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         cr0 |= X86_CR0_ET;
 432
 433         if (cr0 & CR0_RESERVED_BITS) {
 434                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 435                        cr0, kvm_read_cr0(vcpu));
 436                 kvm_inject_gp(vcpu, 0);
 437                 return;
 438         }
 439
 440         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 441                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 442                 kvm_inject_gp(vcpu, 0);
 443                 return;
 444         }
 445
 446         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 447                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 448                        "and a clear PE flag\n");
 449                 kvm_inject_gp(vcpu, 0);
 450                 return;
 451         }
 452
 453         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 454 #ifdef CONFIG_X86_64
 455                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 456                         int cs_db, cs_l;
 457
 458                         if (!is_pae(vcpu)) {
 459                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 460                                        "in long mode while PAE is disabled\n");
 461                                 kvm_inject_gp(vcpu, 0);
 462                                 return;
 463                         }
 464                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 465                         if (cs_l) {
 466                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 467                                        "in long mode while CS.L == 1\n");
 468                                 kvm_inject_gp(vcpu, 0);
 469                                 return;
 470
 471                         }
 472                 } else
 473 #endif
 474                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 475                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 476                                "reserved bits\n");
 477                         kvm_inject_gp(vcpu, 0);
 478                         return;
 479                 }
 480
 481         }
 482
 483         kvm_x86_ops->set_cr0(vcpu, cr0);
 484         vcpu->arch.cr0 = cr0;
 485
 486         kvm_mmu_reset_context(vcpu);
 487         return;
 488 }
 489 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 490
 491 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 492 {
 493         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 494 }
 495 EXPORT_SYMBOL_GPL(kvm_lmsw);
 496
 497 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 498 {
 499         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 500         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 501
 502         if (cr4 & CR4_RESERVED_BITS) {
 503                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 504                 kvm_inject_gp(vcpu, 0);
 505                 return;
 506         }
 507
 508         if (is_long_mode(vcpu)) {
 509                 if (!(cr4 & X86_CR4_PAE)) {
 510                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 511                                "in long mode\n");
 512                         kvm_inject_gp(vcpu, 0);
 513                         return;
 514                 }
 515         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 516                    && ((cr4 ^ old_cr4) & pdptr_bits)
 517                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 518                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 519                 kvm_inject_gp(vcpu, 0);
 520                 return;
 521         }
 522
 523         if (cr4 & X86_CR4_VMXE) {
 524                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 525                 kvm_inject_gp(vcpu, 0);
 526                 return;
 527         }
 528         kvm_x86_ops->set_cr4(vcpu, cr4);
 529         vcpu->arch.cr4 = cr4;
 530         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 531         kvm_mmu_reset_context(vcpu);
 532 }
 533 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 534
 535 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 536 {
 537         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 538                 kvm_mmu_sync_roots(vcpu);
 539                 kvm_mmu_flush_tlb(vcpu);
 540                 return;
 541         }
 542
 543         if (is_long_mode(vcpu)) {
 544                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 545                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 546                         kvm_inject_gp(vcpu, 0);
 547                         return;
 548                 }
 549         } else {
 550                 if (is_pae(vcpu)) {
 551                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 552                                 printk(KERN_DEBUG
 553                                        "set_cr3: #GP, reserved bits\n");
 554                                 kvm_inject_gp(vcpu, 0);
 555                                 return;
 556                         }
 557                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 558                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 559                                        "reserved bits\n");
 560                                 kvm_inject_gp(vcpu, 0);
 561                                 return;
 562                         }
 563                 }
 564                 /*
 565                  * We don't check reserved bits in nonpae mode, because
 566                  * this isn't enforced, and VMware depends on this.
 567                  */
 568         }
 569
 570         /*
 571          * Does the new cr3 value map to physical memory? (Note, we
 572          * catch an invalid cr3 even in real-mode, because it would
 573          * cause trouble later on when we turn on paging anyway.)
 574          *
 575          * A real CPU would silently accept an invalid cr3 and would
 576          * attempt to use it - with largely undefined (and often hard
 577          * to debug) behavior on the guest side.
 578          */
 579         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 580                 kvm_inject_gp(vcpu, 0);
 581         else {
 582                 vcpu->arch.cr3 = cr3;
 583                 vcpu->arch.mmu.new_cr3(vcpu);
 584         }
 585 }
 586 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 587
 588 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 589 {
 590         if (cr8 & CR8_RESERVED_BITS) {
 591                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 592                 kvm_inject_gp(vcpu, 0);
 593                 return;
 594         }
 595         if (irqchip_in_kernel(vcpu->kvm))
 596                 kvm_lapic_set_tpr(vcpu, cr8);
 597         else
 598                 vcpu->arch.cr8 = cr8;
 599 }
 600 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 601
 602 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 603 {
 604         if (irqchip_in_kernel(vcpu->kvm))
 605                 return kvm_lapic_get_cr8(vcpu);
 606         else
 607                 return vcpu->arch.cr8;
 608 }
 609 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 610
 611 static inline u32 bit(int bitno)
 612 {
 613         return 1 << (bitno & 31);
 614 }
 615
 616 /*
 617  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 618  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 619  *
 620  * This list is modified at module load time to reflect the
 621  * capabilities of the host cpu. This capabilities test skips MSRs that are
 622  * kvm-specific. Those are put in the beginning of the list.
 623  */
 624
 625 #define KVM_SAVE_MSRS_BEGIN     4
 626 static u32 msrs_to_save[] = {
 627         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 628         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 629         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 630         MSR_K6_STAR,
 631 #ifdef CONFIG_X86_64
 632         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 633 #endif
 634         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 635 };
 636
 637 static unsigned num_msrs_to_save;
 638
 639 static u32 emulated_msrs[] = {
 640         MSR_IA32_MISC_ENABLE,
 641 };
 642
 643 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 644 {
 645         if (efer & efer_reserved_bits) {
 646                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 647                        efer);
 648                 kvm_inject_gp(vcpu, 0);
 649                 return;
 650         }
 651
 652         if (is_paging(vcpu)
 653             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 654                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 655                 kvm_inject_gp(vcpu, 0);
 656                 return;
 657         }
 658
 659         if (efer & EFER_FFXSR) {
 660                 struct kvm_cpuid_entry2 *feat;
 661
 662                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 663                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 664                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 665                         kvm_inject_gp(vcpu, 0);
 666                         return;
 667                 }
 668         }
 669
 670         if (efer & EFER_SVME) {
 671                 struct kvm_cpuid_entry2 *feat;
 672
 673                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 674                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 675                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 676                         kvm_inject_gp(vcpu, 0);
 677                         return;
 678                 }
 679         }
 680
 681         kvm_x86_ops->set_efer(vcpu, efer);
 682
 683         efer &= ~EFER_LMA;
 684         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 685
 686         vcpu->arch.shadow_efer = efer;
 687
 688         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 689         kvm_mmu_reset_context(vcpu);
 690 }
 691
 692 void kvm_enable_efer_bits(u64 mask)
 693 {
 694        efer_reserved_bits &= ~mask;
 695 }
 696 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 697
 698
 699 /*
 700  * Writes msr value into into the appropriate "register".
 701  * Returns 0 on success, non-0 otherwise.
 702  * Assumes vcpu_load() was already called.
 703  */
 704 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 705 {
 706         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 707 }
 708
 709 /*
 710  * Adapt set_msr() to msr_io()'s calling convention
 711  */
 712 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 713 {
 714         return kvm_set_msr(vcpu, index, *data);
 715 }
 716
 717 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 718 {
 719         static int version;
 720         struct pvclock_wall_clock wc;
 721         struct timespec boot;
 722
 723         if (!wall_clock)
 724                 return;
 725
 726         version++;
 727
 728         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 729
 730         /*
 731          * The guest calculates current wall clock time by adding
 732          * system time (updated by kvm_write_guest_time below) to the
 733          * wall clock specified here.  guest system time equals host
 734          * system time for us, thus we must fill in host boot time here.
 735          */
 736         getboottime(&boot);
 737
 738         wc.sec = boot.tv_sec;
 739         wc.nsec = boot.tv_nsec;
 740         wc.version = version;
 741
 742         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 743
 744         version++;
 745         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 746 }
 747
 748 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 749 {
 750         uint32_t quotient, remainder;
 751
 752         /* Don't try to replace with do_div(), this one calculates
 753          * "(dividend << 32) / divisor" */
 754         __asm__ ( "divl %4"
 755                   : "=a" (quotient), "=d" (remainder)
 756                   : "0" (0), "1" (dividend), "r" (divisor) );
 757         return quotient;
 758 }
 759
 760 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 761 {
 762         uint64_t nsecs = 1000000000LL;
 763         int32_t  shift = 0;
 764         uint64_t tps64;
 765         uint32_t tps32;
 766
 767         tps64 = tsc_khz * 1000LL;
 768         while (tps64 > nsecs*2) {
 769                 tps64 >>= 1;
 770                 shift--;
 771         }
 772
 773         tps32 = (uint32_t)tps64;
 774         while (tps32 <= (uint32_t)nsecs) {
 775                 tps32 <<= 1;
 776                 shift++;
 777         }
 778
 779         hv_clock->tsc_shift = shift;
 780         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 781
 782         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 783                  __func__, tsc_khz, hv_clock->tsc_shift,
 784                  hv_clock->tsc_to_system_mul);
 785 }
 786
 787 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 788
 789 static void kvm_write_guest_time(struct kvm_vcpu *v)
 790 {
 791         struct timespec ts;
 792         unsigned long flags;
 793         struct kvm_vcpu_arch *vcpu = &v->arch;
 794         void *shared_kaddr;
 795         unsigned long this_tsc_khz;
 796
 797         if ((!vcpu->time_page))
 798                 return;
 799
 800         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 801         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 802                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 803                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 804         }
 805         put_cpu_var(cpu_tsc_khz);
 806
 807         /* Keep irq disabled to prevent changes to the clock */
 808         local_irq_save(flags);
 809         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 810         ktime_get_ts(&ts);
 811         monotonic_to_bootbased(&ts);
 812         local_irq_restore(flags);
 813
 814         /* With all the info we got, fill in the values */
 815
 816         vcpu->hv_clock.system_time = ts.tv_nsec +
 817                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 818
 819         /*
 820          * The interface expects us to write an even number signaling that the
 821          * update is finished. Since the guest won't see the intermediate
 822          * state, we just increase by 2 at the end.
 823          */
 824         vcpu->hv_clock.version += 2;
 825
 826         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 827
 828         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 829                sizeof(vcpu->hv_clock));
 830
 831         kunmap_atomic(shared_kaddr, KM_USER0);
 832
 833         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 834 }
 835
 836 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 837 {
 838         struct kvm_vcpu_arch *vcpu = &v->arch;
 839
 840         if (!vcpu->time_page)
 841                 return 0;
 842         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 843         return 1;
 844 }
 845
 846 static bool msr_mtrr_valid(unsigned msr)
 847 {
 848         switch (msr) {
 849         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 850         case MSR_MTRRfix64K_00000:
 851         case MSR_MTRRfix16K_80000:
 852         case MSR_MTRRfix16K_A0000:
 853         case MSR_MTRRfix4K_C0000:
 854         case MSR_MTRRfix4K_C8000:
 855         case MSR_MTRRfix4K_D0000:
 856         case MSR_MTRRfix4K_D8000:
 857         case MSR_MTRRfix4K_E0000:
 858         case MSR_MTRRfix4K_E8000:
 859         case MSR_MTRRfix4K_F0000:
 860         case MSR_MTRRfix4K_F8000:
 861         case MSR_MTRRdefType:
 862         case MSR_IA32_CR_PAT:
 863                 return true;
 864         case 0x2f8:
 865                 return true;
 866         }
 867         return false;
 868 }
 869
 870 static bool valid_pat_type(unsigned t)
 871 {
 872         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 873 }
 874
 875 static bool valid_mtrr_type(unsigned t)
 876 {
 877         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 878 }
 879
 880 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 881 {
 882         int i;
 883
 884         if (!msr_mtrr_valid(msr))
 885                 return false;
 886
 887         if (msr == MSR_IA32_CR_PAT) {
 888                 for (i = 0; i < 8; i++)
 889                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 890                                 return false;
 891                 return true;
 892         } else if (msr == MSR_MTRRdefType) {
 893                 if (data & ~0xcff)
 894                         return false;
 895                 return valid_mtrr_type(data & 0xff);
 896         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 897                 for (i = 0; i < 8 ; i++)
 898                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 899                                 return false;
 900                 return true;
 901         }
 902
 903         /* variable MTRRs */
 904         return valid_mtrr_type(data & 0xff);
 905 }
 906
 907 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 908 {
 909         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 910
 911         if (!mtrr_valid(vcpu, msr, data))
 912                 return 1;
 913
 914         if (msr == MSR_MTRRdefType) {
 915                 vcpu->arch.mtrr_state.def_type = data;
 916                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 917         } else if (msr == MSR_MTRRfix64K_00000)
 918                 p[0] = data;
 919         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 920                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 921         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 922                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 923         else if (msr == MSR_IA32_CR_PAT)
 924                 vcpu->arch.pat = data;
 925         else {  /* Variable MTRRs */
 926                 int idx, is_mtrr_mask;
 927                 u64 *pt;
 928
 929                 idx = (msr - 0x200) / 2;
 930                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 931                 if (!is_mtrr_mask)
 932                         pt =
 933                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 934                 else
 935                         pt =
 936                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 937                 *pt = data;
 938         }
 939
 940         kvm_mmu_reset_context(vcpu);
 941         return 0;
 942 }
 943
 944 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 945 {
 946         u64 mcg_cap = vcpu->arch.mcg_cap;
 947         unsigned bank_num = mcg_cap & 0xff;
 948
 949         switch (msr) {
 950         case MSR_IA32_MCG_STATUS:
 951                 vcpu->arch.mcg_status = data;
 952                 break;
 953         case MSR_IA32_MCG_CTL:
 954                 if (!(mcg_cap & MCG_CTL_P))
 955                         return 1;
 956                 if (data != 0 && data != ~(u64)0)
 957                         return -1;
 958                 vcpu->arch.mcg_ctl = data;
 959                 break;
 960         default:
 961                 if (msr >= MSR_IA32_MC0_CTL &&
 962                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 963                         u32 offset = msr - MSR_IA32_MC0_CTL;
 964                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 965                         if ((offset & 0x3) == 0 &&
 966                             data != 0 && data != ~(u64)0)
 967                                 return -1;
 968                         vcpu->arch.mce_banks[offset] = data;
 969                         break;
 970                 }
 971                 return 1;
 972         }
 973         return 0;
 974 }
 975
 976 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 977 {
 978         struct kvm *kvm = vcpu->kvm;
 979         int lm = is_long_mode(vcpu);
 980         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 981                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 982         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 983                 : kvm->arch.xen_hvm_config.blob_size_32;
 984         u32 page_num = data & ~PAGE_MASK;
 985         u64 page_addr = data & PAGE_MASK;
 986         u8 *page;
 987         int r;
 988
 989         r = -E2BIG;
 990         if (page_num >= blob_size)
 991                 goto out;
 992         r = -ENOMEM;
 993         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 994         if (!page)
 995                 goto out;
 996         r = -EFAULT;
 997         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 998                 goto out_free;
 999         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1000                 goto out_free;
1001         r = 0;
1002 out_free:
1003         kfree(page);
1004 out:
1005         return r;
1006 }
1007
1008 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1009 {
1010         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1011 }
1012
1013 static bool kvm_hv_msr_partition_wide(u32 msr)
1014 {
1015         bool r = false;
1016         switch (msr) {
1017         case HV_X64_MSR_GUEST_OS_ID:
1018         case HV_X64_MSR_HYPERCALL:
1019                 r = true;
1020                 break;
1021         }
1022
1023         return r;
1024 }
1025
1026 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1027 {
1028         struct kvm *kvm = vcpu->kvm;
1029
1030         switch (msr) {
1031         case HV_X64_MSR_GUEST_OS_ID:
1032                 kvm->arch.hv_guest_os_id = data;
1033                 /* setting guest os id to zero disables hypercall page */
1034                 if (!kvm->arch.hv_guest_os_id)
1035                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1036                 break;
1037         case HV_X64_MSR_HYPERCALL: {
1038                 u64 gfn;
1039                 unsigned long addr;
1040                 u8 instructions[4];
1041
1042                 /* if guest os id is not set hypercall should remain disabled */
1043                 if (!kvm->arch.hv_guest_os_id)
1044                         break;
1045                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1046                         kvm->arch.hv_hypercall = data;
1047                         break;
1048                 }
1049                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1050                 addr = gfn_to_hva(kvm, gfn);
1051                 if (kvm_is_error_hva(addr))
1052                         return 1;
1053                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1054                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1055                 if (copy_to_user((void __user *)addr, instructions, 4))
1056                         return 1;
1057                 kvm->arch.hv_hypercall = data;
1058                 break;
1059         }
1060         default:
1061                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1062                           "data 0x%llx\n", msr, data);
1063                 return 1;
1064         }
1065         return 0;
1066 }
1067
1068 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1069 {
1070         pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x data 0x%llx\n",
1071                   msr, data);
1072
1073         return 1;
1074 }
1075
1076 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1077 {
1078         switch (msr) {
1079         case MSR_EFER:
1080                 set_efer(vcpu, data);
1081                 break;
1082         case MSR_K7_HWCR:
1083                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1084                 if (data != 0) {
1085                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1086                                 data);
1087                         return 1;
1088                 }
1089                 break;
1090         case MSR_FAM10H_MMIO_CONF_BASE:
1091                 if (data != 0) {
1092                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1093                                 "0x%llx\n", data);
1094                         return 1;
1095                 }
1096                 break;
1097         case MSR_AMD64_NB_CFG:
1098                 break;
1099         case MSR_IA32_DEBUGCTLMSR:
1100                 if (!data) {
1101                         /* We support the non-activated case already */
1102                         break;
1103                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1104                         /* Values other than LBR and BTF are vendor-specific,
1105                            thus reserved and should throw a #GP */
1106                         return 1;
1107                 }
1108                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1109                         __func__, data);
1110                 break;
1111         case MSR_IA32_UCODE_REV:
1112         case MSR_IA32_UCODE_WRITE:
1113         case MSR_VM_HSAVE_PA:
1114         case MSR_AMD64_PATCH_LOADER:
1115                 break;
1116         case 0x200 ... 0x2ff:
1117                 return set_msr_mtrr(vcpu, msr, data);
1118         case MSR_IA32_APICBASE:
1119                 kvm_set_apic_base(vcpu, data);
1120                 break;
1121         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1122                 return kvm_x2apic_msr_write(vcpu, msr, data);
1123         case MSR_IA32_MISC_ENABLE:
1124                 vcpu->arch.ia32_misc_enable_msr = data;
1125                 break;
1126         case MSR_KVM_WALL_CLOCK:
1127                 vcpu->kvm->arch.wall_clock = data;
1128                 kvm_write_wall_clock(vcpu->kvm, data);
1129                 break;
1130         case MSR_KVM_SYSTEM_TIME: {
1131                 if (vcpu->arch.time_page) {
1132                         kvm_release_page_dirty(vcpu->arch.time_page);
1133                         vcpu->arch.time_page = NULL;
1134                 }
1135
1136                 vcpu->arch.time = data;
1137
1138                 /* we verify if the enable bit is set... */
1139                 if (!(data & 1))
1140                         break;
1141
1142                 /* ...but clean it before doing the actual write */
1143                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1144
1145                 vcpu->arch.time_page =
1146                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1147
1148                 if (is_error_page(vcpu->arch.time_page)) {
1149                         kvm_release_page_clean(vcpu->arch.time_page);
1150                         vcpu->arch.time_page = NULL;
1151                 }
1152
1153                 kvm_request_guest_time_update(vcpu);
1154                 break;
1155         }
1156         case MSR_IA32_MCG_CTL:
1157         case MSR_IA32_MCG_STATUS:
1158         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1159                 return set_msr_mce(vcpu, msr, data);
1160
1161         /* Performance counters are not protected by a CPUID bit,
1162          * so we should check all of them in the generic path for the sake of
1163          * cross vendor migration.
1164          * Writing a zero into the event select MSRs disables them,
1165          * which we perfectly emulate ;-). Any other value should be at least
1166          * reported, some guests depend on them.
1167          */
1168         case MSR_P6_EVNTSEL0:
1169         case MSR_P6_EVNTSEL1:
1170         case MSR_K7_EVNTSEL0:
1171         case MSR_K7_EVNTSEL1:
1172         case MSR_K7_EVNTSEL2:
1173         case MSR_K7_EVNTSEL3:
1174                 if (data != 0)
1175                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1176                                 "0x%x data 0x%llx\n", msr, data);
1177                 break;
1178         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1179          * so we ignore writes to make it happy.
1180          */
1181         case MSR_P6_PERFCTR0:
1182         case MSR_P6_PERFCTR1:
1183         case MSR_K7_PERFCTR0:
1184         case MSR_K7_PERFCTR1:
1185         case MSR_K7_PERFCTR2:
1186         case MSR_K7_PERFCTR3:
1187                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1188                         "0x%x data 0x%llx\n", msr, data);
1189                 break;
1190         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1191                 if (kvm_hv_msr_partition_wide(msr)) {
1192                         int r;
1193                         mutex_lock(&vcpu->kvm->lock);
1194                         r = set_msr_hyperv_pw(vcpu, msr, data);
1195                         mutex_unlock(&vcpu->kvm->lock);
1196                         return r;
1197                 } else
1198                         return set_msr_hyperv(vcpu, msr, data);
1199                 break;
1200         default:
1201                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1202                         return xen_hvm_config(vcpu, data);
1203                 if (!ignore_msrs) {
1204                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1205                                 msr, data);
1206                         return 1;
1207                 } else {
1208                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1209                                 msr, data);
1210                         break;
1211                 }
1212         }
1213         return 0;
1214 }
1215 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1216
1217
1218 /*
1219  * Reads an msr value (of 'msr_index') into 'pdata'.
1220  * Returns 0 on success, non-0 otherwise.
1221  * Assumes vcpu_load() was already called.
1222  */
1223 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1224 {
1225         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1226 }
1227
1228 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1229 {
1230         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1231
1232         if (!msr_mtrr_valid(msr))
1233                 return 1;
1234
1235         if (msr == MSR_MTRRdefType)
1236                 *pdata = vcpu->arch.mtrr_state.def_type +
1237                          (vcpu->arch.mtrr_state.enabled << 10);
1238         else if (msr == MSR_MTRRfix64K_00000)
1239                 *pdata = p[0];
1240         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1241                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1242         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1243                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1244         else if (msr == MSR_IA32_CR_PAT)
1245                 *pdata = vcpu->arch.pat;
1246         else {  /* Variable MTRRs */
1247                 int idx, is_mtrr_mask;
1248                 u64 *pt;
1249
1250                 idx = (msr - 0x200) / 2;
1251                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1252                 if (!is_mtrr_mask)
1253                         pt =
1254                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1255                 else
1256                         pt =
1257                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1258                 *pdata = *pt;
1259         }
1260
1261         return 0;
1262 }
1263
1264 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1265 {
1266         u64 data;
1267         u64 mcg_cap = vcpu->arch.mcg_cap;
1268         unsigned bank_num = mcg_cap & 0xff;
1269
1270         switch (msr) {
1271         case MSR_IA32_P5_MC_ADDR:
1272         case MSR_IA32_P5_MC_TYPE:
1273                 data = 0;
1274                 break;
1275         case MSR_IA32_MCG_CAP:
1276                 data = vcpu->arch.mcg_cap;
1277                 break;
1278         case MSR_IA32_MCG_CTL:
1279                 if (!(mcg_cap & MCG_CTL_P))
1280                         return 1;
1281                 data = vcpu->arch.mcg_ctl;
1282                 break;
1283         case MSR_IA32_MCG_STATUS:
1284                 data = vcpu->arch.mcg_status;
1285                 break;
1286         default:
1287                 if (msr >= MSR_IA32_MC0_CTL &&
1288                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1289                         u32 offset = msr - MSR_IA32_MC0_CTL;
1290                         data = vcpu->arch.mce_banks[offset];
1291                         break;
1292                 }
1293                 return 1;
1294         }
1295         *pdata = data;
1296         return 0;
1297 }
1298
1299 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1300 {
1301         u64 data = 0;
1302         struct kvm *kvm = vcpu->kvm;
1303
1304         switch (msr) {
1305         case HV_X64_MSR_GUEST_OS_ID:
1306                 data = kvm->arch.hv_guest_os_id;
1307                 break;
1308         case HV_X64_MSR_HYPERCALL:
1309                 data = kvm->arch.hv_hypercall;
1310                 break;
1311         default:
1312                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1313                 return 1;
1314         }
1315
1316         *pdata = data;
1317         return 0;
1318 }
1319
1320 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1321 {
1322         u64 data = 0;
1323
1324         switch (msr) {
1325         case HV_X64_MSR_VP_INDEX: {
1326                 int r;
1327                 struct kvm_vcpu *v;
1328                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1329                         if (v == vcpu)
1330                                 data = r;
1331                 break;
1332         }
1333         default:
1334                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1335                 return 1;
1336         }
1337         *pdata = data;
1338         return 0;
1339 }
1340
1341 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1342 {
1343         u64 data;
1344
1345         switch (msr) {
1346         case MSR_IA32_PLATFORM_ID:
1347         case MSR_IA32_UCODE_REV:
1348         case MSR_IA32_EBL_CR_POWERON:
1349         case MSR_IA32_DEBUGCTLMSR:
1350         case MSR_IA32_LASTBRANCHFROMIP:
1351         case MSR_IA32_LASTBRANCHTOIP:
1352         case MSR_IA32_LASTINTFROMIP:
1353         case MSR_IA32_LASTINTTOIP:
1354         case MSR_K8_SYSCFG:
1355         case MSR_K7_HWCR:
1356         case MSR_VM_HSAVE_PA:
1357         case MSR_P6_PERFCTR0:
1358         case MSR_P6_PERFCTR1:
1359         case MSR_P6_EVNTSEL0:
1360         case MSR_P6_EVNTSEL1:
1361         case MSR_K7_EVNTSEL0:
1362         case MSR_K7_PERFCTR0:
1363         case MSR_K8_INT_PENDING_MSG:
1364         case MSR_AMD64_NB_CFG:
1365         case MSR_FAM10H_MMIO_CONF_BASE:
1366                 data = 0;
1367                 break;
1368         case MSR_MTRRcap:
1369                 data = 0x500 | KVM_NR_VAR_MTRR;
1370                 break;
1371         case 0x200 ... 0x2ff:
1372                 return get_msr_mtrr(vcpu, msr, pdata);
1373         case 0xcd: /* fsb frequency */
1374                 data = 3;
1375                 break;
1376         case MSR_IA32_APICBASE:
1377                 data = kvm_get_apic_base(vcpu);
1378                 break;
1379         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1380                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1381                 break;
1382         case MSR_IA32_MISC_ENABLE:
1383                 data = vcpu->arch.ia32_misc_enable_msr;
1384                 break;
1385         case MSR_IA32_PERF_STATUS:
1386                 /* TSC increment by tick */
1387                 data = 1000ULL;
1388                 /* CPU multiplier */
1389                 data |= (((uint64_t)4ULL) << 40);
1390                 break;
1391         case MSR_EFER:
1392                 data = vcpu->arch.shadow_efer;
1393                 break;
1394         case MSR_KVM_WALL_CLOCK:
1395                 data = vcpu->kvm->arch.wall_clock;
1396                 break;
1397         case MSR_KVM_SYSTEM_TIME:
1398                 data = vcpu->arch.time;
1399                 break;
1400         case MSR_IA32_P5_MC_ADDR:
1401         case MSR_IA32_P5_MC_TYPE:
1402         case MSR_IA32_MCG_CAP:
1403         case MSR_IA32_MCG_CTL:
1404         case MSR_IA32_MCG_STATUS:
1405         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1406                 return get_msr_mce(vcpu, msr, pdata);
1407         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1408                 if (kvm_hv_msr_partition_wide(msr)) {
1409                         int r;
1410                         mutex_lock(&vcpu->kvm->lock);
1411                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1412                         mutex_unlock(&vcpu->kvm->lock);
1413                         return r;
1414                 } else
1415                         return get_msr_hyperv(vcpu, msr, pdata);
1416                 break;
1417         default:
1418                 if (!ignore_msrs) {
1419                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1420                         return 1;
1421                 } else {
1422                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1423                         data = 0;
1424                 }
1425                 break;
1426         }
1427         *pdata = data;
1428         return 0;
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1431
1432 /*
1433  * Read or write a bunch of msrs. All parameters are kernel addresses.
1434  *
1435  * @return number of msrs set successfully.
1436  */
1437 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1438                     struct kvm_msr_entry *entries,
1439                     int (*do_msr)(struct kvm_vcpu *vcpu,
1440                                   unsigned index, u64 *data))
1441 {
1442         int i, idx;
1443
1444         vcpu_load(vcpu);
1445
1446         idx = srcu_read_lock(&vcpu->kvm->srcu);
1447         for (i = 0; i < msrs->nmsrs; ++i)
1448                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1449                         break;
1450         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1451
1452         vcpu_put(vcpu);
1453
1454         return i;
1455 }
1456
1457 /*
1458  * Read or write a bunch of msrs. Parameters are user addresses.
1459  *
1460  * @return number of msrs set successfully.
1461  */
1462 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1463                   int (*do_msr)(struct kvm_vcpu *vcpu,
1464                                 unsigned index, u64 *data),
1465                   int writeback)
1466 {
1467         struct kvm_msrs msrs;
1468         struct kvm_msr_entry *entries;
1469         int r, n;
1470         unsigned size;
1471
1472         r = -EFAULT;
1473         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1474                 goto out;
1475
1476         r = -E2BIG;
1477         if (msrs.nmsrs >= MAX_IO_MSRS)
1478                 goto out;
1479
1480         r = -ENOMEM;
1481         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1482         entries = vmalloc(size);
1483         if (!entries)
1484                 goto out;
1485
1486         r = -EFAULT;
1487         if (copy_from_user(entries, user_msrs->entries, size))
1488                 goto out_free;
1489
1490         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1491         if (r < 0)
1492                 goto out_free;
1493
1494         r = -EFAULT;
1495         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1496                 goto out_free;
1497
1498         r = n;
1499
1500 out_free:
1501         vfree(entries);
1502 out:
1503         return r;
1504 }
1505
1506 int kvm_dev_ioctl_check_extension(long ext)
1507 {
1508         int r;
1509
1510         switch (ext) {
1511         case KVM_CAP_IRQCHIP:
1512         case KVM_CAP_HLT:
1513         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1514         case KVM_CAP_SET_TSS_ADDR:
1515         case KVM_CAP_EXT_CPUID:
1516         case KVM_CAP_CLOCKSOURCE:
1517         case KVM_CAP_PIT:
1518         case KVM_CAP_NOP_IO_DELAY:
1519         case KVM_CAP_MP_STATE:
1520         case KVM_CAP_SYNC_MMU:
1521         case KVM_CAP_REINJECT_CONTROL:
1522         case KVM_CAP_IRQ_INJECT_STATUS:
1523         case KVM_CAP_ASSIGN_DEV_IRQ:
1524         case KVM_CAP_IRQFD:
1525         case KVM_CAP_IOEVENTFD:
1526         case KVM_CAP_PIT2:
1527         case KVM_CAP_PIT_STATE2:
1528         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1529         case KVM_CAP_XEN_HVM:
1530         case KVM_CAP_ADJUST_CLOCK:
1531         case KVM_CAP_VCPU_EVENTS:
1532         case KVM_CAP_HYPERV:
1533                 r = 1;
1534                 break;
1535         case KVM_CAP_COALESCED_MMIO:
1536                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1537                 break;
1538         case KVM_CAP_VAPIC:
1539                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1540                 break;
1541         case KVM_CAP_NR_VCPUS:
1542                 r = KVM_MAX_VCPUS;
1543                 break;
1544         case KVM_CAP_NR_MEMSLOTS:
1545                 r = KVM_MEMORY_SLOTS;
1546                 break;
1547         case KVM_CAP_PV_MMU:    /* obsolete */
1548                 r = 0;
1549                 break;
1550         case KVM_CAP_IOMMU:
1551                 r = iommu_found();
1552                 break;
1553         case KVM_CAP_MCE:
1554                 r = KVM_MAX_MCE_BANKS;
1555                 break;
1556         default:
1557                 r = 0;
1558                 break;
1559         }
1560         return r;
1561
1562 }
1563
1564 long kvm_arch_dev_ioctl(struct file *filp,
1565                         unsigned int ioctl, unsigned long arg)
1566 {
1567         void __user *argp = (void __user *)arg;
1568         long r;
1569
1570         switch (ioctl) {
1571         case KVM_GET_MSR_INDEX_LIST: {
1572                 struct kvm_msr_list __user *user_msr_list = argp;
1573                 struct kvm_msr_list msr_list;
1574                 unsigned n;
1575
1576                 r = -EFAULT;
1577                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1578                         goto out;
1579                 n = msr_list.nmsrs;
1580                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1581                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1582                         goto out;
1583                 r = -E2BIG;
1584                 if (n < msr_list.nmsrs)
1585                         goto out;
1586                 r = -EFAULT;
1587                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1588                                  num_msrs_to_save * sizeof(u32)))
1589                         goto out;
1590                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1591                                  &emulated_msrs,
1592                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1593                         goto out;
1594                 r = 0;
1595                 break;
1596         }
1597         case KVM_GET_SUPPORTED_CPUID: {
1598                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1599                 struct kvm_cpuid2 cpuid;
1600
1601                 r = -EFAULT;
1602                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1603                         goto out;
1604                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1605                                                       cpuid_arg->entries);
1606                 if (r)
1607                         goto out;
1608
1609                 r = -EFAULT;
1610                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1611                         goto out;
1612                 r = 0;
1613                 break;
1614         }
1615         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1616                 u64 mce_cap;
1617
1618                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1619                 r = -EFAULT;
1620                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1621                         goto out;
1622                 r = 0;
1623                 break;
1624         }
1625         default:
1626                 r = -EINVAL;
1627         }
1628 out:
1629         return r;
1630 }
1631
1632 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1633 {
1634         kvm_x86_ops->vcpu_load(vcpu, cpu);
1635         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1636                 unsigned long khz = cpufreq_quick_get(cpu);
1637                 if (!khz)
1638                         khz = tsc_khz;
1639                 per_cpu(cpu_tsc_khz, cpu) = khz;
1640         }
1641         kvm_request_guest_time_update(vcpu);
1642 }
1643
1644 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1645 {
1646         kvm_put_guest_fpu(vcpu);
1647         kvm_x86_ops->vcpu_put(vcpu);
1648 }
1649
1650 static int is_efer_nx(void)
1651 {
1652         unsigned long long efer = 0;
1653
1654         rdmsrl_safe(MSR_EFER, &efer);
1655         return efer & EFER_NX;
1656 }
1657
1658 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1659 {
1660         int i;
1661         struct kvm_cpuid_entry2 *e, *entry;
1662
1663         entry = NULL;
1664         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1665                 e = &vcpu->arch.cpuid_entries[i];
1666                 if (e->function == 0x80000001) {
1667                         entry = e;
1668                         break;
1669                 }
1670         }
1671         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1672                 entry->edx &= ~(1 << 20);
1673                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1674         }
1675 }
1676
1677 /* when an old userspace process fills a new kernel module */
1678 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1679                                     struct kvm_cpuid *cpuid,
1680                                     struct kvm_cpuid_entry __user *entries)
1681 {
1682         int r, i;
1683         struct kvm_cpuid_entry *cpuid_entries;
1684
1685         r = -E2BIG;
1686         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1687                 goto out;
1688         r = -ENOMEM;
1689         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1690         if (!cpuid_entries)
1691                 goto out;
1692         r = -EFAULT;
1693         if (copy_from_user(cpuid_entries, entries,
1694                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1695                 goto out_free;
1696         for (i = 0; i < cpuid->nent; i++) {
1697                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1698                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1699                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1700                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1701                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1702                 vcpu->arch.cpuid_entries[i].index = 0;
1703                 vcpu->arch.cpuid_entries[i].flags = 0;
1704                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1705                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1706                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1707         }
1708         vcpu->arch.cpuid_nent = cpuid->nent;
1709         cpuid_fix_nx_cap(vcpu);
1710         r = 0;
1711         kvm_apic_set_version(vcpu);
1712         kvm_x86_ops->cpuid_update(vcpu);
1713
1714 out_free:
1715         vfree(cpuid_entries);
1716 out:
1717         return r;
1718 }
1719
1720 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1721                                      struct kvm_cpuid2 *cpuid,
1722                                      struct kvm_cpuid_entry2 __user *entries)
1723 {
1724         int r;
1725
1726         r = -E2BIG;
1727         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1728                 goto out;
1729         r = -EFAULT;
1730         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1731                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1732                 goto out;
1733         vcpu->arch.cpuid_nent = cpuid->nent;
1734         kvm_apic_set_version(vcpu);
1735         kvm_x86_ops->cpuid_update(vcpu);
1736         return 0;
1737
1738 out:
1739         return r;
1740 }
1741
1742 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1743                                      struct kvm_cpuid2 *cpuid,
1744                                      struct kvm_cpuid_entry2 __user *entries)
1745 {
1746         int r;
1747
1748         r = -E2BIG;
1749         if (cpuid->nent < vcpu->arch.cpuid_nent)
1750                 goto out;
1751         r = -EFAULT;
1752         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1753                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1754                 goto out;
1755         return 0;
1756
1757 out:
1758         cpuid->nent = vcpu->arch.cpuid_nent;
1759         return r;
1760 }
1761
1762 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1763                            u32 index)
1764 {
1765         entry->function = function;
1766         entry->index = index;
1767         cpuid_count(entry->function, entry->index,
1768                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1769         entry->flags = 0;
1770 }
1771
1772 #define F(x) bit(X86_FEATURE_##x)
1773
1774 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1775                          u32 index, int *nent, int maxnent)
1776 {
1777         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1778 #ifdef CONFIG_X86_64
1779         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1780                                 ? F(GBPAGES) : 0;
1781         unsigned f_lm = F(LM);
1782 #else
1783         unsigned f_gbpages = 0;
1784         unsigned f_lm = 0;
1785 #endif
1786         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1787
1788         /* cpuid 1.edx */
1789         const u32 kvm_supported_word0_x86_features =
1790                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1791                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1792                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1793                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1794                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1795                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1796                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1797                 0 /* HTT, TM, Reserved, PBE */;
1798         /* cpuid 0x80000001.edx */
1799         const u32 kvm_supported_word1_x86_features =
1800                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1801                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1802                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1803                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1804                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1805                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1806                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1807                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1808         /* cpuid 1.ecx */
1809         const u32 kvm_supported_word4_x86_features =
1810                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1811                 0 /* DS-CPL, VMX, SMX, EST */ |
1812                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1813                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1814                 0 /* Reserved, DCA */ | F(XMM4_1) |
1815                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1816                 0 /* Reserved, XSAVE, OSXSAVE */;
1817         /* cpuid 0x80000001.ecx */
1818         const u32 kvm_supported_word6_x86_features =
1819                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1820                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1821                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1822                 0 /* SKINIT */ | 0 /* WDT */;
1823
1824         /* all calls to cpuid_count() should be made on the same cpu */
1825         get_cpu();
1826         do_cpuid_1_ent(entry, function, index);
1827         ++*nent;
1828
1829         switch (function) {
1830         case 0:
1831                 entry->eax = min(entry->eax, (u32)0xb);
1832                 break;
1833         case 1:
1834                 entry->edx &= kvm_supported_word0_x86_features;
1835                 entry->ecx &= kvm_supported_word4_x86_features;
1836                 /* we support x2apic emulation even if host does not support
1837                  * it since we emulate x2apic in software */
1838                 entry->ecx |= F(X2APIC);
1839                 break;
1840         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1841          * may return different values. This forces us to get_cpu() before
1842          * issuing the first command, and also to emulate this annoying behavior
1843          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1844         case 2: {
1845                 int t, times = entry->eax & 0xff;
1846
1847                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1848                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1849                 for (t = 1; t < times && *nent < maxnent; ++t) {
1850                         do_cpuid_1_ent(&entry[t], function, 0);
1851                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1852                         ++*nent;
1853                 }
1854                 break;
1855         }
1856         /* function 4 and 0xb have additional index. */
1857         case 4: {
1858                 int i, cache_type;
1859
1860                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1861                 /* read more entries until cache_type is zero */
1862                 for (i = 1; *nent < maxnent; ++i) {
1863                         cache_type = entry[i - 1].eax & 0x1f;
1864                         if (!cache_type)
1865                                 break;
1866                         do_cpuid_1_ent(&entry[i], function, i);
1867                         entry[i].flags |=
1868                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1869                         ++*nent;
1870                 }
1871                 break;
1872         }
1873         case 0xb: {
1874                 int i, level_type;
1875
1876                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1877                 /* read more entries until level_type is zero */
1878                 for (i = 1; *nent < maxnent; ++i) {
1879                         level_type = entry[i - 1].ecx & 0xff00;
1880                         if (!level_type)
1881                                 break;
1882                         do_cpuid_1_ent(&entry[i], function, i);
1883                         entry[i].flags |=
1884                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1885                         ++*nent;
1886                 }
1887                 break;
1888         }
1889         case 0x80000000:
1890                 entry->eax = min(entry->eax, 0x8000001a);
1891                 break;
1892         case 0x80000001:
1893                 entry->edx &= kvm_supported_word1_x86_features;
1894                 entry->ecx &= kvm_supported_word6_x86_features;
1895                 break;
1896         }
1897         put_cpu();
1898 }
1899
1900 #undef F
1901
1902 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1903                                      struct kvm_cpuid_entry2 __user *entries)
1904 {
1905         struct kvm_cpuid_entry2 *cpuid_entries;
1906         int limit, nent = 0, r = -E2BIG;
1907         u32 func;
1908
1909         if (cpuid->nent < 1)
1910                 goto out;
1911         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1912                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1913         r = -ENOMEM;
1914         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1915         if (!cpuid_entries)
1916                 goto out;
1917
1918         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1919         limit = cpuid_entries[0].eax;
1920         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1921                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1922                              &nent, cpuid->nent);
1923         r = -E2BIG;
1924         if (nent >= cpuid->nent)
1925                 goto out_free;
1926
1927         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1928         limit = cpuid_entries[nent - 1].eax;
1929         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1930                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1931                              &nent, cpuid->nent);
1932         r = -E2BIG;
1933         if (nent >= cpuid->nent)
1934                 goto out_free;
1935
1936         r = -EFAULT;
1937         if (copy_to_user(entries, cpuid_entries,
1938                          nent * sizeof(struct kvm_cpuid_entry2)))
1939                 goto out_free;
1940         cpuid->nent = nent;
1941         r = 0;
1942
1943 out_free:
1944         vfree(cpuid_entries);
1945 out:
1946         return r;
1947 }
1948
1949 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1950                                     struct kvm_lapic_state *s)
1951 {
1952         vcpu_load(vcpu);
1953         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1954         vcpu_put(vcpu);
1955
1956         return 0;
1957 }
1958
1959 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1960                                     struct kvm_lapic_state *s)
1961 {
1962         vcpu_load(vcpu);
1963         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1964         kvm_apic_post_state_restore(vcpu);
1965         update_cr8_intercept(vcpu);
1966         vcpu_put(vcpu);
1967
1968         return 0;
1969 }
1970
1971 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1972                                     struct kvm_interrupt *irq)
1973 {
1974         if (irq->irq < 0 || irq->irq >= 256)
1975                 return -EINVAL;
1976         if (irqchip_in_kernel(vcpu->kvm))
1977                 return -ENXIO;
1978         vcpu_load(vcpu);
1979
1980         kvm_queue_interrupt(vcpu, irq->irq, false);
1981
1982         vcpu_put(vcpu);
1983
1984         return 0;
1985 }
1986
1987 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1988 {
1989         vcpu_load(vcpu);
1990         kvm_inject_nmi(vcpu);
1991         vcpu_put(vcpu);
1992
1993         return 0;
1994 }
1995
1996 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1997                                            struct kvm_tpr_access_ctl *tac)
1998 {
1999         if (tac->flags)
2000                 return -EINVAL;
2001         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2002         return 0;
2003 }
2004
2005 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2006                                         u64 mcg_cap)
2007 {
2008         int r;
2009         unsigned bank_num = mcg_cap & 0xff, bank;
2010
2011         r = -EINVAL;
2012         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2013                 goto out;
2014         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2015                 goto out;
2016         r = 0;
2017         vcpu->arch.mcg_cap = mcg_cap;
2018         /* Init IA32_MCG_CTL to all 1s */
2019         if (mcg_cap & MCG_CTL_P)
2020                 vcpu->arch.mcg_ctl = ~(u64)0;
2021         /* Init IA32_MCi_CTL to all 1s */
2022         for (bank = 0; bank < bank_num; bank++)
2023                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2024 out:
2025         return r;
2026 }
2027
2028 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2029                                       struct kvm_x86_mce *mce)
2030 {
2031         u64 mcg_cap = vcpu->arch.mcg_cap;
2032         unsigned bank_num = mcg_cap & 0xff;
2033         u64 *banks = vcpu->arch.mce_banks;
2034
2035         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2036                 return -EINVAL;
2037         /*
2038          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2039          * reporting is disabled
2040          */
2041         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2042             vcpu->arch.mcg_ctl != ~(u64)0)
2043                 return 0;
2044         banks += 4 * mce->bank;
2045         /*
2046          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2047          * reporting is disabled for the bank
2048          */
2049         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2050                 return 0;
2051         if (mce->status & MCI_STATUS_UC) {
2052                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2053                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2054                         printk(KERN_DEBUG "kvm: set_mce: "
2055                                "injects mce exception while "
2056                                "previous one is in progress!\n");
2057                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2058                         return 0;
2059                 }
2060                 if (banks[1] & MCI_STATUS_VAL)
2061                         mce->status |= MCI_STATUS_OVER;
2062                 banks[2] = mce->addr;
2063                 banks[3] = mce->misc;
2064                 vcpu->arch.mcg_status = mce->mcg_status;
2065                 banks[1] = mce->status;
2066                 kvm_queue_exception(vcpu, MC_VECTOR);
2067         } else if (!(banks[1] & MCI_STATUS_VAL)
2068                    || !(banks[1] & MCI_STATUS_UC)) {
2069                 if (banks[1] & MCI_STATUS_VAL)
2070                         mce->status |= MCI_STATUS_OVER;
2071                 banks[2] = mce->addr;
2072                 banks[3] = mce->misc;
2073                 banks[1] = mce->status;
2074         } else
2075                 banks[1] |= MCI_STATUS_OVER;
2076         return 0;
2077 }
2078
2079 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2080                                                struct kvm_vcpu_events *events)
2081 {
2082         vcpu_load(vcpu);
2083
2084         events->exception.injected = vcpu->arch.exception.pending;
2085         events->exception.nr = vcpu->arch.exception.nr;
2086         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2087         events->exception.error_code = vcpu->arch.exception.error_code;
2088
2089         events->interrupt.injected = vcpu->arch.interrupt.pending;
2090         events->interrupt.nr = vcpu->arch.interrupt.nr;
2091         events->interrupt.soft = vcpu->arch.interrupt.soft;
2092
2093         events->nmi.injected = vcpu->arch.nmi_injected;
2094         events->nmi.pending = vcpu->arch.nmi_pending;
2095         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2096
2097         events->sipi_vector = vcpu->arch.sipi_vector;
2098
2099         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2100                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2101
2102         vcpu_put(vcpu);
2103 }
2104
2105 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2106                                               struct kvm_vcpu_events *events)
2107 {
2108         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2109                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2110                 return -EINVAL;
2111
2112         vcpu_load(vcpu);
2113
2114         vcpu->arch.exception.pending = events->exception.injected;
2115         vcpu->arch.exception.nr = events->exception.nr;
2116         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2117         vcpu->arch.exception.error_code = events->exception.error_code;
2118
2119         vcpu->arch.interrupt.pending = events->interrupt.injected;
2120         vcpu->arch.interrupt.nr = events->interrupt.nr;
2121         vcpu->arch.interrupt.soft = events->interrupt.soft;
2122         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2123                 kvm_pic_clear_isr_ack(vcpu->kvm);
2124
2125         vcpu->arch.nmi_injected = events->nmi.injected;
2126         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2127                 vcpu->arch.nmi_pending = events->nmi.pending;
2128         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2129
2130         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2131                 vcpu->arch.sipi_vector = events->sipi_vector;
2132
2133         vcpu_put(vcpu);
2134
2135         return 0;
2136 }
2137
2138 long kvm_arch_vcpu_ioctl(struct file *filp,
2139                          unsigned int ioctl, unsigned long arg)
2140 {
2141         struct kvm_vcpu *vcpu = filp->private_data;
2142         void __user *argp = (void __user *)arg;
2143         int r;
2144         struct kvm_lapic_state *lapic = NULL;
2145
2146         switch (ioctl) {
2147         case KVM_GET_LAPIC: {
2148                 r = -EINVAL;
2149                 if (!vcpu->arch.apic)
2150                         goto out;
2151                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2152
2153                 r = -ENOMEM;
2154                 if (!lapic)
2155                         goto out;
2156                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2157                 if (r)
2158                         goto out;
2159                 r = -EFAULT;
2160                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2161                         goto out;
2162                 r = 0;
2163                 break;
2164         }
2165         case KVM_SET_LAPIC: {
2166                 r = -EINVAL;
2167                 if (!vcpu->arch.apic)
2168                         goto out;
2169                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2170                 r = -ENOMEM;
2171                 if (!lapic)
2172                         goto out;
2173                 r = -EFAULT;
2174                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2175                         goto out;
2176                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2177                 if (r)
2178                         goto out;
2179                 r = 0;
2180                 break;
2181         }
2182         case KVM_INTERRUPT: {
2183                 struct kvm_interrupt irq;
2184
2185                 r = -EFAULT;
2186                 if (copy_from_user(&irq, argp, sizeof irq))
2187                         goto out;
2188                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2189                 if (r)
2190                         goto out;
2191                 r = 0;
2192                 break;
2193         }
2194         case KVM_NMI: {
2195                 r = kvm_vcpu_ioctl_nmi(vcpu);
2196                 if (r)
2197                         goto out;
2198                 r = 0;
2199                 break;
2200         }
2201         case KVM_SET_CPUID: {
2202                 struct kvm_cpuid __user *cpuid_arg = argp;
2203                 struct kvm_cpuid cpuid;
2204
2205                 r = -EFAULT;
2206                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2207                         goto out;
2208                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2209                 if (r)
2210                         goto out;
2211                 break;
2212         }
2213         case KVM_SET_CPUID2: {
2214                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2215                 struct kvm_cpuid2 cpuid;
2216
2217                 r = -EFAULT;
2218                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2219                         goto out;
2220                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2221                                               cpuid_arg->entries);
2222                 if (r)
2223                         goto out;
2224                 break;
2225         }
2226         case KVM_GET_CPUID2: {
2227                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2228                 struct kvm_cpuid2 cpuid;
2229
2230                 r = -EFAULT;
2231                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2232                         goto out;
2233                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2234                                               cpuid_arg->entries);
2235                 if (r)
2236                         goto out;
2237                 r = -EFAULT;
2238                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2239                         goto out;
2240                 r = 0;
2241                 break;
2242         }
2243         case KVM_GET_MSRS:
2244                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2245                 break;
2246         case KVM_SET_MSRS:
2247                 r = msr_io(vcpu, argp, do_set_msr, 0);
2248                 break;
2249         case KVM_TPR_ACCESS_REPORTING: {
2250                 struct kvm_tpr_access_ctl tac;
2251
2252                 r = -EFAULT;
2253                 if (copy_from_user(&tac, argp, sizeof tac))
2254                         goto out;
2255                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2256                 if (r)
2257                         goto out;
2258                 r = -EFAULT;
2259                 if (copy_to_user(argp, &tac, sizeof tac))
2260                         goto out;
2261                 r = 0;
2262                 break;
2263         };
2264         case KVM_SET_VAPIC_ADDR: {
2265                 struct kvm_vapic_addr va;
2266
2267                 r = -EINVAL;
2268                 if (!irqchip_in_kernel(vcpu->kvm))
2269                         goto out;
2270                 r = -EFAULT;
2271                 if (copy_from_user(&va, argp, sizeof va))
2272                         goto out;
2273                 r = 0;
2274                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2275                 break;
2276         }
2277         case KVM_X86_SETUP_MCE: {
2278                 u64 mcg_cap;
2279
2280                 r = -EFAULT;
2281                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2282                         goto out;
2283                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2284                 break;
2285         }
2286         case KVM_X86_SET_MCE: {
2287                 struct kvm_x86_mce mce;
2288
2289                 r = -EFAULT;
2290                 if (copy_from_user(&mce, argp, sizeof mce))
2291                         goto out;
2292                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2293                 break;
2294         }
2295         case KVM_GET_VCPU_EVENTS: {
2296                 struct kvm_vcpu_events events;
2297
2298                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2299
2300                 r = -EFAULT;
2301                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2302                         break;
2303                 r = 0;
2304                 break;
2305         }
2306         case KVM_SET_VCPU_EVENTS: {
2307                 struct kvm_vcpu_events events;
2308
2309                 r = -EFAULT;
2310                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2311                         break;
2312
2313                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2314                 break;
2315         }
2316         default:
2317                 r = -EINVAL;
2318         }
2319 out:
2320         kfree(lapic);
2321         return r;
2322 }
2323
2324 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2325 {
2326         int ret;
2327
2328         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2329                 return -1;
2330         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2331         return ret;
2332 }
2333
2334 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2335                                               u64 ident_addr)
2336 {
2337         kvm->arch.ept_identity_map_addr = ident_addr;
2338         return 0;
2339 }
2340
2341 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2342                                           u32 kvm_nr_mmu_pages)
2343 {
2344         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2345                 return -EINVAL;
2346
2347         mutex_lock(&kvm->slots_lock);
2348         spin_lock(&kvm->mmu_lock);
2349
2350         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2351         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2352
2353         spin_unlock(&kvm->mmu_lock);
2354         mutex_unlock(&kvm->slots_lock);
2355         return 0;
2356 }
2357
2358 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2359 {
2360         return kvm->arch.n_alloc_mmu_pages;
2361 }
2362
2363 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2364 {
2365         int i;
2366         struct kvm_mem_alias *alias;
2367         struct kvm_mem_aliases *aliases;
2368
2369         aliases = rcu_dereference(kvm->arch.aliases);
2370
2371         for (i = 0; i < aliases->naliases; ++i) {
2372                 alias = &aliases->aliases[i];
2373                 if (alias->flags & KVM_ALIAS_INVALID)
2374                         continue;
2375                 if (gfn >= alias->base_gfn
2376                     && gfn < alias->base_gfn + alias->npages)
2377                         return alias->target_gfn + gfn - alias->base_gfn;
2378         }
2379         return gfn;
2380 }
2381
2382 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2383 {
2384         int i;
2385         struct kvm_mem_alias *alias;
2386         struct kvm_mem_aliases *aliases;
2387
2388         aliases = rcu_dereference(kvm->arch.aliases);
2389
2390         for (i = 0; i < aliases->naliases; ++i) {
2391                 alias = &aliases->aliases[i];
2392                 if (gfn >= alias->base_gfn
2393                     && gfn < alias->base_gfn + alias->npages)
2394                         return alias->target_gfn + gfn - alias->base_gfn;
2395         }
2396         return gfn;
2397 }
2398
2399 /*
2400  * Set a new alias region.  Aliases map a portion of physical memory into
2401  * another portion.  This is useful for memory windows, for example the PC
2402  * VGA region.
2403  */
2404 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2405                                          struct kvm_memory_alias *alias)
2406 {
2407         int r, n;
2408         struct kvm_mem_alias *p;
2409         struct kvm_mem_aliases *aliases, *old_aliases;
2410
2411         r = -EINVAL;
2412         /* General sanity checks */
2413         if (alias->memory_size & (PAGE_SIZE - 1))
2414                 goto out;
2415         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2416                 goto out;
2417         if (alias->slot >= KVM_ALIAS_SLOTS)
2418                 goto out;
2419         if (alias->guest_phys_addr + alias->memory_size
2420             < alias->guest_phys_addr)
2421                 goto out;
2422         if (alias->target_phys_addr + alias->memory_size
2423             < alias->target_phys_addr)
2424                 goto out;
2425
2426         r = -ENOMEM;
2427         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2428         if (!aliases)
2429                 goto out;
2430
2431         mutex_lock(&kvm->slots_lock);
2432
2433         /* invalidate any gfn reference in case of deletion/shrinking */
2434         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2435         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2436         old_aliases = kvm->arch.aliases;
2437         rcu_assign_pointer(kvm->arch.aliases, aliases);
2438         synchronize_srcu_expedited(&kvm->srcu);
2439         kvm_mmu_zap_all(kvm);
2440         kfree(old_aliases);
2441
2442         r = -ENOMEM;
2443         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2444         if (!aliases)
2445                 goto out_unlock;
2446
2447         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2448
2449         p = &aliases->aliases[alias->slot];
2450         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2451         p->npages = alias->memory_size >> PAGE_SHIFT;
2452         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2453         p->flags &= ~(KVM_ALIAS_INVALID);
2454
2455         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2456                 if (aliases->aliases[n - 1].npages)
2457                         break;
2458         aliases->naliases = n;
2459
2460         old_aliases = kvm->arch.aliases;
2461         rcu_assign_pointer(kvm->arch.aliases, aliases);
2462         synchronize_srcu_expedited(&kvm->srcu);
2463         kfree(old_aliases);
2464         r = 0;
2465
2466 out_unlock:
2467         mutex_unlock(&kvm->slots_lock);
2468 out:
2469         return r;
2470 }
2471
2472 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2473 {
2474         int r;
2475
2476         r = 0;
2477         switch (chip->chip_id) {
2478         case KVM_IRQCHIP_PIC_MASTER:
2479                 memcpy(&chip->chip.pic,
2480                         &pic_irqchip(kvm)->pics[0],
2481                         sizeof(struct kvm_pic_state));
2482                 break;
2483         case KVM_IRQCHIP_PIC_SLAVE:
2484                 memcpy(&chip->chip.pic,
2485                         &pic_irqchip(kvm)->pics[1],
2486                         sizeof(struct kvm_pic_state));
2487                 break;
2488         case KVM_IRQCHIP_IOAPIC:
2489                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2490                 break;
2491         default:
2492                 r = -EINVAL;
2493                 break;
2494         }
2495         return r;
2496 }
2497
2498 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2499 {
2500         int r;
2501
2502         r = 0;
2503         switch (chip->chip_id) {
2504         case KVM_IRQCHIP_PIC_MASTER:
2505                 spin_lock(&pic_irqchip(kvm)->lock);
2506                 memcpy(&pic_irqchip(kvm)->pics[0],
2507                         &chip->chip.pic,
2508                         sizeof(struct kvm_pic_state));
2509                 spin_unlock(&pic_irqchip(kvm)->lock);
2510                 break;
2511         case KVM_IRQCHIP_PIC_SLAVE:
2512                 spin_lock(&pic_irqchip(kvm)->lock);
2513                 memcpy(&pic_irqchip(kvm)->pics[1],
2514                         &chip->chip.pic,
2515                         sizeof(struct kvm_pic_state));
2516                 spin_unlock(&pic_irqchip(kvm)->lock);
2517                 break;
2518         case KVM_IRQCHIP_IOAPIC:
2519                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2520                 break;
2521         default:
2522                 r = -EINVAL;
2523                 break;
2524         }
2525         kvm_pic_update_irq(pic_irqchip(kvm));
2526         return r;
2527 }
2528
2529 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2530 {
2531         int r = 0;
2532
2533         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2534         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2535         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2536         return r;
2537 }
2538
2539 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2540 {
2541         int r = 0;
2542
2543         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2544         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2545         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2546         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2547         return r;
2548 }
2549
2550 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2551 {
2552         int r = 0;
2553
2554         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2555         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2556                 sizeof(ps->channels));
2557         ps->flags = kvm->arch.vpit->pit_state.flags;
2558         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2559         return r;
2560 }
2561
2562 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2563 {
2564         int r = 0, start = 0;
2565         u32 prev_legacy, cur_legacy;
2566         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2567         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2568         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2569         if (!prev_legacy && cur_legacy)
2570                 start = 1;
2571         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2572                sizeof(kvm->arch.vpit->pit_state.channels));
2573         kvm->arch.vpit->pit_state.flags = ps->flags;
2574         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2575         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2576         return r;
2577 }
2578
2579 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2580                                  struct kvm_reinject_control *control)
2581 {
2582         if (!kvm->arch.vpit)
2583                 return -ENXIO;
2584         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2585         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2586         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2587         return 0;
2588 }
2589
2590 /*
2591  * Get (and clear) the dirty memory log for a memory slot.
2592  */
2593 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2594                                       struct kvm_dirty_log *log)
2595 {
2596         int r, n, i;
2597         struct kvm_memory_slot *memslot;
2598         unsigned long is_dirty = 0;
2599         unsigned long *dirty_bitmap = NULL;
2600
2601         mutex_lock(&kvm->slots_lock);
2602
2603         r = -EINVAL;
2604         if (log->slot >= KVM_MEMORY_SLOTS)
2605                 goto out;
2606
2607         memslot = &kvm->memslots->memslots[log->slot];
2608         r = -ENOENT;
2609         if (!memslot->dirty_bitmap)
2610                 goto out;
2611
2612         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2613
2614         r = -ENOMEM;
2615         dirty_bitmap = vmalloc(n);
2616         if (!dirty_bitmap)
2617                 goto out;
2618         memset(dirty_bitmap, 0, n);
2619
2620         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2621                 is_dirty = memslot->dirty_bitmap[i];
2622
2623         /* If nothing is dirty, don't bother messing with page tables. */
2624         if (is_dirty) {
2625                 struct kvm_memslots *slots, *old_slots;
2626
2627                 spin_lock(&kvm->mmu_lock);
2628                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2629                 spin_unlock(&kvm->mmu_lock);
2630
2631                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2632                 if (!slots)
2633                         goto out_free;
2634
2635                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2636                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2637
2638                 old_slots = kvm->memslots;
2639                 rcu_assign_pointer(kvm->memslots, slots);
2640                 synchronize_srcu_expedited(&kvm->srcu);
2641                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2642                 kfree(old_slots);
2643         }
2644
2645         r = 0;
2646         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2647                 r = -EFAULT;
2648 out_free:
2649         vfree(dirty_bitmap);
2650 out:
2651         mutex_unlock(&kvm->slots_lock);
2652         return r;
2653 }
2654
2655 long kvm_arch_vm_ioctl(struct file *filp,
2656                        unsigned int ioctl, unsigned long arg)
2657 {
2658         struct kvm *kvm = filp->private_data;
2659         void __user *argp = (void __user *)arg;
2660         int r = -ENOTTY;
2661         /*
2662          * This union makes it completely explicit to gcc-3.x
2663          * that these two variables' stack usage should be
2664          * combined, not added together.
2665          */
2666         union {
2667                 struct kvm_pit_state ps;
2668                 struct kvm_pit_state2 ps2;
2669                 struct kvm_memory_alias alias;
2670                 struct kvm_pit_config pit_config;
2671         } u;
2672
2673         switch (ioctl) {
2674         case KVM_SET_TSS_ADDR:
2675                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2676                 if (r < 0)
2677                         goto out;
2678                 break;
2679         case KVM_SET_IDENTITY_MAP_ADDR: {
2680                 u64 ident_addr;
2681
2682                 r = -EFAULT;
2683                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2684                         goto out;
2685                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2686                 if (r < 0)
2687                         goto out;
2688                 break;
2689         }
2690         case KVM_SET_MEMORY_REGION: {
2691                 struct kvm_memory_region kvm_mem;
2692                 struct kvm_userspace_memory_region kvm_userspace_mem;
2693
2694                 r = -EFAULT;
2695                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2696                         goto out;
2697                 kvm_userspace_mem.slot = kvm_mem.slot;
2698                 kvm_userspace_mem.flags = kvm_mem.flags;
2699                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2700                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2701                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2702                 if (r)
2703                         goto out;
2704                 break;
2705         }
2706         case KVM_SET_NR_MMU_PAGES:
2707                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2708                 if (r)
2709                         goto out;
2710                 break;
2711         case KVM_GET_NR_MMU_PAGES:
2712                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2713                 break;
2714         case KVM_SET_MEMORY_ALIAS:
2715                 r = -EFAULT;
2716                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2717                         goto out;
2718                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2719                 if (r)
2720                         goto out;
2721                 break;
2722         case KVM_CREATE_IRQCHIP: {
2723                 struct kvm_pic *vpic;
2724
2725                 mutex_lock(&kvm->lock);
2726                 r = -EEXIST;
2727                 if (kvm->arch.vpic)
2728                         goto create_irqchip_unlock;
2729                 r = -ENOMEM;
2730                 vpic = kvm_create_pic(kvm);
2731                 if (vpic) {
2732                         r = kvm_ioapic_init(kvm);
2733                         if (r) {
2734                                 kfree(vpic);
2735                                 goto create_irqchip_unlock;
2736                         }
2737                 } else
2738                         goto create_irqchip_unlock;
2739                 smp_wmb();
2740                 kvm->arch.vpic = vpic;
2741                 smp_wmb();
2742                 r = kvm_setup_default_irq_routing(kvm);
2743                 if (r) {
2744                         mutex_lock(&kvm->irq_lock);
2745                         kfree(kvm->arch.vpic);
2746                         kfree(kvm->arch.vioapic);
2747                         kvm->arch.vpic = NULL;
2748                         kvm->arch.vioapic = NULL;
2749                         mutex_unlock(&kvm->irq_lock);
2750                 }
2751         create_irqchip_unlock:
2752                 mutex_unlock(&kvm->lock);
2753                 break;
2754         }
2755         case KVM_CREATE_PIT:
2756                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2757                 goto create_pit;
2758         case KVM_CREATE_PIT2:
2759                 r = -EFAULT;
2760                 if (copy_from_user(&u.pit_config, argp,
2761                                    sizeof(struct kvm_pit_config)))
2762                         goto out;
2763         create_pit:
2764                 mutex_lock(&kvm->slots_lock);
2765                 r = -EEXIST;
2766                 if (kvm->arch.vpit)
2767                         goto create_pit_unlock;
2768                 r = -ENOMEM;
2769                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2770                 if (kvm->arch.vpit)
2771                         r = 0;
2772         create_pit_unlock:
2773                 mutex_unlock(&kvm->slots_lock);
2774                 break;
2775         case KVM_IRQ_LINE_STATUS:
2776         case KVM_IRQ_LINE: {
2777                 struct kvm_irq_level irq_event;
2778
2779                 r = -EFAULT;
2780                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2781                         goto out;
2782                 if (irqchip_in_kernel(kvm)) {
2783                         __s32 status;
2784                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2785                                         irq_event.irq, irq_event.level);
2786                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2787                                 irq_event.status = status;
2788                                 if (copy_to_user(argp, &irq_event,
2789                                                         sizeof irq_event))
2790                                         goto out;
2791                         }
2792                         r = 0;
2793                 }
2794                 break;
2795         }
2796         case KVM_GET_IRQCHIP: {
2797                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2798                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2799
2800                 r = -ENOMEM;
2801                 if (!chip)
2802                         goto out;
2803                 r = -EFAULT;
2804                 if (copy_from_user(chip, argp, sizeof *chip))
2805                         goto get_irqchip_out;
2806                 r = -ENXIO;
2807                 if (!irqchip_in_kernel(kvm))
2808                         goto get_irqchip_out;
2809                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2810                 if (r)
2811                         goto get_irqchip_out;
2812                 r = -EFAULT;
2813                 if (copy_to_user(argp, chip, sizeof *chip))
2814                         goto get_irqchip_out;
2815                 r = 0;
2816         get_irqchip_out:
2817                 kfree(chip);
2818                 if (r)
2819                         goto out;
2820                 break;
2821         }
2822         case KVM_SET_IRQCHIP: {
2823                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2824                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2825
2826                 r = -ENOMEM;
2827                 if (!chip)
2828                         goto out;
2829                 r = -EFAULT;
2830                 if (copy_from_user(chip, argp, sizeof *chip))
2831                         goto set_irqchip_out;
2832                 r = -ENXIO;
2833                 if (!irqchip_in_kernel(kvm))
2834                         goto set_irqchip_out;
2835                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2836                 if (r)
2837                         goto set_irqchip_out;
2838                 r = 0;
2839         set_irqchip_out:
2840                 kfree(chip);
2841                 if (r)
2842                         goto out;
2843                 break;
2844         }
2845         case KVM_GET_PIT: {
2846                 r = -EFAULT;
2847                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2848                         goto out;
2849                 r = -ENXIO;
2850                 if (!kvm->arch.vpit)
2851                         goto out;
2852                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2853                 if (r)
2854                         goto out;
2855                 r = -EFAULT;
2856                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2857                         goto out;
2858                 r = 0;
2859                 break;
2860         }
2861         case KVM_SET_PIT: {
2862                 r = -EFAULT;
2863                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2864                         goto out;
2865                 r = -ENXIO;
2866                 if (!kvm->arch.vpit)
2867                         goto out;
2868                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2869                 if (r)
2870                         goto out;
2871                 r = 0;
2872                 break;
2873         }
2874         case KVM_GET_PIT2: {
2875                 r = -ENXIO;
2876                 if (!kvm->arch.vpit)
2877                         goto out;
2878                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2879                 if (r)
2880                         goto out;
2881                 r = -EFAULT;
2882                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2883                         goto out;
2884                 r = 0;
2885                 break;
2886         }
2887         case KVM_SET_PIT2: {
2888                 r = -EFAULT;
2889                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2890                         goto out;
2891                 r = -ENXIO;
2892                 if (!kvm->arch.vpit)
2893                         goto out;
2894                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2895                 if (r)
2896                         goto out;
2897                 r = 0;
2898                 break;
2899         }
2900         case KVM_REINJECT_CONTROL: {
2901                 struct kvm_reinject_control control;
2902                 r =  -EFAULT;
2903                 if (copy_from_user(&control, argp, sizeof(control)))
2904                         goto out;
2905                 r = kvm_vm_ioctl_reinject(kvm, &control);
2906                 if (r)
2907                         goto out;
2908                 r = 0;
2909                 break;
2910         }
2911         case KVM_XEN_HVM_CONFIG: {
2912                 r = -EFAULT;
2913                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2914                                    sizeof(struct kvm_xen_hvm_config)))
2915                         goto out;
2916                 r = -EINVAL;
2917                 if (kvm->arch.xen_hvm_config.flags)
2918                         goto out;
2919                 r = 0;
2920                 break;
2921         }
2922         case KVM_SET_CLOCK: {
2923                 struct timespec now;
2924                 struct kvm_clock_data user_ns;
2925                 u64 now_ns;
2926                 s64 delta;
2927
2928                 r = -EFAULT;
2929                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2930                         goto out;
2931
2932                 r = -EINVAL;
2933                 if (user_ns.flags)
2934                         goto out;
2935
2936                 r = 0;
2937                 ktime_get_ts(&now);
2938                 now_ns = timespec_to_ns(&now);
2939                 delta = user_ns.clock - now_ns;
2940                 kvm->arch.kvmclock_offset = delta;
2941                 break;
2942         }
2943         case KVM_GET_CLOCK: {
2944                 struct timespec now;
2945                 struct kvm_clock_data user_ns;
2946                 u64 now_ns;
2947
2948                 ktime_get_ts(&now);
2949                 now_ns = timespec_to_ns(&now);
2950                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2951                 user_ns.flags = 0;
2952
2953                 r = -EFAULT;
2954                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2955                         goto out;
2956                 r = 0;
2957                 break;
2958         }
2959
2960         default:
2961                 ;
2962         }
2963 out:
2964         return r;
2965 }
2966
2967 static void kvm_init_msr_list(void)
2968 {
2969         u32 dummy[2];
2970         unsigned i, j;
2971
2972         /* skip the first msrs in the list. KVM-specific */
2973         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2974                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2975                         continue;
2976                 if (j < i)
2977                         msrs_to_save[j] = msrs_to_save[i];
2978                 j++;
2979         }
2980         num_msrs_to_save = j;
2981 }
2982
2983 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2984                            const void *v)
2985 {
2986         if (vcpu->arch.apic &&
2987             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2988                 return 0;
2989
2990         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
2991 }
2992
2993 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2994 {
2995         if (vcpu->arch.apic &&
2996             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2997                 return 0;
2998
2999         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3000 }
3001
3002 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3003                                struct kvm_vcpu *vcpu)
3004 {
3005         void *data = val;
3006         int r = X86EMUL_CONTINUE;
3007
3008         while (bytes) {
3009                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3010                 unsigned offset = addr & (PAGE_SIZE-1);
3011                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3012                 int ret;
3013
3014                 if (gpa == UNMAPPED_GVA) {
3015                         r = X86EMUL_PROPAGATE_FAULT;
3016                         goto out;
3017                 }
3018                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3019                 if (ret < 0) {
3020                         r = X86EMUL_UNHANDLEABLE;
3021                         goto out;
3022                 }
3023
3024                 bytes -= toread;
3025                 data += toread;
3026                 addr += toread;
3027         }
3028 out:
3029         return r;
3030 }
3031
3032 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3033                                 struct kvm_vcpu *vcpu)
3034 {
3035         void *data = val;
3036         int r = X86EMUL_CONTINUE;
3037
3038         while (bytes) {
3039                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3040                 unsigned offset = addr & (PAGE_SIZE-1);
3041                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3042                 int ret;
3043
3044                 if (gpa == UNMAPPED_GVA) {
3045                         r = X86EMUL_PROPAGATE_FAULT;
3046                         goto out;
3047                 }
3048                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3049                 if (ret < 0) {
3050                         r = X86EMUL_UNHANDLEABLE;
3051                         goto out;
3052                 }
3053
3054                 bytes -= towrite;
3055                 data += towrite;
3056                 addr += towrite;
3057         }
3058 out:
3059         return r;
3060 }
3061
3062
3063 static int emulator_read_emulated(unsigned long addr,
3064                                   void *val,
3065                                   unsigned int bytes,
3066                                   struct kvm_vcpu *vcpu)
3067 {
3068         gpa_t                 gpa;
3069
3070         if (vcpu->mmio_read_completed) {
3071                 memcpy(val, vcpu->mmio_data, bytes);
3072                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3073                                vcpu->mmio_phys_addr, *(u64 *)val);
3074                 vcpu->mmio_read_completed = 0;
3075                 return X86EMUL_CONTINUE;
3076         }
3077
3078         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3079
3080         /* For APIC access vmexit */
3081         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3082                 goto mmio;
3083
3084         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
3085                                 == X86EMUL_CONTINUE)
3086                 return X86EMUL_CONTINUE;
3087         if (gpa == UNMAPPED_GVA)
3088                 return X86EMUL_PROPAGATE_FAULT;
3089
3090 mmio:
3091         /*
3092          * Is this MMIO handled locally?
3093          */
3094         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3095                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3096                 return X86EMUL_CONTINUE;
3097         }
3098
3099         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3100
3101         vcpu->mmio_needed = 1;
3102         vcpu->mmio_phys_addr = gpa;
3103         vcpu->mmio_size = bytes;
3104         vcpu->mmio_is_write = 0;
3105
3106         return X86EMUL_UNHANDLEABLE;
3107 }
3108
3109 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3110                           const void *val, int bytes)
3111 {
3112         int ret;
3113
3114         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3115         if (ret < 0)
3116                 return 0;
3117         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3118         return 1;
3119 }
3120
3121 static int emulator_write_emulated_onepage(unsigned long addr,
3122                                            const void *val,
3123                                            unsigned int bytes,
3124                                            struct kvm_vcpu *vcpu)
3125 {
3126         gpa_t                 gpa;
3127
3128         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3129
3130         if (gpa == UNMAPPED_GVA) {
3131                 kvm_inject_page_fault(vcpu, addr, 2);
3132                 return X86EMUL_PROPAGATE_FAULT;
3133         }
3134
3135         /* For APIC access vmexit */
3136         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3137                 goto mmio;
3138
3139         if (emulator_write_phys(vcpu, gpa, val, bytes))
3140                 return X86EMUL_CONTINUE;
3141
3142 mmio:
3143         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3144         /*
3145          * Is this MMIO handled locally?
3146          */
3147         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3148                 return X86EMUL_CONTINUE;
3149
3150         vcpu->mmio_needed = 1;
3151         vcpu->mmio_phys_addr = gpa;
3152         vcpu->mmio_size = bytes;
3153         vcpu->mmio_is_write = 1;
3154         memcpy(vcpu->mmio_data, val, bytes);
3155
3156         return X86EMUL_CONTINUE;
3157 }
3158
3159 int emulator_write_emulated(unsigned long addr,
3160                                    const void *val,
3161                                    unsigned int bytes,
3162                                    struct kvm_vcpu *vcpu)
3163 {
3164         /* Crossing a page boundary? */
3165         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3166                 int rc, now;
3167
3168                 now = -addr & ~PAGE_MASK;
3169                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3170                 if (rc != X86EMUL_CONTINUE)
3171                         return rc;
3172                 addr += now;
3173                 val += now;
3174                 bytes -= now;
3175         }
3176         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3177 }
3178 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3179
3180 static int emulator_cmpxchg_emulated(unsigned long addr,
3181                                      const void *old,
3182                                      const void *new,
3183                                      unsigned int bytes,
3184                                      struct kvm_vcpu *vcpu)
3185 {
3186         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3187 #ifndef CONFIG_X86_64
3188         /* guests cmpxchg8b have to be emulated atomically */
3189         if (bytes == 8) {
3190                 gpa_t gpa;
3191                 struct page *page;
3192                 char *kaddr;
3193                 u64 val;
3194
3195                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3196
3197                 if (gpa == UNMAPPED_GVA ||
3198                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3199                         goto emul_write;
3200
3201                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3202                         goto emul_write;
3203
3204                 val = *(u64 *)new;
3205
3206                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3207
3208                 kaddr = kmap_atomic(page, KM_USER0);
3209                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3210                 kunmap_atomic(kaddr, KM_USER0);
3211                 kvm_release_page_dirty(page);
3212         }
3213 emul_write:
3214 #endif
3215
3216         return emulator_write_emulated(addr, new, bytes, vcpu);
3217 }
3218
3219 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3220 {
3221         return kvm_x86_ops->get_segment_base(vcpu, seg);
3222 }
3223
3224 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3225 {
3226         kvm_mmu_invlpg(vcpu, address);
3227         return X86EMUL_CONTINUE;
3228 }
3229
3230 int emulate_clts(struct kvm_vcpu *vcpu)
3231 {
3232         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3233         return X86EMUL_CONTINUE;
3234 }
3235
3236 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3237 {
3238         struct kvm_vcpu *vcpu = ctxt->vcpu;
3239
3240         switch (dr) {
3241         case 0 ... 3:
3242                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3243                 return X86EMUL_CONTINUE;
3244         default:
3245                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3246                 return X86EMUL_UNHANDLEABLE;
3247         }
3248 }
3249
3250 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3251 {
3252         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3253         int exception;
3254
3255         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3256         if (exception) {
3257                 /* FIXME: better handling */
3258                 return X86EMUL_UNHANDLEABLE;
3259         }
3260         return X86EMUL_CONTINUE;
3261 }
3262
3263 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3264 {
3265         u8 opcodes[4];
3266         unsigned long rip = kvm_rip_read(vcpu);
3267         unsigned long rip_linear;
3268
3269         if (!printk_ratelimit())
3270                 return;
3271
3272         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3273
3274         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3275
3276         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3277                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3278 }
3279 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3280
3281 static struct x86_emulate_ops emulate_ops = {
3282         .read_std            = kvm_read_guest_virt,
3283         .read_emulated       = emulator_read_emulated,
3284         .write_emulated      = emulator_write_emulated,
3285         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3286 };
3287
3288 static void cache_all_regs(struct kvm_vcpu *vcpu)
3289 {
3290         kvm_register_read(vcpu, VCPU_REGS_RAX);
3291         kvm_register_read(vcpu, VCPU_REGS_RSP);
3292         kvm_register_read(vcpu, VCPU_REGS_RIP);
3293         vcpu->arch.regs_dirty = ~0;
3294 }
3295
3296 int emulate_instruction(struct kvm_vcpu *vcpu,
3297                         unsigned long cr2,
3298                         u16 error_code,
3299                         int emulation_type)
3300 {
3301         int r, shadow_mask;
3302         struct decode_cache *c;
3303         struct kvm_run *run = vcpu->run;
3304
3305         kvm_clear_exception_queue(vcpu);
3306         vcpu->arch.mmio_fault_cr2 = cr2;
3307         /*
3308          * TODO: fix emulate.c to use guest_read/write_register
3309          * instead of direct ->regs accesses, can save hundred cycles
3310          * on Intel for instructions that don't read/change RSP, for
3311          * for example.
3312          */
3313         cache_all_regs(vcpu);
3314
3315         vcpu->mmio_is_write = 0;
3316         vcpu->arch.pio.string = 0;
3317
3318         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3319                 int cs_db, cs_l;
3320                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3321
3322                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3323                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3324                 vcpu->arch.emulate_ctxt.mode =
3325                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3326                         ? X86EMUL_MODE_REAL : cs_l
3327                         ? X86EMUL_MODE_PROT64 : cs_db
3328                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3329
3330                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3331
3332                 /* Only allow emulation of specific instructions on #UD
3333                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3334                 c = &vcpu->arch.emulate_ctxt.decode;
3335                 if (emulation_type & EMULTYPE_TRAP_UD) {
3336                         if (!c->twobyte)
3337                                 return EMULATE_FAIL;
3338                         switch (c->b) {
3339                         case 0x01: /* VMMCALL */
3340                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3341                                         return EMULATE_FAIL;
3342                                 break;
3343                         case 0x34: /* sysenter */
3344                         case 0x35: /* sysexit */
3345                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3346                                         return EMULATE_FAIL;
3347                                 break;
3348                         case 0x05: /* syscall */
3349                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3350                                         return EMULATE_FAIL;
3351                                 break;
3352                         default:
3353                                 return EMULATE_FAIL;
3354                         }
3355
3356                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3357                                 return EMULATE_FAIL;
3358                 }
3359
3360                 ++vcpu->stat.insn_emulation;
3361                 if (r)  {
3362                         ++vcpu->stat.insn_emulation_fail;
3363                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3364                                 return EMULATE_DONE;
3365                         return EMULATE_FAIL;
3366                 }
3367         }
3368
3369         if (emulation_type & EMULTYPE_SKIP) {
3370                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3371                 return EMULATE_DONE;
3372         }
3373
3374         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3375         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3376
3377         if (r == 0)
3378                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3379
3380         if (vcpu->arch.pio.string)
3381                 return EMULATE_DO_MMIO;
3382
3383         if ((r || vcpu->mmio_is_write) && run) {
3384                 run->exit_reason = KVM_EXIT_MMIO;
3385                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3386                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3387                 run->mmio.len = vcpu->mmio_size;
3388                 run->mmio.is_write = vcpu->mmio_is_write;
3389         }
3390
3391         if (r) {
3392                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3393                         return EMULATE_DONE;
3394                 if (!vcpu->mmio_needed) {
3395                         kvm_report_emulation_failure(vcpu, "mmio");
3396                         return EMULATE_FAIL;
3397                 }
3398                 return EMULATE_DO_MMIO;
3399         }
3400
3401         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3402
3403         if (vcpu->mmio_is_write) {
3404                 vcpu->mmio_needed = 0;
3405                 return EMULATE_DO_MMIO;
3406         }
3407
3408         return EMULATE_DONE;
3409 }
3410 EXPORT_SYMBOL_GPL(emulate_instruction);
3411
3412 static int pio_copy_data(struct kvm_vcpu *vcpu)
3413 {
3414         void *p = vcpu->arch.pio_data;
3415         gva_t q = vcpu->arch.pio.guest_gva;
3416         unsigned bytes;
3417         int ret;
3418
3419         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3420         if (vcpu->arch.pio.in)
3421                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3422         else
3423                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3424         return ret;
3425 }
3426
3427 int complete_pio(struct kvm_vcpu *vcpu)
3428 {
3429         struct kvm_pio_request *io = &vcpu->arch.pio;
3430         long delta;
3431         int r;
3432         unsigned long val;
3433
3434         if (!io->string) {
3435                 if (io->in) {
3436                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3437                         memcpy(&val, vcpu->arch.pio_data, io->size);
3438                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3439                 }
3440         } else {
3441                 if (io->in) {
3442                         r = pio_copy_data(vcpu);
3443                         if (r)
3444                                 return r;
3445                 }
3446
3447                 delta = 1;
3448                 if (io->rep) {
3449                         delta *= io->cur_count;
3450                         /*
3451                          * The size of the register should really depend on
3452                          * current address size.
3453                          */
3454                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3455                         val -= delta;
3456                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3457                 }
3458                 if (io->down)
3459                         delta = -delta;
3460                 delta *= io->size;
3461                 if (io->in) {
3462                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3463                         val += delta;
3464                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3465                 } else {
3466                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3467                         val += delta;
3468                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3469                 }
3470         }
3471
3472         io->count -= io->cur_count;
3473         io->cur_count = 0;
3474
3475         return 0;
3476 }
3477
3478 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3479 {
3480         /* TODO: String I/O for in kernel device */
3481         int r;
3482
3483         if (vcpu->arch.pio.in)
3484                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3485                                     vcpu->arch.pio.size, pd);
3486         else
3487                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3488                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3489                                      pd);
3490         return r;
3491 }
3492
3493 static int pio_string_write(struct kvm_vcpu *vcpu)
3494 {
3495         struct kvm_pio_request *io = &vcpu->arch.pio;
3496         void *pd = vcpu->arch.pio_data;
3497         int i, r = 0;
3498
3499         for (i = 0; i < io->cur_count; i++) {
3500                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3501                                      io->port, io->size, pd)) {
3502                         r = -EOPNOTSUPP;
3503                         break;
3504                 }
3505                 pd += io->size;
3506         }
3507         return r;
3508 }
3509
3510 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3511 {
3512         unsigned long val;
3513
3514         vcpu->run->exit_reason = KVM_EXIT_IO;
3515         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3516         vcpu->run->io.size = vcpu->arch.pio.size = size;
3517         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3518         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3519         vcpu->run->io.port = vcpu->arch.pio.port = port;
3520         vcpu->arch.pio.in = in;
3521         vcpu->arch.pio.string = 0;
3522         vcpu->arch.pio.down = 0;
3523         vcpu->arch.pio.rep = 0;
3524
3525         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3526                       size, 1);
3527
3528         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3529         memcpy(vcpu->arch.pio_data, &val, 4);
3530
3531         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3532                 complete_pio(vcpu);
3533                 return 1;
3534         }
3535         return 0;
3536 }
3537 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3538
3539 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3540                   int size, unsigned long count, int down,
3541                   gva_t address, int rep, unsigned port)
3542 {
3543         unsigned now, in_page;
3544         int ret = 0;
3545
3546         vcpu->run->exit_reason = KVM_EXIT_IO;
3547         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3548         vcpu->run->io.size = vcpu->arch.pio.size = size;
3549         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3550         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3551         vcpu->run->io.port = vcpu->arch.pio.port = port;
3552         vcpu->arch.pio.in = in;
3553         vcpu->arch.pio.string = 1;
3554         vcpu->arch.pio.down = down;
3555         vcpu->arch.pio.rep = rep;
3556
3557         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3558                       size, count);
3559
3560         if (!count) {
3561                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3562                 return 1;
3563         }
3564
3565         if (!down)
3566                 in_page = PAGE_SIZE - offset_in_page(address);
3567         else
3568                 in_page = offset_in_page(address) + size;
3569         now = min(count, (unsigned long)in_page / size);
3570         if (!now)
3571                 now = 1;
3572         if (down) {
3573                 /*
3574                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3575                  */
3576                 pr_unimpl(vcpu, "guest string pio down\n");
3577                 kvm_inject_gp(vcpu, 0);
3578                 return 1;
3579         }
3580         vcpu->run->io.count = now;
3581         vcpu->arch.pio.cur_count = now;
3582
3583         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3584                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3585
3586         vcpu->arch.pio.guest_gva = address;
3587
3588         if (!vcpu->arch.pio.in) {
3589                 /* string PIO write */
3590                 ret = pio_copy_data(vcpu);
3591                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3592                         kvm_inject_gp(vcpu, 0);
3593                         return 1;
3594                 }
3595                 if (ret == 0 && !pio_string_write(vcpu)) {
3596                         complete_pio(vcpu);
3597                         if (vcpu->arch.pio.count == 0)
3598                                 ret = 1;
3599                 }
3600         }
3601         /* no string PIO read support yet */
3602
3603         return ret;
3604 }
3605 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3606
3607 static void bounce_off(void *info)
3608 {
3609         /* nothing */
3610 }
3611
3612 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3613                                      void *data)
3614 {
3615         struct cpufreq_freqs *freq = data;
3616         struct kvm *kvm;
3617         struct kvm_vcpu *vcpu;
3618         int i, send_ipi = 0;
3619
3620         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3621                 return 0;
3622         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3623                 return 0;
3624         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3625
3626         spin_lock(&kvm_lock);
3627         list_for_each_entry(kvm, &vm_list, vm_list) {
3628                 kvm_for_each_vcpu(i, vcpu, kvm) {
3629                         if (vcpu->cpu != freq->cpu)
3630                                 continue;
3631                         if (!kvm_request_guest_time_update(vcpu))
3632                                 continue;
3633                         if (vcpu->cpu != smp_processor_id())
3634                                 send_ipi++;
3635                 }
3636         }
3637         spin_unlock(&kvm_lock);
3638
3639         if (freq->old < freq->new && send_ipi) {
3640                 /*
3641                  * We upscale the frequency.  Must make the guest
3642                  * doesn't see old kvmclock values while running with
3643                  * the new frequency, otherwise we risk the guest sees
3644                  * time go backwards.
3645                  *
3646                  * In case we update the frequency for another cpu
3647                  * (which might be in guest context) send an interrupt
3648                  * to kick the cpu out of guest context.  Next time
3649                  * guest context is entered kvmclock will be updated,
3650                  * so the guest will not see stale values.
3651                  */
3652                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3653         }
3654         return 0;
3655 }
3656
3657 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3658         .notifier_call  = kvmclock_cpufreq_notifier
3659 };
3660
3661 static void kvm_timer_init(void)
3662 {
3663         int cpu;
3664
3665         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3666                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3667                                           CPUFREQ_TRANSITION_NOTIFIER);
3668                 for_each_online_cpu(cpu) {
3669                         unsigned long khz = cpufreq_get(cpu);
3670                         if (!khz)
3671                                 khz = tsc_khz;
3672                         per_cpu(cpu_tsc_khz, cpu) = khz;
3673                 }
3674         } else {
3675                 for_each_possible_cpu(cpu)
3676                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3677         }
3678 }
3679
3680 int kvm_arch_init(void *opaque)
3681 {
3682         int r;
3683         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3684
3685         if (kvm_x86_ops) {
3686                 printk(KERN_ERR "kvm: already loaded the other module\n");
3687                 r = -EEXIST;
3688                 goto out;
3689         }
3690
3691         if (!ops->cpu_has_kvm_support()) {
3692                 printk(KERN_ERR "kvm: no hardware support\n");
3693                 r = -EOPNOTSUPP;
3694                 goto out;
3695         }
3696         if (ops->disabled_by_bios()) {
3697                 printk(KERN_ERR "kvm: disabled by bios\n");
3698                 r = -EOPNOTSUPP;
3699                 goto out;
3700         }
3701
3702         r = kvm_mmu_module_init();
3703         if (r)
3704                 goto out;
3705
3706         kvm_init_msr_list();
3707
3708         kvm_x86_ops = ops;
3709         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3710         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3711         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3712                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3713
3714         kvm_timer_init();
3715
3716         return 0;
3717
3718 out:
3719         return r;
3720 }
3721
3722 void kvm_arch_exit(void)
3723 {
3724         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3725                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3726                                             CPUFREQ_TRANSITION_NOTIFIER);
3727         kvm_x86_ops = NULL;
3728         kvm_mmu_module_exit();
3729 }
3730
3731 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3732 {
3733         ++vcpu->stat.halt_exits;
3734         if (irqchip_in_kernel(vcpu->kvm)) {
3735                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3736                 return 1;
3737         } else {
3738                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3739                 return 0;
3740         }
3741 }
3742 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3743
3744 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3745                            unsigned long a1)
3746 {
3747         if (is_long_mode(vcpu))
3748                 return a0;
3749         else
3750                 return a0 | ((gpa_t)a1 << 32);
3751 }
3752
3753 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3754 {
3755         u64 param, ingpa, outgpa, ret;
3756         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3757         bool fast, longmode;
3758         int cs_db, cs_l;
3759
3760         /*
3761          * hypercall generates UD from non zero cpl and real mode
3762          * per HYPER-V spec
3763          */
3764         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
3765             !kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
3766                 kvm_queue_exception(vcpu, UD_VECTOR);
3767                 return 0;
3768         }
3769
3770         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3771         longmode = is_long_mode(vcpu) && cs_l == 1;
3772
3773         if (!longmode) {
3774                 param = (kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3775                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffff);
3776                 ingpa = (kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3777                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffff);
3778                 outgpa = (kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3779                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffff);
3780         }
3781 #ifdef CONFIG_X86_64
3782         else {
3783                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3784                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3785                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3786         }
3787 #endif
3788
3789         code = param & 0xffff;
3790         fast = (param >> 16) & 0x1;
3791         rep_cnt = (param >> 32) & 0xfff;
3792         rep_idx = (param >> 48) & 0xfff;
3793
3794         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3795
3796         res = HV_STATUS_INVALID_HYPERCALL_CODE;
3797
3798         ret = res | (((u64)rep_done & 0xfff) << 32);
3799         if (longmode) {
3800                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3801         } else {
3802                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3803                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3804         }
3805
3806         return 1;
3807 }
3808
3809 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3810 {
3811         unsigned long nr, a0, a1, a2, a3, ret;
3812         int r = 1;
3813
3814         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3815                 return kvm_hv_hypercall(vcpu);
3816
3817         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3818         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3819         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3820         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3821         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3822
3823         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3824
3825         if (!is_long_mode(vcpu)) {
3826                 nr &= 0xFFFFFFFF;
3827                 a0 &= 0xFFFFFFFF;
3828                 a1 &= 0xFFFFFFFF;
3829                 a2 &= 0xFFFFFFFF;
3830                 a3 &= 0xFFFFFFFF;
3831         }
3832
3833         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3834                 ret = -KVM_EPERM;
3835                 goto out;
3836         }
3837
3838         switch (nr) {
3839         case KVM_HC_VAPIC_POLL_IRQ:
3840                 ret = 0;
3841                 break;
3842         case KVM_HC_MMU_OP:
3843                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3844                 break;
3845         default:
3846                 ret = -KVM_ENOSYS;
3847                 break;
3848         }
3849 out:
3850         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3851         ++vcpu->stat.hypercalls;
3852         return r;
3853 }
3854 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3855
3856 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3857 {
3858         char instruction[3];
3859         int ret = 0;
3860         unsigned long rip = kvm_rip_read(vcpu);
3861
3862
3863         /*
3864          * Blow out the MMU to ensure that no other VCPU has an active mapping
3865          * to ensure that the updated hypercall appears atomically across all
3866          * VCPUs.
3867          */
3868         kvm_mmu_zap_all(vcpu->kvm);
3869
3870         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3871         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3872             != X86EMUL_CONTINUE)
3873                 ret = -EFAULT;
3874
3875         return ret;
3876 }
3877
3878 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3879 {
3880         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3881 }
3882
3883 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3884 {
3885         struct descriptor_table dt = { limit, base };
3886
3887         kvm_x86_ops->set_gdt(vcpu, &dt);
3888 }
3889
3890 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3891 {
3892         struct descriptor_table dt = { limit, base };
3893
3894         kvm_x86_ops->set_idt(vcpu, &dt);
3895 }
3896
3897 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3898                    unsigned long *rflags)
3899 {
3900         kvm_lmsw(vcpu, msw);
3901         *rflags = kvm_get_rflags(vcpu);
3902 }
3903
3904 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3905 {
3906         unsigned long value;
3907
3908         switch (cr) {
3909         case 0:
3910                 value = kvm_read_cr0(vcpu);
3911                 break;
3912         case 2:
3913                 value = vcpu->arch.cr2;
3914                 break;
3915         case 3:
3916                 value = vcpu->arch.cr3;
3917                 break;
3918         case 4:
3919                 value = kvm_read_cr4(vcpu);
3920                 break;
3921         case 8:
3922                 value = kvm_get_cr8(vcpu);
3923                 break;
3924         default:
3925                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3926                 return 0;
3927         }
3928
3929         return value;
3930 }
3931
3932 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3933                      unsigned long *rflags)
3934 {
3935         switch (cr) {
3936         case 0:
3937                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3938                 *rflags = kvm_get_rflags(vcpu);
3939                 break;
3940         case 2:
3941                 vcpu->arch.cr2 = val;
3942                 break;
3943         case 3:
3944                 kvm_set_cr3(vcpu, val);
3945                 break;
3946         case 4:
3947                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3948                 break;
3949         case 8:
3950                 kvm_set_cr8(vcpu, val & 0xfUL);
3951                 break;
3952         default:
3953                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3954         }
3955 }
3956
3957 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3958 {
3959         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3960         int j, nent = vcpu->arch.cpuid_nent;
3961
3962         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3963         /* when no next entry is found, the current entry[i] is reselected */
3964         for (j = i + 1; ; j = (j + 1) % nent) {
3965                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3966                 if (ej->function == e->function) {
3967                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3968                         return j;
3969                 }
3970         }
3971         return 0; /* silence gcc, even though control never reaches here */
3972 }
3973
3974 /* find an entry with matching function, matching index (if needed), and that
3975  * should be read next (if it's stateful) */
3976 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3977         u32 function, u32 index)
3978 {
3979         if (e->function != function)
3980                 return 0;
3981         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3982                 return 0;
3983         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3984             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3985                 return 0;
3986         return 1;
3987 }
3988
3989 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3990                                               u32 function, u32 index)
3991 {
3992         int i;
3993         struct kvm_cpuid_entry2 *best = NULL;
3994
3995         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3996                 struct kvm_cpuid_entry2 *e;
3997
3998                 e = &vcpu->arch.cpuid_entries[i];
3999                 if (is_matching_cpuid_entry(e, function, index)) {
4000                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4001                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4002                         best = e;
4003                         break;
4004                 }
4005                 /*
4006                  * Both basic or both extended?
4007                  */
4008                 if (((e->function ^ function) & 0x80000000) == 0)
4009                         if (!best || e->function > best->function)
4010                                 best = e;
4011         }
4012         return best;
4013 }
4014 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4015
4016 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4017 {
4018         struct kvm_cpuid_entry2 *best;
4019
4020         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4021         if (best)
4022                 return best->eax & 0xff;
4023         return 36;
4024 }
4025
4026 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4027 {
4028         u32 function, index;
4029         struct kvm_cpuid_entry2 *best;
4030
4031         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4032         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4033         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4034         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4035         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4036         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4037         best = kvm_find_cpuid_entry(vcpu, function, index);
4038         if (best) {
4039                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4040                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4041                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4042                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4043         }
4044         kvm_x86_ops->skip_emulated_instruction(vcpu);
4045         trace_kvm_cpuid(function,
4046                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4047                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4048                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4049                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4050 }
4051 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4052
4053 /*
4054  * Check if userspace requested an interrupt window, and that the
4055  * interrupt window is open.
4056  *
4057  * No need to exit to userspace if we already have an interrupt queued.
4058  */
4059 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4060 {
4061         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4062                 vcpu->run->request_interrupt_window &&
4063                 kvm_arch_interrupt_allowed(vcpu));
4064 }
4065
4066 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4067 {
4068         struct kvm_run *kvm_run = vcpu->run;
4069
4070         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4071         kvm_run->cr8 = kvm_get_cr8(vcpu);
4072         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4073         if (irqchip_in_kernel(vcpu->kvm))
4074                 kvm_run->ready_for_interrupt_injection = 1;
4075         else
4076                 kvm_run->ready_for_interrupt_injection =
4077                         kvm_arch_interrupt_allowed(vcpu) &&
4078                         !kvm_cpu_has_interrupt(vcpu) &&
4079                         !kvm_event_needs_reinjection(vcpu);
4080 }
4081
4082 static void vapic_enter(struct kvm_vcpu *vcpu)
4083 {
4084         struct kvm_lapic *apic = vcpu->arch.apic;
4085         struct page *page;
4086
4087         if (!apic || !apic->vapic_addr)
4088                 return;
4089
4090         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4091
4092         vcpu->arch.apic->vapic_page = page;
4093 }
4094
4095 static void vapic_exit(struct kvm_vcpu *vcpu)
4096 {
4097         struct kvm_lapic *apic = vcpu->arch.apic;
4098         int idx;
4099
4100         if (!apic || !apic->vapic_addr)
4101                 return;
4102
4103         idx = srcu_read_lock(&vcpu->kvm->srcu);
4104         kvm_release_page_dirty(apic->vapic_page);
4105         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4106         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4107 }
4108
4109 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4110 {
4111         int max_irr, tpr;
4112
4113         if (!kvm_x86_ops->update_cr8_intercept)
4114                 return;
4115
4116         if (!vcpu->arch.apic)
4117                 return;
4118
4119         if (!vcpu->arch.apic->vapic_addr)
4120                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4121         else
4122                 max_irr = -1;
4123
4124         if (max_irr != -1)
4125                 max_irr >>= 4;
4126
4127         tpr = kvm_lapic_get_cr8(vcpu);
4128
4129         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4130 }
4131
4132 static void inject_pending_event(struct kvm_vcpu *vcpu)
4133 {
4134         /* try to reinject previous events if any */
4135         if (vcpu->arch.exception.pending) {
4136                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4137                                           vcpu->arch.exception.has_error_code,
4138                                           vcpu->arch.exception.error_code);
4139                 return;
4140         }
4141
4142         if (vcpu->arch.nmi_injected) {
4143                 kvm_x86_ops->set_nmi(vcpu);
4144                 return;
4145         }
4146
4147         if (vcpu->arch.interrupt.pending) {
4148                 kvm_x86_ops->set_irq(vcpu);
4149                 return;
4150         }
4151
4152         /* try to inject new event if pending */
4153         if (vcpu->arch.nmi_pending) {
4154                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4155                         vcpu->arch.nmi_pending = false;
4156                         vcpu->arch.nmi_injected = true;
4157                         kvm_x86_ops->set_nmi(vcpu);
4158                 }
4159         } else if (kvm_cpu_has_interrupt(vcpu)) {
4160                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4161                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4162                                             false);
4163                         kvm_x86_ops->set_irq(vcpu);
4164                 }
4165         }
4166 }
4167
4168 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4169 {
4170         int r;
4171         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4172                 vcpu->run->request_interrupt_window;
4173
4174         if (vcpu->requests)
4175                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4176                         kvm_mmu_unload(vcpu);
4177
4178         r = kvm_mmu_reload(vcpu);
4179         if (unlikely(r))
4180                 goto out;
4181
4182         if (vcpu->requests) {
4183                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4184                         __kvm_migrate_timers(vcpu);
4185                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4186                         kvm_write_guest_time(vcpu);
4187                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4188                         kvm_mmu_sync_roots(vcpu);
4189                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4190                         kvm_x86_ops->tlb_flush(vcpu);
4191                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4192                                        &vcpu->requests)) {
4193                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4194                         r = 0;
4195                         goto out;
4196                 }
4197                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4198                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4199                         r = 0;
4200                         goto out;
4201                 }
4202                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4203                         vcpu->fpu_active = 0;
4204                         kvm_x86_ops->fpu_deactivate(vcpu);
4205                 }
4206         }
4207
4208         preempt_disable();
4209
4210         kvm_x86_ops->prepare_guest_switch(vcpu);
4211         kvm_load_guest_fpu(vcpu);
4212
4213         local_irq_disable();
4214
4215         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4216         smp_mb__after_clear_bit();
4217
4218         if (vcpu->requests || need_resched() || signal_pending(current)) {
4219                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4220                 local_irq_enable();
4221                 preempt_enable();
4222                 r = 1;
4223                 goto out;
4224         }
4225
4226         inject_pending_event(vcpu);
4227
4228         /* enable NMI/IRQ window open exits if needed */
4229         if (vcpu->arch.nmi_pending)
4230                 kvm_x86_ops->enable_nmi_window(vcpu);
4231         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4232                 kvm_x86_ops->enable_irq_window(vcpu);
4233
4234         if (kvm_lapic_enabled(vcpu)) {
4235                 update_cr8_intercept(vcpu);
4236                 kvm_lapic_sync_to_vapic(vcpu);
4237         }
4238
4239         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4240
4241         kvm_guest_enter();
4242
4243         if (unlikely(vcpu->arch.switch_db_regs)) {
4244                 set_debugreg(0, 7);
4245                 set_debugreg(vcpu->arch.eff_db[0], 0);
4246                 set_debugreg(vcpu->arch.eff_db[1], 1);
4247                 set_debugreg(vcpu->arch.eff_db[2], 2);
4248                 set_debugreg(vcpu->arch.eff_db[3], 3);
4249         }
4250
4251         trace_kvm_entry(vcpu->vcpu_id);
4252         kvm_x86_ops->run(vcpu);
4253
4254         /*
4255          * If the guest has used debug registers, at least dr7
4256          * will be disabled while returning to the host.
4257          * If we don't have active breakpoints in the host, we don't
4258          * care about the messed up debug address registers. But if
4259          * we have some of them active, restore the old state.
4260          */
4261         if (hw_breakpoint_active())
4262                 hw_breakpoint_restore();
4263
4264         set_bit(KVM_REQ_KICK, &vcpu->requests);
4265         local_irq_enable();
4266
4267         ++vcpu->stat.exits;
4268
4269         /*
4270          * We must have an instruction between local_irq_enable() and
4271          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4272          * the interrupt shadow.  The stat.exits increment will do nicely.
4273          * But we need to prevent reordering, hence this barrier():
4274          */
4275         barrier();
4276
4277         kvm_guest_exit();
4278
4279         preempt_enable();
4280
4281         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4282
4283         /*
4284          * Profile KVM exit RIPs:
4285          */
4286         if (unlikely(prof_on == KVM_PROFILING)) {
4287                 unsigned long rip = kvm_rip_read(vcpu);
4288                 profile_hit(KVM_PROFILING, (void *)rip);
4289         }
4290
4291
4292         kvm_lapic_sync_from_vapic(vcpu);
4293
4294         r = kvm_x86_ops->handle_exit(vcpu);
4295 out:
4296         return r;
4297 }
4298
4299
4300 static int __vcpu_run(struct kvm_vcpu *vcpu)
4301 {
4302         int r;
4303         struct kvm *kvm = vcpu->kvm;
4304
4305         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4306                 pr_debug("vcpu %d received sipi with vector # %x\n",
4307                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4308                 kvm_lapic_reset(vcpu);
4309                 r = kvm_arch_vcpu_reset(vcpu);
4310                 if (r)
4311                         return r;
4312                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4313         }
4314
4315         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4316         vapic_enter(vcpu);
4317
4318         r = 1;
4319         while (r > 0) {
4320                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4321                         r = vcpu_enter_guest(vcpu);
4322                 else {
4323                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4324                         kvm_vcpu_block(vcpu);
4325                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4326                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4327                         {
4328                                 switch(vcpu->arch.mp_state) {
4329                                 case KVM_MP_STATE_HALTED:
4330                                         vcpu->arch.mp_state =
4331                                                 KVM_MP_STATE_RUNNABLE;
4332                                 case KVM_MP_STATE_RUNNABLE:
4333                                         break;
4334                                 case KVM_MP_STATE_SIPI_RECEIVED:
4335                                 default:
4336                                         r = -EINTR;
4337                                         break;
4338                                 }
4339                         }
4340                 }
4341
4342                 if (r <= 0)
4343                         break;
4344
4345                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4346                 if (kvm_cpu_has_pending_timer(vcpu))
4347                         kvm_inject_pending_timer_irqs(vcpu);
4348
4349                 if (dm_request_for_irq_injection(vcpu)) {
4350                         r = -EINTR;
4351                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4352                         ++vcpu->stat.request_irq_exits;
4353                 }
4354                 if (signal_pending(current)) {
4355                         r = -EINTR;
4356                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4357                         ++vcpu->stat.signal_exits;
4358                 }
4359                 if (need_resched()) {
4360                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4361                         kvm_resched(vcpu);
4362                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4363                 }
4364         }
4365
4366         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4367         post_kvm_run_save(vcpu);
4368
4369         vapic_exit(vcpu);
4370
4371         return r;
4372 }
4373
4374 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4375 {
4376         int r;
4377         sigset_t sigsaved;
4378
4379         vcpu_load(vcpu);
4380
4381         if (vcpu->sigset_active)
4382                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4383
4384         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4385                 kvm_vcpu_block(vcpu);
4386                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4387                 r = -EAGAIN;
4388                 goto out;
4389         }
4390
4391         /* re-sync apic's tpr */
4392         if (!irqchip_in_kernel(vcpu->kvm))
4393                 kvm_set_cr8(vcpu, kvm_run->cr8);
4394
4395         if (vcpu->arch.pio.cur_count) {
4396                 r = complete_pio(vcpu);
4397                 if (r)
4398                         goto out;
4399         }
4400         if (vcpu->mmio_needed) {
4401                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4402                 vcpu->mmio_read_completed = 1;
4403                 vcpu->mmio_needed = 0;
4404
4405                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4406                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4407                                         EMULTYPE_NO_DECODE);
4408                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4409                 if (r == EMULATE_DO_MMIO) {
4410                         /*
4411                          * Read-modify-write.  Back to userspace.
4412                          */
4413                         r = 0;
4414                         goto out;
4415                 }
4416         }
4417         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4418                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4419                                      kvm_run->hypercall.ret);
4420
4421         r = __vcpu_run(vcpu);
4422
4423 out:
4424         if (vcpu->sigset_active)
4425                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4426
4427         vcpu_put(vcpu);
4428         return r;
4429 }
4430
4431 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4432 {
4433         vcpu_load(vcpu);
4434
4435         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4436         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4437         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4438         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4439         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4440         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4441         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4442         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4443 #ifdef CONFIG_X86_64
4444         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4445         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4446         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4447         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4448         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4449         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4450         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4451         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4452 #endif
4453
4454         regs->rip = kvm_rip_read(vcpu);
4455         regs->rflags = kvm_get_rflags(vcpu);
4456
4457         vcpu_put(vcpu);
4458
4459         return 0;
4460 }
4461
4462 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4463 {
4464         vcpu_load(vcpu);
4465
4466         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4467         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4468         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4469         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4470         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4471         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4472         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4473         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4474 #ifdef CONFIG_X86_64
4475         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4476         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4477         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4478         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4479         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4480         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4481         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4482         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4483 #endif
4484
4485         kvm_rip_write(vcpu, regs->rip);
4486         kvm_set_rflags(vcpu, regs->rflags);
4487
4488         vcpu->arch.exception.pending = false;
4489
4490         vcpu_put(vcpu);
4491
4492         return 0;
4493 }
4494
4495 void kvm_get_segment(struct kvm_vcpu *vcpu,
4496                      struct kvm_segment *var, int seg)
4497 {
4498         kvm_x86_ops->get_segment(vcpu, var, seg);
4499 }
4500
4501 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4502 {
4503         struct kvm_segment cs;
4504
4505         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4506         *db = cs.db;
4507         *l = cs.l;
4508 }
4509 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4510
4511 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4512                                   struct kvm_sregs *sregs)
4513 {
4514         struct descriptor_table dt;
4515
4516         vcpu_load(vcpu);
4517
4518         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4519         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4520         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4521         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4522         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4523         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4524
4525         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4526         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4527
4528         kvm_x86_ops->get_idt(vcpu, &dt);
4529         sregs->idt.limit = dt.limit;
4530         sregs->idt.base = dt.base;
4531         kvm_x86_ops->get_gdt(vcpu, &dt);
4532         sregs->gdt.limit = dt.limit;
4533         sregs->gdt.base = dt.base;
4534
4535         sregs->cr0 = kvm_read_cr0(vcpu);
4536         sregs->cr2 = vcpu->arch.cr2;
4537         sregs->cr3 = vcpu->arch.cr3;
4538         sregs->cr4 = kvm_read_cr4(vcpu);
4539         sregs->cr8 = kvm_get_cr8(vcpu);
4540         sregs->efer = vcpu->arch.shadow_efer;
4541         sregs->apic_base = kvm_get_apic_base(vcpu);
4542
4543         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4544
4545         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4546                 set_bit(vcpu->arch.interrupt.nr,
4547                         (unsigned long *)sregs->interrupt_bitmap);
4548
4549         vcpu_put(vcpu);
4550
4551         return 0;
4552 }
4553
4554 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4555                                     struct kvm_mp_state *mp_state)
4556 {
4557         vcpu_load(vcpu);
4558         mp_state->mp_state = vcpu->arch.mp_state;
4559         vcpu_put(vcpu);
4560         return 0;
4561 }
4562
4563 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4564                                     struct kvm_mp_state *mp_state)
4565 {
4566         vcpu_load(vcpu);
4567         vcpu->arch.mp_state = mp_state->mp_state;
4568         vcpu_put(vcpu);
4569         return 0;
4570 }
4571
4572 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4573                         struct kvm_segment *var, int seg)
4574 {
4575         kvm_x86_ops->set_segment(vcpu, var, seg);
4576 }
4577
4578 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4579                                    struct kvm_segment *kvm_desct)
4580 {
4581         kvm_desct->base = get_desc_base(seg_desc);
4582         kvm_desct->limit = get_desc_limit(seg_desc);
4583         if (seg_desc->g) {
4584                 kvm_desct->limit <<= 12;
4585                 kvm_desct->limit |= 0xfff;
4586         }
4587         kvm_desct->selector = selector;
4588         kvm_desct->type = seg_desc->type;
4589         kvm_desct->present = seg_desc->p;
4590         kvm_desct->dpl = seg_desc->dpl;
4591         kvm_desct->db = seg_desc->d;
4592         kvm_desct->s = seg_desc->s;
4593         kvm_desct->l = seg_desc->l;
4594         kvm_desct->g = seg_desc->g;
4595         kvm_desct->avl = seg_desc->avl;
4596         if (!selector)
4597                 kvm_desct->unusable = 1;
4598         else
4599                 kvm_desct->unusable = 0;
4600         kvm_desct->padding = 0;
4601 }
4602
4603 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4604                                           u16 selector,
4605                                           struct descriptor_table *dtable)
4606 {
4607         if (selector & 1 << 2) {
4608                 struct kvm_segment kvm_seg;
4609
4610                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4611
4612                 if (kvm_seg.unusable)
4613                         dtable->limit = 0;
4614                 else
4615                         dtable->limit = kvm_seg.limit;
4616                 dtable->base = kvm_seg.base;
4617         }
4618         else
4619                 kvm_x86_ops->get_gdt(vcpu, dtable);
4620 }
4621
4622 /* allowed just for 8 bytes segments */
4623 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4624                                          struct desc_struct *seg_desc)
4625 {
4626         struct descriptor_table dtable;
4627         u16 index = selector >> 3;
4628
4629         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4630
4631         if (dtable.limit < index * 8 + 7) {
4632                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4633                 return 1;
4634         }
4635         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4636 }
4637
4638 /* allowed just for 8 bytes segments */
4639 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4640                                          struct desc_struct *seg_desc)
4641 {
4642         struct descriptor_table dtable;
4643         u16 index = selector >> 3;
4644
4645         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4646
4647         if (dtable.limit < index * 8 + 7)
4648                 return 1;
4649         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4650 }
4651
4652 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4653                              struct desc_struct *seg_desc)
4654 {
4655         u32 base_addr = get_desc_base(seg_desc);
4656
4657         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4658 }
4659
4660 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4661 {
4662         struct kvm_segment kvm_seg;
4663
4664         kvm_get_segment(vcpu, &kvm_seg, seg);
4665         return kvm_seg.selector;
4666 }
4667
4668 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4669                                                 u16 selector,
4670                                                 struct kvm_segment *kvm_seg)
4671 {
4672         struct desc_struct seg_desc;
4673
4674         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4675                 return 1;
4676         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4677         return 0;
4678 }
4679
4680 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4681 {
4682         struct kvm_segment segvar = {
4683                 .base = selector << 4,
4684                 .limit = 0xffff,
4685                 .selector = selector,
4686                 .type = 3,
4687                 .present = 1,
4688                 .dpl = 3,
4689                 .db = 0,
4690                 .s = 1,
4691                 .l = 0,
4692                 .g = 0,
4693                 .avl = 0,
4694                 .unusable = 0,
4695         };
4696         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4697         return 0;
4698 }
4699
4700 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4701 {
4702         return (seg != VCPU_SREG_LDTR) &&
4703                 (seg != VCPU_SREG_TR) &&
4704                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4705 }
4706
4707 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4708                                          u16 selector)
4709 {
4710         /* NULL selector is not valid for CS and SS */
4711         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4712                 if (!selector)
4713                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4714 }
4715
4716 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4717                                 int type_bits, int seg)
4718 {
4719         struct kvm_segment kvm_seg;
4720
4721         if (is_vm86_segment(vcpu, seg) || !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
4722                 return kvm_load_realmode_segment(vcpu, selector, seg);
4723         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4724                 return 1;
4725
4726         kvm_check_segment_descriptor(vcpu, seg, selector);
4727         kvm_seg.type |= type_bits;
4728
4729         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4730             seg != VCPU_SREG_LDTR)
4731                 if (!kvm_seg.s)
4732                         kvm_seg.unusable = 1;
4733
4734         kvm_set_segment(vcpu, &kvm_seg, seg);
4735         return 0;
4736 }
4737
4738 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4739                                 struct tss_segment_32 *tss)
4740 {
4741         tss->cr3 = vcpu->arch.cr3;
4742         tss->eip = kvm_rip_read(vcpu);
4743         tss->eflags = kvm_get_rflags(vcpu);
4744         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4745         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4746         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4747         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4748         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4749         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4750         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4751         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4752         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4753         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4754         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4755         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4756         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4757         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4758         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4759 }
4760
4761 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4762                                   struct tss_segment_32 *tss)
4763 {
4764         kvm_set_cr3(vcpu, tss->cr3);
4765
4766         kvm_rip_write(vcpu, tss->eip);
4767         kvm_set_rflags(vcpu, tss->eflags | 2);
4768
4769         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4770         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4771         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4772         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4773         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4774         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4775         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4776         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4777
4778         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4779                 return 1;
4780
4781         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4782                 return 1;
4783
4784         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4785                 return 1;
4786
4787         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4788                 return 1;
4789
4790         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4791                 return 1;
4792
4793         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4794                 return 1;
4795
4796         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4797                 return 1;
4798         return 0;
4799 }
4800
4801 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4802                                 struct tss_segment_16 *tss)
4803 {
4804         tss->ip = kvm_rip_read(vcpu);
4805         tss->flag = kvm_get_rflags(vcpu);
4806         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4807         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4808         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4809         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4810         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4811         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4812         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4813         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4814
4815         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4816         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4817         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4818         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4819         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4820 }
4821
4822 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4823                                  struct tss_segment_16 *tss)
4824 {
4825         kvm_rip_write(vcpu, tss->ip);
4826         kvm_set_rflags(vcpu, tss->flag | 2);
4827         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4828         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4829         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4830         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4831         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4832         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4833         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4834         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4835
4836         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4837                 return 1;
4838
4839         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4840                 return 1;
4841
4842         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4843                 return 1;
4844
4845         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4846                 return 1;
4847
4848         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4849                 return 1;
4850         return 0;
4851 }
4852
4853 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4854                               u16 old_tss_sel, u32 old_tss_base,
4855                               struct desc_struct *nseg_desc)
4856 {
4857         struct tss_segment_16 tss_segment_16;
4858         int ret = 0;
4859
4860         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4861                            sizeof tss_segment_16))
4862                 goto out;
4863
4864         save_state_to_tss16(vcpu, &tss_segment_16);
4865
4866         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4867                             sizeof tss_segment_16))
4868                 goto out;
4869
4870         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4871                            &tss_segment_16, sizeof tss_segment_16))
4872                 goto out;
4873
4874         if (old_tss_sel != 0xffff) {
4875                 tss_segment_16.prev_task_link = old_tss_sel;
4876
4877                 if (kvm_write_guest(vcpu->kvm,
4878                                     get_tss_base_addr(vcpu, nseg_desc),
4879                                     &tss_segment_16.prev_task_link,
4880                                     sizeof tss_segment_16.prev_task_link))
4881                         goto out;
4882         }
4883
4884         if (load_state_from_tss16(vcpu, &tss_segment_16))
4885                 goto out;
4886
4887         ret = 1;
4888 out:
4889         return ret;
4890 }
4891
4892 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4893                        u16 old_tss_sel, u32 old_tss_base,
4894                        struct desc_struct *nseg_desc)
4895 {
4896         struct tss_segment_32 tss_segment_32;
4897         int ret = 0;
4898
4899         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4900                            sizeof tss_segment_32))
4901                 goto out;
4902
4903         save_state_to_tss32(vcpu, &tss_segment_32);
4904
4905         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4906                             sizeof tss_segment_32))
4907                 goto out;
4908
4909         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4910                            &tss_segment_32, sizeof tss_segment_32))
4911                 goto out;
4912
4913         if (old_tss_sel != 0xffff) {
4914                 tss_segment_32.prev_task_link = old_tss_sel;
4915
4916                 if (kvm_write_guest(vcpu->kvm,
4917                                     get_tss_base_addr(vcpu, nseg_desc),
4918                                     &tss_segment_32.prev_task_link,
4919                                     sizeof tss_segment_32.prev_task_link))
4920                         goto out;
4921         }
4922
4923         if (load_state_from_tss32(vcpu, &tss_segment_32))
4924                 goto out;
4925
4926         ret = 1;
4927 out:
4928         return ret;
4929 }
4930
4931 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4932 {
4933         struct kvm_segment tr_seg;
4934         struct desc_struct cseg_desc;
4935         struct desc_struct nseg_desc;
4936         int ret = 0;
4937         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4938         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4939
4940         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4941
4942         /* FIXME: Handle errors. Failure to read either TSS or their
4943          * descriptors should generate a pagefault.
4944          */
4945         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4946                 goto out;
4947
4948         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4949                 goto out;
4950
4951         if (reason != TASK_SWITCH_IRET) {
4952                 int cpl;
4953
4954                 cpl = kvm_x86_ops->get_cpl(vcpu);
4955                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4956                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4957                         return 1;
4958                 }
4959         }
4960
4961         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4962                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4963                 return 1;
4964         }
4965
4966         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4967                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4968                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4969         }
4970
4971         if (reason == TASK_SWITCH_IRET) {
4972                 u32 eflags = kvm_get_rflags(vcpu);
4973                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4974         }
4975
4976         /* set back link to prev task only if NT bit is set in eflags
4977            note that old_tss_sel is not used afetr this point */
4978         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4979                 old_tss_sel = 0xffff;
4980
4981         if (nseg_desc.type & 8)
4982                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4983                                          old_tss_base, &nseg_desc);
4984         else
4985                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4986                                          old_tss_base, &nseg_desc);
4987
4988         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4989                 u32 eflags = kvm_get_rflags(vcpu);
4990                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4991         }
4992
4993         if (reason != TASK_SWITCH_IRET) {
4994                 nseg_desc.type |= (1 << 1);
4995                 save_guest_segment_descriptor(vcpu, tss_selector,
4996                                               &nseg_desc);
4997         }
4998
4999         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5000         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5001         tr_seg.type = 11;
5002         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5003 out:
5004         return ret;
5005 }
5006 EXPORT_SYMBOL_GPL(kvm_task_switch);
5007
5008 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5009                                   struct kvm_sregs *sregs)
5010 {
5011         int mmu_reset_needed = 0;
5012         int pending_vec, max_bits;
5013         struct descriptor_table dt;
5014
5015         vcpu_load(vcpu);
5016
5017         dt.limit = sregs->idt.limit;
5018         dt.base = sregs->idt.base;
5019         kvm_x86_ops->set_idt(vcpu, &dt);
5020         dt.limit = sregs->gdt.limit;
5021         dt.base = sregs->gdt.base;
5022         kvm_x86_ops->set_gdt(vcpu, &dt);
5023
5024         vcpu->arch.cr2 = sregs->cr2;
5025         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5026         vcpu->arch.cr3 = sregs->cr3;
5027
5028         kvm_set_cr8(vcpu, sregs->cr8);
5029
5030         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
5031         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5032         kvm_set_apic_base(vcpu, sregs->apic_base);
5033
5034         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5035         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5036         vcpu->arch.cr0 = sregs->cr0;
5037
5038         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5039         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5040         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5041                 load_pdptrs(vcpu, vcpu->arch.cr3);
5042                 mmu_reset_needed = 1;
5043         }
5044
5045         if (mmu_reset_needed)
5046                 kvm_mmu_reset_context(vcpu);
5047
5048         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5049         pending_vec = find_first_bit(
5050                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5051         if (pending_vec < max_bits) {
5052                 kvm_queue_interrupt(vcpu, pending_vec, false);
5053                 pr_debug("Set back pending irq %d\n", pending_vec);
5054                 if (irqchip_in_kernel(vcpu->kvm))
5055                         kvm_pic_clear_isr_ack(vcpu->kvm);
5056         }
5057
5058         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5059         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5060         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5061         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5062         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5063         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5064
5065         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5066         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5067
5068         update_cr8_intercept(vcpu);
5069
5070         /* Older userspace won't unhalt the vcpu on reset. */
5071         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5072             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5073             !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
5074                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5075
5076         vcpu_put(vcpu);
5077
5078         return 0;
5079 }
5080
5081 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5082                                         struct kvm_guest_debug *dbg)
5083 {
5084         unsigned long rflags;
5085         int i, r;
5086
5087         vcpu_load(vcpu);
5088
5089         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5090                 r = -EBUSY;
5091                 if (vcpu->arch.exception.pending)
5092                         goto unlock_out;
5093                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5094                         kvm_queue_exception(vcpu, DB_VECTOR);
5095                 else
5096                         kvm_queue_exception(vcpu, BP_VECTOR);
5097         }
5098
5099         /*
5100          * Read rflags as long as potentially injected trace flags are still
5101          * filtered out.
5102          */
5103         rflags = kvm_get_rflags(vcpu);
5104
5105         vcpu->guest_debug = dbg->control;
5106         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5107                 vcpu->guest_debug = 0;
5108
5109         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5110                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5111                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5112                 vcpu->arch.switch_db_regs =
5113                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5114         } else {
5115                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5116                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5117                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5118         }
5119
5120         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5121                 vcpu->arch.singlestep_cs =
5122                         get_segment_selector(vcpu, VCPU_SREG_CS);
5123                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5124         }
5125
5126         /*
5127          * Trigger an rflags update that will inject or remove the trace
5128          * flags.
5129          */
5130         kvm_set_rflags(vcpu, rflags);
5131
5132         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5133
5134         r = 0;
5135
5136 unlock_out:
5137         vcpu_put(vcpu);
5138
5139         return r;
5140 }
5141
5142 /*
5143  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5144  * we have asm/x86/processor.h
5145  */
5146 struct fxsave {
5147         u16     cwd;
5148         u16     swd;
5149         u16     twd;
5150         u16     fop;
5151         u64     rip;
5152         u64     rdp;
5153         u32     mxcsr;
5154         u32     mxcsr_mask;
5155         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5156 #ifdef CONFIG_X86_64
5157         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5158 #else
5159         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5160 #endif
5161 };
5162
5163 /*
5164  * Translate a guest virtual address to a guest physical address.
5165  */
5166 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5167                                     struct kvm_translation *tr)
5168 {
5169         unsigned long vaddr = tr->linear_address;
5170         gpa_t gpa;
5171         int idx;
5172
5173         vcpu_load(vcpu);
5174         idx = srcu_read_lock(&vcpu->kvm->srcu);
5175         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
5176         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5177         tr->physical_address = gpa;
5178         tr->valid = gpa != UNMAPPED_GVA;
5179         tr->writeable = 1;
5180         tr->usermode = 0;
5181         vcpu_put(vcpu);
5182
5183         return 0;
5184 }
5185
5186 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5187 {
5188         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5189
5190         vcpu_load(vcpu);
5191
5192         memcpy(fpu->fpr, fxsave->st_space, 128);
5193         fpu->fcw = fxsave->cwd;
5194         fpu->fsw = fxsave->swd;
5195         fpu->ftwx = fxsave->twd;
5196         fpu->last_opcode = fxsave->fop;
5197         fpu->last_ip = fxsave->rip;
5198         fpu->last_dp = fxsave->rdp;
5199         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5200
5201         vcpu_put(vcpu);
5202
5203         return 0;
5204 }
5205
5206 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5207 {
5208         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5209
5210         vcpu_load(vcpu);
5211
5212         memcpy(fxsave->st_space, fpu->fpr, 128);
5213         fxsave->cwd = fpu->fcw;
5214         fxsave->swd = fpu->fsw;
5215         fxsave->twd = fpu->ftwx;
5216         fxsave->fop = fpu->last_opcode;
5217         fxsave->rip = fpu->last_ip;
5218         fxsave->rdp = fpu->last_dp;
5219         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5220
5221         vcpu_put(vcpu);
5222
5223         return 0;
5224 }
5225
5226 void fx_init(struct kvm_vcpu *vcpu)
5227 {
5228         unsigned after_mxcsr_mask;
5229
5230         /*
5231          * Touch the fpu the first time in non atomic context as if
5232          * this is the first fpu instruction the exception handler
5233          * will fire before the instruction returns and it'll have to
5234          * allocate ram with GFP_KERNEL.
5235          */
5236         if (!used_math())
5237                 kvm_fx_save(&vcpu->arch.host_fx_image);
5238
5239         /* Initialize guest FPU by resetting ours and saving into guest's */
5240         preempt_disable();
5241         kvm_fx_save(&vcpu->arch.host_fx_image);
5242         kvm_fx_finit();
5243         kvm_fx_save(&vcpu->arch.guest_fx_image);
5244         kvm_fx_restore(&vcpu->arch.host_fx_image);
5245         preempt_enable();
5246
5247         vcpu->arch.cr0 |= X86_CR0_ET;
5248         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5249         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5250         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5251                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5252 }
5253 EXPORT_SYMBOL_GPL(fx_init);
5254
5255 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5256 {
5257         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
5258                 return;
5259
5260         vcpu->guest_fpu_loaded = 1;
5261         kvm_fx_save(&vcpu->arch.host_fx_image);
5262         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5263 }
5264 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
5265
5266 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5267 {
5268         if (!vcpu->guest_fpu_loaded)
5269                 return;
5270
5271         vcpu->guest_fpu_loaded = 0;
5272         kvm_fx_save(&vcpu->arch.guest_fx_image);
5273         kvm_fx_restore(&vcpu->arch.host_fx_image);
5274         ++vcpu->stat.fpu_reload;
5275         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5276 }
5277 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
5278
5279 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5280 {
5281         if (vcpu->arch.time_page) {
5282                 kvm_release_page_dirty(vcpu->arch.time_page);
5283                 vcpu->arch.time_page = NULL;
5284         }
5285
5286         kvm_x86_ops->vcpu_free(vcpu);
5287 }
5288
5289 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5290                                                 unsigned int id)
5291 {
5292         return kvm_x86_ops->vcpu_create(kvm, id);
5293 }
5294
5295 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5296 {
5297         int r;
5298
5299         /* We do fxsave: this must be aligned. */
5300         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5301
5302         vcpu->arch.mtrr_state.have_fixed = 1;
5303         vcpu_load(vcpu);
5304         r = kvm_arch_vcpu_reset(vcpu);
5305         if (r == 0)
5306                 r = kvm_mmu_setup(vcpu);
5307         vcpu_put(vcpu);
5308         if (r < 0)
5309                 goto free_vcpu;
5310
5311         return 0;
5312 free_vcpu:
5313         kvm_x86_ops->vcpu_free(vcpu);
5314         return r;
5315 }
5316
5317 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5318 {
5319         vcpu_load(vcpu);
5320         kvm_mmu_unload(vcpu);
5321         vcpu_put(vcpu);
5322
5323         kvm_x86_ops->vcpu_free(vcpu);
5324 }
5325
5326 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5327 {
5328         vcpu->arch.nmi_pending = false;
5329         vcpu->arch.nmi_injected = false;
5330
5331         vcpu->arch.switch_db_regs = 0;
5332         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5333         vcpu->arch.dr6 = DR6_FIXED_1;
5334         vcpu->arch.dr7 = DR7_FIXED_1;
5335
5336         return kvm_x86_ops->vcpu_reset(vcpu);
5337 }
5338
5339 int kvm_arch_hardware_enable(void *garbage)
5340 {
5341         /*
5342          * Since this may be called from a hotplug notifcation,
5343          * we can't get the CPU frequency directly.
5344          */
5345         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5346                 int cpu = raw_smp_processor_id();
5347                 per_cpu(cpu_tsc_khz, cpu) = 0;
5348         }
5349
5350         kvm_shared_msr_cpu_online();
5351
5352         return kvm_x86_ops->hardware_enable(garbage);
5353 }
5354
5355 void kvm_arch_hardware_disable(void *garbage)
5356 {
5357         kvm_x86_ops->hardware_disable(garbage);
5358         drop_user_return_notifiers(garbage);
5359 }
5360
5361 int kvm_arch_hardware_setup(void)
5362 {
5363         return kvm_x86_ops->hardware_setup();
5364 }
5365
5366 void kvm_arch_hardware_unsetup(void)
5367 {
5368         kvm_x86_ops->hardware_unsetup();
5369 }
5370
5371 void kvm_arch_check_processor_compat(void *rtn)
5372 {
5373         kvm_x86_ops->check_processor_compatibility(rtn);
5374 }
5375
5376 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5377 {
5378         struct page *page;
5379         struct kvm *kvm;
5380         int r;
5381
5382         BUG_ON(vcpu->kvm == NULL);
5383         kvm = vcpu->kvm;
5384
5385         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5386         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5387                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5388         else
5389                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5390
5391         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5392         if (!page) {
5393                 r = -ENOMEM;
5394                 goto fail;
5395         }
5396         vcpu->arch.pio_data = page_address(page);
5397
5398         r = kvm_mmu_create(vcpu);
5399         if (r < 0)
5400                 goto fail_free_pio_data;
5401
5402         if (irqchip_in_kernel(kvm)) {
5403                 r = kvm_create_lapic(vcpu);
5404                 if (r < 0)
5405                         goto fail_mmu_destroy;
5406         }
5407
5408         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5409                                        GFP_KERNEL);
5410         if (!vcpu->arch.mce_banks) {
5411                 r = -ENOMEM;
5412                 goto fail_free_lapic;
5413         }
5414         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5415
5416         return 0;
5417 fail_free_lapic:
5418         kvm_free_lapic(vcpu);
5419 fail_mmu_destroy:
5420         kvm_mmu_destroy(vcpu);
5421 fail_free_pio_data:
5422         free_page((unsigned long)vcpu->arch.pio_data);
5423 fail:
5424         return r;
5425 }
5426
5427 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5428 {
5429         int idx;
5430
5431         kfree(vcpu->arch.mce_banks);
5432         kvm_free_lapic(vcpu);
5433         idx = srcu_read_lock(&vcpu->kvm->srcu);
5434         kvm_mmu_destroy(vcpu);
5435         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5436         free_page((unsigned long)vcpu->arch.pio_data);
5437 }
5438
5439 struct  kvm *kvm_arch_create_vm(void)
5440 {
5441         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5442
5443         if (!kvm)
5444                 return ERR_PTR(-ENOMEM);
5445
5446         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5447         if (!kvm->arch.aliases) {
5448                 kfree(kvm);
5449                 return ERR_PTR(-ENOMEM);
5450         }
5451
5452         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5453         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5454
5455         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5456         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5457
5458         rdtscll(kvm->arch.vm_init_tsc);
5459
5460         return kvm;
5461 }
5462
5463 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5464 {
5465         vcpu_load(vcpu);
5466         kvm_mmu_unload(vcpu);
5467         vcpu_put(vcpu);
5468 }
5469
5470 static void kvm_free_vcpus(struct kvm *kvm)
5471 {
5472         unsigned int i;
5473         struct kvm_vcpu *vcpu;
5474
5475         /*
5476          * Unpin any mmu pages first.
5477          */
5478         kvm_for_each_vcpu(i, vcpu, kvm)
5479                 kvm_unload_vcpu_mmu(vcpu);
5480         kvm_for_each_vcpu(i, vcpu, kvm)
5481                 kvm_arch_vcpu_free(vcpu);
5482
5483         mutex_lock(&kvm->lock);
5484         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5485                 kvm->vcpus[i] = NULL;
5486
5487         atomic_set(&kvm->online_vcpus, 0);
5488         mutex_unlock(&kvm->lock);
5489 }
5490
5491 void kvm_arch_sync_events(struct kvm *kvm)
5492 {
5493         kvm_free_all_assigned_devices(kvm);
5494 }
5495
5496 void kvm_arch_destroy_vm(struct kvm *kvm)
5497 {
5498         kvm_iommu_unmap_guest(kvm);
5499         kvm_free_pit(kvm);
5500         kfree(kvm->arch.vpic);
5501         kfree(kvm->arch.vioapic);
5502         kvm_free_vcpus(kvm);
5503         kvm_free_physmem(kvm);
5504         if (kvm->arch.apic_access_page)
5505                 put_page(kvm->arch.apic_access_page);
5506         if (kvm->arch.ept_identity_pagetable)
5507                 put_page(kvm->arch.ept_identity_pagetable);
5508         kfree(kvm->arch.aliases);
5509         kfree(kvm);
5510 }
5511
5512 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5513                                 struct kvm_memory_slot *memslot,
5514                                 struct kvm_memory_slot old,
5515                                 struct kvm_userspace_memory_region *mem,
5516                                 int user_alloc)
5517 {
5518         int npages = memslot->npages;
5519
5520         /*To keep backward compatibility with older userspace,
5521          *x86 needs to hanlde !user_alloc case.
5522          */
5523         if (!user_alloc) {
5524                 if (npages && !old.rmap) {
5525                         unsigned long userspace_addr;
5526
5527                         down_write(&current->mm->mmap_sem);
5528                         userspace_addr = do_mmap(NULL, 0,
5529                                                  npages * PAGE_SIZE,
5530                                                  PROT_READ | PROT_WRITE,
5531                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5532                                                  0);
5533                         up_write(&current->mm->mmap_sem);
5534
5535                         if (IS_ERR((void *)userspace_addr))
5536                                 return PTR_ERR((void *)userspace_addr);
5537
5538                         memslot->userspace_addr = userspace_addr;
5539                 }
5540         }
5541
5542
5543         return 0;
5544 }
5545
5546 void kvm_arch_commit_memory_region(struct kvm *kvm,
5547                                 struct kvm_userspace_memory_region *mem,
5548                                 struct kvm_memory_slot old,
5549                                 int user_alloc)
5550 {
5551
5552         int npages = mem->memory_size >> PAGE_SHIFT;
5553
5554         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5555                 int ret;
5556
5557                 down_write(&current->mm->mmap_sem);
5558                 ret = do_munmap(current->mm, old.userspace_addr,
5559                                 old.npages * PAGE_SIZE);
5560                 up_write(&current->mm->mmap_sem);
5561                 if (ret < 0)
5562                         printk(KERN_WARNING
5563                                "kvm_vm_ioctl_set_memory_region: "
5564                                "failed to munmap memory\n");
5565         }
5566
5567         spin_lock(&kvm->mmu_lock);
5568         if (!kvm->arch.n_requested_mmu_pages) {
5569                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5570                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5571         }
5572
5573         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5574         spin_unlock(&kvm->mmu_lock);
5575 }
5576
5577 void kvm_arch_flush_shadow(struct kvm *kvm)
5578 {
5579         kvm_mmu_zap_all(kvm);
5580         kvm_reload_remote_mmus(kvm);
5581 }
5582
5583 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5584 {
5585         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5586                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5587                 || vcpu->arch.nmi_pending ||
5588                 (kvm_arch_interrupt_allowed(vcpu) &&
5589                  kvm_cpu_has_interrupt(vcpu));
5590 }
5591
5592 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5593 {
5594         int me;
5595         int cpu = vcpu->cpu;
5596
5597         if (waitqueue_active(&vcpu->wq)) {
5598                 wake_up_interruptible(&vcpu->wq);
5599                 ++vcpu->stat.halt_wakeup;
5600         }
5601
5602         me = get_cpu();
5603         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5604                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5605                         smp_send_reschedule(cpu);
5606         put_cpu();
5607 }
5608
5609 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5610 {
5611         return kvm_x86_ops->interrupt_allowed(vcpu);
5612 }
5613
5614 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5615 {
5616         unsigned long rflags;
5617
5618         rflags = kvm_x86_ops->get_rflags(vcpu);
5619         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5620                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5621         return rflags;
5622 }
5623 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5624
5625 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5626 {
5627         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5628             vcpu->arch.singlestep_cs ==
5629                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5630             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5631                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5632         kvm_x86_ops->set_rflags(vcpu, rflags);
5633 }
5634 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5635
5636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5646 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);