 /* find an entry with matching function, matching index (if needed), and that
3780  * should be read next (if it's stateful) */
3781 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3782         u32 function, u32 index)
3783 {
3784         if (e->function != function)
3785                 return 0;
3786         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3787                 return 0;
3788         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3789             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3790                 return 0;
3791         return 1;
3792 }
3793
3794 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3795                                               u32 function, u32 index)
3796 {
3797         int i;
3798         struct kvm_cpuid_entry2 *best = NULL;
3799
3800         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3801                 struct kvm_cpuid_entry2 *e;
3802
3803                 e = &vcpu->arch.cpuid_entries[i];
3804                 if (is_matching_cpuid_entry(e, function, index)) {
3805                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3806                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3807                         best = e;
3808                         break;
3809                 }
3810                 /*
3811                  * Both basic or both extended?
3812                  */
3813                 if (((e->function ^ function) & 0x80000000) == 0)
3814                         if (!best || e->function > best->function)
3815                                 best = e;
3816         }
3817         return best;
3818 }
3819 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
3820
3821 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3822 {
3823         struct kvm_cpuid_entry2 *best;
3824
3825         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3826         if (best)
3827                 return best->eax & 0xff;
3828         return 36;
3829 }
3830
3831 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3832 {
3833         u32 function, index;
3834         struct kvm_cpuid_entry2 *best;
3835
3836         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3837         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3838         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3839         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3840         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3841         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3842         best = kvm_find_cpuid_entry(vcpu, function, index);
3843         if (best) {
3844                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3845                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3846                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3847                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3848         }
3849         kvm_x86_ops->skip_emulated_instruction(vcpu);
3850         trace_kvm_cpuid(function,
3851                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3852                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3853                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3854                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3855 }
3856 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3857
3858 /*
3859  * Check if userspace requested an interrupt window, and that the
3860  * interrupt window is open.
3861  *
3862  * No need to exit to userspace if we already have an interrupt queued.
3863  */
3864 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3865 {
3866         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3867                 vcpu->run->request_interrupt_window &&
3868                 kvm_arch_interrupt_allowed(vcpu));
3869 }
3870
3871 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3872 {
3873         struct kvm_run *kvm_run = vcpu->run;
3874
3875         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3876         kvm_run->cr8 = kvm_get_cr8(vcpu);
3877         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3878         if (irqchip_in_kernel(vcpu->kvm))
3879                 kvm_run->ready_for_interrupt_injection = 1;
3880         else
3881                 kvm_run->ready_for_interrupt_injection =
3882                         kvm_arch_interrupt_allowed(vcpu) &&
3883                         !kvm_cpu_has_interrupt(vcpu) &&
3884                         !kvm_event_needs_reinjection(vcpu);
3885 }
3886
3887 static void vapic_enter(struct kvm_vcpu *vcpu)
3888 {
3889         struct kvm_lapic *apic = vcpu->arch.apic;
3890         struct page *page;
3891
3892         if (!apic || !apic->vapic_addr)
3893                 return;
3894
3895         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3896
3897         vcpu->arch.apic->vapic_page = page;
3898 }
3899
3900 static void vapic_exit(struct kvm_vcpu *vcpu)
3901 {
3902         struct kvm_lapic *apic = vcpu->arch.apic;
3903         int idx;
3904
3905         if (!apic || !apic->vapic_addr)
3906                 return;
3907
3908         idx = srcu_read_lock(&vcpu->kvm->srcu);
3909         kvm_release_page_dirty(apic->vapic_page);
3910         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3911         srcu_read_unlock(&vcpu->kvm->srcu, idx);
3912 }
3913
3914 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3915 {
3916         int max_irr, tpr;
3917
3918         if (!kvm_x86_ops->update_cr8_intercept)
3919                 return;
3920
3921         if (!vcpu->arch.apic)
3922                 return;
3923
3924         if (!vcpu->arch.apic->vapic_addr)
3925                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3926         else
3927                 max_irr = -1;
3928
3929         if (max_irr != -1)
3930                 max_irr >>= 4;
3931
3932         tpr = kvm_lapic_get_cr8(vcpu);
3933
3934         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3935 }
3936
3937 static void inject_pending_event(struct kvm_vcpu *vcpu)
3938 {
3939         /* try to reinject previous events if any */
3940         if (vcpu->arch.exception.pending) {
3941                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3942                                           vcpu->arch.exception.has_error_code,
3943                                           vcpu->arch.exception.error_code);
3944                 return;
3945         }
3946
3947         if (vcpu->arch.nmi_injected) {
3948                 kvm_x86_ops->set_nmi(vcpu);
3949                 return;
3950         }
3951
3952         if (vcpu->arch.interrupt.pending) {
3953                 kvm_x86_ops->set_irq(vcpu);
3954                 return;
3955         }
3956
3957         /* try to inject new event if pending */
3958         if (vcpu->arch.nmi_pending) {
3959                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3960                         vcpu->arch.nmi_pending = false;
3961                         vcpu->arch.nmi_injected = true;
3962                         kvm_x86_ops->set_nmi(vcpu);
3963                 }
3964         } else if (kvm_cpu_has_interrupt(vcpu)) {
3965                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3966                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3967                                             false);
3968                         kvm_x86_ops->set_irq(vcpu);
3969                 }
3970         }
3971 }
3972
3973 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3974 {
3975         int r;
3976         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3977                 vcpu->run->request_interrupt_window;
3978
3979         if (vcpu->requests)
3980                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3981                         kvm_mmu_unload(vcpu);
3982
3983         r = kvm_mmu_reload(vcpu);
3984         if (unlikely(r))
3985                 goto out;
3986
3987         if (vcpu->requests) {
3988                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3989                         __kvm_migrate_timers(vcpu);
3990                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3991                         kvm_write_guest_time(vcpu);
3992                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3993                         kvm_mmu_sync_roots(vcpu);
3994                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3995                         kvm_x86_ops->tlb_flush(vcpu);
3996                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3997                                        &vcpu->requests)) {
3998                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3999                         r = 0;
4000                         goto out;
4001                 }
4002                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4003                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4004                         r = 0;
4005                         goto out;
4006                 }
4007         }
4008
4009         preempt_disable();
4010
4011         kvm_x86_ops->prepare_guest_switch(vcpu);
4012         kvm_load_guest_fpu(vcpu);
4013
4014         local_irq_disable();
4015
4016         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4017         smp_mb__after_clear_bit();
4018
4019         if (vcpu->requests || need_resched() || signal_pending(current)) {
4020                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4021                 local_irq_enable();
4022                 preempt_enable();
4023                 r = 1;
4024                 goto out;
4025         }
4026
4027         inject_pending_event(vcpu);
4028
4029         /* enable NMI/IRQ window open exits if needed */
4030         if (vcpu->arch.nmi_pending)
4031                 kvm_x86_ops->enable_nmi_window(vcpu);
4032         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4033                 kvm_x86_ops->enable_irq_window(vcpu);
4034
4035         if (kvm_lapic_enabled(vcpu)) {
4036                 update_cr8_intercept(vcpu);
4037                 kvm_lapic_sync_to_vapic(vcpu);
4038         }
4039
4040         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4041
4042         kvm_guest_enter();
4043
4044         if (unlikely(vcpu->arch.switch_db_regs)) {
4045                 set_debugreg(0, 7);
4046                 set_debugreg(vcpu->arch.eff_db[0], 0);
4047                 set_debugreg(vcpu->arch.eff_db[1], 1);
4048                 set_debugreg(vcpu->arch.eff_db[2], 2);
4049                 set_debugreg(vcpu->arch.eff_db[3], 3);
4050         }
4051
4052         trace_kvm_entry(vcpu->vcpu_id);
4053         kvm_x86_ops->run(vcpu);
4054
4055         /*
4056          * If the guest has used debug registers, at least dr7
4057          * will be disabled while returning to the host.
4058          * If we don't have active breakpoints in the host, we don't
4059          * care about the messed up debug address registers. But if
4060          * we have some of them active, restore the old state.
4061          */
4062         if (hw_breakpoint_active())
4063                 hw_breakpoint_restore();
4064
4065         set_bit(KVM_REQ_KICK, &vcpu->requests);
4066         local_irq_enable();
4067
4068         ++vcpu->stat.exits;
4069
4070         /*
4071          * We must have an instruction between local_irq_enable() and
4072          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4073          * the interrupt shadow.  The stat.exits increment will do nicely.
4074          * But we need to prevent reordering, hence this barrier():
4075          */
4076         barrier();
4077
4078         kvm_guest_exit();
4079
4080         preempt_enable();
4081
4082         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4083
4084         /*
4085          * Profile KVM exit RIPs:
4086          */
4087         if (unlikely(prof_on == KVM_PROFILING)) {
4088                 unsigned long rip = kvm_rip_read(vcpu);
4089                 profile_hit(KVM_PROFILING, (void *)rip);
4090         }
4091
4092
4093         kvm_lapic_sync_from_vapic(vcpu);
4094
4095         r = kvm_x86_ops->handle_exit(vcpu);
4096 out:
4097         return r;
4098 }
4099
4100
4101 static int __vcpu_run(struct kvm_vcpu *vcpu)
4102 {
4103         int r;
4104         struct kvm *kvm = vcpu->kvm;
4105
4106         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4107                 pr_debug("vcpu %d received sipi with vector # %x\n",
4108                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4109                 kvm_lapic_reset(vcpu);
4110                 r = kvm_arch_vcpu_reset(vcpu);
4111                 if (r)
4112                         return r;
4113                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4114         }
4115
4116         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4117         vapic_enter(vcpu);
4118
4119         r = 1;
4120         while (r > 0) {
4121                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4122                         r = vcpu_enter_guest(vcpu);
4123                 else {
4124                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4125                         kvm_vcpu_block(vcpu);
4126                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4127                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4128                         {
4129                                 switch(vcpu->arch.mp_state) {
4130                                 case KVM_MP_STATE_HALTED:
4131                                         vcpu->arch.mp_state =
4132                                                 KVM_MP_STATE_RUNNABLE;
4133                                 case KVM_MP_STATE_RUNNABLE:
4134                                         break;
4135                                 case KVM_MP_STATE_SIPI_RECEIVED:
4136                                 default:
4137                                         r = -EINTR;
4138                                         break;
4139                                 }
4140                         }
4141                 }
4142
4143                 if (r <= 0)
4144                         break;
4145
4146                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4147                 if (kvm_cpu_has_pending_timer(vcpu))
4148                         kvm_inject_pending_timer_irqs(vcpu);
4149
4150                 if (dm_request_for_irq_injection(vcpu)) {
4151                         r = -EINTR;
4152                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4153                         ++vcpu->stat.request_irq_exits;
4154                 }
4155                 if (signal_pending(current)) {
4156                         r = -EINTR;
4157                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4158                         ++vcpu->stat.signal_exits;
4159                 }
4160                 if (need_resched()) {
4161                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4162                         kvm_resched(vcpu);
4163                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4164                 }
4165         }
4166
4167         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4168         post_kvm_run_save(vcpu);
4169
4170         vapic_exit(vcpu);
4171
4172         return r;
4173 }
4174
4175 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4176 {
4177         int r;
4178         sigset_t sigsaved;
4179
4180         vcpu_load(vcpu);
4181
4182         if (vcpu->sigset_active)
4183                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4184
4185         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4186                 kvm_vcpu_block(vcpu);
4187                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4188                 r = -EAGAIN;
4189                 goto out;
4190         }
4191
4192         /* re-sync apic's tpr */
4193         if (!irqchip_in_kernel(vcpu->kvm))
4194                 kvm_set_cr8(vcpu, kvm_run->cr8);
4195
4196         if (vcpu->arch.pio.cur_count) {
4197                 r = complete_pio(vcpu);
4198                 if (r)
4199                         goto out;
4200         }
4201         if (vcpu->mmio_needed) {
4202                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4203                 vcpu->mmio_read_completed = 1;
4204                 vcpu->mmio_needed = 0;
4205
4206                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4207                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4208                                         EMULTYPE_NO_DECODE);
4209                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4210                 if (r == EMULATE_DO_MMIO) {
4211                         /*
4212                          * Read-modify-write.  Back to userspace.
4213                          */
4214                         r = 0;
4215                         goto out;
4216                 }
4217         }
4218         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4219                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4220                                      kvm_run->hypercall.ret);
4221
4222         r = __vcpu_run(vcpu);
4223
4224 out:
4225         if (vcpu->sigset_active)
4226                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4227
4228         vcpu_put(vcpu);
4229         return r;
4230 }
4231
4232 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4233 {
4234         vcpu_load(vcpu);
4235
4236         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4237         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4238         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4239         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4240         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4241         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4242         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4243         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4244 #ifdef CONFIG_X86_64
4245         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4246         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4247         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4248         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4249         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4250         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4251         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4252         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4253 #endif
4254
4255         regs->rip = kvm_rip_read(vcpu);
4256         regs->rflags = kvm_get_rflags(vcpu);
4257
4258         vcpu_put(vcpu);
4259
4260         return 0;
4261 }
4262
4263 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4264 {
4265         vcpu_load(vcpu);
4266
4267         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4268         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4269         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4270         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4271         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4272         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4273         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4274         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4275 #ifdef CONFIG_X86_64
4276         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4277         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4278         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4279         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4280         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4281         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4282         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4283         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4284 #endif
4285
4286         kvm_rip_write(vcpu, regs->rip);
4287         kvm_set_rflags(vcpu, regs->rflags);
4288
4289         vcpu->arch.exception.pending = false;
4290
4291         vcpu_put(vcpu);
4292
4293         return 0;
4294 }
4295
4296 void kvm_get_segment(struct kvm_vcpu *vcpu,
4297                      struct kvm_segment *var, int seg)
4298 {
4299         kvm_x86_ops->get_segment(vcpu, var, seg);
4300 }
4301
4302 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4303 {
4304         struct kvm_segment cs;
4305
4306         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4307         *db = cs.db;
4308         *l = cs.l;
4309 }
4310 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4311
4312 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4313                                   struct kvm_sregs *sregs)
4314 {
4315         struct descriptor_table dt;
4316
4317         vcpu_load(vcpu);
4318
4319         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4320         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4321         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4322         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4323         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4324         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4325
4326         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4327         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4328
4329         kvm_x86_ops->get_idt(vcpu, &dt);
4330         sregs->idt.limit = dt.limit;
4331         sregs->idt.base = dt.base;
4332         kvm_x86_ops->get_gdt(vcpu, &dt);
4333         sregs->gdt.limit = dt.limit;
4334         sregs->gdt.base = dt.base;
4335
4336         sregs->cr0 = vcpu->arch.cr0;
4337         sregs->cr2 = vcpu->arch.cr2;
4338         sregs->cr3 = vcpu->arch.cr3;
4339         sregs->cr4 = kvm_read_cr4(vcpu);
4340         sregs->cr8 = kvm_get_cr8(vcpu);
4341         sregs->efer = vcpu->arch.shadow_efer;
4342         sregs->apic_base = kvm_get_apic_base(vcpu);
4343
4344         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4345
4346         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4347                 set_bit(vcpu->arch.interrupt.nr,
4348                         (unsigned long *)sregs->interrupt_bitmap);
4349
4350         vcpu_put(vcpu);
4351
4352         return 0;
4353 }
4354
4355 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4356                                     struct kvm_mp_state *mp_state)
4357 {
4358         vcpu_load(vcpu);
4359         mp_state->mp_state = vcpu->arch.mp_state;
4360         vcpu_put(vcpu);
4361         return 0;
4362 }
4363
4364 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4365                                     struct kvm_mp_state *mp_state)
4366 {
4367         vcpu_load(vcpu);
4368         vcpu->arch.mp_state = mp_state->mp_state;
4369         vcpu_put(vcpu);
4370         return 0;
4371 }
4372
4373 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4374                         struct kvm_segment *var, int seg)
4375 {
4376         kvm_x86_ops->set_segment(vcpu, var, seg);
4377 }
4378
4379 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4380                                    struct kvm_segment *kvm_desct)
4381 {
4382         kvm_desct->base = get_desc_base(seg_desc);
4383         kvm_desct->limit = get_desc_limit(seg_desc);
4384         if (seg_desc->g) {
4385                 kvm_desct->limit <<= 12;
4386                 kvm_desct->limit |= 0xfff;
4387         }
4388         kvm_desct->selector = selector;
4389         kvm_desct->type = seg_desc->type;
4390         kvm_desct->present = seg_desc->p;
4391         kvm_desct->dpl = seg_desc->dpl;
4392         kvm_desct->db = seg_desc->d;
4393         kvm_desct->s = seg_desc->s;
4394         kvm_desct->l = seg_desc->l;
4395         kvm_desct->g = seg_desc->g;
4396         kvm_desct->avl = seg_desc->avl;
4397         if (!selector)
4398                 kvm_desct->unusable = 1;
4399         else
4400                 kvm_desct->unusable = 0;
4401         kvm_desct->padding = 0;
4402 }
4403
4404 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4405                                           u16 selector,
4406                                           struct descriptor_table *dtable)
4407 {
4408         if (selector & 1 << 2) {
4409                 struct kvm_segment kvm_seg;
4410
4411                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4412
4413                 if (kvm_seg.unusable)
4414                         dtable->limit = 0;
4415                 else
4416                         dtable->limit = kvm_seg.limit;
4417                 dtable->base = kvm_seg.base;
4418         }
4419         else
4420                 kvm_x86_ops->get_gdt(vcpu, dtable);
4421 }
4422
4423 /* allowed just for 8 bytes segments */
4424 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4425                                          struct desc_struct *seg_desc)
4426 {
4427         struct descriptor_table dtable;
4428         u16 index = selector >> 3;
4429
4430         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4431
4432         if (dtable.limit < index * 8 + 7) {
4433                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4434                 return 1;
4435         }
4436         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4437 }
4438
4439 /* allowed just for 8 bytes segments */
4440 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4441                                          struct desc_struct *seg_desc)
4442 {
4443         struct descriptor_table dtable;
4444         u16 index = selector >> 3;
4445
4446         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4447
4448         if (dtable.limit < index * 8 + 7)
4449                 return 1;
4450         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4451 }
4452
4453 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4454                              struct desc_struct *seg_desc)
4455 {
4456         u32 base_addr = get_desc_base(seg_desc);
4457
4458         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4459 }
4460
4461 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4462 {
4463         struct kvm_segment kvm_seg;
4464
4465         kvm_get_segment(vcpu, &kvm_seg, seg);
4466         return kvm_seg.selector;
4467 }
4468
4469 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4470                                                 u16 selector,
4471                                                 struct kvm_segment *kvm_seg)
4472 {
4473         struct desc_struct seg_desc;
4474
4475         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4476                 return 1;
4477         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4478         return 0;
4479 }
4480
4481 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4482 {
4483         struct kvm_segment segvar = {
4484                 .base = selector << 4,
4485                 .limit = 0xffff,
4486                 .selector = selector,
4487                 .type = 3,
4488                 .present = 1,
4489                 .dpl = 3,
4490                 .db = 0,
4491                 .s = 1,
4492                 .l = 0,
4493                 .g = 0,
4494                 .avl = 0,
4495                 .unusable = 0,
4496         };
4497         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4498         return 0;
4499 }
4500
4501 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4502 {
4503         return (seg != VCPU_SREG_LDTR) &&
4504                 (seg != VCPU_SREG_TR) &&
4505                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4506 }
4507
4508 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4509                                          u16 selector)
4510 {
4511         /* NULL selector is not valid for CS and SS */
4512         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4513                 if (!selector)
4514                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4515 }
4516
4517 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4518                                 int type_bits, int seg)
4519 {
4520         struct kvm_segment kvm_seg;
4521
4522         if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4523                 return kvm_load_realmode_segment(vcpu, selector, seg);
4524         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4525                 return 1;
4526
4527         kvm_check_segment_descriptor(vcpu, seg, selector);
4528         kvm_seg.type |= type_bits;
4529
4530         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4531             seg != VCPU_SREG_LDTR)
4532                 if (!kvm_seg.s)
4533                         kvm_seg.unusable = 1;
4534
4535         kvm_set_segment(vcpu, &kvm_seg, seg);
4536         return 0;
4537 }
4538
4539 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4540                                 struct tss_segment_32 *tss)
4541 {
4542         tss->cr3 = vcpu->arch.cr3;
4543         tss->eip = kvm_rip_read(vcpu);
4544         tss->eflags = kvm_get_rflags(vcpu);
4545         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4546         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4547         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4548         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4549         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4550         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4551         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4552         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4553         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4554         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4555         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4556         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4557         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4558         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4559         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4560 }
4561
4562 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4563                                   struct tss_segment_32 *tss)
4564 {
4565         kvm_set_cr3(vcpu, tss->cr3);
4566
4567         kvm_rip_write(vcpu, tss->eip);
4568         kvm_set_rflags(vcpu, tss->eflags | 2);
4569
4570         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4571         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4572         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4573         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4574         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4575         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4576         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4577         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4578
4579         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4580                 return 1;
4581
4582         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4583                 return 1;
4584
4585         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4586                 return 1;
4587
4588         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4589                 return 1;
4590
4591         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4592                 return 1;
4593
4594         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4595                 return 1;
4596
4597         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4598                 return 1;
4599         return 0;
4600 }
4601
4602 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4603                                 struct tss_segment_16 *tss)
4604 {
4605         tss->ip = kvm_rip_read(vcpu);
4606         tss->flag = kvm_get_rflags(vcpu);
4607         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4608         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4609         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4610         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4611         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4612         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4613         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4614         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4615
4616         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4617         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4618         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4619         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4620         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4621 }
4622
4623 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4624                                  struct tss_segment_16 *tss)
4625 {
4626         kvm_rip_write(vcpu, tss->ip);
4627         kvm_set_rflags(vcpu, tss->flag | 2);
4628         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4629         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4630         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4631         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4632         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4633         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4634         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4635         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4636
4637         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4638                 return 1;
4639
4640         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4641                 return 1;
4642
4643         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4644                 return 1;
4645
4646         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4647                 return 1;
4648
4649         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4650                 return 1;
4651         return 0;
4652 }
4653
4654 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4655                               u16 old_tss_sel, u32 old_tss_base,
4656                               struct desc_struct *nseg_desc)
4657 {
4658         struct tss_segment_16 tss_segment_16;
4659         int ret = 0;
4660
4661         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4662                            sizeof tss_segment_16))
4663                 goto out;
4664
4665         save_state_to_tss16(vcpu, &tss_segment_16);
4666
4667         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4668                             sizeof tss_segment_16))
4669                 goto out;
4670
4671         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4672                            &tss_segment_16, sizeof tss_segment_16))
4673                 goto out;
4674
4675         if (old_tss_sel != 0xffff) {
4676                 tss_segment_16.prev_task_link = old_tss_sel;
4677
4678                 if (kvm_write_guest(vcpu->kvm,
4679                                     get_tss_base_addr(vcpu, nseg_desc),
4680                                     &tss_segment_16.prev_task_link,
4681                                     sizeof tss_segment_16.prev_task_link))
4682                         goto out;
4683         }
4684
4685         if (load_state_from_tss16(vcpu, &tss_segment_16))
4686                 goto out;
4687
4688         ret = 1;
4689 out:
4690         return ret;
4691 }
4692
4693 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4694                        u16 old_tss_sel, u32 old_tss_base,
4695                        struct desc_struct *nseg_desc)
4696 {
4697         struct tss_segment_32 tss_segment_32;
4698         int ret = 0;
4699
4700         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4701                            sizeof tss_segment_32))
4702                 goto out;
4703
4704         save_state_to_tss32(vcpu, &tss_segment_32);
4705
4706         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4707                             sizeof tss_segment_32))
4708                 goto out;
4709
4710         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4711                            &tss_segment_32, sizeof tss_segment_32))
4712                 goto out;
4713
4714         if (old_tss_sel != 0xffff) {
4715                 tss_segment_32.prev_task_link = old_tss_sel;
4716
4717                 if (kvm_write_guest(vcpu->kvm,
4718                                     get_tss_base_addr(vcpu, nseg_desc),
4719                                     &tss_segment_32.prev_task_link,
4720                                     sizeof tss_segment_32.prev_task_link))
4721                         goto out;
4722         }
4723
4724         if (load_state_from_tss32(vcpu, &tss_segment_32))
4725                 goto out;
4726
4727         ret = 1;
4728 out:
4729         return ret;
4730 }
4731
4732 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4733 {
4734         struct kvm_segment tr_seg;
4735         struct desc_struct cseg_desc;
4736         struct desc_struct nseg_desc;
4737         int ret = 0;
4738         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4739         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4740
4741         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4742
4743         /* FIXME: Handle errors. Failure to read either TSS or their
4744          * descriptors should generate a pagefault.
4745          */
4746         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4747                 goto out;
4748
4749         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4750                 goto out;
4751
4752         if (reason != TASK_SWITCH_IRET) {
4753                 int cpl;
4754
4755                 cpl = kvm_x86_ops->get_cpl(vcpu);
4756                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4757                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4758                         return 1;
4759                 }
4760         }
4761
4762         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4763                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4764                 return 1;
4765         }
4766
4767         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4768                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4769                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4770         }
4771
4772         if (reason == TASK_SWITCH_IRET) {
4773                 u32 eflags = kvm_get_rflags(vcpu);
4774                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4775         }
4776
4777         /* set back link to prev task only if NT bit is set in eflags
4778            note that old_tss_sel is not used afetr this point */
4779         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4780                 old_tss_sel = 0xffff;
4781
4782         if (nseg_desc.type & 8)
4783                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4784                                          old_tss_base, &nseg_desc);
4785         else
4786                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4787                                          old_tss_base, &nseg_desc);
4788
4789         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4790                 u32 eflags = kvm_get_rflags(vcpu);
4791                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4792         }
4793
4794         if (reason != TASK_SWITCH_IRET) {
4795                 nseg_desc.type |= (1 << 1);
4796                 save_guest_segment_descriptor(vcpu, tss_selector,
4797                                               &nseg_desc);
4798         }
4799
4800         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4801         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4802         tr_seg.type = 11;
4803         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4804 out:
4805         return ret;
4806 }
4807 EXPORT_SYMBOL_GPL(kvm_task_switch);
4808
4809 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4810                                   struct kvm_sregs *sregs)
4811 {
4812         int mmu_reset_needed = 0;
4813         int pending_vec, max_bits;
4814         struct descriptor_table dt;
4815
4816         vcpu_load(vcpu);
4817
4818         dt.limit = sregs->idt.limit;
4819         dt.base = sregs->idt.base;
4820         kvm_x86_ops->set_idt(vcpu, &dt);
4821         dt.limit = sregs->gdt.limit;
4822         dt.base = sregs->gdt.base;
4823         kvm_x86_ops->set_gdt(vcpu, &dt);
4824
4825         vcpu->arch.cr2 = sregs->cr2;
4826         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4827         vcpu->arch.cr3 = sregs->cr3;
4828
4829         kvm_set_cr8(vcpu, sregs->cr8);
4830
4831         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4832         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4833         kvm_set_apic_base(vcpu, sregs->apic_base);
4834
4835         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4836         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4837         vcpu->arch.cr0 = sregs->cr0;
4838
4839         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4840         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4841         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4842                 load_pdptrs(vcpu, vcpu->arch.cr3);
4843                 mmu_reset_needed = 1;
4844         }
4845
4846         if (mmu_reset_needed)
4847                 kvm_mmu_reset_context(vcpu);
4848
4849         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4850         pending_vec = find_first_bit(
4851                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4852         if (pending_vec < max_bits) {
4853                 kvm_queue_interrupt(vcpu, pending_vec, false);
4854                 pr_debug("Set back pending irq %d\n", pending_vec);
4855                 if (irqchip_in_kernel(vcpu->kvm))
4856                         kvm_pic_clear_isr_ack(vcpu->kvm);
4857         }
4858
4859         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4860         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4861         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4862         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4863         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4864         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4865
4866         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4867         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4868
4869         update_cr8_intercept(vcpu);
4870
4871         /* Older userspace won't unhalt the vcpu on reset. */
4872         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4873             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4874             !(vcpu->arch.cr0 & X86_CR0_PE))
4875                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4876
4877         vcpu_put(vcpu);
4878
4879         return 0;
4880 }
4881
4882 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4883                                         struct kvm_guest_debug *dbg)
4884 {
4885         unsigned long rflags;
4886         int i, r;
4887
4888         vcpu_load(vcpu);
4889
4890         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4891                 r = -EBUSY;
4892                 if (vcpu->arch.exception.pending)
4893                         goto unlock_out;
4894                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4895                         kvm_queue_exception(vcpu, DB_VECTOR);
4896                 else
4897                         kvm_queue_exception(vcpu, BP_VECTOR);
4898         }
4899
4900         /*
4901          * Read rflags as long as potentially injected trace flags are still
4902          * filtered out.
4903          */
4904         rflags = kvm_get_rflags(vcpu);
4905
4906         vcpu->guest_debug = dbg->control;
4907         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4908                 vcpu->guest_debug = 0;
4909
4910         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4911                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4912                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4913                 vcpu->arch.switch_db_regs =
4914                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4915         } else {
4916                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4917                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4918                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4919         }
4920
4921         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4922                 vcpu->arch.singlestep_cs =
4923                         get_segment_selector(vcpu, VCPU_SREG_CS);
4924                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4925         }
4926
4927         /*
4928          * Trigger an rflags update that will inject or remove the trace
4929          * flags.
4930          */
4931         kvm_set_rflags(vcpu, rflags);
4932
4933         kvm_x86_ops->set_guest_debug(vcpu, dbg);
4934
4935         r = 0;
4936
4937 unlock_out:
4938         vcpu_put(vcpu);
4939
4940         return r;
4941 }
4942
4943 /*
4944  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4945  * we have asm/x86/processor.h
4946  */
4947 struct fxsave {
4948         u16     cwd;
4949         u16     swd;
4950         u16     twd;
4951         u16     fop;
4952         u64     rip;
4953         u64     rdp;
4954         u32     mxcsr;
4955         u32     mxcsr_mask;
4956         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4957 #ifdef CONFIG_X86_64
4958         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4959 #else
4960         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4961 #endif
4962 };
4963
4964 /*
4965  * Translate a guest virtual address to a guest physical address.
4966  */
4967 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4968                                     struct kvm_translation *tr)
4969 {
4970         unsigned long vaddr = tr->linear_address;
4971         gpa_t gpa;
4972         int idx;
4973
4974         vcpu_load(vcpu);
4975         idx = srcu_read_lock(&vcpu->kvm->srcu);
4976         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4977         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4978         tr->physical_address = gpa;
4979         tr->valid = gpa != UNMAPPED_GVA;
4980         tr->writeable = 1;
4981         tr->usermode = 0;
4982         vcpu_put(vcpu);
4983
4984         return 0;
4985 }
4986
4987 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4988 {
4989         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4990
4991         vcpu_load(vcpu);
4992
4993         memcpy(fpu->fpr, fxsave->st_space, 128);
4994         fpu->fcw = fxsave->cwd;
4995         fpu->fsw = fxsave->swd;
4996         fpu->ftwx = fxsave->twd;
4997         fpu->last_opcode = fxsave->fop;
4998         fpu->last_ip = fxsave->rip;
4999         fpu->last_dp = fxsave->rdp;
5000         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5001
5002         vcpu_put(vcpu);
5003
5004         return 0;
5005 }
5006
5007 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5008 {
5009         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5010
5011         vcpu_load(vcpu);
5012
5013         memcpy(fxsave->st_space, fpu->fpr, 128);
5014         fxsave->cwd = fpu->fcw;
5015         fxsave->swd = fpu->fsw;
5016         fxsave->twd = fpu->ftwx;
5017         fxsave->fop = fpu->last_opcode;
5018         fxsave->rip = fpu->last_ip;
5019         fxsave->rdp = fpu->last_dp;
5020         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5021
5022         vcpu_put(vcpu);
5023
5024         return 0;
5025 }
5026
5027 void fx_init(struct kvm_vcpu *vcpu)
5028 {
5029         unsigned after_mxcsr_mask;
5030
5031         /*
5032          * Touch the fpu the first time in non atomic context as if
5033          * this is the first fpu instruction the exception handler
5034          * will fire before the instruction returns and it'll have to
5035          * allocate ram with GFP_KERNEL.
5036          */
5037         if (!used_math())
5038                 kvm_fx_save(&vcpu->arch.host_fx_image);
5039
5040         /* Initialize guest FPU by resetting ours and saving into guest's */
5041         preempt_disable();
5042         kvm_fx_save(&vcpu->arch.host_fx_image);
5043         kvm_fx_finit();
5044         kvm_fx_save(&vcpu->arch.guest_fx_image);
5045         kvm_fx_restore(&vcpu->arch.host_fx_image);
5046         preempt_enable();
5047
5048         vcpu->arch.cr0 |= X86_CR0_ET;
5049         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5050         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5051         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5052                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5053 }
5054 EXPORT_SYMBOL_GPL(fx_init);
5055
5056 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5057 {
5058         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
5059                 return;
5060
5061         vcpu->guest_fpu_loaded = 1;
5062         kvm_fx_save(&vcpu->arch.host_fx_image);
5063         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5064 }
5065 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
5066
5067 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5068 {
5069         if (!vcpu->guest_fpu_loaded)
5070                 return;
5071
5072         vcpu->guest_fpu_loaded = 0;
5073         kvm_fx_save(&vcpu->arch.guest_fx_image);
5074         kvm_fx_restore(&vcpu->arch.host_fx_image);
5075         ++vcpu->stat.fpu_reload;
5076 }
5077 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
5078
5079 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5080 {
5081         if (vcpu->arch.time_page) {
5082                 kvm_release_page_dirty(vcpu->arch.time_page);
5083                 vcpu->arch.time_page = NULL;
5084         }
5085
5086         kvm_x86_ops->vcpu_free(vcpu);
5087 }
5088
5089 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5090                                                 unsigned int id)
5091 {
5092         return kvm_x86_ops->vcpu_create(kvm, id);
5093 }
5094
5095 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5096 {
5097         int r;
5098
5099         /* We do fxsave: this must be aligned. */
5100         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5101
5102         vcpu->arch.mtrr_state.have_fixed = 1;
5103         vcpu_load(vcpu);
5104         r = kvm_arch_vcpu_reset(vcpu);
5105         if (r == 0)
5106                 r = kvm_mmu_setup(vcpu);
5107         vcpu_put(vcpu);
5108         if (r < 0)
5109                 goto free_vcpu;
5110
5111         return 0;
5112 free_vcpu:
5113         kvm_x86_ops->vcpu_free(vcpu);
5114         return r;
5115 }
5116
5117 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5118 {
5119         vcpu_load(vcpu);
5120         kvm_mmu_unload(vcpu);
5121         vcpu_put(vcpu);
5122
5123         kvm_x86_ops->vcpu_free(vcpu);
5124 }
5125
5126 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5127 {
5128         vcpu->arch.nmi_pending = false;
5129         vcpu->arch.nmi_injected = false;
5130
5131         vcpu->arch.switch_db_regs = 0;
5132         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5133         vcpu->arch.dr6 = DR6_FIXED_1;
5134         vcpu->arch.dr7 = DR7_FIXED_1;
5135
5136         return kvm_x86_ops->vcpu_reset(vcpu);
5137 }
5138
5139 int kvm_arch_hardware_enable(void *garbage)
5140 {
5141         /*
5142          * Since this may be called from a hotplug notifcation,
5143          * we can't get the CPU frequency directly.
5144          */
5145         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5146                 int cpu = raw_smp_processor_id();
5147                 per_cpu(cpu_tsc_khz, cpu) = 0;
5148         }
5149
5150         kvm_shared_msr_cpu_online();
5151
5152         return kvm_x86_ops->hardware_enable(garbage);
5153 }
5154
5155 void kvm_arch_hardware_disable(void *garbage)
5156 {
5157         kvm_x86_ops->hardware_disable(garbage);
5158         drop_user_return_notifiers(garbage);
5159 }
5160
5161 int kvm_arch_hardware_setup(void)
5162 {
5163         return kvm_x86_ops->hardware_setup();
5164 }
5165
5166 void kvm_arch_hardware_unsetup(void)
5167 {
5168         kvm_x86_ops->hardware_unsetup();
5169 }
5170
5171 void kvm_arch_check_processor_compat(void *rtn)
5172 {
5173         kvm_x86_ops->check_processor_compatibility(rtn);
5174 }
5175
5176 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5177 {
5178         struct page *page;
5179         struct kvm *kvm;
5180         int r;
5181
5182         BUG_ON(vcpu->kvm == NULL);
5183         kvm = vcpu->kvm;
5184
5185         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5186         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5187                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5188         else
5189                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5190
5191         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5192         if (!page) {
5193                 r = -ENOMEM;
5194                 goto fail;
5195         }
5196         vcpu->arch.pio_data = page_address(page);
5197
5198         r = kvm_mmu_create(vcpu);
5199         if (r < 0)
5200                 goto fail_free_pio_data;
5201
5202         if (irqchip_in_kernel(kvm)) {
5203                 r = kvm_create_lapic(vcpu);
5204                 if (r < 0)
5205                         goto fail_mmu_destroy;
5206         }
5207
5208         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5209                                        GFP_KERNEL);
5210         if (!vcpu->arch.mce_banks) {
5211                 r = -ENOMEM;
5212                 goto fail_free_lapic;
5213         }
5214         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5215
5216         return 0;
5217 fail_free_lapic:
5218         kvm_free_lapic(vcpu);
5219 fail_mmu_destroy:
5220         kvm_mmu_destroy(vcpu);
5221 fail_free_pio_data:
5222         free_page((unsigned long)vcpu->arch.pio_data);
5223 fail:
5224         return r;
5225 }
5226
5227 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5228 {
5229         int idx;
5230
5231         kfree(vcpu->arch.mce_banks);
5232         kvm_free_lapic(vcpu);
5233         idx = srcu_read_lock(&vcpu->kvm->srcu);
5234         kvm_mmu_destroy(vcpu);
5235         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5236         free_page((unsigned long)vcpu->arch.pio_data);
5237 }
5238
5239 struct  kvm *kvm_arch_create_vm(void)
5240 {
5241         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5242
5243         if (!kvm)
5244                 return ERR_PTR(-ENOMEM);
5245
5246         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5247         if (!kvm->arch.aliases) {
5248                 kfree(kvm);
5249                 return ERR_PTR(-ENOMEM);
5250         }
5251
5252         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5253         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5254
5255         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5256         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5257
5258         rdtscll(kvm->arch.vm_init_tsc);
5259
5260         return kvm;
5261 }
5262
5263 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5264 {
5265         vcpu_load(vcpu);
5266         kvm_mmu_unload(vcpu);
5267         vcpu_put(vcpu);
5268 }
5269
5270 static void kvm_free_vcpus(struct kvm *kvm)
5271 {
5272         unsigned int i;
5273         struct kvm_vcpu *vcpu;
5274
5275         /*
5276          * Unpin any mmu pages first.
5277          */
5278         kvm_for_each_vcpu(i, vcpu, kvm)
5279                 kvm_unload_vcpu_mmu(vcpu);
5280         kvm_for_each_vcpu(i, vcpu, kvm)
5281                 kvm_arch_vcpu_free(vcpu);
5282
5283         mutex_lock(&kvm->lock);
5284         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5285                 kvm->vcpus[i] = NULL;
5286
5287         atomic_set(&kvm->online_vcpus, 0);
5288         mutex_unlock(&kvm->lock);
5289 }
5290
5291 void kvm_arch_sync_events(struct kvm *kvm)
5292 {
5293         kvm_free_all_assigned_devices(kvm);
5294 }
5295
5296 void kvm_arch_destroy_vm(struct kvm *kvm)
5297 {
5298         kvm_iommu_unmap_guest(kvm);
5299         kvm_free_pit(kvm);
5300         kfree(kvm->arch.vpic);
5301         kfree(kvm->arch.vioapic);
5302         kvm_free_vcpus(kvm);
5303         kvm_free_physmem(kvm);
5304         if (kvm->arch.apic_access_page)
5305                 put_page(kvm->arch.apic_access_page);
5306         if (kvm->arch.ept_identity_pagetable)
5307                 put_page(kvm->arch.ept_identity_pagetable);
5308         kfree(kvm->arch.aliases);
5309         kfree(kvm);
5310 }
5311
5312 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5313                                 struct kvm_memory_slot *memslot,
5314                                 struct kvm_memory_slot old,
5315                                 struct kvm_userspace_memory_region *mem,
5316                                 int user_alloc)
5317 {
5318         int npages = memslot->npages;
5319
5320         /*To keep backward compatibility with older userspace,
5321          *x86 needs to hanlde !user_alloc case.
5322          */
5323         if (!user_alloc) {
5324                 if (npages && !old.rmap) {
5325                         unsigned long userspace_addr;
5326
5327                         down_write(&current->mm->mmap_sem);
5328                         userspace_addr = do_mmap(NULL, 0,
5329                                                  npages * PAGE_SIZE,
5330                                                  PROT_READ | PROT_WRITE,
5331                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5332                                                  0);
5333                         up_write(&current->mm->mmap_sem);
5334
5335                         if (IS_ERR((void *)userspace_addr))
5336                                 return PTR_ERR((void *)userspace_addr);
5337
5338                         memslot->userspace_addr = userspace_addr;
5339                 }
5340         }
5341
5342
5343         return 0;
5344 }
5345
5346 void kvm_arch_commit_memory_region(struct kvm *kvm,
5347                                 struct kvm_userspace_memory_region *mem,
5348                                 struct kvm_memory_slot old,
5349                                 int user_alloc)
5350 {
5351
5352         int npages = mem->memory_size >> PAGE_SHIFT;
5353
5354         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5355                 int ret;
5356
5357                 down_write(&current->mm->mmap_sem);
5358                 ret = do_munmap(current->mm, old.userspace_addr,
5359                                 old.npages * PAGE_SIZE);
5360                 up_write(&current->mm->mmap_sem);
5361                 if (ret < 0)
5362                         printk(KERN_WARNING
5363                                "kvm_vm_ioctl_set_memory_region: "
5364                                "failed to munmap memory\n");
5365         }
5366
5367         spin_lock(&kvm->mmu_lock);
5368         if (!kvm->arch.n_requested_mmu_pages) {
5369                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5370                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5371         }
5372
5373         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5374         spin_unlock(&kvm->mmu_lock);
5375 }
5376
5377 void kvm_arch_flush_shadow(struct kvm *kvm)
5378 {
5379         kvm_mmu_zap_all(kvm);
5380         kvm_reload_remote_mmus(kvm);
5381 }
5382
5383 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5384 {
5385         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5386                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5387                 || vcpu->arch.nmi_pending ||
5388                 (kvm_arch_interrupt_allowed(vcpu) &&
5389                  kvm_cpu_has_interrupt(vcpu));
5390 }
5391
5392 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5393 {
5394         int me;
5395         int cpu = vcpu->cpu;
5396
5397         if (waitqueue_active(&vcpu->wq)) {
5398                 wake_up_interruptible(&vcpu->wq);
5399                 ++vcpu->stat.halt_wakeup;
5400         }
5401
5402         me = get_cpu();
5403         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5404                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5405                         smp_send_reschedule(cpu);
5406         put_cpu();
5407 }
5408
5409 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5410 {
5411         return kvm_x86_ops->interrupt_allowed(vcpu);
5412 }
5413
5414 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5415 {
5416         unsigned long rflags;
5417
5418         rflags = kvm_x86_ops->get_rflags(vcpu);
5419         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5420                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5421         return rflags;
5422 }
5423 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5424
5425 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5426 {
5427         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5428             vcpu->arch.singlestep_cs ==
5429                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5430             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5431                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5432         kvm_x86_ops->set_rflags(vcpu, rflags);
5433 }
5434 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5435
5436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);