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     5
 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         HV_X64_MSR_APIC_ASSIST_PAGE,
 630         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 631         MSR_K6_STAR,
 632 #ifdef CONFIG_X86_64
 633         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 634 #endif
 635         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 636 };
 637
 638 static unsigned num_msrs_to_save;
 639
 640 static u32 emulated_msrs[] = {
 641         MSR_IA32_MISC_ENABLE,
 642 };
 643
 644 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 645 {
 646         if (efer & efer_reserved_bits) {
 647                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 648                        efer);
 649                 kvm_inject_gp(vcpu, 0);
 650                 return;
 651         }
 652
 653         if (is_paging(vcpu)
 654             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 655                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 656                 kvm_inject_gp(vcpu, 0);
 657                 return;
 658         }
 659
 660         if (efer & EFER_FFXSR) {
 661                 struct kvm_cpuid_entry2 *feat;
 662
 663                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 664                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 665                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 666                         kvm_inject_gp(vcpu, 0);
 667                         return;
 668                 }
 669         }
 670
 671         if (efer & EFER_SVME) {
 672                 struct kvm_cpuid_entry2 *feat;
 673
 674                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 675                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 676                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 677                         kvm_inject_gp(vcpu, 0);
 678                         return;
 679                 }
 680         }
 681
 682         kvm_x86_ops->set_efer(vcpu, efer);
 683
 684         efer &= ~EFER_LMA;
 685         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 686
 687         vcpu->arch.shadow_efer = efer;
 688
 689         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 690         kvm_mmu_reset_context(vcpu);
 691 }
 692
 693 void kvm_enable_efer_bits(u64 mask)
 694 {
 695        efer_reserved_bits &= ~mask;
 696 }
 697 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 698
 699
 700 /*
 701  * Writes msr value into into the appropriate "register".
 702  * Returns 0 on success, non-0 otherwise.
 703  * Assumes vcpu_load() was already called.
 704  */
 705 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 706 {
 707         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 708 }
 709
 710 /*
 711  * Adapt set_msr() to msr_io()'s calling convention
 712  */
 713 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 714 {
 715         return kvm_set_msr(vcpu, index, *data);
 716 }
 717
 718 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 719 {
 720         static int version;
 721         struct pvclock_wall_clock wc;
 722         struct timespec boot;
 723
 724         if (!wall_clock)
 725                 return;
 726
 727         version++;
 728
 729         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 730
 731         /*
 732          * The guest calculates current wall clock time by adding
 733          * system time (updated by kvm_write_guest_time below) to the
 734          * wall clock specified here.  guest system time equals host
 735          * system time for us, thus we must fill in host boot time here.
 736          */
 737         getboottime(&boot);
 738
 739         wc.sec = boot.tv_sec;
 740         wc.nsec = boot.tv_nsec;
 741         wc.version = version;
 742
 743         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 744
 745         version++;
 746         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 747 }
 748
 749 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 750 {
 751         uint32_t quotient, remainder;
 752
 753         /* Don't try to replace with do_div(), this one calculates
 754          * "(dividend << 32) / divisor" */
 755         __asm__ ( "divl %4"
 756                   : "=a" (quotient), "=d" (remainder)
 757                   : "0" (0), "1" (dividend), "r" (divisor) );
 758         return quotient;
 759 }
 760
 761 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 762 {
 763         uint64_t nsecs = 1000000000LL;
 764         int32_t  shift = 0;
 765         uint64_t tps64;
 766         uint32_t tps32;
 767
 768         tps64 = tsc_khz * 1000LL;
 769         while (tps64 > nsecs*2) {
 770                 tps64 >>= 1;
 771                 shift--;
 772         }
 773
 774         tps32 = (uint32_t)tps64;
 775         while (tps32 <= (uint32_t)nsecs) {
 776                 tps32 <<= 1;
 777                 shift++;
 778         }
 779
 780         hv_clock->tsc_shift = shift;
 781         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 782
 783         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 784                  __func__, tsc_khz, hv_clock->tsc_shift,
 785                  hv_clock->tsc_to_system_mul);
 786 }
 787
 788 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 789
 790 static void kvm_write_guest_time(struct kvm_vcpu *v)
 791 {
 792         struct timespec ts;
 793         unsigned long flags;
 794         struct kvm_vcpu_arch *vcpu = &v->arch;
 795         void *shared_kaddr;
 796         unsigned long this_tsc_khz;
 797
 798         if ((!vcpu->time_page))
 799                 return;
 800
 801         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 802         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 803                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 804                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 805         }
 806         put_cpu_var(cpu_tsc_khz);
 807
 808         /* Keep irq disabled to prevent changes to the clock */
 809         local_irq_save(flags);
 810         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 811         ktime_get_ts(&ts);
 812         monotonic_to_bootbased(&ts);
 813         local_irq_restore(flags);
 814
 815         /* With all the info we got, fill in the values */
 816
 817         vcpu->hv_clock.system_time = ts.tv_nsec +
 818                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 819
 820         /*
 821          * The interface expects us to write an even number signaling that the
 822          * update is finished. Since the guest won't see the intermediate
 823          * state, we just increase by 2 at the end.
 824          */
 825         vcpu->hv_clock.version += 2;
 826
 827         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 828
 829         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 830                sizeof(vcpu->hv_clock));
 831
 832         kunmap_atomic(shared_kaddr, KM_USER0);
 833
 834         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 835 }
 836
 837 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 838 {
 839         struct kvm_vcpu_arch *vcpu = &v->arch;
 840
 841         if (!vcpu->time_page)
 842                 return 0;
 843         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 844         return 1;
 845 }
 846
 847 static bool msr_mtrr_valid(unsigned msr)
 848 {
 849         switch (msr) {
 850         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 851         case MSR_MTRRfix64K_00000:
 852         case MSR_MTRRfix16K_80000:
 853         case MSR_MTRRfix16K_A0000:
 854         case MSR_MTRRfix4K_C0000:
 855         case MSR_MTRRfix4K_C8000:
 856         case MSR_MTRRfix4K_D0000:
 857         case MSR_MTRRfix4K_D8000:
 858         case MSR_MTRRfix4K_E0000:
 859         case MSR_MTRRfix4K_E8000:
 860         case MSR_MTRRfix4K_F0000:
 861         case MSR_MTRRfix4K_F8000:
 862         case MSR_MTRRdefType:
 863         case MSR_IA32_CR_PAT:
 864                 return true;
 865         case 0x2f8:
 866                 return true;
 867         }
 868         return false;
 869 }
 870
 871 static bool valid_pat_type(unsigned t)
 872 {
 873         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 874 }
 875
 876 static bool valid_mtrr_type(unsigned t)
 877 {
 878         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 879 }
 880
 881 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 882 {
 883         int i;
 884
 885         if (!msr_mtrr_valid(msr))
 886                 return false;
 887
 888         if (msr == MSR_IA32_CR_PAT) {
 889                 for (i = 0; i < 8; i++)
 890                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 891                                 return false;
 892                 return true;
 893         } else if (msr == MSR_MTRRdefType) {
 894                 if (data & ~0xcff)
 895                         return false;
 896                 return valid_mtrr_type(data & 0xff);
 897         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 898                 for (i = 0; i < 8 ; i++)
 899                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 900                                 return false;
 901                 return true;
 902         }
 903
 904         /* variable MTRRs */
 905         return valid_mtrr_type(data & 0xff);
 906 }
 907
 908 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 909 {
 910         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 911
 912         if (!mtrr_valid(vcpu, msr, data))
 913                 return 1;
 914
 915         if (msr == MSR_MTRRdefType) {
 916                 vcpu->arch.mtrr_state.def_type = data;
 917                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 918         } else if (msr == MSR_MTRRfix64K_00000)
 919                 p[0] = data;
 920         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 921                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 922         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 923                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 924         else if (msr == MSR_IA32_CR_PAT)
 925                 vcpu->arch.pat = data;
 926         else {  /* Variable MTRRs */
 927                 int idx, is_mtrr_mask;
 928                 u64 *pt;
 929
 930                 idx = (msr - 0x200) / 2;
 931                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 932                 if (!is_mtrr_mask)
 933                         pt =
 934                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 935                 else
 936                         pt =
 937                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 938                 *pt = data;
 939         }
 940
 941         kvm_mmu_reset_context(vcpu);
 942         return 0;
 943 }
 944
 945 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 946 {
 947         u64 mcg_cap = vcpu->arch.mcg_cap;
 948         unsigned bank_num = mcg_cap & 0xff;
 949
 950         switch (msr) {
 951         case MSR_IA32_MCG_STATUS:
 952                 vcpu->arch.mcg_status = data;
 953                 break;
 954         case MSR_IA32_MCG_CTL:
 955                 if (!(mcg_cap & MCG_CTL_P))
 956                         return 1;
 957                 if (data != 0 && data != ~(u64)0)
 958                         return -1;
 959                 vcpu->arch.mcg_ctl = data;
 960                 break;
 961         default:
 962                 if (msr >= MSR_IA32_MC0_CTL &&
 963                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 964                         u32 offset = msr - MSR_IA32_MC0_CTL;
 965                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 966                         if ((offset & 0x3) == 0 &&
 967                             data != 0 && data != ~(u64)0)
 968                                 return -1;
 969                         vcpu->arch.mce_banks[offset] = data;
 970                         break;
 971                 }
 972                 return 1;
 973         }
 974         return 0;
 975 }
 976
 977 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 978 {
 979         struct kvm *kvm = vcpu->kvm;
 980         int lm = is_long_mode(vcpu);
 981         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 982                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 983         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 984                 : kvm->arch.xen_hvm_config.blob_size_32;
 985         u32 page_num = data & ~PAGE_MASK;
 986         u64 page_addr = data & PAGE_MASK;
 987         u8 *page;
 988         int r;
 989
 990         r = -E2BIG;
 991         if (page_num >= blob_size)
 992                 goto out;
 993         r = -ENOMEM;
 994         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 995         if (!page)
 996                 goto out;
 997         r = -EFAULT;
 998         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 999                 goto out_free;
1000         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1001                 goto out_free;
1002         r = 0;
1003 out_free:
1004         kfree(page);
1005 out:
1006         return r;
1007 }
1008
1009 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1010 {
1011         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1012 }
1013
1014 static bool kvm_hv_msr_partition_wide(u32 msr)
1015 {
1016         bool r = false;
1017         switch (msr) {
1018         case HV_X64_MSR_GUEST_OS_ID:
1019         case HV_X64_MSR_HYPERCALL:
1020                 r = true;
1021                 break;
1022         }
1023
1024         return r;
1025 }
1026
1027 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1028 {
1029         struct kvm *kvm = vcpu->kvm;
1030
1031         switch (msr) {
1032         case HV_X64_MSR_GUEST_OS_ID:
1033                 kvm->arch.hv_guest_os_id = data;
1034                 /* setting guest os id to zero disables hypercall page */
1035                 if (!kvm->arch.hv_guest_os_id)
1036                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1037                 break;
1038         case HV_X64_MSR_HYPERCALL: {
1039                 u64 gfn;
1040                 unsigned long addr;
1041                 u8 instructions[4];
1042
1043                 /* if guest os id is not set hypercall should remain disabled */
1044                 if (!kvm->arch.hv_guest_os_id)
1045                         break;
1046                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1047                         kvm->arch.hv_hypercall = data;
1048                         break;
1049                 }
1050                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1051                 addr = gfn_to_hva(kvm, gfn);
1052                 if (kvm_is_error_hva(addr))
1053                         return 1;
1054                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1055                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1056                 if (copy_to_user((void __user *)addr, instructions, 4))
1057                         return 1;
1058                 kvm->arch.hv_hypercall = data;
1059                 break;
1060         }
1061         default:
1062                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1063                           "data 0x%llx\n", msr, data);
1064                 return 1;
1065         }
1066         return 0;
1067 }
1068
1069 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1070 {
1071         switch (msr) {
1072         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1073                 unsigned long addr;
1074
1075                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1076                         vcpu->arch.hv_vapic = data;
1077                         break;
1078                 }
1079                 addr = gfn_to_hva(vcpu->kvm, data >>
1080                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1081                 if (kvm_is_error_hva(addr))
1082                         return 1;
1083                 if (clear_user((void __user *)addr, PAGE_SIZE))
1084                         return 1;
1085                 vcpu->arch.hv_vapic = data;
1086                 break;
1087         }
1088         case HV_X64_MSR_EOI:
1089                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1090         case HV_X64_MSR_ICR:
1091                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1092         case HV_X64_MSR_TPR:
1093                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1094         default:
1095                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1096                           "data 0x%llx\n", msr, data);
1097                 return 1;
1098         }
1099
1100         return 0;
1101 }
1102
1103 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1104 {
1105         switch (msr) {
1106         case MSR_EFER:
1107                 set_efer(vcpu, data);
1108                 break;
1109         case MSR_K7_HWCR:
1110                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1111                 if (data != 0) {
1112                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1113                                 data);
1114                         return 1;
1115                 }
1116                 break;
1117         case MSR_FAM10H_MMIO_CONF_BASE:
1118                 if (data != 0) {
1119                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1120                                 "0x%llx\n", data);
1121                         return 1;
1122                 }
1123                 break;
1124         case MSR_AMD64_NB_CFG:
1125                 break;
1126         case MSR_IA32_DEBUGCTLMSR:
1127                 if (!data) {
1128                         /* We support the non-activated case already */
1129                         break;
1130                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1131                         /* Values other than LBR and BTF are vendor-specific,
1132                            thus reserved and should throw a #GP */
1133                         return 1;
1134                 }
1135                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1136                         __func__, data);
1137                 break;
1138         case MSR_IA32_UCODE_REV:
1139         case MSR_IA32_UCODE_WRITE:
1140         case MSR_VM_HSAVE_PA:
1141         case MSR_AMD64_PATCH_LOADER:
1142                 break;
1143         case 0x200 ... 0x2ff:
1144                 return set_msr_mtrr(vcpu, msr, data);
1145         case MSR_IA32_APICBASE:
1146                 kvm_set_apic_base(vcpu, data);
1147                 break;
1148         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1149                 return kvm_x2apic_msr_write(vcpu, msr, data);
1150         case MSR_IA32_MISC_ENABLE:
1151                 vcpu->arch.ia32_misc_enable_msr = data;
1152                 break;
1153         case MSR_KVM_WALL_CLOCK:
1154                 vcpu->kvm->arch.wall_clock = data;
1155                 kvm_write_wall_clock(vcpu->kvm, data);
1156                 break;
1157         case MSR_KVM_SYSTEM_TIME: {
1158                 if (vcpu->arch.time_page) {
1159                         kvm_release_page_dirty(vcpu->arch.time_page);
1160                         vcpu->arch.time_page = NULL;
1161                 }
1162
1163                 vcpu->arch.time = data;
1164
1165                 /* we verify if the enable bit is set... */
1166                 if (!(data & 1))
1167                         break;
1168
1169                 /* ...but clean it before doing the actual write */
1170                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1171
1172                 vcpu->arch.time_page =
1173                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1174
1175                 if (is_error_page(vcpu->arch.time_page)) {
1176                         kvm_release_page_clean(vcpu->arch.time_page);
1177                         vcpu->arch.time_page = NULL;
1178                 }
1179
1180                 kvm_request_guest_time_update(vcpu);
1181                 break;
1182         }
1183         case MSR_IA32_MCG_CTL:
1184         case MSR_IA32_MCG_STATUS:
1185         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1186                 return set_msr_mce(vcpu, msr, data);
1187
1188         /* Performance counters are not protected by a CPUID bit,
1189          * so we should check all of them in the generic path for the sake of
1190          * cross vendor migration.
1191          * Writing a zero into the event select MSRs disables them,
1192          * which we perfectly emulate ;-). Any other value should be at least
1193          * reported, some guests depend on them.
1194          */
1195         case MSR_P6_EVNTSEL0:
1196         case MSR_P6_EVNTSEL1:
1197         case MSR_K7_EVNTSEL0:
1198         case MSR_K7_EVNTSEL1:
1199         case MSR_K7_EVNTSEL2:
1200         case MSR_K7_EVNTSEL3:
1201                 if (data != 0)
1202                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1203                                 "0x%x data 0x%llx\n", msr, data);
1204                 break;
1205         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1206          * so we ignore writes to make it happy.
1207          */
1208         case MSR_P6_PERFCTR0:
1209         case MSR_P6_PERFCTR1:
1210         case MSR_K7_PERFCTR0:
1211         case MSR_K7_PERFCTR1:
1212         case MSR_K7_PERFCTR2:
1213         case MSR_K7_PERFCTR3:
1214                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1215                         "0x%x data 0x%llx\n", msr, data);
1216                 break;
1217         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1218                 if (kvm_hv_msr_partition_wide(msr)) {
1219                         int r;
1220                         mutex_lock(&vcpu->kvm->lock);
1221                         r = set_msr_hyperv_pw(vcpu, msr, data);
1222                         mutex_unlock(&vcpu->kvm->lock);
1223                         return r;
1224                 } else
1225                         return set_msr_hyperv(vcpu, msr, data);
1226                 break;
1227         default:
1228                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1229                         return xen_hvm_config(vcpu, data);
1230                 if (!ignore_msrs) {
1231                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1232                                 msr, data);
1233                         return 1;
1234                 } else {
1235                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1236                                 msr, data);
1237                         break;
1238                 }
1239         }
1240         return 0;
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1243
1244
1245 /*
1246  * Reads an msr value (of 'msr_index') into 'pdata'.
1247  * Returns 0 on success, non-0 otherwise.
1248  * Assumes vcpu_load() was already called.
1249  */
1250 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1251 {
1252         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1253 }
1254
1255 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1256 {
1257         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1258
1259         if (!msr_mtrr_valid(msr))
1260                 return 1;
1261
1262         if (msr == MSR_MTRRdefType)
1263                 *pdata = vcpu->arch.mtrr_state.def_type +
1264                          (vcpu->arch.mtrr_state.enabled << 10);
1265         else if (msr == MSR_MTRRfix64K_00000)
1266                 *pdata = p[0];
1267         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1268                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1269         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1270                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1271         else if (msr == MSR_IA32_CR_PAT)
1272                 *pdata = vcpu->arch.pat;
1273         else {  /* Variable MTRRs */
1274                 int idx, is_mtrr_mask;
1275                 u64 *pt;
1276
1277                 idx = (msr - 0x200) / 2;
1278                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1279                 if (!is_mtrr_mask)
1280                         pt =
1281                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1282                 else
1283                         pt =
1284                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1285                 *pdata = *pt;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1292 {
1293         u64 data;
1294         u64 mcg_cap = vcpu->arch.mcg_cap;
1295         unsigned bank_num = mcg_cap & 0xff;
1296
1297         switch (msr) {
1298         case MSR_IA32_P5_MC_ADDR:
1299         case MSR_IA32_P5_MC_TYPE:
1300                 data = 0;
1301                 break;
1302         case MSR_IA32_MCG_CAP:
1303                 data = vcpu->arch.mcg_cap;
1304                 break;
1305         case MSR_IA32_MCG_CTL:
1306                 if (!(mcg_cap & MCG_CTL_P))
1307                         return 1;
1308                 data = vcpu->arch.mcg_ctl;
1309                 break;
1310         case MSR_IA32_MCG_STATUS:
1311                 data = vcpu->arch.mcg_status;
1312                 break;
1313         default:
1314                 if (msr >= MSR_IA32_MC0_CTL &&
1315                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1316                         u32 offset = msr - MSR_IA32_MC0_CTL;
1317                         data = vcpu->arch.mce_banks[offset];
1318                         break;
1319                 }
1320                 return 1;
1321         }
1322         *pdata = data;
1323         return 0;
1324 }
1325
1326 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1327 {
1328         u64 data = 0;
1329         struct kvm *kvm = vcpu->kvm;
1330
1331         switch (msr) {
1332         case HV_X64_MSR_GUEST_OS_ID:
1333                 data = kvm->arch.hv_guest_os_id;
1334                 break;
1335         case HV_X64_MSR_HYPERCALL:
1336                 data = kvm->arch.hv_hypercall;
1337                 break;
1338         default:
1339                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1340                 return 1;
1341         }
1342
1343         *pdata = data;
1344         return 0;
1345 }
1346
1347 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1348 {
1349         u64 data = 0;
1350
1351         switch (msr) {
1352         case HV_X64_MSR_VP_INDEX: {
1353                 int r;
1354                 struct kvm_vcpu *v;
1355                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1356                         if (v == vcpu)
1357                                 data = r;
1358                 break;
1359         }
1360         case HV_X64_MSR_EOI:
1361                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1362         case HV_X64_MSR_ICR:
1363                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1364         case HV_X64_MSR_TPR:
1365                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1366         default:
1367                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1368                 return 1;
1369         }
1370         *pdata = data;
1371         return 0;
1372 }
1373
1374 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1375 {
1376         u64 data;
1377
1378         switch (msr) {
1379         case MSR_IA32_PLATFORM_ID:
1380         case MSR_IA32_UCODE_REV:
1381         case MSR_IA32_EBL_CR_POWERON:
1382         case MSR_IA32_DEBUGCTLMSR:
1383         case MSR_IA32_LASTBRANCHFROMIP:
1384         case MSR_IA32_LASTBRANCHTOIP:
1385         case MSR_IA32_LASTINTFROMIP:
1386         case MSR_IA32_LASTINTTOIP:
1387         case MSR_K8_SYSCFG:
1388         case MSR_K7_HWCR:
1389         case MSR_VM_HSAVE_PA:
1390         case MSR_P6_PERFCTR0:
1391         case MSR_P6_PERFCTR1:
1392         case MSR_P6_EVNTSEL0:
1393         case MSR_P6_EVNTSEL1:
1394         case MSR_K7_EVNTSEL0:
1395         case MSR_K7_PERFCTR0:
1396         case MSR_K8_INT_PENDING_MSG:
1397         case MSR_AMD64_NB_CFG:
1398         case MSR_FAM10H_MMIO_CONF_BASE:
1399                 data = 0;
1400                 break;
1401         case MSR_MTRRcap:
1402                 data = 0x500 | KVM_NR_VAR_MTRR;
1403                 break;
1404         case 0x200 ... 0x2ff:
1405                 return get_msr_mtrr(vcpu, msr, pdata);
1406         case 0xcd: /* fsb frequency */
1407                 data = 3;
1408                 break;
1409         case MSR_IA32_APICBASE:
1410                 data = kvm_get_apic_base(vcpu);
1411                 break;
1412         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1413                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1414                 break;
1415         case MSR_IA32_MISC_ENABLE:
1416                 data = vcpu->arch.ia32_misc_enable_msr;
1417                 break;
1418         case MSR_IA32_PERF_STATUS:
1419                 /* TSC increment by tick */
1420                 data = 1000ULL;
1421                 /* CPU multiplier */
1422                 data |= (((uint64_t)4ULL) << 40);
1423                 break;
1424         case MSR_EFER:
1425                 data = vcpu->arch.shadow_efer;
1426                 break;
1427         case MSR_KVM_WALL_CLOCK:
1428                 data = vcpu->kvm->arch.wall_clock;
1429                 break;
1430         case MSR_KVM_SYSTEM_TIME:
1431                 data = vcpu->arch.time;
1432                 break;
1433         case MSR_IA32_P5_MC_ADDR:
1434         case MSR_IA32_P5_MC_TYPE:
1435         case MSR_IA32_MCG_CAP:
1436         case MSR_IA32_MCG_CTL:
1437         case MSR_IA32_MCG_STATUS:
1438         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1439                 return get_msr_mce(vcpu, msr, pdata);
1440         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1441                 if (kvm_hv_msr_partition_wide(msr)) {
1442                         int r;
1443                         mutex_lock(&vcpu->kvm->lock);
1444                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1445                         mutex_unlock(&vcpu->kvm->lock);
1446                         return r;
1447                 } else
1448                         return get_msr_hyperv(vcpu, msr, pdata);
1449                 break;
1450         default:
1451                 if (!ignore_msrs) {
1452                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1453                         return 1;
1454                 } else {
1455                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1456                         data = 0;
1457                 }
1458                 break;
1459         }
1460         *pdata = data;
1461         return 0;
1462 }
1463 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1464
1465 /*
1466  * Read or write a bunch of msrs. All parameters are kernel addresses.
1467  *
1468  * @return number of msrs set successfully.
1469  */
1470 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1471                     struct kvm_msr_entry *entries,
1472                     int (*do_msr)(struct kvm_vcpu *vcpu,
1473                                   unsigned index, u64 *data))
1474 {
1475         int i, idx;
1476
1477         vcpu_load(vcpu);
1478
1479         idx = srcu_read_lock(&vcpu->kvm->srcu);
1480         for (i = 0; i < msrs->nmsrs; ++i)
1481                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1482                         break;
1483         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1484
1485         vcpu_put(vcpu);
1486
1487         return i;
1488 }
1489
1490 /*
1491  * Read or write a bunch of msrs. Parameters are user addresses.
1492  *
1493  * @return number of msrs set successfully.
1494  */
1495 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1496                   int (*do_msr)(struct kvm_vcpu *vcpu,
1497                                 unsigned index, u64 *data),
1498                   int writeback)
1499 {
1500         struct kvm_msrs msrs;
1501         struct kvm_msr_entry *entries;
1502         int r, n;
1503         unsigned size;
1504
1505         r = -EFAULT;
1506         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1507                 goto out;
1508
1509         r = -E2BIG;
1510         if (msrs.nmsrs >= MAX_IO_MSRS)
1511                 goto out;
1512
1513         r = -ENOMEM;
1514         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1515         entries = vmalloc(size);
1516         if (!entries)
1517                 goto out;
1518
1519         r = -EFAULT;
1520         if (copy_from_user(entries, user_msrs->entries, size))
1521                 goto out_free;
1522
1523         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1524         if (r < 0)
1525                 goto out_free;
1526
1527         r = -EFAULT;
1528         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1529                 goto out_free;
1530
1531         r = n;
1532
1533 out_free:
1534         vfree(entries);
1535 out:
1536         return r;
1537 }
1538
1539 int kvm_dev_ioctl_check_extension(long ext)
1540 {
1541         int r;
1542
1543         switch (ext) {
1544         case KVM_CAP_IRQCHIP:
1545         case KVM_CAP_HLT:
1546         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1547         case KVM_CAP_SET_TSS_ADDR:
1548         case KVM_CAP_EXT_CPUID:
1549         case KVM_CAP_CLOCKSOURCE:
1550         case KVM_CAP_PIT:
1551         case KVM_CAP_NOP_IO_DELAY:
1552         case KVM_CAP_MP_STATE:
1553         case KVM_CAP_SYNC_MMU:
1554         case KVM_CAP_REINJECT_CONTROL:
1555         case KVM_CAP_IRQ_INJECT_STATUS:
1556         case KVM_CAP_ASSIGN_DEV_IRQ:
1557         case KVM_CAP_IRQFD:
1558         case KVM_CAP_IOEVENTFD:
1559         case KVM_CAP_PIT2:
1560         case KVM_CAP_PIT_STATE2:
1561         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1562         case KVM_CAP_XEN_HVM:
1563         case KVM_CAP_ADJUST_CLOCK:
1564         case KVM_CAP_VCPU_EVENTS:
1565         case KVM_CAP_HYPERV:
1566         case KVM_CAP_HYPERV_VAPIC:
1567         case KVM_CAP_HYPERV_SPIN:
1568                 r = 1;
1569                 break;
1570         case KVM_CAP_COALESCED_MMIO:
1571                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1572                 break;
1573         case KVM_CAP_VAPIC:
1574                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1575                 break;
1576         case KVM_CAP_NR_VCPUS:
1577                 r = KVM_MAX_VCPUS;
1578                 break;
1579         case KVM_CAP_NR_MEMSLOTS:
1580                 r = KVM_MEMORY_SLOTS;
1581                 break;
1582         case KVM_CAP_PV_MMU:    /* obsolete */
1583                 r = 0;
1584                 break;
1585         case KVM_CAP_IOMMU:
1586                 r = iommu_found();
1587                 break;
1588         case KVM_CAP_MCE:
1589                 r = KVM_MAX_MCE_BANKS;
1590                 break;
1591         default:
1592                 r = 0;
1593                 break;
1594         }
1595         return r;
1596
1597 }
1598
1599 long kvm_arch_dev_ioctl(struct file *filp,
1600                         unsigned int ioctl, unsigned long arg)
1601 {
1602         void __user *argp = (void __user *)arg;
1603         long r;
1604
1605         switch (ioctl) {
1606         case KVM_GET_MSR_INDEX_LIST: {
1607                 struct kvm_msr_list __user *user_msr_list = argp;
1608                 struct kvm_msr_list msr_list;
1609                 unsigned n;
1610
1611                 r = -EFAULT;
1612                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1613                         goto out;
1614                 n = msr_list.nmsrs;
1615                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1616                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1617                         goto out;
1618                 r = -E2BIG;
1619                 if (n < msr_list.nmsrs)
1620                         goto out;
1621                 r = -EFAULT;
1622                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1623                                  num_msrs_to_save * sizeof(u32)))
1624                         goto out;
1625                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1626                                  &emulated_msrs,
1627                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1628                         goto out;
1629                 r = 0;
1630                 break;
1631         }
1632         case KVM_GET_SUPPORTED_CPUID: {
1633                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1634                 struct kvm_cpuid2 cpuid;
1635
1636                 r = -EFAULT;
1637                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1638                         goto out;
1639                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1640                                                       cpuid_arg->entries);
1641                 if (r)
1642                         goto out;
1643
1644                 r = -EFAULT;
1645                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1646                         goto out;
1647                 r = 0;
1648                 break;
1649         }
1650         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1651                 u64 mce_cap;
1652
1653                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1654                 r = -EFAULT;
1655                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1656                         goto out;
1657                 r = 0;
1658                 break;
1659         }
1660         default:
1661                 r = -EINVAL;
1662         }
1663 out:
1664         return r;
1665 }
1666
1667 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1668 {
1669         kvm_x86_ops->vcpu_load(vcpu, cpu);
1670         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1671                 unsigned long khz = cpufreq_quick_get(cpu);
1672                 if (!khz)
1673                         khz = tsc_khz;
1674                 per_cpu(cpu_tsc_khz, cpu) = khz;
1675         }
1676         kvm_request_guest_time_update(vcpu);
1677 }
1678
1679 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1680 {
1681         kvm_put_guest_fpu(vcpu);
1682         kvm_x86_ops->vcpu_put(vcpu);
1683 }
1684
1685 static int is_efer_nx(void)
1686 {
1687         unsigned long long efer = 0;
1688
1689         rdmsrl_safe(MSR_EFER, &efer);
1690         return efer & EFER_NX;
1691 }
1692
1693 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1694 {
1695         int i;
1696         struct kvm_cpuid_entry2 *e, *entry;
1697
1698         entry = NULL;
1699         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1700                 e = &vcpu->arch.cpuid_entries[i];
1701                 if (e->function == 0x80000001) {
1702                         entry = e;
1703                         break;
1704                 }
1705         }
1706         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1707                 entry->edx &= ~(1 << 20);
1708                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1709         }
1710 }
1711
1712 /* when an old userspace process fills a new kernel module */
1713 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1714                                     struct kvm_cpuid *cpuid,
1715                                     struct kvm_cpuid_entry __user *entries)
1716 {
1717         int r, i;
1718         struct kvm_cpuid_entry *cpuid_entries;
1719
1720         r = -E2BIG;
1721         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1722                 goto out;
1723         r = -ENOMEM;
1724         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1725         if (!cpuid_entries)
1726                 goto out;
1727         r = -EFAULT;
1728         if (copy_from_user(cpuid_entries, entries,
1729                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1730                 goto out_free;
1731         for (i = 0; i < cpuid->nent; i++) {
1732                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1733                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1734                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1735                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1736                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1737                 vcpu->arch.cpuid_entries[i].index = 0;
1738                 vcpu->arch.cpuid_entries[i].flags = 0;
1739                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1740                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1741                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1742         }
1743         vcpu->arch.cpuid_nent = cpuid->nent;
1744         cpuid_fix_nx_cap(vcpu);
1745         r = 0;
1746         kvm_apic_set_version(vcpu);
1747         kvm_x86_ops->cpuid_update(vcpu);
1748
1749 out_free:
1750         vfree(cpuid_entries);
1751 out:
1752         return r;
1753 }
1754
1755 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1756                                      struct kvm_cpuid2 *cpuid,
1757                                      struct kvm_cpuid_entry2 __user *entries)
1758 {
1759         int r;
1760
1761         r = -E2BIG;
1762         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1763                 goto out;
1764         r = -EFAULT;
1765         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1766                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1767                 goto out;
1768         vcpu->arch.cpuid_nent = cpuid->nent;
1769         kvm_apic_set_version(vcpu);
1770         kvm_x86_ops->cpuid_update(vcpu);
1771         return 0;
1772
1773 out:
1774         return r;
1775 }
1776
1777 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1778                                      struct kvm_cpuid2 *cpuid,
1779                                      struct kvm_cpuid_entry2 __user *entries)
1780 {
1781         int r;
1782
1783         r = -E2BIG;
1784         if (cpuid->nent < vcpu->arch.cpuid_nent)
1785                 goto out;
1786         r = -EFAULT;
1787         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1788                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1789                 goto out;
1790         return 0;
1791
1792 out:
1793         cpuid->nent = vcpu->arch.cpuid_nent;
1794         return r;
1795 }
1796
1797 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1798                            u32 index)
1799 {
1800         entry->function = function;
1801         entry->index = index;
1802         cpuid_count(entry->function, entry->index,
1803                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1804         entry->flags = 0;
1805 }
1806
1807 #define F(x) bit(X86_FEATURE_##x)
1808
1809 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1810                          u32 index, int *nent, int maxnent)
1811 {
1812         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1813 #ifdef CONFIG_X86_64
1814         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1815                                 ? F(GBPAGES) : 0;
1816         unsigned f_lm = F(LM);
1817 #else
1818         unsigned f_gbpages = 0;
1819         unsigned f_lm = 0;
1820 #endif
1821         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1822
1823         /* cpuid 1.edx */
1824         const u32 kvm_supported_word0_x86_features =
1825                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1826                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1827                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1828                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1829                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1830                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1831                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1832                 0 /* HTT, TM, Reserved, PBE */;
1833         /* cpuid 0x80000001.edx */
1834         const u32 kvm_supported_word1_x86_features =
1835                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1836                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1837                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1838                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1839                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1840                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1841                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1842                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1843         /* cpuid 1.ecx */
1844         const u32 kvm_supported_word4_x86_features =
1845                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1846                 0 /* DS-CPL, VMX, SMX, EST */ |
1847                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1848                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1849                 0 /* Reserved, DCA */ | F(XMM4_1) |
1850                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1851                 0 /* Reserved, XSAVE, OSXSAVE */;
1852         /* cpuid 0x80000001.ecx */
1853         const u32 kvm_supported_word6_x86_features =
1854                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1855                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1856                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1857                 0 /* SKINIT */ | 0 /* WDT */;
1858
1859         /* all calls to cpuid_count() should be made on the same cpu */
1860         get_cpu();
1861         do_cpuid_1_ent(entry, function, index);
1862         ++*nent;
1863
1864         switch (function) {
1865         case 0:
1866                 entry->eax = min(entry->eax, (u32)0xb);
1867                 break;
1868         case 1:
1869                 entry->edx &= kvm_supported_word0_x86_features;
1870                 entry->ecx &= kvm_supported_word4_x86_features;
1871                 /* we support x2apic emulation even if host does not support
1872                  * it since we emulate x2apic in software */
1873                 entry->ecx |= F(X2APIC);
1874                 break;
1875         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1876          * may return different values. This forces us to get_cpu() before
1877          * issuing the first command, and also to emulate this annoying behavior
1878          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1879         case 2: {
1880                 int t, times = entry->eax & 0xff;
1881
1882                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1883                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1884                 for (t = 1; t < times && *nent < maxnent; ++t) {
1885                         do_cpuid_1_ent(&entry[t], function, 0);
1886                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1887                         ++*nent;
1888                 }
1889                 break;
1890         }
1891         /* function 4 and 0xb have additional index. */
1892         case 4: {
1893                 int i, cache_type;
1894
1895                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1896                 /* read more entries until cache_type is zero */
1897                 for (i = 1; *nent < maxnent; ++i) {
1898                         cache_type = entry[i - 1].eax & 0x1f;
1899                         if (!cache_type)
1900                                 break;
1901                         do_cpuid_1_ent(&entry[i], function, i);
1902                         entry[i].flags |=
1903                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1904                         ++*nent;
1905                 }
1906                 break;
1907         }
1908         case 0xb: {
1909                 int i, level_type;
1910
1911                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1912                 /* read more entries until level_type is zero */
1913                 for (i = 1; *nent < maxnent; ++i) {
1914                         level_type = entry[i - 1].ecx & 0xff00;
1915                         if (!level_type)
1916                                 break;
1917                         do_cpuid_1_ent(&entry[i], function, i);
1918                         entry[i].flags |=
1919                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1920                         ++*nent;
1921                 }
1922                 break;
1923         }
1924         case 0x80000000:
1925                 entry->eax = min(entry->eax, 0x8000001a);
1926                 break;
1927         case 0x80000001:
1928                 entry->edx &= kvm_supported_word1_x86_features;
1929                 entry->ecx &= kvm_supported_word6_x86_features;
1930                 break;
1931         }
1932         put_cpu();
1933 }
1934
1935 #undef F
1936
1937 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1938                                      struct kvm_cpuid_entry2 __user *entries)
1939 {
1940         struct kvm_cpuid_entry2 *cpuid_entries;
1941         int limit, nent = 0, r = -E2BIG;
1942         u32 func;
1943
1944         if (cpuid->nent < 1)
1945                 goto out;
1946         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1947                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1948         r = -ENOMEM;
1949         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1950         if (!cpuid_entries)
1951                 goto out;
1952
1953         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1954         limit = cpuid_entries[0].eax;
1955         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1956                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1957                              &nent, cpuid->nent);
1958         r = -E2BIG;
1959         if (nent >= cpuid->nent)
1960                 goto out_free;
1961
1962         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1963         limit = cpuid_entries[nent - 1].eax;
1964         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1965                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1966                              &nent, cpuid->nent);
1967         r = -E2BIG;
1968         if (nent >= cpuid->nent)
1969                 goto out_free;
1970
1971         r = -EFAULT;
1972         if (copy_to_user(entries, cpuid_entries,
1973                          nent * sizeof(struct kvm_cpuid_entry2)))
1974                 goto out_free;
1975         cpuid->nent = nent;
1976         r = 0;
1977
1978 out_free:
1979         vfree(cpuid_entries);
1980 out:
1981         return r;
1982 }
1983
1984 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1985                                     struct kvm_lapic_state *s)
1986 {
1987         vcpu_load(vcpu);
1988         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1989         vcpu_put(vcpu);
1990
1991         return 0;
1992 }
1993
1994 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1995                                     struct kvm_lapic_state *s)
1996 {
1997         vcpu_load(vcpu);
1998         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1999         kvm_apic_post_state_restore(vcpu);
2000         update_cr8_intercept(vcpu);
2001         vcpu_put(vcpu);
2002
2003         return 0;
2004 }
2005
2006 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2007                                     struct kvm_interrupt *irq)
2008 {
2009         if (irq->irq < 0 || irq->irq >= 256)
2010                 return -EINVAL;
2011         if (irqchip_in_kernel(vcpu->kvm))
2012                 return -ENXIO;
2013         vcpu_load(vcpu);
2014
2015         kvm_queue_interrupt(vcpu, irq->irq, false);
2016
2017         vcpu_put(vcpu);
2018
2019         return 0;
2020 }
2021
2022 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2023 {
2024         vcpu_load(vcpu);
2025         kvm_inject_nmi(vcpu);
2026         vcpu_put(vcpu);
2027
2028         return 0;
2029 }
2030
2031 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2032                                            struct kvm_tpr_access_ctl *tac)
2033 {
2034         if (tac->flags)
2035                 return -EINVAL;
2036         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2037         return 0;
2038 }
2039
2040 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2041                                         u64 mcg_cap)
2042 {
2043         int r;
2044         unsigned bank_num = mcg_cap & 0xff, bank;
2045
2046         r = -EINVAL;
2047         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2048                 goto out;
2049         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2050                 goto out;
2051         r = 0;
2052         vcpu->arch.mcg_cap = mcg_cap;
2053         /* Init IA32_MCG_CTL to all 1s */
2054         if (mcg_cap & MCG_CTL_P)
2055                 vcpu->arch.mcg_ctl = ~(u64)0;
2056         /* Init IA32_MCi_CTL to all 1s */
2057         for (bank = 0; bank < bank_num; bank++)
2058                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2059 out:
2060         return r;
2061 }
2062
2063 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2064                                       struct kvm_x86_mce *mce)
2065 {
2066         u64 mcg_cap = vcpu->arch.mcg_cap;
2067         unsigned bank_num = mcg_cap & 0xff;
2068         u64 *banks = vcpu->arch.mce_banks;
2069
2070         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2071                 return -EINVAL;
2072         /*
2073          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2074          * reporting is disabled
2075          */
2076         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2077             vcpu->arch.mcg_ctl != ~(u64)0)
2078                 return 0;
2079         banks += 4 * mce->bank;
2080         /*
2081          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2082          * reporting is disabled for the bank
2083          */
2084         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2085                 return 0;
2086         if (mce->status & MCI_STATUS_UC) {
2087                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2088                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2089                         printk(KERN_DEBUG "kvm: set_mce: "
2090                                "injects mce exception while "
2091                                "previous one is in progress!\n");
2092                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2093                         return 0;
2094                 }
2095                 if (banks[1] & MCI_STATUS_VAL)
2096                         mce->status |= MCI_STATUS_OVER;
2097                 banks[2] = mce->addr;
2098                 banks[3] = mce->misc;
2099                 vcpu->arch.mcg_status = mce->mcg_status;
2100                 banks[1] = mce->status;
2101                 kvm_queue_exception(vcpu, MC_VECTOR);
2102         } else if (!(banks[1] & MCI_STATUS_VAL)
2103                    || !(banks[1] & MCI_STATUS_UC)) {
2104                 if (banks[1] & MCI_STATUS_VAL)
2105                         mce->status |= MCI_STATUS_OVER;
2106                 banks[2] = mce->addr;
2107                 banks[3] = mce->misc;
2108                 banks[1] = mce->status;
2109         } else
2110                 banks[1] |= MCI_STATUS_OVER;
2111         return 0;
2112 }
2113
2114 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2115                                                struct kvm_vcpu_events *events)
2116 {
2117         vcpu_load(vcpu);
2118
2119         events->exception.injected = vcpu->arch.exception.pending;
2120         events->exception.nr = vcpu->arch.exception.nr;
2121         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2122         events->exception.error_code = vcpu->arch.exception.error_code;
2123
2124         events->interrupt.injected = vcpu->arch.interrupt.pending;
2125         events->interrupt.nr = vcpu->arch.interrupt.nr;
2126         events->interrupt.soft = vcpu->arch.interrupt.soft;
2127
2128         events->nmi.injected = vcpu->arch.nmi_injected;
2129         events->nmi.pending = vcpu->arch.nmi_pending;
2130         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2131
2132         events->sipi_vector = vcpu->arch.sipi_vector;
2133
2134         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2135                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2136
2137         vcpu_put(vcpu);
2138 }
2139
2140 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2141                                               struct kvm_vcpu_events *events)
2142 {
2143         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2144                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2145                 return -EINVAL;
2146
2147         vcpu_load(vcpu);
2148
2149         vcpu->arch.exception.pending = events->exception.injected;
2150         vcpu->arch.exception.nr = events->exception.nr;
2151         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2152         vcpu->arch.exception.error_code = events->exception.error_code;
2153
2154         vcpu->arch.interrupt.pending = events->interrupt.injected;
2155         vcpu->arch.interrupt.nr = events->interrupt.nr;
2156         vcpu->arch.interrupt.soft = events->interrupt.soft;
2157         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2158                 kvm_pic_clear_isr_ack(vcpu->kvm);
2159
2160         vcpu->arch.nmi_injected = events->nmi.injected;
2161         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2162                 vcpu->arch.nmi_pending = events->nmi.pending;
2163         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2164
2165         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2166                 vcpu->arch.sipi_vector = events->sipi_vector;
2167
2168         vcpu_put(vcpu);
2169
2170         return 0;
2171 }
2172
2173 long kvm_arch_vcpu_ioctl(struct file *filp,
2174                          unsigned int ioctl, unsigned long arg)
2175 {
2176         struct kvm_vcpu *vcpu = filp->private_data;
2177         void __user *argp = (void __user *)arg;
2178         int r;
2179         struct kvm_lapic_state *lapic = NULL;
2180
2181         switch (ioctl) {
2182         case KVM_GET_LAPIC: {
2183                 r = -EINVAL;
2184                 if (!vcpu->arch.apic)
2185                         goto out;
2186                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2187
2188                 r = -ENOMEM;
2189                 if (!lapic)
2190                         goto out;
2191                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2192                 if (r)
2193                         goto out;
2194                 r = -EFAULT;
2195                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2196                         goto out;
2197                 r = 0;
2198                 break;
2199         }
2200         case KVM_SET_LAPIC: {
2201                 r = -EINVAL;
2202                 if (!vcpu->arch.apic)
2203                         goto out;
2204                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2205                 r = -ENOMEM;
2206                 if (!lapic)
2207                         goto out;
2208                 r = -EFAULT;
2209                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2210                         goto out;
2211                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2212                 if (r)
2213                         goto out;
2214                 r = 0;
2215                 break;
2216         }
2217         case KVM_INTERRUPT: {
2218                 struct kvm_interrupt irq;
2219
2220                 r = -EFAULT;
2221                 if (copy_from_user(&irq, argp, sizeof irq))
2222                         goto out;
2223                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2224                 if (r)
2225                         goto out;
2226                 r = 0;
2227                 break;
2228         }
2229         case KVM_NMI: {
2230                 r = kvm_vcpu_ioctl_nmi(vcpu);
2231                 if (r)
2232                         goto out;
2233                 r = 0;
2234                 break;
2235         }
2236         case KVM_SET_CPUID: {
2237                 struct kvm_cpuid __user *cpuid_arg = argp;
2238                 struct kvm_cpuid cpuid;
2239
2240                 r = -EFAULT;
2241                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2242                         goto out;
2243                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2244                 if (r)
2245                         goto out;
2246                 break;
2247         }
2248         case KVM_SET_CPUID2: {
2249                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2250                 struct kvm_cpuid2 cpuid;
2251
2252                 r = -EFAULT;
2253                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2254                         goto out;
2255                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2256                                               cpuid_arg->entries);
2257                 if (r)
2258                         goto out;
2259                 break;
2260         }
2261         case KVM_GET_CPUID2: {
2262                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2263                 struct kvm_cpuid2 cpuid;
2264
2265                 r = -EFAULT;
2266                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2267                         goto out;
2268                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2269                                               cpuid_arg->entries);
2270                 if (r)
2271                         goto out;
2272                 r = -EFAULT;
2273                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2274                         goto out;
2275                 r = 0;
2276                 break;
2277         }
2278         case KVM_GET_MSRS:
2279                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2280                 break;
2281         case KVM_SET_MSRS:
2282                 r = msr_io(vcpu, argp, do_set_msr, 0);
2283                 break;
2284         case KVM_TPR_ACCESS_REPORTING: {
2285                 struct kvm_tpr_access_ctl tac;
2286
2287                 r = -EFAULT;
2288                 if (copy_from_user(&tac, argp, sizeof tac))
2289                         goto out;
2290                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2291                 if (r)
2292                         goto out;
2293                 r = -EFAULT;
2294                 if (copy_to_user(argp, &tac, sizeof tac))
2295                         goto out;
2296                 r = 0;
2297                 break;
2298         };
2299         case KVM_SET_VAPIC_ADDR: {
2300                 struct kvm_vapic_addr va;
2301
2302                 r = -EINVAL;
2303                 if (!irqchip_in_kernel(vcpu->kvm))
2304                         goto out;
2305                 r = -EFAULT;
2306                 if (copy_from_user(&va, argp, sizeof va))
2307                         goto out;
2308                 r = 0;
2309                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2310                 break;
2311         }
2312         case KVM_X86_SETUP_MCE: {
2313                 u64 mcg_cap;
2314
2315                 r = -EFAULT;
2316                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2317                         goto out;
2318                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2319                 break;
2320         }
2321         case KVM_X86_SET_MCE: {
2322                 struct kvm_x86_mce mce;
2323
2324                 r = -EFAULT;
2325                 if (copy_from_user(&mce, argp, sizeof mce))
2326                         goto out;
2327                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2328                 break;
2329         }
2330         case KVM_GET_VCPU_EVENTS: {
2331                 struct kvm_vcpu_events events;
2332
2333                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2334
2335                 r = -EFAULT;
2336                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2337                         break;
2338                 r = 0;
2339                 break;
2340         }
2341         case KVM_SET_VCPU_EVENTS: {
2342                 struct kvm_vcpu_events events;
2343
2344                 r = -EFAULT;
2345                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2346                         break;
2347
2348                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2349                 break;
2350         }
2351         default:
2352                 r = -EINVAL;
2353         }
2354 out:
2355         kfree(lapic);
2356         return r;
2357 }
2358
2359 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2360 {
2361         int ret;
2362
2363         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2364                 return -1;
2365         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2366         return ret;
2367 }
2368
2369 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2370                                               u64 ident_addr)
2371 {
2372         kvm->arch.ept_identity_map_addr = ident_addr;
2373         return 0;
2374 }
2375
2376 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2377                                           u32 kvm_nr_mmu_pages)
2378 {
2379         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2380                 return -EINVAL;
2381
2382         mutex_lock(&kvm->slots_lock);
2383         spin_lock(&kvm->mmu_lock);
2384
2385         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2386         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2387
2388         spin_unlock(&kvm->mmu_lock);
2389         mutex_unlock(&kvm->slots_lock);
2390         return 0;
2391 }
2392
2393 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2394 {
2395         return kvm->arch.n_alloc_mmu_pages;
2396 }
2397
2398 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2399 {
2400         int i;
2401         struct kvm_mem_alias *alias;
2402         struct kvm_mem_aliases *aliases;
2403
2404         aliases = rcu_dereference(kvm->arch.aliases);
2405
2406         for (i = 0; i < aliases->naliases; ++i) {
2407                 alias = &aliases->aliases[i];
2408                 if (alias->flags & KVM_ALIAS_INVALID)
2409                         continue;
2410                 if (gfn >= alias->base_gfn
2411                     && gfn < alias->base_gfn + alias->npages)
2412                         return alias->target_gfn + gfn - alias->base_gfn;
2413         }
2414         return gfn;
2415 }
2416
2417 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2418 {
2419         int i;
2420         struct kvm_mem_alias *alias;
2421         struct kvm_mem_aliases *aliases;
2422
2423         aliases = rcu_dereference(kvm->arch.aliases);
2424
2425         for (i = 0; i < aliases->naliases; ++i) {
2426                 alias = &aliases->aliases[i];
2427                 if (gfn >= alias->base_gfn
2428                     && gfn < alias->base_gfn + alias->npages)
2429                         return alias->target_gfn + gfn - alias->base_gfn;
2430         }
2431         return gfn;
2432 }
2433
2434 /*
2435  * Set a new alias region.  Aliases map a portion of physical memory into
2436  * another portion.  This is useful for memory windows, for example the PC
2437  * VGA region.
2438  */
2439 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2440                                          struct kvm_memory_alias *alias)
2441 {
2442         int r, n;
2443         struct kvm_mem_alias *p;
2444         struct kvm_mem_aliases *aliases, *old_aliases;
2445
2446         r = -EINVAL;
2447         /* General sanity checks */
2448         if (alias->memory_size & (PAGE_SIZE - 1))
2449                 goto out;
2450         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2451                 goto out;
2452         if (alias->slot >= KVM_ALIAS_SLOTS)
2453                 goto out;
2454         if (alias->guest_phys_addr + alias->memory_size
2455             < alias->guest_phys_addr)
2456                 goto out;
2457         if (alias->target_phys_addr + alias->memory_size
2458             < alias->target_phys_addr)
2459                 goto out;
2460
2461         r = -ENOMEM;
2462         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2463         if (!aliases)
2464                 goto out;
2465
2466         mutex_lock(&kvm->slots_lock);
2467
2468         /* invalidate any gfn reference in case of deletion/shrinking */
2469         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2470         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2471         old_aliases = kvm->arch.aliases;
2472         rcu_assign_pointer(kvm->arch.aliases, aliases);
2473         synchronize_srcu_expedited(&kvm->srcu);
2474         kvm_mmu_zap_all(kvm);
2475         kfree(old_aliases);
2476
2477         r = -ENOMEM;
2478         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2479         if (!aliases)
2480                 goto out_unlock;
2481
2482         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2483
2484         p = &aliases->aliases[alias->slot];
2485         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2486         p->npages = alias->memory_size >> PAGE_SHIFT;
2487         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2488         p->flags &= ~(KVM_ALIAS_INVALID);
2489
2490         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2491                 if (aliases->aliases[n - 1].npages)
2492                         break;
2493         aliases->naliases = n;
2494
2495         old_aliases = kvm->arch.aliases;
2496         rcu_assign_pointer(kvm->arch.aliases, aliases);
2497         synchronize_srcu_expedited(&kvm->srcu);
2498         kfree(old_aliases);
2499         r = 0;
2500
2501 out_unlock:
2502         mutex_unlock(&kvm->slots_lock);
2503 out:
2504         return r;
2505 }
2506
2507 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2508 {
2509         int r;
2510
2511         r = 0;
2512         switch (chip->chip_id) {
2513         case KVM_IRQCHIP_PIC_MASTER:
2514                 memcpy(&chip->chip.pic,
2515                         &pic_irqchip(kvm)->pics[0],
2516                         sizeof(struct kvm_pic_state));
2517                 break;
2518         case KVM_IRQCHIP_PIC_SLAVE:
2519                 memcpy(&chip->chip.pic,
2520                         &pic_irqchip(kvm)->pics[1],
2521                         sizeof(struct kvm_pic_state));
2522                 break;
2523         case KVM_IRQCHIP_IOAPIC:
2524                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2525                 break;
2526         default:
2527                 r = -EINVAL;
2528                 break;
2529         }
2530         return r;
2531 }
2532
2533 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2534 {
2535         int r;
2536
2537         r = 0;
2538         switch (chip->chip_id) {
2539         case KVM_IRQCHIP_PIC_MASTER:
2540                 spin_lock(&pic_irqchip(kvm)->lock);
2541                 memcpy(&pic_irqchip(kvm)->pics[0],
2542                         &chip->chip.pic,
2543                         sizeof(struct kvm_pic_state));
2544                 spin_unlock(&pic_irqchip(kvm)->lock);
2545                 break;
2546         case KVM_IRQCHIP_PIC_SLAVE:
2547                 spin_lock(&pic_irqchip(kvm)->lock);
2548                 memcpy(&pic_irqchip(kvm)->pics[1],
2549                         &chip->chip.pic,
2550                         sizeof(struct kvm_pic_state));
2551                 spin_unlock(&pic_irqchip(kvm)->lock);
2552                 break;
2553         case KVM_IRQCHIP_IOAPIC:
2554                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2555                 break;
2556         default:
2557                 r = -EINVAL;
2558                 break;
2559         }
2560         kvm_pic_update_irq(pic_irqchip(kvm));
2561         return r;
2562 }
2563
2564 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2565 {
2566         int r = 0;
2567
2568         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2569         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2570         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2571         return r;
2572 }
2573
2574 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2575 {
2576         int r = 0;
2577
2578         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2579         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2580         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2581         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2582         return r;
2583 }
2584
2585 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2586 {
2587         int r = 0;
2588
2589         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2590         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2591                 sizeof(ps->channels));
2592         ps->flags = kvm->arch.vpit->pit_state.flags;
2593         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2594         return r;
2595 }
2596
2597 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2598 {
2599         int r = 0, start = 0;
2600         u32 prev_legacy, cur_legacy;
2601         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2602         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2603         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2604         if (!prev_legacy && cur_legacy)
2605                 start = 1;
2606         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2607                sizeof(kvm->arch.vpit->pit_state.channels));
2608         kvm->arch.vpit->pit_state.flags = ps->flags;
2609         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2610         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2611         return r;
2612 }
2613
2614 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2615                                  struct kvm_reinject_control *control)
2616 {
2617         if (!kvm->arch.vpit)
2618                 return -ENXIO;
2619         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2620         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2621         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2622         return 0;
2623 }
2624
2625 /*
2626  * Get (and clear) the dirty memory log for a memory slot.
2627  */
2628 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2629                                       struct kvm_dirty_log *log)
2630 {
2631         int r, n, i;
2632         struct kvm_memory_slot *memslot;
2633         unsigned long is_dirty = 0;
2634         unsigned long *dirty_bitmap = NULL;
2635
2636         mutex_lock(&kvm->slots_lock);
2637
2638         r = -EINVAL;
2639         if (log->slot >= KVM_MEMORY_SLOTS)
2640                 goto out;
2641
2642         memslot = &kvm->memslots->memslots[log->slot];
2643         r = -ENOENT;
2644         if (!memslot->dirty_bitmap)
2645                 goto out;
2646
2647         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2648
2649         r = -ENOMEM;
2650         dirty_bitmap = vmalloc(n);
2651         if (!dirty_bitmap)
2652                 goto out;
2653         memset(dirty_bitmap, 0, n);
2654
2655         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2656                 is_dirty = memslot->dirty_bitmap[i];
2657
2658         /* If nothing is dirty, don't bother messing with page tables. */
2659         if (is_dirty) {
2660                 struct kvm_memslots *slots, *old_slots;
2661
2662                 spin_lock(&kvm->mmu_lock);
2663                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2664                 spin_unlock(&kvm->mmu_lock);
2665
2666                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2667                 if (!slots)
2668                         goto out_free;
2669
2670                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2671                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2672
2673                 old_slots = kvm->memslots;
2674                 rcu_assign_pointer(kvm->memslots, slots);
2675                 synchronize_srcu_expedited(&kvm->srcu);
2676                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2677                 kfree(old_slots);
2678         }
2679
2680         r = 0;
2681         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2682                 r = -EFAULT;
2683 out_free:
2684         vfree(dirty_bitmap);
2685 out:
2686         mutex_unlock(&kvm->slots_lock);
2687         return r;
2688 }
2689
2690 long kvm_arch_vm_ioctl(struct file *filp,
2691                        unsigned int ioctl, unsigned long arg)
2692 {
2693         struct kvm *kvm = filp->private_data;
2694         void __user *argp = (void __user *)arg;
2695         int r = -ENOTTY;
2696         /*
2697          * This union makes it completely explicit to gcc-3.x
2698          * that these two variables' stack usage should be
2699          * combined, not added together.
2700          */
2701         union {
2702                 struct kvm_pit_state ps;
2703                 struct kvm_pit_state2 ps2;
2704                 struct kvm_memory_alias alias;
2705                 struct kvm_pit_config pit_config;
2706         } u;
2707
2708         switch (ioctl) {
2709         case KVM_SET_TSS_ADDR:
2710                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2711                 if (r < 0)
2712                         goto out;
2713                 break;
2714         case KVM_SET_IDENTITY_MAP_ADDR: {
2715                 u64 ident_addr;
2716
2717                 r = -EFAULT;
2718                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2719                         goto out;
2720                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2721                 if (r < 0)
2722                         goto out;
2723                 break;
2724         }
2725         case KVM_SET_MEMORY_REGION: {
2726                 struct kvm_memory_region kvm_mem;
2727                 struct kvm_userspace_memory_region kvm_userspace_mem;
2728
2729                 r = -EFAULT;
2730                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2731                         goto out;
2732                 kvm_userspace_mem.slot = kvm_mem.slot;
2733                 kvm_userspace_mem.flags = kvm_mem.flags;
2734                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2735                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2736                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2737                 if (r)
2738                         goto out;
2739                 break;
2740         }
2741         case KVM_SET_NR_MMU_PAGES:
2742                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2743                 if (r)
2744                         goto out;
2745                 break;
2746         case KVM_GET_NR_MMU_PAGES:
2747                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2748                 break;
2749         case KVM_SET_MEMORY_ALIAS:
2750                 r = -EFAULT;
2751                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2752                         goto out;
2753                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2754                 if (r)
2755                         goto out;
2756                 break;
2757         case KVM_CREATE_IRQCHIP: {
2758                 struct kvm_pic *vpic;
2759
2760                 mutex_lock(&kvm->lock);
2761                 r = -EEXIST;
2762                 if (kvm->arch.vpic)
2763                         goto create_irqchip_unlock;
2764                 r = -ENOMEM;
2765                 vpic = kvm_create_pic(kvm);
2766                 if (vpic) {
2767                         r = kvm_ioapic_init(kvm);
2768                         if (r) {
2769                                 kfree(vpic);
2770                                 goto create_irqchip_unlock;
2771                         }
2772                 } else
2773                         goto create_irqchip_unlock;
2774                 smp_wmb();
2775                 kvm->arch.vpic = vpic;
2776                 smp_wmb();
2777                 r = kvm_setup_default_irq_routing(kvm);
2778                 if (r) {
2779                         mutex_lock(&kvm->irq_lock);
2780                         kfree(kvm->arch.vpic);
2781                         kfree(kvm->arch.vioapic);
2782                         kvm->arch.vpic = NULL;
2783                         kvm->arch.vioapic = NULL;
2784                         mutex_unlock(&kvm->irq_lock);
2785                 }
2786         create_irqchip_unlock:
2787                 mutex_unlock(&kvm->lock);
2788                 break;
2789         }
2790         case KVM_CREATE_PIT:
2791                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2792                 goto create_pit;
2793         case KVM_CREATE_PIT2:
2794                 r = -EFAULT;
2795                 if (copy_from_user(&u.pit_config, argp,
2796                                    sizeof(struct kvm_pit_config)))
2797                         goto out;
2798         create_pit:
2799                 mutex_lock(&kvm->slots_lock);
2800                 r = -EEXIST;
2801                 if (kvm->arch.vpit)
2802                         goto create_pit_unlock;
2803                 r = -ENOMEM;
2804                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2805                 if (kvm->arch.vpit)
2806                         r = 0;
2807         create_pit_unlock:
2808                 mutex_unlock(&kvm->slots_lock);
2809                 break;
2810         case KVM_IRQ_LINE_STATUS:
2811         case KVM_IRQ_LINE: {
2812                 struct kvm_irq_level irq_event;
2813
2814                 r = -EFAULT;
2815                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2816                         goto out;
2817                 if (irqchip_in_kernel(kvm)) {
2818                         __s32 status;
2819                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2820                                         irq_event.irq, irq_event.level);
2821                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2822                                 irq_event.status = status;
2823                                 if (copy_to_user(argp, &irq_event,
2824                                                         sizeof irq_event))
2825                                         goto out;
2826                         }
2827                         r = 0;
2828                 }
2829                 break;
2830         }
2831         case KVM_GET_IRQCHIP: {
2832                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2833                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2834
2835                 r = -ENOMEM;
2836                 if (!chip)
2837                         goto out;
2838                 r = -EFAULT;
2839                 if (copy_from_user(chip, argp, sizeof *chip))
2840                         goto get_irqchip_out;
2841                 r = -ENXIO;
2842                 if (!irqchip_in_kernel(kvm))
2843                         goto get_irqchip_out;
2844                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2845                 if (r)
2846                         goto get_irqchip_out;
2847                 r = -EFAULT;
2848                 if (copy_to_user(argp, chip, sizeof *chip))
2849                         goto get_irqchip_out;
2850                 r = 0;
2851         get_irqchip_out:
2852                 kfree(chip);
2853                 if (r)
2854                         goto out;
2855                 break;
2856         }
2857         case KVM_SET_IRQCHIP: {
2858                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2859                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2860
2861                 r = -ENOMEM;
2862                 if (!chip)
2863                         goto out;
2864                 r = -EFAULT;
2865                 if (copy_from_user(chip, argp, sizeof *chip))
2866                         goto set_irqchip_out;
2867                 r = -ENXIO;
2868                 if (!irqchip_in_kernel(kvm))
2869                         goto set_irqchip_out;
2870                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2871                 if (r)
2872                         goto set_irqchip_out;
2873                 r = 0;
2874         set_irqchip_out:
2875                 kfree(chip);
2876                 if (r)
2877                         goto out;
2878                 break;
2879         }
2880         case KVM_GET_PIT: {
2881                 r = -EFAULT;
2882                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2883                         goto out;
2884                 r = -ENXIO;
2885                 if (!kvm->arch.vpit)
2886                         goto out;
2887                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2888                 if (r)
2889                         goto out;
2890                 r = -EFAULT;
2891                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2892                         goto out;
2893                 r = 0;
2894                 break;
2895         }
2896         case KVM_SET_PIT: {
2897                 r = -EFAULT;
2898                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2899                         goto out;
2900                 r = -ENXIO;
2901                 if (!kvm->arch.vpit)
2902                         goto out;
2903                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2904                 if (r)
2905                         goto out;
2906                 r = 0;
2907                 break;
2908         }
2909         case KVM_GET_PIT2: {
2910                 r = -ENXIO;
2911                 if (!kvm->arch.vpit)
2912                         goto out;
2913                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2914                 if (r)
2915                         goto out;
2916                 r = -EFAULT;
2917                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2918                         goto out;
2919                 r = 0;
2920                 break;
2921         }
2922         case KVM_SET_PIT2: {
2923                 r = -EFAULT;
2924                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2925                         goto out;
2926                 r = -ENXIO;
2927                 if (!kvm->arch.vpit)
2928                         goto out;
2929                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2930                 if (r)
2931                         goto out;
2932                 r = 0;
2933                 break;
2934         }
2935         case KVM_REINJECT_CONTROL: {
2936                 struct kvm_reinject_control control;
2937                 r =  -EFAULT;
2938                 if (copy_from_user(&control, argp, sizeof(control)))
2939                         goto out;
2940                 r = kvm_vm_ioctl_reinject(kvm, &control);
2941                 if (r)
2942                         goto out;
2943                 r = 0;
2944                 break;
2945         }
2946         case KVM_XEN_HVM_CONFIG: {
2947                 r = -EFAULT;
2948                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2949                                    sizeof(struct kvm_xen_hvm_config)))
2950                         goto out;
2951                 r = -EINVAL;
2952                 if (kvm->arch.xen_hvm_config.flags)
2953                         goto out;
2954                 r = 0;
2955                 break;
2956         }
2957         case KVM_SET_CLOCK: {
2958                 struct timespec now;
2959                 struct kvm_clock_data user_ns;
2960                 u64 now_ns;
2961                 s64 delta;
2962
2963                 r = -EFAULT;
2964                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2965                         goto out;
2966
2967                 r = -EINVAL;
2968                 if (user_ns.flags)
2969                         goto out;
2970
2971                 r = 0;
2972                 ktime_get_ts(&now);
2973                 now_ns = timespec_to_ns(&now);
2974                 delta = user_ns.clock - now_ns;
2975                 kvm->arch.kvmclock_offset = delta;
2976                 break;
2977         }
2978         case KVM_GET_CLOCK: {
2979                 struct timespec now;
2980                 struct kvm_clock_data user_ns;
2981                 u64 now_ns;
2982
2983                 ktime_get_ts(&now);
2984                 now_ns = timespec_to_ns(&now);
2985                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2986                 user_ns.flags = 0;
2987
2988                 r = -EFAULT;
2989                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2990                         goto out;
2991                 r = 0;
2992                 break;
2993         }
2994
2995         default:
2996                 ;
2997         }
2998 out:
2999         return r;
3000 }
3001
3002 static void kvm_init_msr_list(void)
3003 {
3004         u32 dummy[2];
3005         unsigned i, j;
3006
3007         /* skip the first msrs in the list. KVM-specific */
3008         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3009                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3010                         continue;
3011                 if (j < i)
3012                         msrs_to_save[j] = msrs_to_save[i];
3013                 j++;
3014         }
3015         num_msrs_to_save = j;
3016 }
3017
3018 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3019                            const void *v)
3020 {
3021         if (vcpu->arch.apic &&
3022             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3023                 return 0;
3024
3025         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3026 }
3027
3028 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3029 {
3030         if (vcpu->arch.apic &&
3031             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3032                 return 0;
3033
3034         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3035 }
3036
3037 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3038                                struct kvm_vcpu *vcpu)
3039 {
3040         void *data = val;
3041         int r = X86EMUL_CONTINUE;
3042
3043         while (bytes) {
3044                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3045                 unsigned offset = addr & (PAGE_SIZE-1);
3046                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3047                 int ret;
3048
3049                 if (gpa == UNMAPPED_GVA) {
3050                         r = X86EMUL_PROPAGATE_FAULT;
3051                         goto out;
3052                 }
3053                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3054                 if (ret < 0) {
3055                         r = X86EMUL_UNHANDLEABLE;
3056                         goto out;
3057                 }
3058
3059                 bytes -= toread;
3060                 data += toread;
3061                 addr += toread;
3062         }
3063 out:
3064         return r;
3065 }
3066
3067 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3068                                 struct kvm_vcpu *vcpu)
3069 {
3070         void *data = val;
3071         int r = X86EMUL_CONTINUE;
3072
3073         while (bytes) {
3074                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3075                 unsigned offset = addr & (PAGE_SIZE-1);
3076                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3077                 int ret;
3078
3079                 if (gpa == UNMAPPED_GVA) {
3080                         r = X86EMUL_PROPAGATE_FAULT;
3081                         goto out;
3082                 }
3083                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3084                 if (ret < 0) {
3085                         r = X86EMUL_UNHANDLEABLE;
3086                         goto out;
3087                 }
3088
3089                 bytes -= towrite;
3090                 data += towrite;
3091                 addr += towrite;
3092         }
3093 out:
3094         return r;
3095 }
3096
3097
3098 static int emulator_read_emulated(unsigned long addr,
3099                                   void *val,
3100                                   unsigned int bytes,
3101                                   struct kvm_vcpu *vcpu)
3102 {
3103         gpa_t                 gpa;
3104
3105         if (vcpu->mmio_read_completed) {
3106                 memcpy(val, vcpu->mmio_data, bytes);
3107                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3108                                vcpu->mmio_phys_addr, *(u64 *)val);
3109                 vcpu->mmio_read_completed = 0;
3110                 return X86EMUL_CONTINUE;
3111         }
3112
3113         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3114
3115         /* For APIC access vmexit */
3116         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3117                 goto mmio;
3118
3119         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
3120                                 == X86EMUL_CONTINUE)
3121                 return X86EMUL_CONTINUE;
3122         if (gpa == UNMAPPED_GVA)
3123                 return X86EMUL_PROPAGATE_FAULT;
3124
3125 mmio:
3126         /*
3127          * Is this MMIO handled locally?
3128          */
3129         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3130                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3131                 return X86EMUL_CONTINUE;
3132         }
3133
3134         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3135
3136         vcpu->mmio_needed = 1;
3137         vcpu->mmio_phys_addr = gpa;
3138         vcpu->mmio_size = bytes;
3139         vcpu->mmio_is_write = 0;
3140
3141         return X86EMUL_UNHANDLEABLE;
3142 }
3143
3144 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3145                           const void *val, int bytes)
3146 {
3147         int ret;
3148
3149         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3150         if (ret < 0)
3151                 return 0;
3152         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3153         return 1;
3154 }
3155
3156 static int emulator_write_emulated_onepage(unsigned long addr,
3157                                            const void *val,
3158                                            unsigned int bytes,
3159                                            struct kvm_vcpu *vcpu)
3160 {
3161         gpa_t                 gpa;
3162
3163         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3164
3165         if (gpa == UNMAPPED_GVA) {
3166                 kvm_inject_page_fault(vcpu, addr, 2);
3167                 return X86EMUL_PROPAGATE_FAULT;
3168         }
3169
3170         /* For APIC access vmexit */
3171         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3172                 goto mmio;
3173
3174         if (emulator_write_phys(vcpu, gpa, val, bytes))
3175                 return X86EMUL_CONTINUE;
3176
3177 mmio:
3178         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3179         /*
3180          * Is this MMIO handled locally?
3181          */
3182         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3183                 return X86EMUL_CONTINUE;
3184
3185         vcpu->mmio_needed = 1;
3186         vcpu->mmio_phys_addr = gpa;
3187         vcpu->mmio_size = bytes;
3188         vcpu->mmio_is_write = 1;
3189         memcpy(vcpu->mmio_data, val, bytes);
3190
3191         return X86EMUL_CONTINUE;
3192 }
3193
3194 int emulator_write_emulated(unsigned long addr,
3195                                    const void *val,
3196                                    unsigned int bytes,
3197                                    struct kvm_vcpu *vcpu)
3198 {
3199         /* Crossing a page boundary? */
3200         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3201                 int rc, now;
3202
3203                 now = -addr & ~PAGE_MASK;
3204                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3205                 if (rc != X86EMUL_CONTINUE)
3206                         return rc;
3207                 addr += now;
3208                 val += now;
3209                 bytes -= now;
3210         }
3211         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3212 }
3213 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3214
3215 static int emulator_cmpxchg_emulated(unsigned long addr,
3216                                      const void *old,
3217                                      const void *new,
3218                                      unsigned int bytes,
3219                                      struct kvm_vcpu *vcpu)
3220 {
3221         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3222 #ifndef CONFIG_X86_64
3223         /* guests cmpxchg8b have to be emulated atomically */
3224         if (bytes == 8) {
3225                 gpa_t gpa;
3226                 struct page *page;
3227                 char *kaddr;
3228                 u64 val;
3229
3230                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3231
3232                 if (gpa == UNMAPPED_GVA ||
3233                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3234                         goto emul_write;
3235
3236                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3237                         goto emul_write;
3238
3239                 val = *(u64 *)new;
3240
3241                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3242
3243                 kaddr = kmap_atomic(page, KM_USER0);
3244                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3245                 kunmap_atomic(kaddr, KM_USER0);
3246                 kvm_release_page_dirty(page);
3247         }
3248 emul_write:
3249 #endif
3250
3251         return emulator_write_emulated(addr, new, bytes, vcpu);
3252 }
3253
3254 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3255 {
3256         return kvm_x86_ops->get_segment_base(vcpu, seg);
3257 }
3258
3259 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3260 {
3261         kvm_mmu_invlpg(vcpu, address);
3262         return X86EMUL_CONTINUE;
3263 }
3264
3265 int emulate_clts(struct kvm_vcpu *vcpu)
3266 {
3267         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3268         return X86EMUL_CONTINUE;
3269 }
3270
3271 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3272 {
3273         struct kvm_vcpu *vcpu = ctxt->vcpu;
3274
3275         switch (dr) {
3276         case 0 ... 3:
3277                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3278                 return X86EMUL_CONTINUE;
3279         default:
3280                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3281                 return X86EMUL_UNHANDLEABLE;
3282         }
3283 }
3284
3285 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3286 {
3287         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3288         int exception;
3289
3290         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3291         if (exception) {
3292                 /* FIXME: better handling */
3293                 return X86EMUL_UNHANDLEABLE;
3294         }
3295         return X86EMUL_CONTINUE;
3296 }
3297
3298 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3299 {
3300         u8 opcodes[4];
3301         unsigned long rip = kvm_rip_read(vcpu);
3302         unsigned long rip_linear;
3303
3304         if (!printk_ratelimit())
3305                 return;
3306
3307         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3308
3309         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3310
3311         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3312                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3313 }
3314 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3315
3316 static struct x86_emulate_ops emulate_ops = {
3317         .read_std            = kvm_read_guest_virt,
3318         .read_emulated       = emulator_read_emulated,
3319         .write_emulated      = emulator_write_emulated,
3320         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3321 };
3322
3323 static void cache_all_regs(struct kvm_vcpu *vcpu)
3324 {
3325         kvm_register_read(vcpu, VCPU_REGS_RAX);
3326         kvm_register_read(vcpu, VCPU_REGS_RSP);
3327         kvm_register_read(vcpu, VCPU_REGS_RIP);
3328         vcpu->arch.regs_dirty = ~0;
3329 }
3330
3331 int emulate_instruction(struct kvm_vcpu *vcpu,
3332                         unsigned long cr2,
3333                         u16 error_code,
3334                         int emulation_type)
3335 {
3336         int r, shadow_mask;
3337         struct decode_cache *c;
3338         struct kvm_run *run = vcpu->run;
3339
3340         kvm_clear_exception_queue(vcpu);
3341         vcpu->arch.mmio_fault_cr2 = cr2;
3342         /*
3343          * TODO: fix emulate.c to use guest_read/write_register
3344          * instead of direct ->regs accesses, can save hundred cycles
3345          * on Intel for instructions that don't read/change RSP, for
3346          * for example.
3347          */
3348         cache_all_regs(vcpu);
3349
3350         vcpu->mmio_is_write = 0;
3351         vcpu->arch.pio.string = 0;
3352
3353         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3354                 int cs_db, cs_l;
3355                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3356
3357                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3358                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3359                 vcpu->arch.emulate_ctxt.mode =
3360                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3361                         ? X86EMUL_MODE_REAL : cs_l
3362                         ? X86EMUL_MODE_PROT64 : cs_db
3363                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3364
3365                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3366
3367                 /* Only allow emulation of specific instructions on #UD
3368                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3369                 c = &vcpu->arch.emulate_ctxt.decode;
3370                 if (emulation_type & EMULTYPE_TRAP_UD) {
3371                         if (!c->twobyte)
3372                                 return EMULATE_FAIL;
3373                         switch (c->b) {
3374                         case 0x01: /* VMMCALL */
3375                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3376                                         return EMULATE_FAIL;
3377                                 break;
3378                         case 0x34: /* sysenter */
3379                         case 0x35: /* sysexit */
3380                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3381                                         return EMULATE_FAIL;
3382                                 break;
3383                         case 0x05: /* syscall */
3384                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3385                                         return EMULATE_FAIL;
3386                                 break;
3387                         default:
3388                                 return EMULATE_FAIL;
3389                         }
3390
3391                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3392                                 return EMULATE_FAIL;
3393                 }
3394
3395                 ++vcpu->stat.insn_emulation;
3396                 if (r)  {
3397                         ++vcpu->stat.insn_emulation_fail;
3398                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3399                                 return EMULATE_DONE;
3400                         return EMULATE_FAIL;
3401                 }
3402         }
3403
3404         if (emulation_type & EMULTYPE_SKIP) {
3405                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3406                 return EMULATE_DONE;
3407         }
3408
3409         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3410         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3411
3412         if (r == 0)
3413                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3414
3415         if (vcpu->arch.pio.string)
3416                 return EMULATE_DO_MMIO;
3417
3418         if ((r || vcpu->mmio_is_write) && run) {
3419                 run->exit_reason = KVM_EXIT_MMIO;
3420                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3421                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3422                 run->mmio.len = vcpu->mmio_size;
3423                 run->mmio.is_write = vcpu->mmio_is_write;
3424         }
3425
3426         if (r) {
3427                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3428                         return EMULATE_DONE;
3429                 if (!vcpu->mmio_needed) {
3430                         kvm_report_emulation_failure(vcpu, "mmio");
3431                         return EMULATE_FAIL;
3432                 }
3433                 return EMULATE_DO_MMIO;
3434         }
3435
3436         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3437
3438         if (vcpu->mmio_is_write) {
3439                 vcpu->mmio_needed = 0;
3440                 return EMULATE_DO_MMIO;
3441         }
3442
3443         return EMULATE_DONE;
3444 }
3445 EXPORT_SYMBOL_GPL(emulate_instruction);
3446
3447 static int pio_copy_data(struct kvm_vcpu *vcpu)
3448 {
3449         void *p = vcpu->arch.pio_data;
3450         gva_t q = vcpu->arch.pio.guest_gva;
3451         unsigned bytes;
3452         int ret;
3453
3454         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3455         if (vcpu->arch.pio.in)
3456                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3457         else
3458                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3459         return ret;
3460 }
3461
3462 int complete_pio(struct kvm_vcpu *vcpu)
3463 {
3464         struct kvm_pio_request *io = &vcpu->arch.pio;
3465         long delta;
3466         int r;
3467         unsigned long val;
3468
3469         if (!io->string) {
3470                 if (io->in) {
3471                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3472                         memcpy(&val, vcpu->arch.pio_data, io->size);
3473                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3474                 }
3475         } else {
3476                 if (io->in) {
3477                         r = pio_copy_data(vcpu);
3478                         if (r)
3479                                 return r;
3480                 }
3481
3482                 delta = 1;
3483                 if (io->rep) {
3484                         delta *= io->cur_count;
3485                         /*
3486                          * The size of the register should really depend on
3487                          * current address size.
3488                          */
3489                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3490                         val -= delta;
3491                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3492                 }
3493                 if (io->down)
3494                         delta = -delta;
3495                 delta *= io->size;
3496                 if (io->in) {
3497                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3498                         val += delta;
3499                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3500                 } else {
3501                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3502                         val += delta;
3503                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3504                 }
3505         }
3506
3507         io->count -= io->cur_count;
3508         io->cur_count = 0;
3509
3510         return 0;
3511 }
3512
3513 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3514 {
3515         /* TODO: String I/O for in kernel device */
3516         int r;
3517
3518         if (vcpu->arch.pio.in)
3519                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3520                                     vcpu->arch.pio.size, pd);
3521         else
3522                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3523                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3524                                      pd);
3525         return r;
3526 }
3527
3528 static int pio_string_write(struct kvm_vcpu *vcpu)
3529 {
3530         struct kvm_pio_request *io = &vcpu->arch.pio;
3531         void *pd = vcpu->arch.pio_data;
3532         int i, r = 0;
3533
3534         for (i = 0; i < io->cur_count; i++) {
3535                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3536                                      io->port, io->size, pd)) {
3537                         r = -EOPNOTSUPP;
3538                         break;
3539                 }
3540                 pd += io->size;
3541         }
3542         return r;
3543 }
3544
3545 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3546 {
3547         unsigned long val;
3548
3549         vcpu->run->exit_reason = KVM_EXIT_IO;
3550         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3551         vcpu->run->io.size = vcpu->arch.pio.size = size;
3552         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3553         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3554         vcpu->run->io.port = vcpu->arch.pio.port = port;
3555         vcpu->arch.pio.in = in;
3556         vcpu->arch.pio.string = 0;
3557         vcpu->arch.pio.down = 0;
3558         vcpu->arch.pio.rep = 0;
3559
3560         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3561                       size, 1);
3562
3563         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3564         memcpy(vcpu->arch.pio_data, &val, 4);
3565
3566         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3567                 complete_pio(vcpu);
3568                 return 1;
3569         }
3570         return 0;
3571 }
3572 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3573
3574 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3575                   int size, unsigned long count, int down,
3576                   gva_t address, int rep, unsigned port)
3577 {
3578         unsigned now, in_page;
3579         int ret = 0;
3580
3581         vcpu->run->exit_reason = KVM_EXIT_IO;
3582         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3583         vcpu->run->io.size = vcpu->arch.pio.size = size;
3584         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3585         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3586         vcpu->run->io.port = vcpu->arch.pio.port = port;
3587         vcpu->arch.pio.in = in;
3588         vcpu->arch.pio.string = 1;
3589         vcpu->arch.pio.down = down;
3590         vcpu->arch.pio.rep = rep;
3591
3592         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3593                       size, count);
3594
3595         if (!count) {
3596                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3597                 return 1;
3598         }
3599
3600         if (!down)
3601                 in_page = PAGE_SIZE - offset_in_page(address);
3602         else
3603                 in_page = offset_in_page(address) + size;
3604         now = min(count, (unsigned long)in_page / size);
3605         if (!now)
3606                 now = 1;
3607         if (down) {
3608                 /*
3609                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3610                  */
3611                 pr_unimpl(vcpu, "guest string pio down\n");
3612                 kvm_inject_gp(vcpu, 0);
3613                 return 1;
3614         }
3615         vcpu->run->io.count = now;
3616         vcpu->arch.pio.cur_count = now;
3617
3618         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3619                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3620
3621         vcpu->arch.pio.guest_gva = address;
3622
3623         if (!vcpu->arch.pio.in) {
3624                 /* string PIO write */
3625                 ret = pio_copy_data(vcpu);
3626                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3627                         kvm_inject_gp(vcpu, 0);
3628                         return 1;
3629                 }
3630                 if (ret == 0 && !pio_string_write(vcpu)) {
3631                         complete_pio(vcpu);
3632                         if (vcpu->arch.pio.count == 0)
3633                                 ret = 1;
3634                 }
3635         }
3636         /* no string PIO read support yet */
3637
3638         return ret;
3639 }
3640 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3641
3642 static void bounce_off(void *info)
3643 {
3644         /* nothing */
3645 }
3646
3647 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3648                                      void *data)
3649 {
3650         struct cpufreq_freqs *freq = data;
3651         struct kvm *kvm;
3652         struct kvm_vcpu *vcpu;
3653         int i, send_ipi = 0;
3654
3655         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3656                 return 0;
3657         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3658                 return 0;
3659         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3660
3661         spin_lock(&kvm_lock);
3662         list_for_each_entry(kvm, &vm_list, vm_list) {
3663                 kvm_for_each_vcpu(i, vcpu, kvm) {
3664                         if (vcpu->cpu != freq->cpu)
3665                                 continue;
3666                         if (!kvm_request_guest_time_update(vcpu))
3667                                 continue;
3668                         if (vcpu->cpu != smp_processor_id())
3669                                 send_ipi++;
3670                 }
3671         }
3672         spin_unlock(&kvm_lock);
3673
3674         if (freq->old < freq->new && send_ipi) {
3675                 /*
3676                  * We upscale the frequency.  Must make the guest
3677                  * doesn't see old kvmclock values while running with
3678                  * the new frequency, otherwise we risk the guest sees
3679                  * time go backwards.
3680                  *
3681                  * In case we update the frequency for another cpu
3682                  * (which might be in guest context) send an interrupt
3683                  * to kick the cpu out of guest context.  Next time
3684                  * guest context is entered kvmclock will be updated,
3685                  * so the guest will not see stale values.
3686                  */
3687                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3688         }
3689         return 0;
3690 }
3691
3692 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3693         .notifier_call  = kvmclock_cpufreq_notifier
3694 };
3695
3696 static void kvm_timer_init(void)
3697 {
3698         int cpu;
3699
3700         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3701                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3702                                           CPUFREQ_TRANSITION_NOTIFIER);
3703                 for_each_online_cpu(cpu) {
3704                         unsigned long khz = cpufreq_get(cpu);
3705                         if (!khz)
3706                                 khz = tsc_khz;
3707                         per_cpu(cpu_tsc_khz, cpu) = khz;
3708                 }
3709         } else {
3710                 for_each_possible_cpu(cpu)
3711                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3712         }
3713 }
3714
3715 int kvm_arch_init(void *opaque)
3716 {
3717         int r;
3718         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3719
3720         if (kvm_x86_ops) {
3721                 printk(KERN_ERR "kvm: already loaded the other module\n");
3722                 r = -EEXIST;
3723                 goto out;
3724         }
3725
3726         if (!ops->cpu_has_kvm_support()) {
3727                 printk(KERN_ERR "kvm: no hardware support\n");
3728                 r = -EOPNOTSUPP;
3729                 goto out;
3730         }
3731         if (ops->disabled_by_bios()) {
3732                 printk(KERN_ERR "kvm: disabled by bios\n");
3733                 r = -EOPNOTSUPP;
3734                 goto out;
3735         }
3736
3737         r = kvm_mmu_module_init();
3738         if (r)
3739                 goto out;
3740
3741         kvm_init_msr_list();
3742
3743         kvm_x86_ops = ops;
3744         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3745         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3746         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3747                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3748
3749         kvm_timer_init();
3750
3751         return 0;
3752
3753 out:
3754         return r;
3755 }
3756
3757 void kvm_arch_exit(void)
3758 {
3759         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3760                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3761                                             CPUFREQ_TRANSITION_NOTIFIER);
3762         kvm_x86_ops = NULL;
3763         kvm_mmu_module_exit();
3764 }
3765
3766 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3767 {
3768         ++vcpu->stat.halt_exits;
3769         if (irqchip_in_kernel(vcpu->kvm)) {
3770                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3771                 return 1;
3772         } else {
3773                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3774                 return 0;
3775         }
3776 }
3777 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3778
3779 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3780                            unsigned long a1)
3781 {
3782         if (is_long_mode(vcpu))
3783                 return a0;
3784         else
3785                 return a0 | ((gpa_t)a1 << 32);
3786 }
3787
3788 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3789 {
3790         u64 param, ingpa, outgpa, ret;
3791         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3792         bool fast, longmode;
3793         int cs_db, cs_l;
3794
3795         /*
3796          * hypercall generates UD from non zero cpl and real mode
3797          * per HYPER-V spec
3798          */
3799         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
3800             !kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
3801                 kvm_queue_exception(vcpu, UD_VECTOR);
3802                 return 0;
3803         }
3804
3805         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3806         longmode = is_long_mode(vcpu) && cs_l == 1;
3807
3808         if (!longmode) {
3809                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3810                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3811                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3812                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3813                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3814                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3815         }
3816 #ifdef CONFIG_X86_64
3817         else {
3818                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3819                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3820                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3821         }
3822 #endif
3823
3824         code = param & 0xffff;
3825         fast = (param >> 16) & 0x1;
3826         rep_cnt = (param >> 32) & 0xfff;
3827         rep_idx = (param >> 48) & 0xfff;
3828
3829         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3830
3831         switch (code) {
3832         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3833                 kvm_vcpu_on_spin(vcpu);
3834                 break;
3835         default:
3836                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3837                 break;
3838         }
3839
3840         ret = res | (((u64)rep_done & 0xfff) << 32);
3841         if (longmode) {
3842                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3843         } else {
3844                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3845                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3846         }
3847
3848         return 1;
3849 }
3850
3851 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3852 {
3853         unsigned long nr, a0, a1, a2, a3, ret;
3854         int r = 1;
3855
3856         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3857                 return kvm_hv_hypercall(vcpu);
3858
3859         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3860         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3861         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3862         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3863         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3864
3865         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3866
3867         if (!is_long_mode(vcpu)) {
3868                 nr &= 0xFFFFFFFF;
3869                 a0 &= 0xFFFFFFFF;
3870                 a1 &= 0xFFFFFFFF;
3871                 a2 &= 0xFFFFFFFF;
3872                 a3 &= 0xFFFFFFFF;
3873         }
3874
3875         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3876                 ret = -KVM_EPERM;
3877                 goto out;
3878         }
3879
3880         switch (nr) {
3881         case KVM_HC_VAPIC_POLL_IRQ:
3882                 ret = 0;
3883                 break;
3884         case KVM_HC_MMU_OP:
3885                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3886                 break;
3887         default:
3888                 ret = -KVM_ENOSYS;
3889                 break;
3890         }
3891 out:
3892         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3893         ++vcpu->stat.hypercalls;
3894         return r;
3895 }
3896 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3897
3898 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3899 {
3900         char instruction[3];
3901         int ret = 0;
3902         unsigned long rip = kvm_rip_read(vcpu);
3903
3904
3905         /*
3906          * Blow out the MMU to ensure that no other VCPU has an active mapping
3907          * to ensure that the updated hypercall appears atomically across all
3908          * VCPUs.
3909          */
3910         kvm_mmu_zap_all(vcpu->kvm);
3911
3912         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3913         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3914             != X86EMUL_CONTINUE)
3915                 ret = -EFAULT;
3916
3917         return ret;
3918 }
3919
3920 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3921 {
3922         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3923 }
3924
3925 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3926 {
3927         struct descriptor_table dt = { limit, base };
3928
3929         kvm_x86_ops->set_gdt(vcpu, &dt);
3930 }
3931
3932 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3933 {
3934         struct descriptor_table dt = { limit, base };
3935
3936         kvm_x86_ops->set_idt(vcpu, &dt);
3937 }
3938
3939 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3940                    unsigned long *rflags)
3941 {
3942         kvm_lmsw(vcpu, msw);
3943         *rflags = kvm_get_rflags(vcpu);
3944 }
3945
3946 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3947 {
3948         unsigned long value;
3949
3950         switch (cr) {
3951         case 0:
3952                 value = kvm_read_cr0(vcpu);
3953                 break;
3954         case 2:
3955                 value = vcpu->arch.cr2;
3956                 break;
3957         case 3:
3958                 value = vcpu->arch.cr3;
3959                 break;
3960         case 4:
3961                 value = kvm_read_cr4(vcpu);
3962                 break;
3963         case 8:
3964                 value = kvm_get_cr8(vcpu);
3965                 break;
3966         default:
3967                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3968                 return 0;
3969         }
3970
3971         return value;
3972 }
3973
3974 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3975                      unsigned long *rflags)
3976 {
3977         switch (cr) {
3978         case 0:
3979                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3980                 *rflags = kvm_get_rflags(vcpu);
3981                 break;
3982         case 2:
3983                 vcpu->arch.cr2 = val;
3984                 break;
3985         case 3:
3986                 kvm_set_cr3(vcpu, val);
3987                 break;
3988         case 4:
3989                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3990                 break;
3991         case 8:
3992                 kvm_set_cr8(vcpu, val & 0xfUL);
3993                 break;
3994         default:
3995                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3996         }
3997 }
3998
3999 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4000 {
4001         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4002         int j, nent = vcpu->arch.cpuid_nent;
4003
4004         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4005         /* when no next entry is found, the current entry[i] is reselected */
4006         for (j = i + 1; ; j = (j + 1) % nent) {
4007                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4008                 if (ej->function == e->function) {
4009                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4010                         return j;
4011                 }
4012         }
4013         return 0; /* silence gcc, even though control never reaches here */
4014 }
4015
4016 /* find an entry with matching function, matching index (if needed), and that
4017  * should be read next (if it's stateful) */
4018 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4019         u32 function, u32 index)
4020 {
4021         if (e->function != function)
4022                 return 0;
4023         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4024                 return 0;
4025         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4026             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4027                 return 0;
4028         return 1;
4029 }
4030
4031 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4032                                               u32 function, u32 index)
4033 {
4034         int i;
4035         struct kvm_cpuid_entry2 *best = NULL;
4036
4037         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4038                 struct kvm_cpuid_entry2 *e;
4039
4040                 e = &vcpu->arch.cpuid_entries[i];
4041                 if (is_matching_cpuid_entry(e, function, index)) {
4042                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4043                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4044                         best = e;
4045                         break;
4046                 }
4047                 /*
4048                  * Both basic or both extended?
4049                  */
4050                 if (((e->function ^ function) & 0x80000000) == 0)
4051                         if (!best || e->function > best->function)
4052                                 best = e;
4053         }
4054         return best;
4055 }
4056 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4057
4058 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4059 {
4060         struct kvm_cpuid_entry2 *best;
4061
4062         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4063         if (best)
4064                 return best->eax & 0xff;
4065         return 36;
4066 }
4067
4068 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4069 {
4070         u32 function, index;
4071         struct kvm_cpuid_entry2 *best;
4072
4073         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4074         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4075         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4076         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4077         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4078         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4079         best = kvm_find_cpuid_entry(vcpu, function, index);
4080         if (best) {
4081                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4082                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4083                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4084                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4085         }
4086         kvm_x86_ops->skip_emulated_instruction(vcpu);
4087         trace_kvm_cpuid(function,
4088                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4089                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4090                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4091                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4092 }
4093 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4094
4095 /*
4096  * Check if userspace requested an interrupt window, and that the
4097  * interrupt window is open.
4098  *
4099  * No need to exit to userspace if we already have an interrupt queued.
4100  */
4101 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4102 {
4103         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4104                 vcpu->run->request_interrupt_window &&
4105                 kvm_arch_interrupt_allowed(vcpu));
4106 }
4107
4108 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4109 {
4110         struct kvm_run *kvm_run = vcpu->run;
4111
4112         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4113         kvm_run->cr8 = kvm_get_cr8(vcpu);
4114         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4115         if (irqchip_in_kernel(vcpu->kvm))
4116                 kvm_run->ready_for_interrupt_injection = 1;
4117         else
4118                 kvm_run->ready_for_interrupt_injection =
4119                         kvm_arch_interrupt_allowed(vcpu) &&
4120                         !kvm_cpu_has_interrupt(vcpu) &&
4121                         !kvm_event_needs_reinjection(vcpu);
4122 }
4123
4124 static void vapic_enter(struct kvm_vcpu *vcpu)
4125 {
4126         struct kvm_lapic *apic = vcpu->arch.apic;
4127         struct page *page;
4128
4129         if (!apic || !apic->vapic_addr)
4130                 return;
4131
4132         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4133
4134         vcpu->arch.apic->vapic_page = page;
4135 }
4136
4137 static void vapic_exit(struct kvm_vcpu *vcpu)
4138 {
4139         struct kvm_lapic *apic = vcpu->arch.apic;
4140         int idx;
4141
4142         if (!apic || !apic->vapic_addr)
4143                 return;
4144
4145         idx = srcu_read_lock(&vcpu->kvm->srcu);
4146         kvm_release_page_dirty(apic->vapic_page);
4147         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4148         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4149 }
4150
4151 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4152 {
4153         int max_irr, tpr;
4154
4155         if (!kvm_x86_ops->update_cr8_intercept)
4156                 return;
4157
4158         if (!vcpu->arch.apic)
4159                 return;
4160
4161         if (!vcpu->arch.apic->vapic_addr)
4162                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4163         else
4164                 max_irr = -1;
4165
4166         if (max_irr != -1)
4167                 max_irr >>= 4;
4168
4169         tpr = kvm_lapic_get_cr8(vcpu);
4170
4171         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4172 }
4173
4174 static void inject_pending_event(struct kvm_vcpu *vcpu)
4175 {
4176         /* try to reinject previous events if any */
4177         if (vcpu->arch.exception.pending) {
4178                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4179                                           vcpu->arch.exception.has_error_code,
4180                                           vcpu->arch.exception.error_code);
4181                 return;
4182         }
4183
4184         if (vcpu->arch.nmi_injected) {
4185                 kvm_x86_ops->set_nmi(vcpu);
4186                 return;
4187         }
4188
4189         if (vcpu->arch.interrupt.pending) {
4190                 kvm_x86_ops->set_irq(vcpu);
4191                 return;
4192         }
4193
4194         /* try to inject new event if pending */
4195         if (vcpu->arch.nmi_pending) {
4196                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4197                         vcpu->arch.nmi_pending = false;
4198                         vcpu->arch.nmi_injected = true;
4199                         kvm_x86_ops->set_nmi(vcpu);
4200                 }
4201         } else if (kvm_cpu_has_interrupt(vcpu)) {
4202                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4203                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4204                                             false);
4205                         kvm_x86_ops->set_irq(vcpu);
4206                 }
4207         }
4208 }
4209
4210 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4211 {
4212         int r;
4213         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4214                 vcpu->run->request_interrupt_window;
4215
4216         if (vcpu->requests)
4217                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4218                         kvm_mmu_unload(vcpu);
4219
4220         r = kvm_mmu_reload(vcpu);
4221         if (unlikely(r))
4222                 goto out;
4223
4224         if (vcpu->requests) {
4225                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4226                         __kvm_migrate_timers(vcpu);
4227                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4228                         kvm_write_guest_time(vcpu);
4229                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4230                         kvm_mmu_sync_roots(vcpu);
4231                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4232                         kvm_x86_ops->tlb_flush(vcpu);
4233                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4234                                        &vcpu->requests)) {
4235                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4236                         r = 0;
4237                         goto out;
4238                 }
4239                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4240                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4241                         r = 0;
4242                         goto out;
4243                 }
4244                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4245                         vcpu->fpu_active = 0;
4246                         kvm_x86_ops->fpu_deactivate(vcpu);
4247                 }
4248         }
4249
4250         preempt_disable();
4251
4252         kvm_x86_ops->prepare_guest_switch(vcpu);
4253         kvm_load_guest_fpu(vcpu);
4254
4255         local_irq_disable();
4256
4257         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4258         smp_mb__after_clear_bit();
4259
4260         if (vcpu->requests || need_resched() || signal_pending(current)) {
4261                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4262                 local_irq_enable();
4263                 preempt_enable();
4264                 r = 1;
4265                 goto out;
4266         }
4267
4268         inject_pending_event(vcpu);
4269
4270         /* enable NMI/IRQ window open exits if needed */
4271         if (vcpu->arch.nmi_pending)
4272                 kvm_x86_ops->enable_nmi_window(vcpu);
4273         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4274                 kvm_x86_ops->enable_irq_window(vcpu);
4275
4276         if (kvm_lapic_enabled(vcpu)) {
4277                 update_cr8_intercept(vcpu);
4278                 kvm_lapic_sync_to_vapic(vcpu);
4279         }
4280
4281         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4282
4283         kvm_guest_enter();
4284
4285         if (unlikely(vcpu->arch.switch_db_regs)) {
4286                 set_debugreg(0, 7);
4287                 set_debugreg(vcpu->arch.eff_db[0], 0);
4288                 set_debugreg(vcpu->arch.eff_db[1], 1);
4289                 set_debugreg(vcpu->arch.eff_db[2], 2);
4290                 set_debugreg(vcpu->arch.eff_db[3], 3);
4291         }
4292
4293         trace_kvm_entry(vcpu->vcpu_id);
4294         kvm_x86_ops->run(vcpu);
4295
4296         /*
4297          * If the guest has used debug registers, at least dr7
4298          * will be disabled while returning to the host.
4299          * If we don't have active breakpoints in the host, we don't
4300          * care about the messed up debug address registers. But if
4301          * we have some of them active, restore the old state.
4302          */
4303         if (hw_breakpoint_active())
4304                 hw_breakpoint_restore();
4305
4306         set_bit(KVM_REQ_KICK, &vcpu->requests);
4307         local_irq_enable();
4308
4309         ++vcpu->stat.exits;
4310
4311         /*
4312          * We must have an instruction between local_irq_enable() and
4313          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4314          * the interrupt shadow.  The stat.exits increment will do nicely.
4315          * But we need to prevent reordering, hence this barrier():
4316          */
4317         barrier();
4318
4319         kvm_guest_exit();
4320
4321         preempt_enable();
4322
4323         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4324
4325         /*
4326          * Profile KVM exit RIPs:
4327          */
4328         if (unlikely(prof_on == KVM_PROFILING)) {
4329                 unsigned long rip = kvm_rip_read(vcpu);
4330                 profile_hit(KVM_PROFILING, (void *)rip);
4331         }
4332
4333
4334         kvm_lapic_sync_from_vapic(vcpu);
4335
4336         r = kvm_x86_ops->handle_exit(vcpu);
4337 out:
4338         return r;
4339 }
4340
4341
4342 static int __vcpu_run(struct kvm_vcpu *vcpu)
4343 {
4344         int r;
4345         struct kvm *kvm = vcpu->kvm;
4346
4347         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4348                 pr_debug("vcpu %d received sipi with vector # %x\n",
4349                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4350                 kvm_lapic_reset(vcpu);
4351                 r = kvm_arch_vcpu_reset(vcpu);
4352                 if (r)
4353                         return r;
4354                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4355         }
4356
4357         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4358         vapic_enter(vcpu);
4359
4360         r = 1;
4361         while (r > 0) {
4362                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4363                         r = vcpu_enter_guest(vcpu);
4364                 else {
4365                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4366                         kvm_vcpu_block(vcpu);
4367                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4368                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4369                         {
4370                                 switch(vcpu->arch.mp_state) {
4371                                 case KVM_MP_STATE_HALTED:
4372                                         vcpu->arch.mp_state =
4373                                                 KVM_MP_STATE_RUNNABLE;
4374                                 case KVM_MP_STATE_RUNNABLE:
4375                                         break;
4376                                 case KVM_MP_STATE_SIPI_RECEIVED:
4377                                 default:
4378                                         r = -EINTR;
4379                                         break;
4380                                 }
4381                         }
4382                 }
4383
4384                 if (r <= 0)
4385                         break;
4386
4387                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4388                 if (kvm_cpu_has_pending_timer(vcpu))
4389                         kvm_inject_pending_timer_irqs(vcpu);
4390
4391                 if (dm_request_for_irq_injection(vcpu)) {
4392                         r = -EINTR;
4393                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4394                         ++vcpu->stat.request_irq_exits;
4395                 }
4396                 if (signal_pending(current)) {
4397                         r = -EINTR;
4398                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4399                         ++vcpu->stat.signal_exits;
4400                 }
4401                 if (need_resched()) {
4402                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4403                         kvm_resched(vcpu);
4404                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4405                 }
4406         }
4407
4408         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4409         post_kvm_run_save(vcpu);
4410
4411         vapic_exit(vcpu);
4412
4413         return r;
4414 }
4415
4416 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4417 {
4418         int r;
4419         sigset_t sigsaved;
4420
4421         vcpu_load(vcpu);
4422
4423         if (vcpu->sigset_active)
4424                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4425
4426         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4427                 kvm_vcpu_block(vcpu);
4428                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4429                 r = -EAGAIN;
4430                 goto out;
4431         }
4432
4433         /* re-sync apic's tpr */
4434         if (!irqchip_in_kernel(vcpu->kvm))
4435                 kvm_set_cr8(vcpu, kvm_run->cr8);
4436
4437         if (vcpu->arch.pio.cur_count) {
4438                 r = complete_pio(vcpu);
4439                 if (r)
4440                         goto out;
4441         }
4442         if (vcpu->mmio_needed) {
4443                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4444                 vcpu->mmio_read_completed = 1;
4445                 vcpu->mmio_needed = 0;
4446
4447                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4448                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4449                                         EMULTYPE_NO_DECODE);
4450                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4451                 if (r == EMULATE_DO_MMIO) {
4452                         /*
4453                          * Read-modify-write.  Back to userspace.
4454                          */
4455                         r = 0;
4456                         goto out;
4457                 }
4458         }
4459         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4460                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4461                                      kvm_run->hypercall.ret);
4462
4463         r = __vcpu_run(vcpu);
4464
4465 out:
4466         if (vcpu->sigset_active)
4467                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4468
4469         vcpu_put(vcpu);
4470         return r;
4471 }
4472
4473 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4474 {
4475         vcpu_load(vcpu);
4476
4477         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4478         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4479         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4480         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4481         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4482         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4483         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4484         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4485 #ifdef CONFIG_X86_64
4486         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4487         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4488         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4489         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4490         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4491         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4492         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4493         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4494 #endif
4495
4496         regs->rip = kvm_rip_read(vcpu);
4497         regs->rflags = kvm_get_rflags(vcpu);
4498
4499         vcpu_put(vcpu);
4500
4501         return 0;
4502 }
4503
4504 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4505 {
4506         vcpu_load(vcpu);
4507
4508         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4509         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4510         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4511         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4512         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4513         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4514         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4515         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4516 #ifdef CONFIG_X86_64
4517         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4518         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4519         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4520         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4521         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4522         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4523         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4524         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4525 #endif
4526
4527         kvm_rip_write(vcpu, regs->rip);
4528         kvm_set_rflags(vcpu, regs->rflags);
4529
4530         vcpu->arch.exception.pending = false;
4531
4532         vcpu_put(vcpu);
4533
4534         return 0;
4535 }
4536
4537 void kvm_get_segment(struct kvm_vcpu *vcpu,
4538                      struct kvm_segment *var, int seg)
4539 {
4540         kvm_x86_ops->get_segment(vcpu, var, seg);
4541 }
4542
4543 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4544 {
4545         struct kvm_segment cs;
4546
4547         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4548         *db = cs.db;
4549         *l = cs.l;
4550 }
4551 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4552
4553 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4554                                   struct kvm_sregs *sregs)
4555 {
4556         struct descriptor_table dt;
4557
4558         vcpu_load(vcpu);
4559
4560         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4561         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4562         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4563         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4564         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4565         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4566
4567         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4568         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4569
4570         kvm_x86_ops->get_idt(vcpu, &dt);
4571         sregs->idt.limit = dt.limit;
4572         sregs->idt.base = dt.base;
4573         kvm_x86_ops->get_gdt(vcpu, &dt);
4574         sregs->gdt.limit = dt.limit;
4575         sregs->gdt.base = dt.base;
4576
4577         sregs->cr0 = kvm_read_cr0(vcpu);
4578         sregs->cr2 = vcpu->arch.cr2;
4579         sregs->cr3 = vcpu->arch.cr3;
4580         sregs->cr4 = kvm_read_cr4(vcpu);
4581         sregs->cr8 = kvm_get_cr8(vcpu);
4582         sregs->efer = vcpu->arch.shadow_efer;
4583         sregs->apic_base = kvm_get_apic_base(vcpu);
4584
4585         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4586
4587         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4588                 set_bit(vcpu->arch.interrupt.nr,
4589                         (unsigned long *)sregs->interrupt_bitmap);
4590
4591         vcpu_put(vcpu);
4592
4593         return 0;
4594 }
4595
4596 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4597                                     struct kvm_mp_state *mp_state)
4598 {
4599         vcpu_load(vcpu);
4600         mp_state->mp_state = vcpu->arch.mp_state;
4601         vcpu_put(vcpu);
4602         return 0;
4603 }
4604
4605 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4606                                     struct kvm_mp_state *mp_state)
4607 {
4608         vcpu_load(vcpu);
4609         vcpu->arch.mp_state = mp_state->mp_state;
4610         vcpu_put(vcpu);
4611         return 0;
4612 }
4613
4614 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4615                         struct kvm_segment *var, int seg)
4616 {
4617         kvm_x86_ops->set_segment(vcpu, var, seg);
4618 }
4619
4620 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4621                                    struct kvm_segment *kvm_desct)
4622 {
4623         kvm_desct->base = get_desc_base(seg_desc);
4624         kvm_desct->limit = get_desc_limit(seg_desc);
4625         if (seg_desc->g) {
4626                 kvm_desct->limit <<= 12;
4627                 kvm_desct->limit |= 0xfff;
4628         }
4629         kvm_desct->selector = selector;
4630         kvm_desct->type = seg_desc->type;
4631         kvm_desct->present = seg_desc->p;
4632         kvm_desct->dpl = seg_desc->dpl;
4633         kvm_desct->db = seg_desc->d;
4634         kvm_desct->s = seg_desc->s;
4635         kvm_desct->l = seg_desc->l;
4636         kvm_desct->g = seg_desc->g;
4637         kvm_desct->avl = seg_desc->avl;
4638         if (!selector)
4639                 kvm_desct->unusable = 1;
4640         else
4641                 kvm_desct->unusable = 0;
4642         kvm_desct->padding = 0;
4643 }
4644
4645 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4646                                           u16 selector,
4647                                           struct descriptor_table *dtable)
4648 {
4649         if (selector & 1 << 2) {
4650                 struct kvm_segment kvm_seg;
4651
4652                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4653
4654                 if (kvm_seg.unusable)
4655                         dtable->limit = 0;
4656                 else
4657                         dtable->limit = kvm_seg.limit;
4658                 dtable->base = kvm_seg.base;
4659         }
4660         else
4661                 kvm_x86_ops->get_gdt(vcpu, dtable);
4662 }
4663
4664 /* allowed just for 8 bytes segments */
4665 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4666                                          struct desc_struct *seg_desc)
4667 {
4668         struct descriptor_table dtable;
4669         u16 index = selector >> 3;
4670
4671         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4672
4673         if (dtable.limit < index * 8 + 7) {
4674                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4675                 return 1;
4676         }
4677         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4678 }
4679
4680 /* allowed just for 8 bytes segments */
4681 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4682                                          struct desc_struct *seg_desc)
4683 {
4684         struct descriptor_table dtable;
4685         u16 index = selector >> 3;
4686
4687         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4688
4689         if (dtable.limit < index * 8 + 7)
4690                 return 1;
4691         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4692 }
4693
4694 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4695                              struct desc_struct *seg_desc)
4696 {
4697         u32 base_addr = get_desc_base(seg_desc);
4698
4699         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4700 }
4701
4702 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4703 {
4704         struct kvm_segment kvm_seg;
4705
4706         kvm_get_segment(vcpu, &kvm_seg, seg);
4707         return kvm_seg.selector;
4708 }
4709
4710 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4711                                                 u16 selector,
4712                                                 struct kvm_segment *kvm_seg)
4713 {
4714         struct desc_struct seg_desc;
4715
4716         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4717                 return 1;
4718         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4719         return 0;
4720 }
4721
4722 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4723 {
4724         struct kvm_segment segvar = {
4725                 .base = selector << 4,
4726                 .limit = 0xffff,
4727                 .selector = selector,
4728                 .type = 3,
4729                 .present = 1,
4730                 .dpl = 3,
4731                 .db = 0,
4732                 .s = 1,
4733                 .l = 0,
4734                 .g = 0,
4735                 .avl = 0,
4736                 .unusable = 0,
4737         };
4738         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4739         return 0;
4740 }
4741
4742 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4743 {
4744         return (seg != VCPU_SREG_LDTR) &&
4745                 (seg != VCPU_SREG_TR) &&
4746                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4747 }
4748
4749 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4750                                          u16 selector)
4751 {
4752         /* NULL selector is not valid for CS and SS */
4753         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4754                 if (!selector)
4755                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4756 }
4757
4758 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4759                                 int type_bits, int seg)
4760 {
4761         struct kvm_segment kvm_seg;
4762
4763         if (is_vm86_segment(vcpu, seg) || !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
4764                 return kvm_load_realmode_segment(vcpu, selector, seg);
4765         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4766                 return 1;
4767
4768         kvm_check_segment_descriptor(vcpu, seg, selector);
4769         kvm_seg.type |= type_bits;
4770
4771         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4772             seg != VCPU_SREG_LDTR)
4773                 if (!kvm_seg.s)
4774                         kvm_seg.unusable = 1;
4775
4776         kvm_set_segment(vcpu, &kvm_seg, seg);
4777         return 0;
4778 }
4779
4780 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4781                                 struct tss_segment_32 *tss)
4782 {
4783         tss->cr3 = vcpu->arch.cr3;
4784         tss->eip = kvm_rip_read(vcpu);
4785         tss->eflags = kvm_get_rflags(vcpu);
4786         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4787         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4788         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4789         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4790         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4791         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4792         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4793         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4794         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4795         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4796         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4797         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4798         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4799         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4800         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4801 }
4802
4803 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4804                                   struct tss_segment_32 *tss)
4805 {
4806         kvm_set_cr3(vcpu, tss->cr3);
4807
4808         kvm_rip_write(vcpu, tss->eip);
4809         kvm_set_rflags(vcpu, tss->eflags | 2);
4810
4811         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4812         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4813         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4814         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4815         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4816         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4817         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4818         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4819
4820         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4821                 return 1;
4822
4823         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4824                 return 1;
4825
4826         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4827                 return 1;
4828
4829         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4830                 return 1;
4831
4832         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4833                 return 1;
4834
4835         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4836                 return 1;
4837
4838         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4839                 return 1;
4840         return 0;
4841 }
4842
4843 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4844                                 struct tss_segment_16 *tss)
4845 {
4846         tss->ip = kvm_rip_read(vcpu);
4847         tss->flag = kvm_get_rflags(vcpu);
4848         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4849         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4850         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4851         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4852         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4853         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4854         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4855         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4856
4857         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4858         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4859         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4860         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4861         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4862 }
4863
4864 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4865                                  struct tss_segment_16 *tss)
4866 {
4867         kvm_rip_write(vcpu, tss->ip);
4868         kvm_set_rflags(vcpu, tss->flag | 2);
4869         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4870         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4871         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4872         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4873         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4874         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4875         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4876         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4877
4878         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4879                 return 1;
4880
4881         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4882                 return 1;
4883
4884         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4885                 return 1;
4886
4887         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4888                 return 1;
4889
4890         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4891                 return 1;
4892         return 0;
4893 }
4894
4895 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4896                               u16 old_tss_sel, u32 old_tss_base,
4897                               struct desc_struct *nseg_desc)
4898 {
4899         struct tss_segment_16 tss_segment_16;
4900         int ret = 0;
4901
4902         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4903                            sizeof tss_segment_16))
4904                 goto out;
4905
4906         save_state_to_tss16(vcpu, &tss_segment_16);
4907
4908         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4909                             sizeof tss_segment_16))
4910                 goto out;
4911
4912         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4913                            &tss_segment_16, sizeof tss_segment_16))
4914                 goto out;
4915
4916         if (old_tss_sel != 0xffff) {
4917                 tss_segment_16.prev_task_link = old_tss_sel;
4918
4919                 if (kvm_write_guest(vcpu->kvm,
4920                                     get_tss_base_addr(vcpu, nseg_desc),
4921                                     &tss_segment_16.prev_task_link,
4922                                     sizeof tss_segment_16.prev_task_link))
4923                         goto out;
4924         }
4925
4926         if (load_state_from_tss16(vcpu, &tss_segment_16))
4927                 goto out;
4928
4929         ret = 1;
4930 out:
4931         return ret;
4932 }
4933
4934 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4935                        u16 old_tss_sel, u32 old_tss_base,
4936                        struct desc_struct *nseg_desc)
4937 {
4938         struct tss_segment_32 tss_segment_32;
4939         int ret = 0;
4940
4941         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4942                            sizeof tss_segment_32))
4943                 goto out;
4944
4945         save_state_to_tss32(vcpu, &tss_segment_32);
4946
4947         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4948                             sizeof tss_segment_32))
4949                 goto out;
4950
4951         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4952                            &tss_segment_32, sizeof tss_segment_32))
4953                 goto out;
4954
4955         if (old_tss_sel != 0xffff) {
4956                 tss_segment_32.prev_task_link = old_tss_sel;
4957
4958                 if (kvm_write_guest(vcpu->kvm,
4959                                     get_tss_base_addr(vcpu, nseg_desc),
4960                                     &tss_segment_32.prev_task_link,
4961                                     sizeof tss_segment_32.prev_task_link))
4962                         goto out;
4963         }
4964
4965         if (load_state_from_tss32(vcpu, &tss_segment_32))
4966                 goto out;
4967
4968         ret = 1;
4969 out:
4970         return ret;
4971 }
4972
4973 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4974 {
4975         struct kvm_segment tr_seg;
4976         struct desc_struct cseg_desc;
4977         struct desc_struct nseg_desc;
4978         int ret = 0;
4979         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4980         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4981
4982         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4983
4984         /* FIXME: Handle errors. Failure to read either TSS or their
4985          * descriptors should generate a pagefault.
4986          */
4987         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4988                 goto out;
4989
4990         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4991                 goto out;
4992
4993         if (reason != TASK_SWITCH_IRET) {
4994                 int cpl;
4995
4996                 cpl = kvm_x86_ops->get_cpl(vcpu);
4997                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4998                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4999                         return 1;
5000                 }
5001         }
5002
5003         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
5004                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5005                 return 1;
5006         }
5007
5008         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5009                 cseg_desc.type &= ~(1 << 1); //clear the B flag
5010                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5011         }
5012
5013         if (reason == TASK_SWITCH_IRET) {
5014                 u32 eflags = kvm_get_rflags(vcpu);
5015                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5016         }
5017
5018         /* set back link to prev task only if NT bit is set in eflags
5019            note that old_tss_sel is not used afetr this point */
5020         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5021                 old_tss_sel = 0xffff;
5022
5023         if (nseg_desc.type & 8)
5024                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5025                                          old_tss_base, &nseg_desc);
5026         else
5027                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5028                                          old_tss_base, &nseg_desc);
5029
5030         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5031                 u32 eflags = kvm_get_rflags(vcpu);
5032                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5033         }
5034
5035         if (reason != TASK_SWITCH_IRET) {
5036                 nseg_desc.type |= (1 << 1);
5037                 save_guest_segment_descriptor(vcpu, tss_selector,
5038                                               &nseg_desc);
5039         }
5040
5041         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5042         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5043         tr_seg.type = 11;
5044         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5045 out:
5046         return ret;
5047 }
5048 EXPORT_SYMBOL_GPL(kvm_task_switch);
5049
5050 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5051                                   struct kvm_sregs *sregs)
5052 {
5053         int mmu_reset_needed = 0;
5054         int pending_vec, max_bits;
5055         struct descriptor_table dt;
5056
5057         vcpu_load(vcpu);
5058
5059         dt.limit = sregs->idt.limit;
5060         dt.base = sregs->idt.base;
5061         kvm_x86_ops->set_idt(vcpu, &dt);
5062         dt.limit = sregs->gdt.limit;
5063         dt.base = sregs->gdt.base;
5064         kvm_x86_ops->set_gdt(vcpu, &dt);
5065
5066         vcpu->arch.cr2 = sregs->cr2;
5067         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5068         vcpu->arch.cr3 = sregs->cr3;
5069
5070         kvm_set_cr8(vcpu, sregs->cr8);
5071
5072         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
5073         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5074         kvm_set_apic_base(vcpu, sregs->apic_base);
5075
5076         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5077         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5078         vcpu->arch.cr0 = sregs->cr0;
5079
5080         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5081         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5082         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5083                 load_pdptrs(vcpu, vcpu->arch.cr3);
5084                 mmu_reset_needed = 1;
5085         }
5086
5087         if (mmu_reset_needed)
5088                 kvm_mmu_reset_context(vcpu);
5089
5090         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5091         pending_vec = find_first_bit(
5092                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5093         if (pending_vec < max_bits) {
5094                 kvm_queue_interrupt(vcpu, pending_vec, false);
5095                 pr_debug("Set back pending irq %d\n", pending_vec);
5096                 if (irqchip_in_kernel(vcpu->kvm))
5097                         kvm_pic_clear_isr_ack(vcpu->kvm);
5098         }
5099
5100         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5101         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5102         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5103         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5104         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5105         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5106
5107         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5108         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5109
5110         update_cr8_intercept(vcpu);
5111
5112         /* Older userspace won't unhalt the vcpu on reset. */
5113         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5114             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5115             !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
5116                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5117
5118         vcpu_put(vcpu);
5119
5120         return 0;
5121 }
5122
5123 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5124                                         struct kvm_guest_debug *dbg)
5125 {
5126         unsigned long rflags;
5127         int i, r;
5128
5129         vcpu_load(vcpu);
5130
5131         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5132                 r = -EBUSY;
5133                 if (vcpu->arch.exception.pending)
5134                         goto unlock_out;
5135                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5136                         kvm_queue_exception(vcpu, DB_VECTOR);
5137                 else
5138                         kvm_queue_exception(vcpu, BP_VECTOR);
5139         }
5140
5141         /*
5142          * Read rflags as long as potentially injected trace flags are still
5143          * filtered out.
5144          */
5145         rflags = kvm_get_rflags(vcpu);
5146
5147         vcpu->guest_debug = dbg->control;
5148         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5149                 vcpu->guest_debug = 0;
5150
5151         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5152                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5153                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5154                 vcpu->arch.switch_db_regs =
5155                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5156         } else {
5157                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5158                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5159                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5160         }
5161
5162         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5163                 vcpu->arch.singlestep_cs =
5164                         get_segment_selector(vcpu, VCPU_SREG_CS);
5165                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5166         }
5167
5168         /*
5169          * Trigger an rflags update that will inject or remove the trace
5170          * flags.
5171          */
5172         kvm_set_rflags(vcpu, rflags);
5173
5174         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5175
5176         r = 0;
5177
5178 unlock_out:
5179         vcpu_put(vcpu);
5180
5181         return r;
5182 }
5183
5184 /*
5185  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5186  * we have asm/x86/processor.h
5187  */
5188 struct fxsave {
5189         u16     cwd;
5190         u16     swd;
5191         u16     twd;
5192         u16     fop;
5193         u64     rip;
5194         u64     rdp;
5195         u32     mxcsr;
5196         u32     mxcsr_mask;
5197         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5198 #ifdef CONFIG_X86_64
5199         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5200 #else
5201         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5202 #endif
5203 };
5204
5205 /*
5206  * Translate a guest virtual address to a guest physical address.
5207  */
5208 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5209                                     struct kvm_translation *tr)
5210 {
5211         unsigned long vaddr = tr->linear_address;
5212         gpa_t gpa;
5213         int idx;
5214
5215         vcpu_load(vcpu);
5216         idx = srcu_read_lock(&vcpu->kvm->srcu);
5217         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
5218         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5219         tr->physical_address = gpa;
5220         tr->valid = gpa != UNMAPPED_GVA;
5221         tr->writeable = 1;
5222         tr->usermode = 0;
5223         vcpu_put(vcpu);
5224
5225         return 0;
5226 }
5227
5228 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5229 {
5230         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5231
5232         vcpu_load(vcpu);
5233
5234         memcpy(fpu->fpr, fxsave->st_space, 128);
5235         fpu->fcw = fxsave->cwd;
5236         fpu->fsw = fxsave->swd;
5237         fpu->ftwx = fxsave->twd;
5238         fpu->last_opcode = fxsave->fop;
5239         fpu->last_ip = fxsave->rip;
5240         fpu->last_dp = fxsave->rdp;
5241         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5242
5243         vcpu_put(vcpu);
5244
5245         return 0;
5246 }
5247
5248 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5249 {
5250         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5251
5252         vcpu_load(vcpu);
5253
5254         memcpy(fxsave->st_space, fpu->fpr, 128);
5255         fxsave->cwd = fpu->fcw;
5256         fxsave->swd = fpu->fsw;
5257         fxsave->twd = fpu->ftwx;
5258         fxsave->fop = fpu->last_opcode;
5259         fxsave->rip = fpu->last_ip;
5260         fxsave->rdp = fpu->last_dp;
5261         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5262
5263         vcpu_put(vcpu);
5264
5265         return 0;
5266 }
5267
5268 void fx_init(struct kvm_vcpu *vcpu)
5269 {
5270         unsigned after_mxcsr_mask;
5271
5272         /*
5273          * Touch the fpu the first time in non atomic context as if
5274          * this is the first fpu instruction the exception handler
5275          * will fire before the instruction returns and it'll have to
5276          * allocate ram with GFP_KERNEL.
5277          */
5278         if (!used_math())
5279                 kvm_fx_save(&vcpu->arch.host_fx_image);
5280
5281         /* Initialize guest FPU by resetting ours and saving into guest's */
5282         preempt_disable();
5283         kvm_fx_save(&vcpu->arch.host_fx_image);
5284         kvm_fx_finit();
5285         kvm_fx_save(&vcpu->arch.guest_fx_image);
5286         kvm_fx_restore(&vcpu->arch.host_fx_image);
5287         preempt_enable();
5288
5289         vcpu->arch.cr0 |= X86_CR0_ET;
5290         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5291         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5292         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5293                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5294 }
5295 EXPORT_SYMBOL_GPL(fx_init);
5296
5297 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5298 {
5299         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
5300                 return;
5301
5302         vcpu->guest_fpu_loaded = 1;
5303         kvm_fx_save(&vcpu->arch.host_fx_image);
5304         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5305 }
5306 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
5307
5308 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5309 {
5310         if (!vcpu->guest_fpu_loaded)
5311                 return;
5312
5313         vcpu->guest_fpu_loaded = 0;
5314         kvm_fx_save(&vcpu->arch.guest_fx_image);
5315         kvm_fx_restore(&vcpu->arch.host_fx_image);
5316         ++vcpu->stat.fpu_reload;
5317         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5318 }
5319 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
5320
5321 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5322 {
5323         if (vcpu->arch.time_page) {
5324                 kvm_release_page_dirty(vcpu->arch.time_page);
5325                 vcpu->arch.time_page = NULL;
5326         }
5327
5328         kvm_x86_ops->vcpu_free(vcpu);
5329 }
5330
5331 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5332                                                 unsigned int id)
5333 {
5334         return kvm_x86_ops->vcpu_create(kvm, id);
5335 }
5336
5337 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5338 {
5339         int r;
5340
5341         /* We do fxsave: this must be aligned. */
5342         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5343
5344         vcpu->arch.mtrr_state.have_fixed = 1;
5345         vcpu_load(vcpu);
5346         r = kvm_arch_vcpu_reset(vcpu);
5347         if (r == 0)
5348                 r = kvm_mmu_setup(vcpu);
5349         vcpu_put(vcpu);
5350         if (r < 0)
5351                 goto free_vcpu;
5352
5353         return 0;
5354 free_vcpu:
5355         kvm_x86_ops->vcpu_free(vcpu);
5356         return r;
5357 }
5358
5359 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5360 {
5361         vcpu_load(vcpu);
5362         kvm_mmu_unload(vcpu);
5363         vcpu_put(vcpu);
5364
5365         kvm_x86_ops->vcpu_free(vcpu);
5366 }
5367
5368 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5369 {
5370         vcpu->arch.nmi_pending = false;
5371         vcpu->arch.nmi_injected = false;
5372
5373         vcpu->arch.switch_db_regs = 0;
5374         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5375         vcpu->arch.dr6 = DR6_FIXED_1;
5376         vcpu->arch.dr7 = DR7_FIXED_1;
5377
5378         return kvm_x86_ops->vcpu_reset(vcpu);
5379 }
5380
5381 int kvm_arch_hardware_enable(void *garbage)
5382 {
5383         /*
5384          * Since this may be called from a hotplug notifcation,
5385          * we can't get the CPU frequency directly.
5386          */
5387         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5388                 int cpu = raw_smp_processor_id();
5389                 per_cpu(cpu_tsc_khz, cpu) = 0;
5390         }
5391
5392         kvm_shared_msr_cpu_online();
5393
5394         return kvm_x86_ops->hardware_enable(garbage);
5395 }
5396
5397 void kvm_arch_hardware_disable(void *garbage)
5398 {
5399         kvm_x86_ops->hardware_disable(garbage);
5400         drop_user_return_notifiers(garbage);
5401 }
5402
5403 int kvm_arch_hardware_setup(void)
5404 {
5405         return kvm_x86_ops->hardware_setup();
5406 }
5407
5408 void kvm_arch_hardware_unsetup(void)
5409 {
5410         kvm_x86_ops->hardware_unsetup();
5411 }
5412
5413 void kvm_arch_check_processor_compat(void *rtn)
5414 {
5415         kvm_x86_ops->check_processor_compatibility(rtn);
5416 }
5417
5418 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5419 {
5420         struct page *page;
5421         struct kvm *kvm;
5422         int r;
5423
5424         BUG_ON(vcpu->kvm == NULL);
5425         kvm = vcpu->kvm;
5426
5427         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5428         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5429                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5430         else
5431                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5432
5433         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5434         if (!page) {
5435                 r = -ENOMEM;
5436                 goto fail;
5437         }
5438         vcpu->arch.pio_data = page_address(page);
5439
5440         r = kvm_mmu_create(vcpu);
5441         if (r < 0)
5442                 goto fail_free_pio_data;
5443
5444         if (irqchip_in_kernel(kvm)) {
5445                 r = kvm_create_lapic(vcpu);
5446                 if (r < 0)
5447                         goto fail_mmu_destroy;
5448         }
5449
5450         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5451                                        GFP_KERNEL);
5452         if (!vcpu->arch.mce_banks) {
5453                 r = -ENOMEM;
5454                 goto fail_free_lapic;
5455         }
5456         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5457
5458         return 0;
5459 fail_free_lapic:
5460         kvm_free_lapic(vcpu);
5461 fail_mmu_destroy:
5462         kvm_mmu_destroy(vcpu);
5463 fail_free_pio_data:
5464         free_page((unsigned long)vcpu->arch.pio_data);
5465 fail:
5466         return r;
5467 }
5468
5469 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5470 {
5471         int idx;
5472
5473         kfree(vcpu->arch.mce_banks);
5474         kvm_free_lapic(vcpu);
5475         idx = srcu_read_lock(&vcpu->kvm->srcu);
5476         kvm_mmu_destroy(vcpu);
5477         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5478         free_page((unsigned long)vcpu->arch.pio_data);
5479 }
5480
5481 struct  kvm *kvm_arch_create_vm(void)
5482 {
5483         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5484
5485         if (!kvm)
5486                 return ERR_PTR(-ENOMEM);
5487
5488         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5489         if (!kvm->arch.aliases) {
5490                 kfree(kvm);
5491                 return ERR_PTR(-ENOMEM);
5492         }
5493
5494         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5495         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5496
5497         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5498         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5499
5500         rdtscll(kvm->arch.vm_init_tsc);
5501
5502         return kvm;
5503 }
5504
5505 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5506 {
5507         vcpu_load(vcpu);
5508         kvm_mmu_unload(vcpu);
5509         vcpu_put(vcpu);
5510 }
5511
5512 static void kvm_free_vcpus(struct kvm *kvm)
5513 {
5514         unsigned int i;
5515         struct kvm_vcpu *vcpu;
5516
5517         /*
5518          * Unpin any mmu pages first.
5519          */
5520         kvm_for_each_vcpu(i, vcpu, kvm)
5521                 kvm_unload_vcpu_mmu(vcpu);
5522         kvm_for_each_vcpu(i, vcpu, kvm)
5523                 kvm_arch_vcpu_free(vcpu);
5524
5525         mutex_lock(&kvm->lock);
5526         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5527                 kvm->vcpus[i] = NULL;
5528
5529         atomic_set(&kvm->online_vcpus, 0);
5530         mutex_unlock(&kvm->lock);
5531 }
5532
5533 void kvm_arch_sync_events(struct kvm *kvm)
5534 {
5535         kvm_free_all_assigned_devices(kvm);
5536 }
5537
5538 void kvm_arch_destroy_vm(struct kvm *kvm)
5539 {
5540         kvm_iommu_unmap_guest(kvm);
5541         kvm_free_pit(kvm);
5542         kfree(kvm->arch.vpic);
5543         kfree(kvm->arch.vioapic);
5544         kvm_free_vcpus(kvm);
5545         kvm_free_physmem(kvm);
5546         if (kvm->arch.apic_access_page)
5547                 put_page(kvm->arch.apic_access_page);
5548         if (kvm->arch.ept_identity_pagetable)
5549                 put_page(kvm->arch.ept_identity_pagetable);
5550         cleanup_srcu_struct(&kvm->srcu);
5551         kfree(kvm->arch.aliases);
5552         kfree(kvm);
5553 }
5554
5555 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5556                                 struct kvm_memory_slot *memslot,
5557                                 struct kvm_memory_slot old,
5558                                 struct kvm_userspace_memory_region *mem,
5559                                 int user_alloc)
5560 {
5561         int npages = memslot->npages;
5562
5563         /*To keep backward compatibility with older userspace,
5564          *x86 needs to hanlde !user_alloc case.
5565          */
5566         if (!user_alloc) {
5567                 if (npages && !old.rmap) {
5568                         unsigned long userspace_addr;
5569
5570                         down_write(&current->mm->mmap_sem);
5571                         userspace_addr = do_mmap(NULL, 0,
5572                                                  npages * PAGE_SIZE,
5573                                                  PROT_READ | PROT_WRITE,
5574                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5575                                                  0);
5576                         up_write(&current->mm->mmap_sem);
5577
5578                         if (IS_ERR((void *)userspace_addr))
5579                                 return PTR_ERR((void *)userspace_addr);
5580
5581                         memslot->userspace_addr = userspace_addr;
5582                 }
5583         }
5584
5585
5586         return 0;
5587 }
5588
5589 void kvm_arch_commit_memory_region(struct kvm *kvm,
5590                                 struct kvm_userspace_memory_region *mem,
5591                                 struct kvm_memory_slot old,
5592                                 int user_alloc)
5593 {
5594
5595         int npages = mem->memory_size >> PAGE_SHIFT;
5596
5597         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5598                 int ret;
5599
5600                 down_write(&current->mm->mmap_sem);
5601                 ret = do_munmap(current->mm, old.userspace_addr,
5602                                 old.npages * PAGE_SIZE);
5603                 up_write(&current->mm->mmap_sem);
5604                 if (ret < 0)
5605                         printk(KERN_WARNING
5606                                "kvm_vm_ioctl_set_memory_region: "
5607                                "failed to munmap memory\n");
5608         }
5609
5610         spin_lock(&kvm->mmu_lock);
5611         if (!kvm->arch.n_requested_mmu_pages) {
5612                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5613                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5614         }
5615
5616         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5617         spin_unlock(&kvm->mmu_lock);
5618 }
5619
5620 void kvm_arch_flush_shadow(struct kvm *kvm)
5621 {
5622         kvm_mmu_zap_all(kvm);
5623         kvm_reload_remote_mmus(kvm);
5624 }
5625
5626 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5627 {
5628         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5629                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5630                 || vcpu->arch.nmi_pending ||
5631                 (kvm_arch_interrupt_allowed(vcpu) &&
5632                  kvm_cpu_has_interrupt(vcpu));
5633 }
5634
5635 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5636 {
5637         int me;
5638         int cpu = vcpu->cpu;
5639
5640         if (waitqueue_active(&vcpu->wq)) {
5641                 wake_up_interruptible(&vcpu->wq);
5642                 ++vcpu->stat.halt_wakeup;
5643         }
5644
5645         me = get_cpu();
5646         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5647                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5648                         smp_send_reschedule(cpu);
5649         put_cpu();
5650 }
5651
5652 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5653 {
5654         return kvm_x86_ops->interrupt_allowed(vcpu);
5655 }
5656
5657 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5658 {
5659         unsigned long rflags;
5660
5661         rflags = kvm_x86_ops->get_rflags(vcpu);
5662         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5663                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5664         return rflags;
5665 }
5666 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5667
5668 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5669 {
5670         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5671             vcpu->arch.singlestep_cs ==
5672                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5673             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5674                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5675         kvm_x86_ops->set_rflags(vcpu, rflags);
5676 }
5677 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5678
5679 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5680 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5681 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5682 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5683 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5684 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5685 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5686 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5687 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5688 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5689 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);