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
  41 #include <asm/uaccess.h>
  42 #include <asm/msr.h>
  43 #include <asm/desc.h>
  44 #include <asm/mtrr.h>
  45
  46 #define MAX_IO_MSRS 256
  47 #define CR0_RESERVED_BITS                                               \
  48         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  49                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  50                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  51 #define CR4_RESERVED_BITS                                               \
  52         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  53                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  54                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  55                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  56
  57 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  58 /* EFER defaults:
  59  * - enable syscall per default because its emulated by KVM
  60  * - enable LME and LMA per default on 64 bit KVM
  61  */
  62 #ifdef CONFIG_X86_64
  63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  64 #else
  65 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  66 #endif
  67
  68 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  69 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  70
  71 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  72                                     struct kvm_cpuid_entry2 __user *entries);
  73 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
  74                                               u32 function, u32 index);
  75
  76 struct kvm_x86_ops *kvm_x86_ops;
  77 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  78
  79 struct kvm_stats_debugfs_item debugfs_entries[] = {
  80         { "pf_fixed", VCPU_STAT(pf_fixed) },
  81         { "pf_guest", VCPU_STAT(pf_guest) },
  82         { "tlb_flush", VCPU_STAT(tlb_flush) },
  83         { "invlpg", VCPU_STAT(invlpg) },
  84         { "exits", VCPU_STAT(exits) },
  85         { "io_exits", VCPU_STAT(io_exits) },
  86         { "mmio_exits", VCPU_STAT(mmio_exits) },
  87         { "signal_exits", VCPU_STAT(signal_exits) },
  88         { "irq_window", VCPU_STAT(irq_window_exits) },
  89         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  90         { "halt_exits", VCPU_STAT(halt_exits) },
  91         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  92         { "hypercalls", VCPU_STAT(hypercalls) },
  93         { "request_irq", VCPU_STAT(request_irq_exits) },
  94         { "irq_exits", VCPU_STAT(irq_exits) },
  95         { "host_state_reload", VCPU_STAT(host_state_reload) },
  96         { "efer_reload", VCPU_STAT(efer_reload) },
  97         { "fpu_reload", VCPU_STAT(fpu_reload) },
  98         { "insn_emulation", VCPU_STAT(insn_emulation) },
  99         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 100         { "irq_injections", VCPU_STAT(irq_injections) },
 101         { "nmi_injections", VCPU_STAT(nmi_injections) },
 102         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 103         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 104         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 105         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 106         { "mmu_flooded", VM_STAT(mmu_flooded) },
 107         { "mmu_recycled", VM_STAT(mmu_recycled) },
 108         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 109         { "mmu_unsync", VM_STAT(mmu_unsync) },
 110         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 111         { "largepages", VM_STAT(lpages) },
 112         { NULL }
 113 };
 114
 115 unsigned long segment_base(u16 selector)
 116 {
 117         struct descriptor_table gdt;
 118         struct desc_struct *d;
 119         unsigned long table_base;
 120         unsigned long v;
 121
 122         if (selector == 0)
 123                 return 0;
 124
 125         asm("sgdt %0" : "=m"(gdt));
 126         table_base = gdt.base;
 127
 128         if (selector & 4) {           /* from ldt */
 129                 u16 ldt_selector;
 130
 131                 asm("sldt %0" : "=g"(ldt_selector));
 132                 table_base = segment_base(ldt_selector);
 133         }
 134         d = (struct desc_struct *)(table_base + (selector & ~7));
 135         v = d->base0 | ((unsigned long)d->base1 << 16) |
 136                 ((unsigned long)d->base2 << 24);
 137 #ifdef CONFIG_X86_64
 138         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 139                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 140 #endif
 141         return v;
 142 }
 143 EXPORT_SYMBOL_GPL(segment_base);
 144
 145 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 146 {
 147         if (irqchip_in_kernel(vcpu->kvm))
 148                 return vcpu->arch.apic_base;
 149         else
 150                 return vcpu->arch.apic_base;
 151 }
 152 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 153
 154 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 155 {
 156         /* TODO: reserve bits check */
 157         if (irqchip_in_kernel(vcpu->kvm))
 158                 kvm_lapic_set_base(vcpu, data);
 159         else
 160                 vcpu->arch.apic_base = data;
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 163
 164 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 165 {
 166         WARN_ON(vcpu->arch.exception.pending);
 167         vcpu->arch.exception.pending = true;
 168         vcpu->arch.exception.has_error_code = false;
 169         vcpu->arch.exception.nr = nr;
 170 }
 171 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 172
 173 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 174                            u32 error_code)
 175 {
 176         ++vcpu->stat.pf_guest;
 177
 178         if (vcpu->arch.exception.pending) {
 179                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 180                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 181                                         " double fault 0x%lx\n", addr);
 182                         vcpu->arch.exception.nr = DF_VECTOR;
 183                         vcpu->arch.exception.error_code = 0;
 184                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 185                         /* triple fault -> shutdown */
 186                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 187                 }
 188                 return;
 189         }
 190         vcpu->arch.cr2 = addr;
 191         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 192 }
 193
 194 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 195 {
 196         vcpu->arch.nmi_pending = 1;
 197 }
 198 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 199
 200 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 201 {
 202         WARN_ON(vcpu->arch.exception.pending);
 203         vcpu->arch.exception.pending = true;
 204         vcpu->arch.exception.has_error_code = true;
 205         vcpu->arch.exception.nr = nr;
 206         vcpu->arch.exception.error_code = error_code;
 207 }
 208 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 209
 210 static void __queue_exception(struct kvm_vcpu *vcpu)
 211 {
 212         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 213                                      vcpu->arch.exception.has_error_code,
 214                                      vcpu->arch.exception.error_code);
 215 }
 216
 217 /*
 218  * Load the pae pdptrs.  Return true is they are all valid.
 219  */
 220 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 221 {
 222         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 223         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 224         int i;
 225         int ret;
 226         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 227
 228         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 229                                   offset * sizeof(u64), sizeof(pdpte));
 230         if (ret < 0) {
 231                 ret = 0;
 232                 goto out;
 233         }
 234         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 235                 if (is_present_pte(pdpte[i]) &&
 236                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 237                         ret = 0;
 238                         goto out;
 239                 }
 240         }
 241         ret = 1;
 242
 243         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 244 out:
 245
 246         return ret;
 247 }
 248 EXPORT_SYMBOL_GPL(load_pdptrs);
 249
 250 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 251 {
 252         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 253         bool changed = true;
 254         int r;
 255
 256         if (is_long_mode(vcpu) || !is_pae(vcpu))
 257                 return false;
 258
 259         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 260         if (r < 0)
 261                 goto out;
 262         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 263 out:
 264
 265         return changed;
 266 }
 267
 268 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 269 {
 270         if (cr0 & CR0_RESERVED_BITS) {
 271                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 272                        cr0, vcpu->arch.cr0);
 273                 kvm_inject_gp(vcpu, 0);
 274                 return;
 275         }
 276
 277         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 278                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 279                 kvm_inject_gp(vcpu, 0);
 280                 return;
 281         }
 282
 283         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 284                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 285                        "and a clear PE flag\n");
 286                 kvm_inject_gp(vcpu, 0);
 287                 return;
 288         }
 289
 290         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 291 #ifdef CONFIG_X86_64
 292                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 293                         int cs_db, cs_l;
 294
 295                         if (!is_pae(vcpu)) {
 296                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 297                                        "in long mode while PAE is disabled\n");
 298                                 kvm_inject_gp(vcpu, 0);
 299                                 return;
 300                         }
 301                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 302                         if (cs_l) {
 303                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 304                                        "in long mode while CS.L == 1\n");
 305                                 kvm_inject_gp(vcpu, 0);
 306                                 return;
 307
 308                         }
 309                 } else
 310 #endif
 311                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 312                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 313                                "reserved bits\n");
 314                         kvm_inject_gp(vcpu, 0);
 315                         return;
 316                 }
 317
 318         }
 319
 320         kvm_x86_ops->set_cr0(vcpu, cr0);
 321         vcpu->arch.cr0 = cr0;
 322
 323         kvm_mmu_reset_context(vcpu);
 324         return;
 325 }
 326 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 327
 328 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 329 {
 330         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 331         KVMTRACE_1D(LMSW, vcpu,
 332                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 333                     handler);
 334 }
 335 EXPORT_SYMBOL_GPL(kvm_lmsw);
 336
 337 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 338 {
 339         unsigned long old_cr4 = vcpu->arch.cr4;
 340         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 341
 342         if (cr4 & CR4_RESERVED_BITS) {
 343                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 344                 kvm_inject_gp(vcpu, 0);
 345                 return;
 346         }
 347
 348         if (is_long_mode(vcpu)) {
 349                 if (!(cr4 & X86_CR4_PAE)) {
 350                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 351                                "in long mode\n");
 352                         kvm_inject_gp(vcpu, 0);
 353                         return;
 354                 }
 355         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 356                    && ((cr4 ^ old_cr4) & pdptr_bits)
 357                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 358                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 359                 kvm_inject_gp(vcpu, 0);
 360                 return;
 361         }
 362
 363         if (cr4 & X86_CR4_VMXE) {
 364                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 365                 kvm_inject_gp(vcpu, 0);
 366                 return;
 367         }
 368         kvm_x86_ops->set_cr4(vcpu, cr4);
 369         vcpu->arch.cr4 = cr4;
 370         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 371         kvm_mmu_reset_context(vcpu);
 372 }
 373 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 374
 375 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 376 {
 377         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 378                 kvm_mmu_sync_roots(vcpu);
 379                 kvm_mmu_flush_tlb(vcpu);
 380                 return;
 381         }
 382
 383         if (is_long_mode(vcpu)) {
 384                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 385                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 386                         kvm_inject_gp(vcpu, 0);
 387                         return;
 388                 }
 389         } else {
 390                 if (is_pae(vcpu)) {
 391                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 392                                 printk(KERN_DEBUG
 393                                        "set_cr3: #GP, reserved bits\n");
 394                                 kvm_inject_gp(vcpu, 0);
 395                                 return;
 396                         }
 397                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 398                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 399                                        "reserved bits\n");
 400                                 kvm_inject_gp(vcpu, 0);
 401                                 return;
 402                         }
 403                 }
 404                 /*
 405                  * We don't check reserved bits in nonpae mode, because
 406                  * this isn't enforced, and VMware depends on this.
 407                  */
 408         }
 409
 410         /*
 411          * Does the new cr3 value map to physical memory? (Note, we
 412          * catch an invalid cr3 even in real-mode, because it would
 413          * cause trouble later on when we turn on paging anyway.)
 414          *
 415          * A real CPU would silently accept an invalid cr3 and would
 416          * attempt to use it - with largely undefined (and often hard
 417          * to debug) behavior on the guest side.
 418          */
 419         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 420                 kvm_inject_gp(vcpu, 0);
 421         else {
 422                 vcpu->arch.cr3 = cr3;
 423                 vcpu->arch.mmu.new_cr3(vcpu);
 424         }
 425 }
 426 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 427
 428 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 429 {
 430         if (cr8 & CR8_RESERVED_BITS) {
 431                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 432                 kvm_inject_gp(vcpu, 0);
 433                 return;
 434         }
 435         if (irqchip_in_kernel(vcpu->kvm))
 436                 kvm_lapic_set_tpr(vcpu, cr8);
 437         else
 438                 vcpu->arch.cr8 = cr8;
 439 }
 440 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 441
 442 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 443 {
 444         if (irqchip_in_kernel(vcpu->kvm))
 445                 return kvm_lapic_get_cr8(vcpu);
 446         else
 447                 return vcpu->arch.cr8;
 448 }
 449 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 450
 451 static inline u32 bit(int bitno)
 452 {
 453         return 1 << (bitno & 31);
 454 }
 455
 456 /*
 457  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 458  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 459  *
 460  * This list is modified at module load time to reflect the
 461  * capabilities of the host cpu.
 462  */
 463 static u32 msrs_to_save[] = {
 464         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 465         MSR_K6_STAR,
 466 #ifdef CONFIG_X86_64
 467         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 468 #endif
 469         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 470         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 471 };
 472
 473 static unsigned num_msrs_to_save;
 474
 475 static u32 emulated_msrs[] = {
 476         MSR_IA32_MISC_ENABLE,
 477 };
 478
 479 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 480 {
 481         if (efer & efer_reserved_bits) {
 482                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 483                        efer);
 484                 kvm_inject_gp(vcpu, 0);
 485                 return;
 486         }
 487
 488         if (is_paging(vcpu)
 489             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 490                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 491                 kvm_inject_gp(vcpu, 0);
 492                 return;
 493         }
 494
 495         if (efer & EFER_FFXSR) {
 496                 struct kvm_cpuid_entry2 *feat;
 497
 498                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 499                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 500                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 501                         kvm_inject_gp(vcpu, 0);
 502                         return;
 503                 }
 504         }
 505
 506         if (efer & EFER_SVME) {
 507                 struct kvm_cpuid_entry2 *feat;
 508
 509                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 510                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 511                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 512                         kvm_inject_gp(vcpu, 0);
 513                         return;
 514                 }
 515         }
 516
 517         kvm_x86_ops->set_efer(vcpu, efer);
 518
 519         efer &= ~EFER_LMA;
 520         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 521
 522         vcpu->arch.shadow_efer = efer;
 523
 524         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 525         kvm_mmu_reset_context(vcpu);
 526 }
 527
 528 void kvm_enable_efer_bits(u64 mask)
 529 {
 530        efer_reserved_bits &= ~mask;
 531 }
 532 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 533
 534
 535 /*
 536  * Writes msr value into into the appropriate "register".
 537  * Returns 0 on success, non-0 otherwise.
 538  * Assumes vcpu_load() was already called.
 539  */
 540 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 541 {
 542         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 543 }
 544
 545 /*
 546  * Adapt set_msr() to msr_io()'s calling convention
 547  */
 548 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 549 {
 550         return kvm_set_msr(vcpu, index, *data);
 551 }
 552
 553 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 554 {
 555         static int version;
 556         struct pvclock_wall_clock wc;
 557         struct timespec now, sys, boot;
 558
 559         if (!wall_clock)
 560                 return;
 561
 562         version++;
 563
 564         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 565
 566         /*
 567          * The guest calculates current wall clock time by adding
 568          * system time (updated by kvm_write_guest_time below) to the
 569          * wall clock specified here.  guest system time equals host
 570          * system time for us, thus we must fill in host boot time here.
 571          */
 572         now = current_kernel_time();
 573         ktime_get_ts(&sys);
 574         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 575
 576         wc.sec = boot.tv_sec;
 577         wc.nsec = boot.tv_nsec;
 578         wc.version = version;
 579
 580         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 581
 582         version++;
 583         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 584 }
 585
 586 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 587 {
 588         uint32_t quotient, remainder;
 589
 590         /* Don't try to replace with do_div(), this one calculates
 591          * "(dividend << 32) / divisor" */
 592         __asm__ ( "divl %4"
 593                   : "=a" (quotient), "=d" (remainder)
 594                   : "0" (0), "1" (dividend), "r" (divisor) );
 595         return quotient;
 596 }
 597
 598 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 599 {
 600         uint64_t nsecs = 1000000000LL;
 601         int32_t  shift = 0;
 602         uint64_t tps64;
 603         uint32_t tps32;
 604
 605         tps64 = tsc_khz * 1000LL;
 606         while (tps64 > nsecs*2) {
 607                 tps64 >>= 1;
 608                 shift--;
 609         }
 610
 611         tps32 = (uint32_t)tps64;
 612         while (tps32 <= (uint32_t)nsecs) {
 613                 tps32 <<= 1;
 614                 shift++;
 615         }
 616
 617         hv_clock->tsc_shift = shift;
 618         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 619
 620         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 621                  __func__, tsc_khz, hv_clock->tsc_shift,
 622                  hv_clock->tsc_to_system_mul);
 623 }
 624
 625 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 626
 627 static void kvm_write_guest_time(struct kvm_vcpu *v)
 628 {
 629         struct timespec ts;
 630         unsigned long flags;
 631         struct kvm_vcpu_arch *vcpu = &v->arch;
 632         void *shared_kaddr;
 633         unsigned long this_tsc_khz;
 634
 635         if ((!vcpu->time_page))
 636                 return;
 637
 638         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 639         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 640                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 641                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 642         }
 643         put_cpu_var(cpu_tsc_khz);
 644
 645         /* Keep irq disabled to prevent changes to the clock */
 646         local_irq_save(flags);
 647         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 648                           &vcpu->hv_clock.tsc_timestamp);
 649         ktime_get_ts(&ts);
 650         local_irq_restore(flags);
 651
 652         /* With all the info we got, fill in the values */
 653
 654         vcpu->hv_clock.system_time = ts.tv_nsec +
 655                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 656         /*
 657          * The interface expects us to write an even number signaling that the
 658          * update is finished. Since the guest won't see the intermediate
 659          * state, we just increase by 2 at the end.
 660          */
 661         vcpu->hv_clock.version += 2;
 662
 663         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 664
 665         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 666                sizeof(vcpu->hv_clock));
 667
 668         kunmap_atomic(shared_kaddr, KM_USER0);
 669
 670         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 671 }
 672
 673 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 674 {
 675         struct kvm_vcpu_arch *vcpu = &v->arch;
 676
 677         if (!vcpu->time_page)
 678                 return 0;
 679         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 680         return 1;
 681 }
 682
 683 static bool msr_mtrr_valid(unsigned msr)
 684 {
 685         switch (msr) {
 686         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 687         case MSR_MTRRfix64K_00000:
 688         case MSR_MTRRfix16K_80000:
 689         case MSR_MTRRfix16K_A0000:
 690         case MSR_MTRRfix4K_C0000:
 691         case MSR_MTRRfix4K_C8000:
 692         case MSR_MTRRfix4K_D0000:
 693         case MSR_MTRRfix4K_D8000:
 694         case MSR_MTRRfix4K_E0000:
 695         case MSR_MTRRfix4K_E8000:
 696         case MSR_MTRRfix4K_F0000:
 697         case MSR_MTRRfix4K_F8000:
 698         case MSR_MTRRdefType:
 699         case MSR_IA32_CR_PAT:
 700                 return true;
 701         case 0x2f8:
 702                 return true;
 703         }
 704         return false;
 705 }
 706
 707 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 708 {
 709         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 710
 711         if (!msr_mtrr_valid(msr))
 712                 return 1;
 713
 714         if (msr == MSR_MTRRdefType) {
 715                 vcpu->arch.mtrr_state.def_type = data;
 716                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 717         } else if (msr == MSR_MTRRfix64K_00000)
 718                 p[0] = data;
 719         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 720                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 721         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 722                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 723         else if (msr == MSR_IA32_CR_PAT)
 724                 vcpu->arch.pat = data;
 725         else {  /* Variable MTRRs */
 726                 int idx, is_mtrr_mask;
 727                 u64 *pt;
 728
 729                 idx = (msr - 0x200) / 2;
 730                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 731                 if (!is_mtrr_mask)
 732                         pt =
 733                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 734                 else
 735                         pt =
 736                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 737                 *pt = data;
 738         }
 739
 740         kvm_mmu_reset_context(vcpu);
 741         return 0;
 742 }
 743
 744 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 745 {
 746         switch (msr) {
 747         case MSR_EFER:
 748                 set_efer(vcpu, data);
 749                 break;
 750         case MSR_IA32_MC0_STATUS:
 751                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 752                        __func__, data);
 753                 break;
 754         case MSR_IA32_MCG_STATUS:
 755                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 756                         __func__, data);
 757                 break;
 758         case MSR_IA32_MCG_CTL:
 759                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 760                         __func__, data);
 761                 break;
 762         case MSR_IA32_DEBUGCTLMSR:
 763                 if (!data) {
 764                         /* We support the non-activated case already */
 765                         break;
 766                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 767                         /* Values other than LBR and BTF are vendor-specific,
 768                            thus reserved and should throw a #GP */
 769                         return 1;
 770                 }
 771                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 772                         __func__, data);
 773                 break;
 774         case MSR_IA32_UCODE_REV:
 775         case MSR_IA32_UCODE_WRITE:
 776         case MSR_VM_HSAVE_PA:
 777                 break;
 778         case 0x200 ... 0x2ff:
 779                 return set_msr_mtrr(vcpu, msr, data);
 780         case MSR_IA32_APICBASE:
 781                 kvm_set_apic_base(vcpu, data);
 782                 break;
 783         case MSR_IA32_MISC_ENABLE:
 784                 vcpu->arch.ia32_misc_enable_msr = data;
 785                 break;
 786         case MSR_KVM_WALL_CLOCK:
 787                 vcpu->kvm->arch.wall_clock = data;
 788                 kvm_write_wall_clock(vcpu->kvm, data);
 789                 break;
 790         case MSR_KVM_SYSTEM_TIME: {
 791                 if (vcpu->arch.time_page) {
 792                         kvm_release_page_dirty(vcpu->arch.time_page);
 793                         vcpu->arch.time_page = NULL;
 794                 }
 795
 796                 vcpu->arch.time = data;
 797
 798                 /* we verify if the enable bit is set... */
 799                 if (!(data & 1))
 800                         break;
 801
 802                 /* ...but clean it before doing the actual write */
 803                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 804
 805                 vcpu->arch.time_page =
 806                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 807
 808                 if (is_error_page(vcpu->arch.time_page)) {
 809                         kvm_release_page_clean(vcpu->arch.time_page);
 810                         vcpu->arch.time_page = NULL;
 811                 }
 812
 813                 kvm_request_guest_time_update(vcpu);
 814                 break;
 815         }
 816         default:
 817                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 818                 return 1;
 819         }
 820         return 0;
 821 }
 822 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 823
 824
 825 /*
 826  * Reads an msr value (of 'msr_index') into 'pdata'.
 827  * Returns 0 on success, non-0 otherwise.
 828  * Assumes vcpu_load() was already called.
 829  */
 830 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 831 {
 832         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 833 }
 834
 835 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 836 {
 837         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 838
 839         if (!msr_mtrr_valid(msr))
 840                 return 1;
 841
 842         if (msr == MSR_MTRRdefType)
 843                 *pdata = vcpu->arch.mtrr_state.def_type +
 844                          (vcpu->arch.mtrr_state.enabled << 10);
 845         else if (msr == MSR_MTRRfix64K_00000)
 846                 *pdata = p[0];
 847         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 848                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 849         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 850                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 851         else if (msr == MSR_IA32_CR_PAT)
 852                 *pdata = vcpu->arch.pat;
 853         else {  /* Variable MTRRs */
 854                 int idx, is_mtrr_mask;
 855                 u64 *pt;
 856
 857                 idx = (msr - 0x200) / 2;
 858                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 859                 if (!is_mtrr_mask)
 860                         pt =
 861                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 862                 else
 863                         pt =
 864                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 865                 *pdata = *pt;
 866         }
 867
 868         return 0;
 869 }
 870
 871 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 872 {
 873         u64 data;
 874
 875         switch (msr) {
 876         case 0xc0010010: /* SYSCFG */
 877         case 0xc0010015: /* HWCR */
 878         case MSR_IA32_PLATFORM_ID:
 879         case MSR_IA32_P5_MC_ADDR:
 880         case MSR_IA32_P5_MC_TYPE:
 881         case MSR_IA32_MC0_CTL:
 882         case MSR_IA32_MCG_STATUS:
 883         case MSR_IA32_MCG_CAP:
 884         case MSR_IA32_MCG_CTL:
 885         case MSR_IA32_MC0_MISC:
 886         case MSR_IA32_MC0_MISC+4:
 887         case MSR_IA32_MC0_MISC+8:
 888         case MSR_IA32_MC0_MISC+12:
 889         case MSR_IA32_MC0_MISC+16:
 890         case MSR_IA32_MC0_MISC+20:
 891         case MSR_IA32_UCODE_REV:
 892         case MSR_IA32_EBL_CR_POWERON:
 893         case MSR_IA32_DEBUGCTLMSR:
 894         case MSR_IA32_LASTBRANCHFROMIP:
 895         case MSR_IA32_LASTBRANCHTOIP:
 896         case MSR_IA32_LASTINTFROMIP:
 897         case MSR_IA32_LASTINTTOIP:
 898         case MSR_VM_HSAVE_PA:
 899         case MSR_P6_EVNTSEL0:
 900         case MSR_P6_EVNTSEL1:
 901                 data = 0;
 902                 break;
 903         case MSR_MTRRcap:
 904                 data = 0x500 | KVM_NR_VAR_MTRR;
 905                 break;
 906         case 0x200 ... 0x2ff:
 907                 return get_msr_mtrr(vcpu, msr, pdata);
 908         case 0xcd: /* fsb frequency */
 909                 data = 3;
 910                 break;
 911         case MSR_IA32_APICBASE:
 912                 data = kvm_get_apic_base(vcpu);
 913                 break;
 914         case MSR_IA32_MISC_ENABLE:
 915                 data = vcpu->arch.ia32_misc_enable_msr;
 916                 break;
 917         case MSR_IA32_PERF_STATUS:
 918                 /* TSC increment by tick */
 919                 data = 1000ULL;
 920                 /* CPU multiplier */
 921                 data |= (((uint64_t)4ULL) << 40);
 922                 break;
 923         case MSR_EFER:
 924                 data = vcpu->arch.shadow_efer;
 925                 break;
 926         case MSR_KVM_WALL_CLOCK:
 927                 data = vcpu->kvm->arch.wall_clock;
 928                 break;
 929         case MSR_KVM_SYSTEM_TIME:
 930                 data = vcpu->arch.time;
 931                 break;
 932         default:
 933                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 934                 return 1;
 935         }
 936         *pdata = data;
 937         return 0;
 938 }
 939 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 940
 941 /*
 942  * Read or write a bunch of msrs. All parameters are kernel addresses.
 943  *
 944  * @return number of msrs set successfully.
 945  */
 946 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 947                     struct kvm_msr_entry *entries,
 948                     int (*do_msr)(struct kvm_vcpu *vcpu,
 949                                   unsigned index, u64 *data))
 950 {
 951         int i;
 952
 953         vcpu_load(vcpu);
 954
 955         down_read(&vcpu->kvm->slots_lock);
 956         for (i = 0; i < msrs->nmsrs; ++i)
 957                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 958                         break;
 959         up_read(&vcpu->kvm->slots_lock);
 960
 961         vcpu_put(vcpu);
 962
 963         return i;
 964 }
 965
 966 /*
 967  * Read or write a bunch of msrs. Parameters are user addresses.
 968  *
 969  * @return number of msrs set successfully.
 970  */
 971 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 972                   int (*do_msr)(struct kvm_vcpu *vcpu,
 973                                 unsigned index, u64 *data),
 974                   int writeback)
 975 {
 976         struct kvm_msrs msrs;
 977         struct kvm_msr_entry *entries;
 978         int r, n;
 979         unsigned size;
 980
 981         r = -EFAULT;
 982         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 983                 goto out;
 984
 985         r = -E2BIG;
 986         if (msrs.nmsrs >= MAX_IO_MSRS)
 987                 goto out;
 988
 989         r = -ENOMEM;
 990         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 991         entries = vmalloc(size);
 992         if (!entries)
 993                 goto out;
 994
 995         r = -EFAULT;
 996         if (copy_from_user(entries, user_msrs->entries, size))
 997                 goto out_free;
 998
 999         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1000         if (r < 0)
1001                 goto out_free;
1002
1003         r = -EFAULT;
1004         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1005                 goto out_free;
1006
1007         r = n;
1008
1009 out_free:
1010         vfree(entries);
1011 out:
1012         return r;
1013 }
1014
1015 int kvm_dev_ioctl_check_extension(long ext)
1016 {
1017         int r;
1018
1019         switch (ext) {
1020         case KVM_CAP_IRQCHIP:
1021         case KVM_CAP_HLT:
1022         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1023         case KVM_CAP_SET_TSS_ADDR:
1024         case KVM_CAP_EXT_CPUID:
1025         case KVM_CAP_CLOCKSOURCE:
1026         case KVM_CAP_PIT:
1027         case KVM_CAP_NOP_IO_DELAY:
1028         case KVM_CAP_MP_STATE:
1029         case KVM_CAP_SYNC_MMU:
1030         case KVM_CAP_REINJECT_CONTROL:
1031         case KVM_CAP_IRQ_INJECT_STATUS:
1032         case KVM_CAP_ASSIGN_DEV_IRQ:
1033                 r = 1;
1034                 break;
1035         case KVM_CAP_COALESCED_MMIO:
1036                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1037                 break;
1038         case KVM_CAP_VAPIC:
1039                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1040                 break;
1041         case KVM_CAP_NR_VCPUS:
1042                 r = KVM_MAX_VCPUS;
1043                 break;
1044         case KVM_CAP_NR_MEMSLOTS:
1045                 r = KVM_MEMORY_SLOTS;
1046                 break;
1047         case KVM_CAP_PV_MMU:
1048                 r = !tdp_enabled;
1049                 break;
1050         case KVM_CAP_IOMMU:
1051                 r = iommu_found();
1052                 break;
1053         default:
1054                 r = 0;
1055                 break;
1056         }
1057         return r;
1058
1059 }
1060
1061 long kvm_arch_dev_ioctl(struct file *filp,
1062                         unsigned int ioctl, unsigned long arg)
1063 {
1064         void __user *argp = (void __user *)arg;
1065         long r;
1066
1067         switch (ioctl) {
1068         case KVM_GET_MSR_INDEX_LIST: {
1069                 struct kvm_msr_list __user *user_msr_list = argp;
1070                 struct kvm_msr_list msr_list;
1071                 unsigned n;
1072
1073                 r = -EFAULT;
1074                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1075                         goto out;
1076                 n = msr_list.nmsrs;
1077                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1078                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1079                         goto out;
1080                 r = -E2BIG;
1081                 if (n < num_msrs_to_save)
1082                         goto out;
1083                 r = -EFAULT;
1084                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1085                                  num_msrs_to_save * sizeof(u32)))
1086                         goto out;
1087                 if (copy_to_user(user_msr_list->indices
1088                                  + num_msrs_to_save * sizeof(u32),
1089                                  &emulated_msrs,
1090                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1091                         goto out;
1092                 r = 0;
1093                 break;
1094         }
1095         case KVM_GET_SUPPORTED_CPUID: {
1096                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1097                 struct kvm_cpuid2 cpuid;
1098
1099                 r = -EFAULT;
1100                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1101                         goto out;
1102                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1103                                                       cpuid_arg->entries);
1104                 if (r)
1105                         goto out;
1106
1107                 r = -EFAULT;
1108                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1109                         goto out;
1110                 r = 0;
1111                 break;
1112         }
1113         default:
1114                 r = -EINVAL;
1115         }
1116 out:
1117         return r;
1118 }
1119
1120 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1121 {
1122         kvm_x86_ops->vcpu_load(vcpu, cpu);
1123         kvm_request_guest_time_update(vcpu);
1124 }
1125
1126 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1127 {
1128         kvm_x86_ops->vcpu_put(vcpu);
1129         kvm_put_guest_fpu(vcpu);
1130 }
1131
1132 static int is_efer_nx(void)
1133 {
1134         unsigned long long efer = 0;
1135
1136         rdmsrl_safe(MSR_EFER, &efer);
1137         return efer & EFER_NX;
1138 }
1139
1140 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1141 {
1142         int i;
1143         struct kvm_cpuid_entry2 *e, *entry;
1144
1145         entry = NULL;
1146         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1147                 e = &vcpu->arch.cpuid_entries[i];
1148                 if (e->function == 0x80000001) {
1149                         entry = e;
1150                         break;
1151                 }
1152         }
1153         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1154                 entry->edx &= ~(1 << 20);
1155                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1156         }
1157 }
1158
1159 /* when an old userspace process fills a new kernel module */
1160 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1161                                     struct kvm_cpuid *cpuid,
1162                                     struct kvm_cpuid_entry __user *entries)
1163 {
1164         int r, i;
1165         struct kvm_cpuid_entry *cpuid_entries;
1166
1167         r = -E2BIG;
1168         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1169                 goto out;
1170         r = -ENOMEM;
1171         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1172         if (!cpuid_entries)
1173                 goto out;
1174         r = -EFAULT;
1175         if (copy_from_user(cpuid_entries, entries,
1176                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1177                 goto out_free;
1178         for (i = 0; i < cpuid->nent; i++) {
1179                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1180                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1181                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1182                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1183                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1184                 vcpu->arch.cpuid_entries[i].index = 0;
1185                 vcpu->arch.cpuid_entries[i].flags = 0;
1186                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1187                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1188                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1189         }
1190         vcpu->arch.cpuid_nent = cpuid->nent;
1191         cpuid_fix_nx_cap(vcpu);
1192         r = 0;
1193
1194 out_free:
1195         vfree(cpuid_entries);
1196 out:
1197         return r;
1198 }
1199
1200 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1201                                      struct kvm_cpuid2 *cpuid,
1202                                      struct kvm_cpuid_entry2 __user *entries)
1203 {
1204         int r;
1205
1206         r = -E2BIG;
1207         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1208                 goto out;
1209         r = -EFAULT;
1210         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1211                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1212                 goto out;
1213         vcpu->arch.cpuid_nent = cpuid->nent;
1214         return 0;
1215
1216 out:
1217         return r;
1218 }
1219
1220 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1221                                      struct kvm_cpuid2 *cpuid,
1222                                      struct kvm_cpuid_entry2 __user *entries)
1223 {
1224         int r;
1225
1226         r = -E2BIG;
1227         if (cpuid->nent < vcpu->arch.cpuid_nent)
1228                 goto out;
1229         r = -EFAULT;
1230         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1231                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1232                 goto out;
1233         return 0;
1234
1235 out:
1236         cpuid->nent = vcpu->arch.cpuid_nent;
1237         return r;
1238 }
1239
1240 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1241                            u32 index)
1242 {
1243         entry->function = function;
1244         entry->index = index;
1245         cpuid_count(entry->function, entry->index,
1246                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1247         entry->flags = 0;
1248 }
1249
1250 #define F(x) bit(X86_FEATURE_##x)
1251
1252 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1253                          u32 index, int *nent, int maxnent)
1254 {
1255         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1256 #ifdef CONFIG_X86_64
1257         unsigned f_lm = F(LM);
1258 #else
1259         unsigned f_lm = 0;
1260 #endif
1261
1262         /* cpuid 1.edx */
1263         const u32 kvm_supported_word0_x86_features =
1264                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1265                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1266                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1267                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1268                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1269                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1270                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1271                 0 /* HTT, TM, Reserved, PBE */;
1272         /* cpuid 0x80000001.edx */
1273         const u32 kvm_supported_word1_x86_features =
1274                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1275                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1276                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1277                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1278                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1279                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1280                 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1281                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1282         /* cpuid 1.ecx */
1283         const u32 kvm_supported_word4_x86_features =
1284                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1285                 0 /* DS-CPL, VMX, SMX, EST */ |
1286                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1287                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1288                 0 /* Reserved, DCA */ | F(XMM4_1) |
1289                 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1290                 0 /* Reserved, XSAVE, OSXSAVE */;
1291         /* cpuid 0x80000001.ecx */
1292         const u32 kvm_supported_word6_x86_features =
1293                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1294                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1295                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1296                 0 /* SKINIT */ | 0 /* WDT */;
1297
1298         /* all calls to cpuid_count() should be made on the same cpu */
1299         get_cpu();
1300         do_cpuid_1_ent(entry, function, index);
1301         ++*nent;
1302
1303         switch (function) {
1304         case 0:
1305                 entry->eax = min(entry->eax, (u32)0xb);
1306                 break;
1307         case 1:
1308                 entry->edx &= kvm_supported_word0_x86_features;
1309                 entry->ecx &= kvm_supported_word4_x86_features;
1310                 break;
1311         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1312          * may return different values. This forces us to get_cpu() before
1313          * issuing the first command, and also to emulate this annoying behavior
1314          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1315         case 2: {
1316                 int t, times = entry->eax & 0xff;
1317
1318                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1319                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1320                 for (t = 1; t < times && *nent < maxnent; ++t) {
1321                         do_cpuid_1_ent(&entry[t], function, 0);
1322                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1323                         ++*nent;
1324                 }
1325                 break;
1326         }
1327         /* function 4 and 0xb have additional index. */
1328         case 4: {
1329                 int i, cache_type;
1330
1331                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1332                 /* read more entries until cache_type is zero */
1333                 for (i = 1; *nent < maxnent; ++i) {
1334                         cache_type = entry[i - 1].eax & 0x1f;
1335                         if (!cache_type)
1336                                 break;
1337                         do_cpuid_1_ent(&entry[i], function, i);
1338                         entry[i].flags |=
1339                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1340                         ++*nent;
1341                 }
1342                 break;
1343         }
1344         case 0xb: {
1345                 int i, level_type;
1346
1347                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1348                 /* read more entries until level_type is zero */
1349                 for (i = 1; *nent < maxnent; ++i) {
1350                         level_type = entry[i - 1].ecx & 0xff00;
1351                         if (!level_type)
1352                                 break;
1353                         do_cpuid_1_ent(&entry[i], function, i);
1354                         entry[i].flags |=
1355                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1356                         ++*nent;
1357                 }
1358                 break;
1359         }
1360         case 0x80000000:
1361                 entry->eax = min(entry->eax, 0x8000001a);
1362                 break;
1363         case 0x80000001:
1364                 entry->edx &= kvm_supported_word1_x86_features;
1365                 entry->ecx &= kvm_supported_word6_x86_features;
1366                 break;
1367         }
1368         put_cpu();
1369 }
1370
1371 #undef F
1372
1373 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1374                                      struct kvm_cpuid_entry2 __user *entries)
1375 {
1376         struct kvm_cpuid_entry2 *cpuid_entries;
1377         int limit, nent = 0, r = -E2BIG;
1378         u32 func;
1379
1380         if (cpuid->nent < 1)
1381                 goto out;
1382         r = -ENOMEM;
1383         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1384         if (!cpuid_entries)
1385                 goto out;
1386
1387         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1388         limit = cpuid_entries[0].eax;
1389         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1390                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1391                              &nent, cpuid->nent);
1392         r = -E2BIG;
1393         if (nent >= cpuid->nent)
1394                 goto out_free;
1395
1396         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1397         limit = cpuid_entries[nent - 1].eax;
1398         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1399                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1400                              &nent, cpuid->nent);
1401         r = -EFAULT;
1402         if (copy_to_user(entries, cpuid_entries,
1403                          nent * sizeof(struct kvm_cpuid_entry2)))
1404                 goto out_free;
1405         cpuid->nent = nent;
1406         r = 0;
1407
1408 out_free:
1409         vfree(cpuid_entries);
1410 out:
1411         return r;
1412 }
1413
1414 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1415                                     struct kvm_lapic_state *s)
1416 {
1417         vcpu_load(vcpu);
1418         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1419         vcpu_put(vcpu);
1420
1421         return 0;
1422 }
1423
1424 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1425                                     struct kvm_lapic_state *s)
1426 {
1427         vcpu_load(vcpu);
1428         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1429         kvm_apic_post_state_restore(vcpu);
1430         vcpu_put(vcpu);
1431
1432         return 0;
1433 }
1434
1435 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1436                                     struct kvm_interrupt *irq)
1437 {
1438         if (irq->irq < 0 || irq->irq >= 256)
1439                 return -EINVAL;
1440         if (irqchip_in_kernel(vcpu->kvm))
1441                 return -ENXIO;
1442         vcpu_load(vcpu);
1443
1444         kvm_queue_interrupt(vcpu, irq->irq, false);
1445
1446         vcpu_put(vcpu);
1447
1448         return 0;
1449 }
1450
1451 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1452 {
1453         vcpu_load(vcpu);
1454         kvm_inject_nmi(vcpu);
1455         vcpu_put(vcpu);
1456
1457         return 0;
1458 }
1459
1460 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1461                                            struct kvm_tpr_access_ctl *tac)
1462 {
1463         if (tac->flags)
1464                 return -EINVAL;
1465         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1466         return 0;
1467 }
1468
1469 long kvm_arch_vcpu_ioctl(struct file *filp,
1470                          unsigned int ioctl, unsigned long arg)
1471 {
1472         struct kvm_vcpu *vcpu = filp->private_data;
1473         void __user *argp = (void __user *)arg;
1474         int r;
1475         struct kvm_lapic_state *lapic = NULL;
1476
1477         switch (ioctl) {
1478         case KVM_GET_LAPIC: {
1479                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1480
1481                 r = -ENOMEM;
1482                 if (!lapic)
1483                         goto out;
1484                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1485                 if (r)
1486                         goto out;
1487                 r = -EFAULT;
1488                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1489                         goto out;
1490                 r = 0;
1491                 break;
1492         }
1493         case KVM_SET_LAPIC: {
1494                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1495                 r = -ENOMEM;
1496                 if (!lapic)
1497                         goto out;
1498                 r = -EFAULT;
1499                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1500                         goto out;
1501                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1502                 if (r)
1503                         goto out;
1504                 r = 0;
1505                 break;
1506         }
1507         case KVM_INTERRUPT: {
1508                 struct kvm_interrupt irq;
1509
1510                 r = -EFAULT;
1511                 if (copy_from_user(&irq, argp, sizeof irq))
1512                         goto out;
1513                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1514                 if (r)
1515                         goto out;
1516                 r = 0;
1517                 break;
1518         }
1519         case KVM_NMI: {
1520                 r = kvm_vcpu_ioctl_nmi(vcpu);
1521                 if (r)
1522                         goto out;
1523                 r = 0;
1524                 break;
1525         }
1526         case KVM_SET_CPUID: {
1527                 struct kvm_cpuid __user *cpuid_arg = argp;
1528                 struct kvm_cpuid cpuid;
1529
1530                 r = -EFAULT;
1531                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1532                         goto out;
1533                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1534                 if (r)
1535                         goto out;
1536                 break;
1537         }
1538         case KVM_SET_CPUID2: {
1539                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1540                 struct kvm_cpuid2 cpuid;
1541
1542                 r = -EFAULT;
1543                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1544                         goto out;
1545                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1546                                               cpuid_arg->entries);
1547                 if (r)
1548                         goto out;
1549                 break;
1550         }
1551         case KVM_GET_CPUID2: {
1552                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1553                 struct kvm_cpuid2 cpuid;
1554
1555                 r = -EFAULT;
1556                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1557                         goto out;
1558                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1559                                               cpuid_arg->entries);
1560                 if (r)
1561                         goto out;
1562                 r = -EFAULT;
1563                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1564                         goto out;
1565                 r = 0;
1566                 break;
1567         }
1568         case KVM_GET_MSRS:
1569                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1570                 break;
1571         case KVM_SET_MSRS:
1572                 r = msr_io(vcpu, argp, do_set_msr, 0);
1573                 break;
1574         case KVM_TPR_ACCESS_REPORTING: {
1575                 struct kvm_tpr_access_ctl tac;
1576
1577                 r = -EFAULT;
1578                 if (copy_from_user(&tac, argp, sizeof tac))
1579                         goto out;
1580                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1581                 if (r)
1582                         goto out;
1583                 r = -EFAULT;
1584                 if (copy_to_user(argp, &tac, sizeof tac))
1585                         goto out;
1586                 r = 0;
1587                 break;
1588         };
1589         case KVM_SET_VAPIC_ADDR: {
1590                 struct kvm_vapic_addr va;
1591
1592                 r = -EINVAL;
1593                 if (!irqchip_in_kernel(vcpu->kvm))
1594                         goto out;
1595                 r = -EFAULT;
1596                 if (copy_from_user(&va, argp, sizeof va))
1597                         goto out;
1598                 r = 0;
1599                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1600                 break;
1601         }
1602         default:
1603                 r = -EINVAL;
1604         }
1605 out:
1606         kfree(lapic);
1607         return r;
1608 }
1609
1610 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1611 {
1612         int ret;
1613
1614         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1615                 return -1;
1616         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1617         return ret;
1618 }
1619
1620 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1621                                           u32 kvm_nr_mmu_pages)
1622 {
1623         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1624                 return -EINVAL;
1625
1626         down_write(&kvm->slots_lock);
1627         spin_lock(&kvm->mmu_lock);
1628
1629         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1630         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1631
1632         spin_unlock(&kvm->mmu_lock);
1633         up_write(&kvm->slots_lock);
1634         return 0;
1635 }
1636
1637 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1638 {
1639         return kvm->arch.n_alloc_mmu_pages;
1640 }
1641
1642 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1643 {
1644         int i;
1645         struct kvm_mem_alias *alias;
1646
1647         for (i = 0; i < kvm->arch.naliases; ++i) {
1648                 alias = &kvm->arch.aliases[i];
1649                 if (gfn >= alias->base_gfn
1650                     && gfn < alias->base_gfn + alias->npages)
1651                         return alias->target_gfn + gfn - alias->base_gfn;
1652         }
1653         return gfn;
1654 }
1655
1656 /*
1657  * Set a new alias region.  Aliases map a portion of physical memory into
1658  * another portion.  This is useful for memory windows, for example the PC
1659  * VGA region.
1660  */
1661 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1662                                          struct kvm_memory_alias *alias)
1663 {
1664         int r, n;
1665         struct kvm_mem_alias *p;
1666
1667         r = -EINVAL;
1668         /* General sanity checks */
1669         if (alias->memory_size & (PAGE_SIZE - 1))
1670                 goto out;
1671         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1672                 goto out;
1673         if (alias->slot >= KVM_ALIAS_SLOTS)
1674                 goto out;
1675         if (alias->guest_phys_addr + alias->memory_size
1676             < alias->guest_phys_addr)
1677                 goto out;
1678         if (alias->target_phys_addr + alias->memory_size
1679             < alias->target_phys_addr)
1680                 goto out;
1681
1682         down_write(&kvm->slots_lock);
1683         spin_lock(&kvm->mmu_lock);
1684
1685         p = &kvm->arch.aliases[alias->slot];
1686         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1687         p->npages = alias->memory_size >> PAGE_SHIFT;
1688         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1689
1690         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1691                 if (kvm->arch.aliases[n - 1].npages)
1692                         break;
1693         kvm->arch.naliases = n;
1694
1695         spin_unlock(&kvm->mmu_lock);
1696         kvm_mmu_zap_all(kvm);
1697
1698         up_write(&kvm->slots_lock);
1699
1700         return 0;
1701
1702 out:
1703         return r;
1704 }
1705
1706 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1707 {
1708         int r;
1709
1710         r = 0;
1711         switch (chip->chip_id) {
1712         case KVM_IRQCHIP_PIC_MASTER:
1713                 memcpy(&chip->chip.pic,
1714                         &pic_irqchip(kvm)->pics[0],
1715                         sizeof(struct kvm_pic_state));
1716                 break;
1717         case KVM_IRQCHIP_PIC_SLAVE:
1718                 memcpy(&chip->chip.pic,
1719                         &pic_irqchip(kvm)->pics[1],
1720                         sizeof(struct kvm_pic_state));
1721                 break;
1722         case KVM_IRQCHIP_IOAPIC:
1723                 memcpy(&chip->chip.ioapic,
1724                         ioapic_irqchip(kvm),
1725                         sizeof(struct kvm_ioapic_state));
1726                 break;
1727         default:
1728                 r = -EINVAL;
1729                 break;
1730         }
1731         return r;
1732 }
1733
1734 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1735 {
1736         int r;
1737
1738         r = 0;
1739         switch (chip->chip_id) {
1740         case KVM_IRQCHIP_PIC_MASTER:
1741                 memcpy(&pic_irqchip(kvm)->pics[0],
1742                         &chip->chip.pic,
1743                         sizeof(struct kvm_pic_state));
1744                 break;
1745         case KVM_IRQCHIP_PIC_SLAVE:
1746                 memcpy(&pic_irqchip(kvm)->pics[1],
1747                         &chip->chip.pic,
1748                         sizeof(struct kvm_pic_state));
1749                 break;
1750         case KVM_IRQCHIP_IOAPIC:
1751                 memcpy(ioapic_irqchip(kvm),
1752                         &chip->chip.ioapic,
1753                         sizeof(struct kvm_ioapic_state));
1754                 break;
1755         default:
1756                 r = -EINVAL;
1757                 break;
1758         }
1759         kvm_pic_update_irq(pic_irqchip(kvm));
1760         return r;
1761 }
1762
1763 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1764 {
1765         int r = 0;
1766
1767         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1768         return r;
1769 }
1770
1771 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1772 {
1773         int r = 0;
1774
1775         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1776         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1777         return r;
1778 }
1779
1780 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1781                                  struct kvm_reinject_control *control)
1782 {
1783         if (!kvm->arch.vpit)
1784                 return -ENXIO;
1785         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1786         return 0;
1787 }
1788
1789 /*
1790  * Get (and clear) the dirty memory log for a memory slot.
1791  */
1792 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1793                                       struct kvm_dirty_log *log)
1794 {
1795         int r;
1796         int n;
1797         struct kvm_memory_slot *memslot;
1798         int is_dirty = 0;
1799
1800         down_write(&kvm->slots_lock);
1801
1802         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1803         if (r)
1804                 goto out;
1805
1806         /* If nothing is dirty, don't bother messing with page tables. */
1807         if (is_dirty) {
1808                 spin_lock(&kvm->mmu_lock);
1809                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1810                 spin_unlock(&kvm->mmu_lock);
1811                 kvm_flush_remote_tlbs(kvm);
1812                 memslot = &kvm->memslots[log->slot];
1813                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1814                 memset(memslot->dirty_bitmap, 0, n);
1815         }
1816         r = 0;
1817 out:
1818         up_write(&kvm->slots_lock);
1819         return r;
1820 }
1821
1822 long kvm_arch_vm_ioctl(struct file *filp,
1823                        unsigned int ioctl, unsigned long arg)
1824 {
1825         struct kvm *kvm = filp->private_data;
1826         void __user *argp = (void __user *)arg;
1827         int r = -EINVAL;
1828         /*
1829          * This union makes it completely explicit to gcc-3.x
1830          * that these two variables' stack usage should be
1831          * combined, not added together.
1832          */
1833         union {
1834                 struct kvm_pit_state ps;
1835                 struct kvm_memory_alias alias;
1836         } u;
1837
1838         switch (ioctl) {
1839         case KVM_SET_TSS_ADDR:
1840                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1841                 if (r < 0)
1842                         goto out;
1843                 break;
1844         case KVM_SET_MEMORY_REGION: {
1845                 struct kvm_memory_region kvm_mem;
1846                 struct kvm_userspace_memory_region kvm_userspace_mem;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1850                         goto out;
1851                 kvm_userspace_mem.slot = kvm_mem.slot;
1852                 kvm_userspace_mem.flags = kvm_mem.flags;
1853                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1854                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1855                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1856                 if (r)
1857                         goto out;
1858                 break;
1859         }
1860         case KVM_SET_NR_MMU_PAGES:
1861                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1862                 if (r)
1863                         goto out;
1864                 break;
1865         case KVM_GET_NR_MMU_PAGES:
1866                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1867                 break;
1868         case KVM_SET_MEMORY_ALIAS:
1869                 r = -EFAULT;
1870                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1871                         goto out;
1872                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1873                 if (r)
1874                         goto out;
1875                 break;
1876         case KVM_CREATE_IRQCHIP:
1877                 r = -ENOMEM;
1878                 kvm->arch.vpic = kvm_create_pic(kvm);
1879                 if (kvm->arch.vpic) {
1880                         r = kvm_ioapic_init(kvm);
1881                         if (r) {
1882                                 kfree(kvm->arch.vpic);
1883                                 kvm->arch.vpic = NULL;
1884                                 goto out;
1885                         }
1886                 } else
1887                         goto out;
1888                 r = kvm_setup_default_irq_routing(kvm);
1889                 if (r) {
1890                         kfree(kvm->arch.vpic);
1891                         kfree(kvm->arch.vioapic);
1892                         goto out;
1893                 }
1894                 break;
1895         case KVM_CREATE_PIT:
1896                 mutex_lock(&kvm->lock);
1897                 r = -EEXIST;
1898                 if (kvm->arch.vpit)
1899                         goto create_pit_unlock;
1900                 r = -ENOMEM;
1901                 kvm->arch.vpit = kvm_create_pit(kvm);
1902                 if (kvm->arch.vpit)
1903                         r = 0;
1904         create_pit_unlock:
1905                 mutex_unlock(&kvm->lock);
1906                 break;
1907         case KVM_IRQ_LINE_STATUS:
1908         case KVM_IRQ_LINE: {
1909                 struct kvm_irq_level irq_event;
1910
1911                 r = -EFAULT;
1912                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1913                         goto out;
1914                 if (irqchip_in_kernel(kvm)) {
1915                         __s32 status;
1916                         mutex_lock(&kvm->lock);
1917                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1918                                         irq_event.irq, irq_event.level);
1919                         mutex_unlock(&kvm->lock);
1920                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1921                                 irq_event.status = status;
1922                                 if (copy_to_user(argp, &irq_event,
1923                                                         sizeof irq_event))
1924                                         goto out;
1925                         }
1926                         r = 0;
1927                 }
1928                 break;
1929         }
1930         case KVM_GET_IRQCHIP: {
1931                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1932                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1933
1934                 r = -ENOMEM;
1935                 if (!chip)
1936                         goto out;
1937                 r = -EFAULT;
1938                 if (copy_from_user(chip, argp, sizeof *chip))
1939                         goto get_irqchip_out;
1940                 r = -ENXIO;
1941                 if (!irqchip_in_kernel(kvm))
1942                         goto get_irqchip_out;
1943                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1944                 if (r)
1945                         goto get_irqchip_out;
1946                 r = -EFAULT;
1947                 if (copy_to_user(argp, chip, sizeof *chip))
1948                         goto get_irqchip_out;
1949                 r = 0;
1950         get_irqchip_out:
1951                 kfree(chip);
1952                 if (r)
1953                         goto out;
1954                 break;
1955         }
1956         case KVM_SET_IRQCHIP: {
1957                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1958                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1959
1960                 r = -ENOMEM;
1961                 if (!chip)
1962                         goto out;
1963                 r = -EFAULT;
1964                 if (copy_from_user(chip, argp, sizeof *chip))
1965                         goto set_irqchip_out;
1966                 r = -ENXIO;
1967                 if (!irqchip_in_kernel(kvm))
1968                         goto set_irqchip_out;
1969                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1970                 if (r)
1971                         goto set_irqchip_out;
1972                 r = 0;
1973         set_irqchip_out:
1974                 kfree(chip);
1975                 if (r)
1976                         goto out;
1977                 break;
1978         }
1979         case KVM_GET_PIT: {
1980                 r = -EFAULT;
1981                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1982                         goto out;
1983                 r = -ENXIO;
1984                 if (!kvm->arch.vpit)
1985                         goto out;
1986                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1987                 if (r)
1988                         goto out;
1989                 r = -EFAULT;
1990                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1991                         goto out;
1992                 r = 0;
1993                 break;
1994         }
1995         case KVM_SET_PIT: {
1996                 r = -EFAULT;
1997                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1998                         goto out;
1999                 r = -ENXIO;
2000                 if (!kvm->arch.vpit)
2001                         goto out;
2002                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2003                 if (r)
2004                         goto out;
2005                 r = 0;
2006                 break;
2007         }
2008         case KVM_REINJECT_CONTROL: {
2009                 struct kvm_reinject_control control;
2010                 r =  -EFAULT;
2011                 if (copy_from_user(&control, argp, sizeof(control)))
2012                         goto out;
2013                 r = kvm_vm_ioctl_reinject(kvm, &control);
2014                 if (r)
2015                         goto out;
2016                 r = 0;
2017                 break;
2018         }
2019         default:
2020                 ;
2021         }
2022 out:
2023         return r;
2024 }
2025
2026 static void kvm_init_msr_list(void)
2027 {
2028         u32 dummy[2];
2029         unsigned i, j;
2030
2031         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2032                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2033                         continue;
2034                 if (j < i)
2035                         msrs_to_save[j] = msrs_to_save[i];
2036                 j++;
2037         }
2038         num_msrs_to_save = j;
2039 }
2040
2041 /*
2042  * Only apic need an MMIO device hook, so shortcut now..
2043  */
2044 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2045                                                 gpa_t addr, int len,
2046                                                 int is_write)
2047 {
2048         struct kvm_io_device *dev;
2049
2050         if (vcpu->arch.apic) {
2051                 dev = &vcpu->arch.apic->dev;
2052                 if (dev->in_range(dev, addr, len, is_write))
2053                         return dev;
2054         }
2055         return NULL;
2056 }
2057
2058
2059 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2060                                                 gpa_t addr, int len,
2061                                                 int is_write)
2062 {
2063         struct kvm_io_device *dev;
2064
2065         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2066         if (dev == NULL)
2067                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2068                                           is_write);
2069         return dev;
2070 }
2071
2072 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2073                                struct kvm_vcpu *vcpu)
2074 {
2075         void *data = val;
2076         int r = X86EMUL_CONTINUE;
2077
2078         while (bytes) {
2079                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2080                 unsigned offset = addr & (PAGE_SIZE-1);
2081                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2082                 int ret;
2083
2084                 if (gpa == UNMAPPED_GVA) {
2085                         r = X86EMUL_PROPAGATE_FAULT;
2086                         goto out;
2087                 }
2088                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2089                 if (ret < 0) {
2090                         r = X86EMUL_UNHANDLEABLE;
2091                         goto out;
2092                 }
2093
2094                 bytes -= toread;
2095                 data += toread;
2096                 addr += toread;
2097         }
2098 out:
2099         return r;
2100 }
2101
2102 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2103                                 struct kvm_vcpu *vcpu)
2104 {
2105         void *data = val;
2106         int r = X86EMUL_CONTINUE;
2107
2108         while (bytes) {
2109                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2110                 unsigned offset = addr & (PAGE_SIZE-1);
2111                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2112                 int ret;
2113
2114                 if (gpa == UNMAPPED_GVA) {
2115                         r = X86EMUL_PROPAGATE_FAULT;
2116                         goto out;
2117                 }
2118                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2119                 if (ret < 0) {
2120                         r = X86EMUL_UNHANDLEABLE;
2121                         goto out;
2122                 }
2123
2124                 bytes -= towrite;
2125                 data += towrite;
2126                 addr += towrite;
2127         }
2128 out:
2129         return r;
2130 }
2131
2132
2133 static int emulator_read_emulated(unsigned long addr,
2134                                   void *val,
2135                                   unsigned int bytes,
2136                                   struct kvm_vcpu *vcpu)
2137 {
2138         struct kvm_io_device *mmio_dev;
2139         gpa_t                 gpa;
2140
2141         if (vcpu->mmio_read_completed) {
2142                 memcpy(val, vcpu->mmio_data, bytes);
2143                 vcpu->mmio_read_completed = 0;
2144                 return X86EMUL_CONTINUE;
2145         }
2146
2147         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2148
2149         /* For APIC access vmexit */
2150         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2151                 goto mmio;
2152
2153         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2154                                 == X86EMUL_CONTINUE)
2155                 return X86EMUL_CONTINUE;
2156         if (gpa == UNMAPPED_GVA)
2157                 return X86EMUL_PROPAGATE_FAULT;
2158
2159 mmio:
2160         /*
2161          * Is this MMIO handled locally?
2162          */
2163         mutex_lock(&vcpu->kvm->lock);
2164         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2165         if (mmio_dev) {
2166                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2167                 mutex_unlock(&vcpu->kvm->lock);
2168                 return X86EMUL_CONTINUE;
2169         }
2170         mutex_unlock(&vcpu->kvm->lock);
2171
2172         vcpu->mmio_needed = 1;
2173         vcpu->mmio_phys_addr = gpa;
2174         vcpu->mmio_size = bytes;
2175         vcpu->mmio_is_write = 0;
2176
2177         return X86EMUL_UNHANDLEABLE;
2178 }
2179
2180 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2181                           const void *val, int bytes)
2182 {
2183         int ret;
2184
2185         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2186         if (ret < 0)
2187                 return 0;
2188         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2189         return 1;
2190 }
2191
2192 static int emulator_write_emulated_onepage(unsigned long addr,
2193                                            const void *val,
2194                                            unsigned int bytes,
2195                                            struct kvm_vcpu *vcpu)
2196 {
2197         struct kvm_io_device *mmio_dev;
2198         gpa_t                 gpa;
2199
2200         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2201
2202         if (gpa == UNMAPPED_GVA) {
2203                 kvm_inject_page_fault(vcpu, addr, 2);
2204                 return X86EMUL_PROPAGATE_FAULT;
2205         }
2206
2207         /* For APIC access vmexit */
2208         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2209                 goto mmio;
2210
2211         if (emulator_write_phys(vcpu, gpa, val, bytes))
2212                 return X86EMUL_CONTINUE;
2213
2214 mmio:
2215         /*
2216          * Is this MMIO handled locally?
2217          */
2218         mutex_lock(&vcpu->kvm->lock);
2219         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2220         if (mmio_dev) {
2221                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2222                 mutex_unlock(&vcpu->kvm->lock);
2223                 return X86EMUL_CONTINUE;
2224         }
2225         mutex_unlock(&vcpu->kvm->lock);
2226
2227         vcpu->mmio_needed = 1;
2228         vcpu->mmio_phys_addr = gpa;
2229         vcpu->mmio_size = bytes;
2230         vcpu->mmio_is_write = 1;
2231         memcpy(vcpu->mmio_data, val, bytes);
2232
2233         return X86EMUL_CONTINUE;
2234 }
2235
2236 int emulator_write_emulated(unsigned long addr,
2237                                    const void *val,
2238                                    unsigned int bytes,
2239                                    struct kvm_vcpu *vcpu)
2240 {
2241         /* Crossing a page boundary? */
2242         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2243                 int rc, now;
2244
2245                 now = -addr & ~PAGE_MASK;
2246                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2247                 if (rc != X86EMUL_CONTINUE)
2248                         return rc;
2249                 addr += now;
2250                 val += now;
2251                 bytes -= now;
2252         }
2253         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2254 }
2255 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2256
2257 static int emulator_cmpxchg_emulated(unsigned long addr,
2258                                      const void *old,
2259                                      const void *new,
2260                                      unsigned int bytes,
2261                                      struct kvm_vcpu *vcpu)
2262 {
2263         static int reported;
2264
2265         if (!reported) {
2266                 reported = 1;
2267                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2268         }
2269 #ifndef CONFIG_X86_64
2270         /* guests cmpxchg8b have to be emulated atomically */
2271         if (bytes == 8) {
2272                 gpa_t gpa;
2273                 struct page *page;
2274                 char *kaddr;
2275                 u64 val;
2276
2277                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2278
2279                 if (gpa == UNMAPPED_GVA ||
2280                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2281                         goto emul_write;
2282
2283                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2284                         goto emul_write;
2285
2286                 val = *(u64 *)new;
2287
2288                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2289
2290                 kaddr = kmap_atomic(page, KM_USER0);
2291                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2292                 kunmap_atomic(kaddr, KM_USER0);
2293                 kvm_release_page_dirty(page);
2294         }
2295 emul_write:
2296 #endif
2297
2298         return emulator_write_emulated(addr, new, bytes, vcpu);
2299 }
2300
2301 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2302 {
2303         return kvm_x86_ops->get_segment_base(vcpu, seg);
2304 }
2305
2306 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2307 {
2308         kvm_mmu_invlpg(vcpu, address);
2309         return X86EMUL_CONTINUE;
2310 }
2311
2312 int emulate_clts(struct kvm_vcpu *vcpu)
2313 {
2314         KVMTRACE_0D(CLTS, vcpu, handler);
2315         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2316         return X86EMUL_CONTINUE;
2317 }
2318
2319 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2320 {
2321         struct kvm_vcpu *vcpu = ctxt->vcpu;
2322
2323         switch (dr) {
2324         case 0 ... 3:
2325                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2326                 return X86EMUL_CONTINUE;
2327         default:
2328                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2329                 return X86EMUL_UNHANDLEABLE;
2330         }
2331 }
2332
2333 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2334 {
2335         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2336         int exception;
2337
2338         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2339         if (exception) {
2340                 /* FIXME: better handling */
2341                 return X86EMUL_UNHANDLEABLE;
2342         }
2343         return X86EMUL_CONTINUE;
2344 }
2345
2346 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2347 {
2348         u8 opcodes[4];
2349         unsigned long rip = kvm_rip_read(vcpu);
2350         unsigned long rip_linear;
2351
2352         if (!printk_ratelimit())
2353                 return;
2354
2355         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2356
2357         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2358
2359         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2360                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2361 }
2362 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2363
2364 static struct x86_emulate_ops emulate_ops = {
2365         .read_std            = kvm_read_guest_virt,
2366         .read_emulated       = emulator_read_emulated,
2367         .write_emulated      = emulator_write_emulated,
2368         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2369 };
2370
2371 static void cache_all_regs(struct kvm_vcpu *vcpu)
2372 {
2373         kvm_register_read(vcpu, VCPU_REGS_RAX);
2374         kvm_register_read(vcpu, VCPU_REGS_RSP);
2375         kvm_register_read(vcpu, VCPU_REGS_RIP);
2376         vcpu->arch.regs_dirty = ~0;
2377 }
2378
2379 int emulate_instruction(struct kvm_vcpu *vcpu,
2380                         struct kvm_run *run,
2381                         unsigned long cr2,
2382                         u16 error_code,
2383                         int emulation_type)
2384 {
2385         int r, shadow_mask;
2386         struct decode_cache *c;
2387
2388         kvm_clear_exception_queue(vcpu);
2389         vcpu->arch.mmio_fault_cr2 = cr2;
2390         /*
2391          * TODO: fix x86_emulate.c to use guest_read/write_register
2392          * instead of direct ->regs accesses, can save hundred cycles
2393          * on Intel for instructions that don't read/change RSP, for
2394          * for example.
2395          */
2396         cache_all_regs(vcpu);
2397
2398         vcpu->mmio_is_write = 0;
2399         vcpu->arch.pio.string = 0;
2400
2401         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2402                 int cs_db, cs_l;
2403                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2404
2405                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2406                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2407                 vcpu->arch.emulate_ctxt.mode =
2408                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2409                         ? X86EMUL_MODE_REAL : cs_l
2410                         ? X86EMUL_MODE_PROT64 : cs_db
2411                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2412
2413                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2414
2415                 /* Reject the instructions other than VMCALL/VMMCALL when
2416                  * try to emulate invalid opcode */
2417                 c = &vcpu->arch.emulate_ctxt.decode;
2418                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2419                     (!(c->twobyte && c->b == 0x01 &&
2420                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2421                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2422                         return EMULATE_FAIL;
2423
2424                 ++vcpu->stat.insn_emulation;
2425                 if (r)  {
2426                         ++vcpu->stat.insn_emulation_fail;
2427                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2428                                 return EMULATE_DONE;
2429                         return EMULATE_FAIL;
2430                 }
2431         }
2432
2433         if (emulation_type & EMULTYPE_SKIP) {
2434                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2435                 return EMULATE_DONE;
2436         }
2437
2438         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2439         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2440
2441         if (r == 0)
2442                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2443
2444         if (vcpu->arch.pio.string)
2445                 return EMULATE_DO_MMIO;
2446
2447         if ((r || vcpu->mmio_is_write) && run) {
2448                 run->exit_reason = KVM_EXIT_MMIO;
2449                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2450                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2451                 run->mmio.len = vcpu->mmio_size;
2452                 run->mmio.is_write = vcpu->mmio_is_write;
2453         }
2454
2455         if (r) {
2456                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2457                         return EMULATE_DONE;
2458                 if (!vcpu->mmio_needed) {
2459                         kvm_report_emulation_failure(vcpu, "mmio");
2460                         return EMULATE_FAIL;
2461                 }
2462                 return EMULATE_DO_MMIO;
2463         }
2464
2465         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2466
2467         if (vcpu->mmio_is_write) {
2468                 vcpu->mmio_needed = 0;
2469                 return EMULATE_DO_MMIO;
2470         }
2471
2472         return EMULATE_DONE;
2473 }
2474 EXPORT_SYMBOL_GPL(emulate_instruction);
2475
2476 static int pio_copy_data(struct kvm_vcpu *vcpu)
2477 {
2478         void *p = vcpu->arch.pio_data;
2479         gva_t q = vcpu->arch.pio.guest_gva;
2480         unsigned bytes;
2481         int ret;
2482
2483         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2484         if (vcpu->arch.pio.in)
2485                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2486         else
2487                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2488         return ret;
2489 }
2490
2491 int complete_pio(struct kvm_vcpu *vcpu)
2492 {
2493         struct kvm_pio_request *io = &vcpu->arch.pio;
2494         long delta;
2495         int r;
2496         unsigned long val;
2497
2498         if (!io->string) {
2499                 if (io->in) {
2500                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2501                         memcpy(&val, vcpu->arch.pio_data, io->size);
2502                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2503                 }
2504         } else {
2505                 if (io->in) {
2506                         r = pio_copy_data(vcpu);
2507                         if (r)
2508                                 return r;
2509                 }
2510
2511                 delta = 1;
2512                 if (io->rep) {
2513                         delta *= io->cur_count;
2514                         /*
2515                          * The size of the register should really depend on
2516                          * current address size.
2517                          */
2518                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2519                         val -= delta;
2520                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2521                 }
2522                 if (io->down)
2523                         delta = -delta;
2524                 delta *= io->size;
2525                 if (io->in) {
2526                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2527                         val += delta;
2528                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2529                 } else {
2530                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2531                         val += delta;
2532                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2533                 }
2534         }
2535
2536         io->count -= io->cur_count;
2537         io->cur_count = 0;
2538
2539         return 0;
2540 }
2541
2542 static void kernel_pio(struct kvm_io_device *pio_dev,
2543                        struct kvm_vcpu *vcpu,
2544                        void *pd)
2545 {
2546         /* TODO: String I/O for in kernel device */
2547
2548         mutex_lock(&vcpu->kvm->lock);
2549         if (vcpu->arch.pio.in)
2550                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2551                                   vcpu->arch.pio.size,
2552                                   pd);
2553         else
2554                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2555                                    vcpu->arch.pio.size,
2556                                    pd);
2557         mutex_unlock(&vcpu->kvm->lock);
2558 }
2559
2560 static void pio_string_write(struct kvm_io_device *pio_dev,
2561                              struct kvm_vcpu *vcpu)
2562 {
2563         struct kvm_pio_request *io = &vcpu->arch.pio;
2564         void *pd = vcpu->arch.pio_data;
2565         int i;
2566
2567         mutex_lock(&vcpu->kvm->lock);
2568         for (i = 0; i < io->cur_count; i++) {
2569                 kvm_iodevice_write(pio_dev, io->port,
2570                                    io->size,
2571                                    pd);
2572                 pd += io->size;
2573         }
2574         mutex_unlock(&vcpu->kvm->lock);
2575 }
2576
2577 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2578                                                gpa_t addr, int len,
2579                                                int is_write)
2580 {
2581         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2582 }
2583
2584 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2585                   int size, unsigned port)
2586 {
2587         struct kvm_io_device *pio_dev;
2588         unsigned long val;
2589
2590         vcpu->run->exit_reason = KVM_EXIT_IO;
2591         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2592         vcpu->run->io.size = vcpu->arch.pio.size = size;
2593         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2594         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2595         vcpu->run->io.port = vcpu->arch.pio.port = port;
2596         vcpu->arch.pio.in = in;
2597         vcpu->arch.pio.string = 0;
2598         vcpu->arch.pio.down = 0;
2599         vcpu->arch.pio.rep = 0;
2600
2601         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2602                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2603                             handler);
2604         else
2605                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2606                             handler);
2607
2608         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2609         memcpy(vcpu->arch.pio_data, &val, 4);
2610
2611         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2612         if (pio_dev) {
2613                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2614                 complete_pio(vcpu);
2615                 return 1;
2616         }
2617         return 0;
2618 }
2619 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2620
2621 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2622                   int size, unsigned long count, int down,
2623                   gva_t address, int rep, unsigned port)
2624 {
2625         unsigned now, in_page;
2626         int ret = 0;
2627         struct kvm_io_device *pio_dev;
2628
2629         vcpu->run->exit_reason = KVM_EXIT_IO;
2630         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2631         vcpu->run->io.size = vcpu->arch.pio.size = size;
2632         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2633         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2634         vcpu->run->io.port = vcpu->arch.pio.port = port;
2635         vcpu->arch.pio.in = in;
2636         vcpu->arch.pio.string = 1;
2637         vcpu->arch.pio.down = down;
2638         vcpu->arch.pio.rep = rep;
2639
2640         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2641                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2642                             handler);
2643         else
2644                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2645                             handler);
2646
2647         if (!count) {
2648                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2649                 return 1;
2650         }
2651
2652         if (!down)
2653                 in_page = PAGE_SIZE - offset_in_page(address);
2654         else
2655                 in_page = offset_in_page(address) + size;
2656         now = min(count, (unsigned long)in_page / size);
2657         if (!now)
2658                 now = 1;
2659         if (down) {
2660                 /*
2661                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2662                  */
2663                 pr_unimpl(vcpu, "guest string pio down\n");
2664                 kvm_inject_gp(vcpu, 0);
2665                 return 1;
2666         }
2667         vcpu->run->io.count = now;
2668         vcpu->arch.pio.cur_count = now;
2669
2670         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2671                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2672
2673         vcpu->arch.pio.guest_gva = address;
2674
2675         pio_dev = vcpu_find_pio_dev(vcpu, port,
2676                                     vcpu->arch.pio.cur_count,
2677                                     !vcpu->arch.pio.in);
2678         if (!vcpu->arch.pio.in) {
2679                 /* string PIO write */
2680                 ret = pio_copy_data(vcpu);
2681                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2682                         kvm_inject_gp(vcpu, 0);
2683                         return 1;
2684                 }
2685                 if (ret == 0 && pio_dev) {
2686                         pio_string_write(pio_dev, vcpu);
2687                         complete_pio(vcpu);
2688                         if (vcpu->arch.pio.count == 0)
2689                                 ret = 1;
2690                 }
2691         } else if (pio_dev)
2692                 pr_unimpl(vcpu, "no string pio read support yet, "
2693                        "port %x size %d count %ld\n",
2694                         port, size, count);
2695
2696         return ret;
2697 }
2698 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2699
2700 static void bounce_off(void *info)
2701 {
2702         /* nothing */
2703 }
2704
2705 static unsigned int  ref_freq;
2706 static unsigned long tsc_khz_ref;
2707
2708 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2709                                      void *data)
2710 {
2711         struct cpufreq_freqs *freq = data;
2712         struct kvm *kvm;
2713         struct kvm_vcpu *vcpu;
2714         int i, send_ipi = 0;
2715
2716         if (!ref_freq)
2717                 ref_freq = freq->old;
2718
2719         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2720                 return 0;
2721         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2722                 return 0;
2723         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2724
2725         spin_lock(&kvm_lock);
2726         list_for_each_entry(kvm, &vm_list, vm_list) {
2727                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2728                         vcpu = kvm->vcpus[i];
2729                         if (!vcpu)
2730                                 continue;
2731                         if (vcpu->cpu != freq->cpu)
2732                                 continue;
2733                         if (!kvm_request_guest_time_update(vcpu))
2734                                 continue;
2735                         if (vcpu->cpu != smp_processor_id())
2736                                 send_ipi++;
2737                 }
2738         }
2739         spin_unlock(&kvm_lock);
2740
2741         if (freq->old < freq->new && send_ipi) {
2742                 /*
2743                  * We upscale the frequency.  Must make the guest
2744                  * doesn't see old kvmclock values while running with
2745                  * the new frequency, otherwise we risk the guest sees
2746                  * time go backwards.
2747                  *
2748                  * In case we update the frequency for another cpu
2749                  * (which might be in guest context) send an interrupt
2750                  * to kick the cpu out of guest context.  Next time
2751                  * guest context is entered kvmclock will be updated,
2752                  * so the guest will not see stale values.
2753                  */
2754                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2755         }
2756         return 0;
2757 }
2758
2759 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2760         .notifier_call  = kvmclock_cpufreq_notifier
2761 };
2762
2763 int kvm_arch_init(void *opaque)
2764 {
2765         int r, cpu;
2766         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2767
2768         if (kvm_x86_ops) {
2769                 printk(KERN_ERR "kvm: already loaded the other module\n");
2770                 r = -EEXIST;
2771                 goto out;
2772         }
2773
2774         if (!ops->cpu_has_kvm_support()) {
2775                 printk(KERN_ERR "kvm: no hardware support\n");
2776                 r = -EOPNOTSUPP;
2777                 goto out;
2778         }
2779         if (ops->disabled_by_bios()) {
2780                 printk(KERN_ERR "kvm: disabled by bios\n");
2781                 r = -EOPNOTSUPP;
2782                 goto out;
2783         }
2784
2785         r = kvm_mmu_module_init();
2786         if (r)
2787                 goto out;
2788
2789         kvm_init_msr_list();
2790
2791         kvm_x86_ops = ops;
2792         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2793         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2794         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2795                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2796
2797         for_each_possible_cpu(cpu)
2798                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2799         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2800                 tsc_khz_ref = tsc_khz;
2801                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2802                                           CPUFREQ_TRANSITION_NOTIFIER);
2803         }
2804
2805         return 0;
2806
2807 out:
2808         return r;
2809 }
2810
2811 void kvm_arch_exit(void)
2812 {
2813         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2814                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2815                                             CPUFREQ_TRANSITION_NOTIFIER);
2816         kvm_x86_ops = NULL;
2817         kvm_mmu_module_exit();
2818 }
2819
2820 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2821 {
2822         ++vcpu->stat.halt_exits;
2823         KVMTRACE_0D(HLT, vcpu, handler);
2824         if (irqchip_in_kernel(vcpu->kvm)) {
2825                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2826                 return 1;
2827         } else {
2828                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2829                 return 0;
2830         }
2831 }
2832 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2833
2834 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2835                            unsigned long a1)
2836 {
2837         if (is_long_mode(vcpu))
2838                 return a0;
2839         else
2840                 return a0 | ((gpa_t)a1 << 32);
2841 }
2842
2843 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2844 {
2845         unsigned long nr, a0, a1, a2, a3, ret;
2846         int r = 1;
2847
2848         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2849         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2850         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2851         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2852         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2853
2854         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2855
2856         if (!is_long_mode(vcpu)) {
2857                 nr &= 0xFFFFFFFF;
2858                 a0 &= 0xFFFFFFFF;
2859                 a1 &= 0xFFFFFFFF;
2860                 a2 &= 0xFFFFFFFF;
2861                 a3 &= 0xFFFFFFFF;
2862         }
2863
2864         switch (nr) {
2865         case KVM_HC_VAPIC_POLL_IRQ:
2866                 ret = 0;
2867                 break;
2868         case KVM_HC_MMU_OP:
2869                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2870                 break;
2871         default:
2872                 ret = -KVM_ENOSYS;
2873                 break;
2874         }
2875         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2876         ++vcpu->stat.hypercalls;
2877         return r;
2878 }
2879 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2880
2881 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2882 {
2883         char instruction[3];
2884         int ret = 0;
2885         unsigned long rip = kvm_rip_read(vcpu);
2886
2887
2888         /*
2889          * Blow out the MMU to ensure that no other VCPU has an active mapping
2890          * to ensure that the updated hypercall appears atomically across all
2891          * VCPUs.
2892          */
2893         kvm_mmu_zap_all(vcpu->kvm);
2894
2895         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2896         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2897             != X86EMUL_CONTINUE)
2898                 ret = -EFAULT;
2899
2900         return ret;
2901 }
2902
2903 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2904 {
2905         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2906 }
2907
2908 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2909 {
2910         struct descriptor_table dt = { limit, base };
2911
2912         kvm_x86_ops->set_gdt(vcpu, &dt);
2913 }
2914
2915 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2916 {
2917         struct descriptor_table dt = { limit, base };
2918
2919         kvm_x86_ops->set_idt(vcpu, &dt);
2920 }
2921
2922 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2923                    unsigned long *rflags)
2924 {
2925         kvm_lmsw(vcpu, msw);
2926         *rflags = kvm_x86_ops->get_rflags(vcpu);
2927 }
2928
2929 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2930 {
2931         unsigned long value;
2932
2933         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2934         switch (cr) {
2935         case 0:
2936                 value = vcpu->arch.cr0;
2937                 break;
2938         case 2:
2939                 value = vcpu->arch.cr2;
2940                 break;
2941         case 3:
2942                 value = vcpu->arch.cr3;
2943                 break;
2944         case 4:
2945                 value = vcpu->arch.cr4;
2946                 break;
2947         case 8:
2948                 value = kvm_get_cr8(vcpu);
2949                 break;
2950         default:
2951                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2952                 return 0;
2953         }
2954         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2955                     (u32)((u64)value >> 32), handler);
2956
2957         return value;
2958 }
2959
2960 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2961                      unsigned long *rflags)
2962 {
2963         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2964                     (u32)((u64)val >> 32), handler);
2965
2966         switch (cr) {
2967         case 0:
2968                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2969                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2970                 break;
2971         case 2:
2972                 vcpu->arch.cr2 = val;
2973                 break;
2974         case 3:
2975                 kvm_set_cr3(vcpu, val);
2976                 break;
2977         case 4:
2978                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2979                 break;
2980         case 8:
2981                 kvm_set_cr8(vcpu, val & 0xfUL);
2982                 break;
2983         default:
2984                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2985         }
2986 }
2987
2988 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2989 {
2990         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2991         int j, nent = vcpu->arch.cpuid_nent;
2992
2993         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2994         /* when no next entry is found, the current entry[i] is reselected */
2995         for (j = i + 1; ; j = (j + 1) % nent) {
2996                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2997                 if (ej->function == e->function) {
2998                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2999                         return j;
3000                 }
3001         }
3002         return 0; /* silence gcc, even though control never reaches here */
3003 }
3004
3005 /* find an entry with matching function, matching index (if needed), and that
3006  * should be read next (if it's stateful) */
3007 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3008         u32 function, u32 index)
3009 {
3010         if (e->function != function)
3011                 return 0;
3012         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3013                 return 0;
3014         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3015             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3016                 return 0;
3017         return 1;
3018 }
3019
3020 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3021                                               u32 function, u32 index)
3022 {
3023         int i;
3024         struct kvm_cpuid_entry2 *best = NULL;
3025
3026         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3027                 struct kvm_cpuid_entry2 *e;
3028
3029                 e = &vcpu->arch.cpuid_entries[i];
3030                 if (is_matching_cpuid_entry(e, function, index)) {
3031                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3032                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3033                         best = e;
3034                         break;
3035                 }
3036                 /*
3037                  * Both basic or both extended?
3038                  */
3039                 if (((e->function ^ function) & 0x80000000) == 0)
3040                         if (!best || e->function > best->function)
3041                                 best = e;
3042         }
3043         return best;
3044 }
3045
3046 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3047 {
3048         struct kvm_cpuid_entry2 *best;
3049
3050         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3051         if (best)
3052                 return best->eax & 0xff;
3053         return 36;
3054 }
3055
3056 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3057 {
3058         u32 function, index;
3059         struct kvm_cpuid_entry2 *best;
3060
3061         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3062         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3063         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3064         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3065         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3066         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3067         best = kvm_find_cpuid_entry(vcpu, function, index);
3068         if (best) {
3069                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3070                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3071                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3072                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3073         }
3074         kvm_x86_ops->skip_emulated_instruction(vcpu);
3075         KVMTRACE_5D(CPUID, vcpu, function,
3076                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3077                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3078                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3079                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3080 }
3081 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3082
3083 /*
3084  * Check if userspace requested an interrupt window, and that the
3085  * interrupt window is open.
3086  *
3087  * No need to exit to userspace if we already have an interrupt queued.
3088  */
3089 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3090                                           struct kvm_run *kvm_run)
3091 {
3092         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3093                 kvm_run->request_interrupt_window &&
3094                 kvm_arch_interrupt_allowed(vcpu));
3095 }
3096
3097 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3098                               struct kvm_run *kvm_run)
3099 {
3100         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3101         kvm_run->cr8 = kvm_get_cr8(vcpu);
3102         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3103         if (irqchip_in_kernel(vcpu->kvm))
3104                 kvm_run->ready_for_interrupt_injection = 1;
3105         else
3106                 kvm_run->ready_for_interrupt_injection =
3107                         kvm_arch_interrupt_allowed(vcpu) &&
3108                         !kvm_cpu_has_interrupt(vcpu) &&
3109                         !kvm_event_needs_reinjection(vcpu);
3110 }
3111
3112 static void vapic_enter(struct kvm_vcpu *vcpu)
3113 {
3114         struct kvm_lapic *apic = vcpu->arch.apic;
3115         struct page *page;
3116
3117         if (!apic || !apic->vapic_addr)
3118                 return;
3119
3120         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3121
3122         vcpu->arch.apic->vapic_page = page;
3123 }
3124
3125 static void vapic_exit(struct kvm_vcpu *vcpu)
3126 {
3127         struct kvm_lapic *apic = vcpu->arch.apic;
3128
3129         if (!apic || !apic->vapic_addr)
3130                 return;
3131
3132         down_read(&vcpu->kvm->slots_lock);
3133         kvm_release_page_dirty(apic->vapic_page);
3134         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3135         up_read(&vcpu->kvm->slots_lock);
3136 }
3137
3138 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3139 {
3140         int max_irr, tpr;
3141
3142         if (!kvm_x86_ops->update_cr8_intercept)
3143                 return;
3144
3145         if (!vcpu->arch.apic->vapic_addr)
3146                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3147         else
3148                 max_irr = -1;
3149
3150         if (max_irr != -1)
3151                 max_irr >>= 4;
3152
3153         tpr = kvm_lapic_get_cr8(vcpu);
3154
3155         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3156 }
3157
3158 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3159 {
3160         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3161                 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
3162
3163         /* try to reinject previous events if any */
3164         if (vcpu->arch.nmi_injected) {
3165                 kvm_x86_ops->set_nmi(vcpu);
3166                 return;
3167         }
3168
3169         if (vcpu->arch.interrupt.pending) {
3170                 kvm_x86_ops->set_irq(vcpu);
3171                 return;
3172         }
3173
3174         /* try to inject new event if pending */
3175         if (vcpu->arch.nmi_pending) {
3176                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3177                         vcpu->arch.nmi_pending = false;
3178                         vcpu->arch.nmi_injected = true;
3179                         kvm_x86_ops->set_nmi(vcpu);
3180                 }
3181         } else if (kvm_cpu_has_interrupt(vcpu)) {
3182                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3183                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3184                                             false);
3185                         kvm_x86_ops->set_irq(vcpu);
3186                 }
3187         }
3188 }
3189
3190 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3191 {
3192         int r;
3193         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3194                 kvm_run->request_interrupt_window;
3195
3196         if (vcpu->requests)
3197                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3198                         kvm_mmu_unload(vcpu);
3199
3200         r = kvm_mmu_reload(vcpu);
3201         if (unlikely(r))
3202                 goto out;
3203
3204         if (vcpu->requests) {
3205                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3206                         __kvm_migrate_timers(vcpu);
3207                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3208                         kvm_write_guest_time(vcpu);
3209                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3210                         kvm_mmu_sync_roots(vcpu);
3211                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3212                         kvm_x86_ops->tlb_flush(vcpu);
3213                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3214                                        &vcpu->requests)) {
3215                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3216                         r = 0;
3217                         goto out;
3218                 }
3219                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3220                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3221                         r = 0;
3222                         goto out;
3223                 }
3224         }
3225
3226         preempt_disable();
3227
3228         kvm_x86_ops->prepare_guest_switch(vcpu);
3229         kvm_load_guest_fpu(vcpu);
3230
3231         local_irq_disable();
3232
3233         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3234         smp_mb__after_clear_bit();
3235
3236         if (vcpu->requests || need_resched() || signal_pending(current)) {
3237                 local_irq_enable();
3238                 preempt_enable();
3239                 r = 1;
3240                 goto out;
3241         }
3242
3243         if (vcpu->arch.exception.pending)
3244                 __queue_exception(vcpu);
3245         else
3246                 inject_pending_irq(vcpu, kvm_run);
3247
3248         /* enable NMI/IRQ window open exits if needed */
3249         if (vcpu->arch.nmi_pending)
3250                 kvm_x86_ops->enable_nmi_window(vcpu);
3251         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3252                 kvm_x86_ops->enable_irq_window(vcpu);
3253
3254         if (kvm_lapic_enabled(vcpu)) {
3255                 update_cr8_intercept(vcpu);
3256                 kvm_lapic_sync_to_vapic(vcpu);
3257         }
3258
3259         up_read(&vcpu->kvm->slots_lock);
3260
3261         kvm_guest_enter();
3262
3263         get_debugreg(vcpu->arch.host_dr6, 6);
3264         get_debugreg(vcpu->arch.host_dr7, 7);
3265         if (unlikely(vcpu->arch.switch_db_regs)) {
3266                 get_debugreg(vcpu->arch.host_db[0], 0);
3267                 get_debugreg(vcpu->arch.host_db[1], 1);
3268                 get_debugreg(vcpu->arch.host_db[2], 2);
3269                 get_debugreg(vcpu->arch.host_db[3], 3);
3270
3271                 set_debugreg(0, 7);
3272                 set_debugreg(vcpu->arch.eff_db[0], 0);
3273                 set_debugreg(vcpu->arch.eff_db[1], 1);
3274                 set_debugreg(vcpu->arch.eff_db[2], 2);
3275                 set_debugreg(vcpu->arch.eff_db[3], 3);
3276         }
3277
3278         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3279         kvm_x86_ops->run(vcpu, kvm_run);
3280
3281         if (unlikely(vcpu->arch.switch_db_regs)) {
3282                 set_debugreg(0, 7);
3283                 set_debugreg(vcpu->arch.host_db[0], 0);
3284                 set_debugreg(vcpu->arch.host_db[1], 1);
3285                 set_debugreg(vcpu->arch.host_db[2], 2);
3286                 set_debugreg(vcpu->arch.host_db[3], 3);
3287         }
3288         set_debugreg(vcpu->arch.host_dr6, 6);
3289         set_debugreg(vcpu->arch.host_dr7, 7);
3290
3291         set_bit(KVM_REQ_KICK, &vcpu->requests);
3292         local_irq_enable();
3293
3294         ++vcpu->stat.exits;
3295
3296         /*
3297          * We must have an instruction between local_irq_enable() and
3298          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3299          * the interrupt shadow.  The stat.exits increment will do nicely.
3300          * But we need to prevent reordering, hence this barrier():
3301          */
3302         barrier();
3303
3304         kvm_guest_exit();
3305
3306         preempt_enable();
3307
3308         down_read(&vcpu->kvm->slots_lock);
3309
3310         /*
3311          * Profile KVM exit RIPs:
3312          */
3313         if (unlikely(prof_on == KVM_PROFILING)) {
3314                 unsigned long rip = kvm_rip_read(vcpu);
3315                 profile_hit(KVM_PROFILING, (void *)rip);
3316         }
3317
3318
3319         kvm_lapic_sync_from_vapic(vcpu);
3320
3321         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3322 out:
3323         return r;
3324 }
3325
3326
3327 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3328 {
3329         int r;
3330
3331         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3332                 pr_debug("vcpu %d received sipi with vector # %x\n",
3333                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3334                 kvm_lapic_reset(vcpu);
3335                 r = kvm_arch_vcpu_reset(vcpu);
3336                 if (r)
3337                         return r;
3338                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3339         }
3340
3341         down_read(&vcpu->kvm->slots_lock);
3342         vapic_enter(vcpu);
3343
3344         r = 1;
3345         while (r > 0) {
3346                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3347                         r = vcpu_enter_guest(vcpu, kvm_run);
3348                 else {
3349                         up_read(&vcpu->kvm->slots_lock);
3350                         kvm_vcpu_block(vcpu);
3351                         down_read(&vcpu->kvm->slots_lock);
3352                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3353                         {
3354                                 switch(vcpu->arch.mp_state) {
3355                                 case KVM_MP_STATE_HALTED:
3356                                         vcpu->arch.mp_state =
3357                                                 KVM_MP_STATE_RUNNABLE;
3358                                 case KVM_MP_STATE_RUNNABLE:
3359                                         break;
3360                                 case KVM_MP_STATE_SIPI_RECEIVED:
3361                                 default:
3362                                         r = -EINTR;
3363                                         break;
3364                                 }
3365                         }
3366                 }
3367
3368                 if (r <= 0)
3369                         break;
3370
3371                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3372                 if (kvm_cpu_has_pending_timer(vcpu))
3373                         kvm_inject_pending_timer_irqs(vcpu);
3374
3375                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3376                         r = -EINTR;
3377                         kvm_run->exit_reason = KVM_EXIT_INTR;
3378                         ++vcpu->stat.request_irq_exits;
3379                 }
3380                 if (signal_pending(current)) {
3381                         r = -EINTR;
3382                         kvm_run->exit_reason = KVM_EXIT_INTR;
3383                         ++vcpu->stat.signal_exits;
3384                 }
3385                 if (need_resched()) {
3386                         up_read(&vcpu->kvm->slots_lock);
3387                         kvm_resched(vcpu);
3388                         down_read(&vcpu->kvm->slots_lock);
3389                 }
3390         }
3391
3392         up_read(&vcpu->kvm->slots_lock);
3393         post_kvm_run_save(vcpu, kvm_run);
3394
3395         vapic_exit(vcpu);
3396
3397         return r;
3398 }
3399
3400 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3401 {
3402         int r;
3403         sigset_t sigsaved;
3404
3405         vcpu_load(vcpu);
3406
3407         if (vcpu->sigset_active)
3408                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3409
3410         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3411                 kvm_vcpu_block(vcpu);
3412                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3413                 r = -EAGAIN;
3414                 goto out;
3415         }
3416
3417         /* re-sync apic's tpr */
3418         if (!irqchip_in_kernel(vcpu->kvm))
3419                 kvm_set_cr8(vcpu, kvm_run->cr8);
3420
3421         if (vcpu->arch.pio.cur_count) {
3422                 r = complete_pio(vcpu);
3423                 if (r)
3424                         goto out;
3425         }
3426 #if CONFIG_HAS_IOMEM
3427         if (vcpu->mmio_needed) {
3428                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3429                 vcpu->mmio_read_completed = 1;
3430                 vcpu->mmio_needed = 0;
3431
3432                 down_read(&vcpu->kvm->slots_lock);
3433                 r = emulate_instruction(vcpu, kvm_run,
3434                                         vcpu->arch.mmio_fault_cr2, 0,
3435                                         EMULTYPE_NO_DECODE);
3436                 up_read(&vcpu->kvm->slots_lock);
3437                 if (r == EMULATE_DO_MMIO) {
3438                         /*
3439                          * Read-modify-write.  Back to userspace.
3440                          */
3441                         r = 0;
3442                         goto out;
3443                 }
3444         }
3445 #endif
3446         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3447                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3448                                      kvm_run->hypercall.ret);
3449
3450         r = __vcpu_run(vcpu, kvm_run);
3451
3452 out:
3453         if (vcpu->sigset_active)
3454                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3455
3456         vcpu_put(vcpu);
3457         return r;
3458 }
3459
3460 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3461 {
3462         vcpu_load(vcpu);
3463
3464         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3465         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3466         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3467         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3468         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3469         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3470         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3471         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3472 #ifdef CONFIG_X86_64
3473         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3474         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3475         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3476         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3477         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3478         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3479         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3480         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3481 #endif
3482
3483         regs->rip = kvm_rip_read(vcpu);
3484         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3485
3486         /*
3487          * Don't leak debug flags in case they were set for guest debugging
3488          */
3489         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3490                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3491
3492         vcpu_put(vcpu);
3493
3494         return 0;
3495 }
3496
3497 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3498 {
3499         vcpu_load(vcpu);
3500
3501         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3502         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3503         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3504         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3505         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3506         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3507         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3508         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3509 #ifdef CONFIG_X86_64
3510         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3511         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3512         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3513         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3514         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3515         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3516         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3517         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3518
3519 #endif
3520
3521         kvm_rip_write(vcpu, regs->rip);
3522         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3523
3524
3525         vcpu->arch.exception.pending = false;
3526
3527         vcpu_put(vcpu);
3528
3529         return 0;
3530 }
3531
3532 void kvm_get_segment(struct kvm_vcpu *vcpu,
3533                      struct kvm_segment *var, int seg)
3534 {
3535         kvm_x86_ops->get_segment(vcpu, var, seg);
3536 }
3537
3538 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3539 {
3540         struct kvm_segment cs;
3541
3542         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3543         *db = cs.db;
3544         *l = cs.l;
3545 }
3546 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3547
3548 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3549                                   struct kvm_sregs *sregs)
3550 {
3551         struct descriptor_table dt;
3552
3553         vcpu_load(vcpu);
3554
3555         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3556         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3557         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3558         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3559         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3560         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3561
3562         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3563         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3564
3565         kvm_x86_ops->get_idt(vcpu, &dt);
3566         sregs->idt.limit = dt.limit;
3567         sregs->idt.base = dt.base;
3568         kvm_x86_ops->get_gdt(vcpu, &dt);
3569         sregs->gdt.limit = dt.limit;
3570         sregs->gdt.base = dt.base;
3571
3572         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3573         sregs->cr0 = vcpu->arch.cr0;
3574         sregs->cr2 = vcpu->arch.cr2;
3575         sregs->cr3 = vcpu->arch.cr3;
3576         sregs->cr4 = vcpu->arch.cr4;
3577         sregs->cr8 = kvm_get_cr8(vcpu);
3578         sregs->efer = vcpu->arch.shadow_efer;
3579         sregs->apic_base = kvm_get_apic_base(vcpu);
3580
3581         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3582
3583         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3584                 set_bit(vcpu->arch.interrupt.nr,
3585                         (unsigned long *)sregs->interrupt_bitmap);
3586
3587         vcpu_put(vcpu);
3588
3589         return 0;
3590 }
3591
3592 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3593                                     struct kvm_mp_state *mp_state)
3594 {
3595         vcpu_load(vcpu);
3596         mp_state->mp_state = vcpu->arch.mp_state;
3597         vcpu_put(vcpu);
3598         return 0;
3599 }
3600
3601 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3602                                     struct kvm_mp_state *mp_state)
3603 {
3604         vcpu_load(vcpu);
3605         vcpu->arch.mp_state = mp_state->mp_state;
3606         vcpu_put(vcpu);
3607         return 0;
3608 }
3609
3610 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3611                         struct kvm_segment *var, int seg)
3612 {
3613         kvm_x86_ops->set_segment(vcpu, var, seg);
3614 }
3615
3616 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3617                                    struct kvm_segment *kvm_desct)
3618 {
3619         kvm_desct->base = seg_desc->base0;
3620         kvm_desct->base |= seg_desc->base1 << 16;
3621         kvm_desct->base |= seg_desc->base2 << 24;
3622         kvm_desct->limit = seg_desc->limit0;
3623         kvm_desct->limit |= seg_desc->limit << 16;
3624         if (seg_desc->g) {
3625                 kvm_desct->limit <<= 12;
3626                 kvm_desct->limit |= 0xfff;
3627         }
3628         kvm_desct->selector = selector;
3629         kvm_desct->type = seg_desc->type;
3630         kvm_desct->present = seg_desc->p;
3631         kvm_desct->dpl = seg_desc->dpl;
3632         kvm_desct->db = seg_desc->d;
3633         kvm_desct->s = seg_desc->s;
3634         kvm_desct->l = seg_desc->l;
3635         kvm_desct->g = seg_desc->g;
3636         kvm_desct->avl = seg_desc->avl;
3637         if (!selector)
3638                 kvm_desct->unusable = 1;
3639         else
3640                 kvm_desct->unusable = 0;
3641         kvm_desct->padding = 0;
3642 }
3643
3644 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3645                                           u16 selector,
3646                                           struct descriptor_table *dtable)
3647 {
3648         if (selector & 1 << 2) {
3649                 struct kvm_segment kvm_seg;
3650
3651                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3652
3653                 if (kvm_seg.unusable)
3654                         dtable->limit = 0;
3655                 else
3656                         dtable->limit = kvm_seg.limit;
3657                 dtable->base = kvm_seg.base;
3658         }
3659         else
3660                 kvm_x86_ops->get_gdt(vcpu, dtable);
3661 }
3662
3663 /* allowed just for 8 bytes segments */
3664 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3665                                          struct desc_struct *seg_desc)
3666 {
3667         gpa_t gpa;
3668         struct descriptor_table dtable;
3669         u16 index = selector >> 3;
3670
3671         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3672
3673         if (dtable.limit < index * 8 + 7) {
3674                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3675                 return 1;
3676         }
3677         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3678         gpa += index * 8;
3679         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3680 }
3681
3682 /* allowed just for 8 bytes segments */
3683 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3684                                          struct desc_struct *seg_desc)
3685 {
3686         gpa_t gpa;
3687         struct descriptor_table dtable;
3688         u16 index = selector >> 3;
3689
3690         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3691
3692         if (dtable.limit < index * 8 + 7)
3693                 return 1;
3694         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3695         gpa += index * 8;
3696         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3697 }
3698
3699 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3700                              struct desc_struct *seg_desc)
3701 {
3702         u32 base_addr;
3703
3704         base_addr = seg_desc->base0;
3705         base_addr |= (seg_desc->base1 << 16);
3706         base_addr |= (seg_desc->base2 << 24);
3707
3708         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3709 }
3710
3711 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3712 {
3713         struct kvm_segment kvm_seg;
3714
3715         kvm_get_segment(vcpu, &kvm_seg, seg);
3716         return kvm_seg.selector;
3717 }
3718
3719 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3720                                                 u16 selector,
3721                                                 struct kvm_segment *kvm_seg)
3722 {
3723         struct desc_struct seg_desc;
3724
3725         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3726                 return 1;
3727         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3728         return 0;
3729 }
3730
3731 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3732 {
3733         struct kvm_segment segvar = {
3734                 .base = selector << 4,
3735                 .limit = 0xffff,
3736                 .selector = selector,
3737                 .type = 3,
3738                 .present = 1,
3739                 .dpl = 3,
3740                 .db = 0,
3741                 .s = 1,
3742                 .l = 0,
3743                 .g = 0,
3744                 .avl = 0,
3745                 .unusable = 0,
3746         };
3747         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3748         return 0;
3749 }
3750
3751 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3752                                 int type_bits, int seg)
3753 {
3754         struct kvm_segment kvm_seg;
3755
3756         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3757                 return kvm_load_realmode_segment(vcpu, selector, seg);
3758         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3759                 return 1;
3760         kvm_seg.type |= type_bits;
3761
3762         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3763             seg != VCPU_SREG_LDTR)
3764                 if (!kvm_seg.s)
3765                         kvm_seg.unusable = 1;
3766
3767         kvm_set_segment(vcpu, &kvm_seg, seg);
3768         return 0;
3769 }
3770
3771 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3772                                 struct tss_segment_32 *tss)
3773 {
3774         tss->cr3 = vcpu->arch.cr3;
3775         tss->eip = kvm_rip_read(vcpu);
3776         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3777         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3778         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3779         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3780         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3781         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3782         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3783         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3784         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3785         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3786         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3787         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3788         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3789         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3790         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3791         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3792 }
3793
3794 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3795                                   struct tss_segment_32 *tss)
3796 {
3797         kvm_set_cr3(vcpu, tss->cr3);
3798
3799         kvm_rip_write(vcpu, tss->eip);
3800         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3801
3802         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3803         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3804         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3805         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3806         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3807         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3808         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3809         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3810
3811         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3812                 return 1;
3813
3814         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3815                 return 1;
3816
3817         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3818                 return 1;
3819
3820         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3821                 return 1;
3822
3823         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3824                 return 1;
3825
3826         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3827                 return 1;
3828
3829         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3830                 return 1;
3831         return 0;
3832 }
3833
3834 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3835                                 struct tss_segment_16 *tss)
3836 {
3837         tss->ip = kvm_rip_read(vcpu);
3838         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3839         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3840         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3841         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3842         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3843         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3844         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3845         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3846         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3847
3848         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3849         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3850         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3851         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3852         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3853         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3854 }
3855
3856 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3857                                  struct tss_segment_16 *tss)
3858 {
3859         kvm_rip_write(vcpu, tss->ip);
3860         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3861         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3862         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3863         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3864         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3865         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3866         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3867         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3868         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3869
3870         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3871                 return 1;
3872
3873         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3874                 return 1;
3875
3876         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3877                 return 1;
3878
3879         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3880                 return 1;
3881
3882         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3883                 return 1;
3884         return 0;
3885 }
3886
3887 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3888                               u16 old_tss_sel, u32 old_tss_base,
3889                               struct desc_struct *nseg_desc)
3890 {
3891         struct tss_segment_16 tss_segment_16;
3892         int ret = 0;
3893
3894         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3895                            sizeof tss_segment_16))
3896                 goto out;
3897
3898         save_state_to_tss16(vcpu, &tss_segment_16);
3899
3900         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3901                             sizeof tss_segment_16))
3902                 goto out;
3903
3904         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3905                            &tss_segment_16, sizeof tss_segment_16))
3906                 goto out;
3907
3908         if (old_tss_sel != 0xffff) {
3909                 tss_segment_16.prev_task_link = old_tss_sel;
3910
3911                 if (kvm_write_guest(vcpu->kvm,
3912                                     get_tss_base_addr(vcpu, nseg_desc),
3913                                     &tss_segment_16.prev_task_link,
3914                                     sizeof tss_segment_16.prev_task_link))
3915                         goto out;
3916         }
3917
3918         if (load_state_from_tss16(vcpu, &tss_segment_16))
3919                 goto out;
3920
3921         ret = 1;
3922 out:
3923         return ret;
3924 }
3925
3926 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3927                        u16 old_tss_sel, u32 old_tss_base,
3928                        struct desc_struct *nseg_desc)
3929 {
3930         struct tss_segment_32 tss_segment_32;
3931         int ret = 0;
3932
3933         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3934                            sizeof tss_segment_32))
3935                 goto out;
3936
3937         save_state_to_tss32(vcpu, &tss_segment_32);
3938
3939         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3940                             sizeof tss_segment_32))
3941                 goto out;
3942
3943         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3944                            &tss_segment_32, sizeof tss_segment_32))
3945                 goto out;
3946
3947         if (old_tss_sel != 0xffff) {
3948                 tss_segment_32.prev_task_link = old_tss_sel;
3949
3950                 if (kvm_write_guest(vcpu->kvm,
3951                                     get_tss_base_addr(vcpu, nseg_desc),
3952                                     &tss_segment_32.prev_task_link,
3953                                     sizeof tss_segment_32.prev_task_link))
3954                         goto out;
3955         }
3956
3957         if (load_state_from_tss32(vcpu, &tss_segment_32))
3958                 goto out;
3959
3960         ret = 1;
3961 out:
3962         return ret;
3963 }
3964
3965 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3966 {
3967         struct kvm_segment tr_seg;
3968         struct desc_struct cseg_desc;
3969         struct desc_struct nseg_desc;
3970         int ret = 0;
3971         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3972         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3973
3974         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3975
3976         /* FIXME: Handle errors. Failure to read either TSS or their
3977          * descriptors should generate a pagefault.
3978          */
3979         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3980                 goto out;
3981
3982         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3983                 goto out;
3984
3985         if (reason != TASK_SWITCH_IRET) {
3986                 int cpl;
3987
3988                 cpl = kvm_x86_ops->get_cpl(vcpu);
3989                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3990                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3991                         return 1;
3992                 }
3993         }
3994
3995         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3996                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3997                 return 1;
3998         }
3999
4000         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4001                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4002                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4003         }
4004
4005         if (reason == TASK_SWITCH_IRET) {
4006                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4007                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4008         }
4009
4010         /* set back link to prev task only if NT bit is set in eflags
4011            note that old_tss_sel is not used afetr this point */
4012         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4013                 old_tss_sel = 0xffff;
4014
4015         /* set back link to prev task only if NT bit is set in eflags
4016            note that old_tss_sel is not used afetr this point */
4017         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4018                 old_tss_sel = 0xffff;
4019
4020         if (nseg_desc.type & 8)
4021                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4022                                          old_tss_base, &nseg_desc);
4023         else
4024                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4025                                          old_tss_base, &nseg_desc);
4026
4027         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4028                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4029                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4030         }
4031
4032         if (reason != TASK_SWITCH_IRET) {
4033                 nseg_desc.type |= (1 << 1);
4034                 save_guest_segment_descriptor(vcpu, tss_selector,
4035                                               &nseg_desc);
4036         }
4037
4038         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4039         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4040         tr_seg.type = 11;
4041         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4042 out:
4043         return ret;
4044 }
4045 EXPORT_SYMBOL_GPL(kvm_task_switch);
4046
4047 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4048                                   struct kvm_sregs *sregs)
4049 {
4050         int mmu_reset_needed = 0;
4051         int pending_vec, max_bits;
4052         struct descriptor_table dt;
4053
4054         vcpu_load(vcpu);
4055
4056         dt.limit = sregs->idt.limit;
4057         dt.base = sregs->idt.base;
4058         kvm_x86_ops->set_idt(vcpu, &dt);
4059         dt.limit = sregs->gdt.limit;
4060         dt.base = sregs->gdt.base;
4061         kvm_x86_ops->set_gdt(vcpu, &dt);
4062
4063         vcpu->arch.cr2 = sregs->cr2;
4064         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4065
4066         down_read(&vcpu->kvm->slots_lock);
4067         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4068                 vcpu->arch.cr3 = sregs->cr3;
4069         else
4070                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4071         up_read(&vcpu->kvm->slots_lock);
4072
4073         kvm_set_cr8(vcpu, sregs->cr8);
4074
4075         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4076         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4077         kvm_set_apic_base(vcpu, sregs->apic_base);
4078
4079         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4080
4081         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4082         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4083         vcpu->arch.cr0 = sregs->cr0;
4084
4085         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4086         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4087         if (!is_long_mode(vcpu) && is_pae(vcpu))
4088                 load_pdptrs(vcpu, vcpu->arch.cr3);
4089
4090         if (mmu_reset_needed)
4091                 kvm_mmu_reset_context(vcpu);
4092
4093         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4094         pending_vec = find_first_bit(
4095                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4096         if (pending_vec < max_bits) {
4097                 kvm_queue_interrupt(vcpu, pending_vec, false);
4098                 pr_debug("Set back pending irq %d\n", pending_vec);
4099                 if (irqchip_in_kernel(vcpu->kvm))
4100                         kvm_pic_clear_isr_ack(vcpu->kvm);
4101         }
4102
4103         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4104         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4105         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4106         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4107         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4108         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4109
4110         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4111         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4112
4113         /* Older userspace won't unhalt the vcpu on reset. */
4114         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4115             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4116             !(vcpu->arch.cr0 & X86_CR0_PE))
4117                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4118
4119         vcpu_put(vcpu);
4120
4121         return 0;
4122 }
4123
4124 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4125                                         struct kvm_guest_debug *dbg)
4126 {
4127         int i, r;
4128
4129         vcpu_load(vcpu);
4130
4131         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4132             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4133                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4134                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4135                 vcpu->arch.switch_db_regs =
4136                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4137         } else {
4138                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4139                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4140                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4141         }
4142
4143         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4144
4145         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4146                 kvm_queue_exception(vcpu, DB_VECTOR);
4147         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4148                 kvm_queue_exception(vcpu, BP_VECTOR);
4149
4150         vcpu_put(vcpu);
4151
4152         return r;
4153 }
4154
4155 /*
4156  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4157  * we have asm/x86/processor.h
4158  */
4159 struct fxsave {
4160         u16     cwd;
4161         u16     swd;
4162         u16     twd;
4163         u16     fop;
4164         u64     rip;
4165         u64     rdp;
4166         u32     mxcsr;
4167         u32     mxcsr_mask;
4168         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4169 #ifdef CONFIG_X86_64
4170         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4171 #else
4172         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4173 #endif
4174 };
4175
4176 /*
4177  * Translate a guest virtual address to a guest physical address.
4178  */
4179 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4180                                     struct kvm_translation *tr)
4181 {
4182         unsigned long vaddr = tr->linear_address;
4183         gpa_t gpa;
4184
4185         vcpu_load(vcpu);
4186         down_read(&vcpu->kvm->slots_lock);
4187         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4188         up_read(&vcpu->kvm->slots_lock);
4189         tr->physical_address = gpa;
4190         tr->valid = gpa != UNMAPPED_GVA;
4191         tr->writeable = 1;
4192         tr->usermode = 0;
4193         vcpu_put(vcpu);
4194
4195         return 0;
4196 }
4197
4198 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4199 {
4200         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4201
4202         vcpu_load(vcpu);
4203
4204         memcpy(fpu->fpr, fxsave->st_space, 128);
4205         fpu->fcw = fxsave->cwd;
4206         fpu->fsw = fxsave->swd;
4207         fpu->ftwx = fxsave->twd;
4208         fpu->last_opcode = fxsave->fop;
4209         fpu->last_ip = fxsave->rip;
4210         fpu->last_dp = fxsave->rdp;
4211         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4212
4213         vcpu_put(vcpu);
4214
4215         return 0;
4216 }
4217
4218 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4219 {
4220         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4221
4222         vcpu_load(vcpu);
4223
4224         memcpy(fxsave->st_space, fpu->fpr, 128);
4225         fxsave->cwd = fpu->fcw;
4226         fxsave->swd = fpu->fsw;
4227         fxsave->twd = fpu->ftwx;
4228         fxsave->fop = fpu->last_opcode;
4229         fxsave->rip = fpu->last_ip;
4230         fxsave->rdp = fpu->last_dp;
4231         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4232
4233         vcpu_put(vcpu);
4234
4235         return 0;
4236 }
4237
4238 void fx_init(struct kvm_vcpu *vcpu)
4239 {
4240         unsigned after_mxcsr_mask;
4241
4242         /*
4243          * Touch the fpu the first time in non atomic context as if
4244          * this is the first fpu instruction the exception handler
4245          * will fire before the instruction returns and it'll have to
4246          * allocate ram with GFP_KERNEL.
4247          */
4248         if (!used_math())
4249                 kvm_fx_save(&vcpu->arch.host_fx_image);
4250
4251         /* Initialize guest FPU by resetting ours and saving into guest's */
4252         preempt_disable();
4253         kvm_fx_save(&vcpu->arch.host_fx_image);
4254         kvm_fx_finit();
4255         kvm_fx_save(&vcpu->arch.guest_fx_image);
4256         kvm_fx_restore(&vcpu->arch.host_fx_image);
4257         preempt_enable();
4258
4259         vcpu->arch.cr0 |= X86_CR0_ET;
4260         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4261         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4262         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4263                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4264 }
4265 EXPORT_SYMBOL_GPL(fx_init);
4266
4267 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4268 {
4269         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4270                 return;
4271
4272         vcpu->guest_fpu_loaded = 1;
4273         kvm_fx_save(&vcpu->arch.host_fx_image);
4274         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4275 }
4276 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4277
4278 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4279 {
4280         if (!vcpu->guest_fpu_loaded)
4281                 return;
4282
4283         vcpu->guest_fpu_loaded = 0;
4284         kvm_fx_save(&vcpu->arch.guest_fx_image);
4285         kvm_fx_restore(&vcpu->arch.host_fx_image);
4286         ++vcpu->stat.fpu_reload;
4287 }
4288 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4289
4290 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4291 {
4292         if (vcpu->arch.time_page) {
4293                 kvm_release_page_dirty(vcpu->arch.time_page);
4294                 vcpu->arch.time_page = NULL;
4295         }
4296
4297         kvm_x86_ops->vcpu_free(vcpu);
4298 }
4299
4300 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4301                                                 unsigned int id)
4302 {
4303         return kvm_x86_ops->vcpu_create(kvm, id);
4304 }
4305
4306 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4307 {
4308         int r;
4309
4310         /* We do fxsave: this must be aligned. */
4311         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4312
4313         vcpu->arch.mtrr_state.have_fixed = 1;
4314         vcpu_load(vcpu);
4315         r = kvm_arch_vcpu_reset(vcpu);
4316         if (r == 0)
4317                 r = kvm_mmu_setup(vcpu);
4318         vcpu_put(vcpu);
4319         if (r < 0)
4320                 goto free_vcpu;
4321
4322         return 0;
4323 free_vcpu:
4324         kvm_x86_ops->vcpu_free(vcpu);
4325         return r;
4326 }
4327
4328 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4329 {
4330         vcpu_load(vcpu);
4331         kvm_mmu_unload(vcpu);
4332         vcpu_put(vcpu);
4333
4334         kvm_x86_ops->vcpu_free(vcpu);
4335 }
4336
4337 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4338 {
4339         vcpu->arch.nmi_pending = false;
4340         vcpu->arch.nmi_injected = false;
4341
4342         vcpu->arch.switch_db_regs = 0;
4343         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4344         vcpu->arch.dr6 = DR6_FIXED_1;
4345         vcpu->arch.dr7 = DR7_FIXED_1;
4346
4347         return kvm_x86_ops->vcpu_reset(vcpu);
4348 }
4349
4350 void kvm_arch_hardware_enable(void *garbage)
4351 {
4352         kvm_x86_ops->hardware_enable(garbage);
4353 }
4354
4355 void kvm_arch_hardware_disable(void *garbage)
4356 {
4357         kvm_x86_ops->hardware_disable(garbage);
4358 }
4359
4360 int kvm_arch_hardware_setup(void)
4361 {
4362         return kvm_x86_ops->hardware_setup();
4363 }
4364
4365 void kvm_arch_hardware_unsetup(void)
4366 {
4367         kvm_x86_ops->hardware_unsetup();
4368 }
4369
4370 void kvm_arch_check_processor_compat(void *rtn)
4371 {
4372         kvm_x86_ops->check_processor_compatibility(rtn);
4373 }
4374
4375 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4376 {
4377         struct page *page;
4378         struct kvm *kvm;
4379         int r;
4380
4381         BUG_ON(vcpu->kvm == NULL);
4382         kvm = vcpu->kvm;
4383
4384         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4385         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4386                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4387         else
4388                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4389
4390         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4391         if (!page) {
4392                 r = -ENOMEM;
4393                 goto fail;
4394         }
4395         vcpu->arch.pio_data = page_address(page);
4396
4397         r = kvm_mmu_create(vcpu);
4398         if (r < 0)
4399                 goto fail_free_pio_data;
4400
4401         if (irqchip_in_kernel(kvm)) {
4402                 r = kvm_create_lapic(vcpu);
4403                 if (r < 0)
4404                         goto fail_mmu_destroy;
4405         }
4406
4407         return 0;
4408
4409 fail_mmu_destroy:
4410         kvm_mmu_destroy(vcpu);
4411 fail_free_pio_data:
4412         free_page((unsigned long)vcpu->arch.pio_data);
4413 fail:
4414         return r;
4415 }
4416
4417 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4418 {
4419         kvm_free_lapic(vcpu);
4420         down_read(&vcpu->kvm->slots_lock);
4421         kvm_mmu_destroy(vcpu);
4422         up_read(&vcpu->kvm->slots_lock);
4423         free_page((unsigned long)vcpu->arch.pio_data);
4424 }
4425
4426 struct  kvm *kvm_arch_create_vm(void)
4427 {
4428         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4429
4430         if (!kvm)
4431                 return ERR_PTR(-ENOMEM);
4432
4433         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4434         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4435
4436         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4437         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4438
4439         rdtscll(kvm->arch.vm_init_tsc);
4440
4441         return kvm;
4442 }
4443
4444 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4445 {
4446         vcpu_load(vcpu);
4447         kvm_mmu_unload(vcpu);
4448         vcpu_put(vcpu);
4449 }
4450
4451 static void kvm_free_vcpus(struct kvm *kvm)
4452 {
4453         unsigned int i;
4454
4455         /*
4456          * Unpin any mmu pages first.
4457          */
4458         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4459                 if (kvm->vcpus[i])
4460                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4461         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4462                 if (kvm->vcpus[i]) {
4463                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4464                         kvm->vcpus[i] = NULL;
4465                 }
4466         }
4467
4468 }
4469
4470 void kvm_arch_sync_events(struct kvm *kvm)
4471 {
4472         kvm_free_all_assigned_devices(kvm);
4473 }
4474
4475 void kvm_arch_destroy_vm(struct kvm *kvm)
4476 {
4477         kvm_iommu_unmap_guest(kvm);
4478         kvm_free_pit(kvm);
4479         kfree(kvm->arch.vpic);
4480         kfree(kvm->arch.vioapic);
4481         kvm_free_vcpus(kvm);
4482         kvm_free_physmem(kvm);
4483         if (kvm->arch.apic_access_page)
4484                 put_page(kvm->arch.apic_access_page);
4485         if (kvm->arch.ept_identity_pagetable)
4486                 put_page(kvm->arch.ept_identity_pagetable);
4487         kfree(kvm);
4488 }
4489
4490 int kvm_arch_set_memory_region(struct kvm *kvm,
4491                                 struct kvm_userspace_memory_region *mem,
4492                                 struct kvm_memory_slot old,
4493                                 int user_alloc)
4494 {
4495         int npages = mem->memory_size >> PAGE_SHIFT;
4496         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4497
4498         /*To keep backward compatibility with older userspace,
4499          *x86 needs to hanlde !user_alloc case.
4500          */
4501         if (!user_alloc) {
4502                 if (npages && !old.rmap) {
4503                         unsigned long userspace_addr;
4504
4505                         down_write(&current->mm->mmap_sem);
4506                         userspace_addr = do_mmap(NULL, 0,
4507                                                  npages * PAGE_SIZE,
4508                                                  PROT_READ | PROT_WRITE,
4509                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4510                                                  0);
4511                         up_write(&current->mm->mmap_sem);
4512
4513                         if (IS_ERR((void *)userspace_addr))
4514                                 return PTR_ERR((void *)userspace_addr);
4515
4516                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4517                         spin_lock(&kvm->mmu_lock);
4518                         memslot->userspace_addr = userspace_addr;
4519                         spin_unlock(&kvm->mmu_lock);
4520                 } else {
4521                         if (!old.user_alloc && old.rmap) {
4522                                 int ret;
4523
4524                                 down_write(&current->mm->mmap_sem);
4525                                 ret = do_munmap(current->mm, old.userspace_addr,
4526                                                 old.npages * PAGE_SIZE);
4527                                 up_write(&current->mm->mmap_sem);
4528                                 if (ret < 0)
4529                                         printk(KERN_WARNING
4530                                        "kvm_vm_ioctl_set_memory_region: "
4531                                        "failed to munmap memory\n");
4532                         }
4533                 }
4534         }
4535
4536         spin_lock(&kvm->mmu_lock);
4537         if (!kvm->arch.n_requested_mmu_pages) {
4538                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4539                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4540         }
4541
4542         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4543         spin_unlock(&kvm->mmu_lock);
4544         kvm_flush_remote_tlbs(kvm);
4545
4546         return 0;
4547 }
4548
4549 void kvm_arch_flush_shadow(struct kvm *kvm)
4550 {
4551         kvm_mmu_zap_all(kvm);
4552         kvm_reload_remote_mmus(kvm);
4553 }
4554
4555 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4556 {
4557         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4558                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4559                || vcpu->arch.nmi_pending;
4560 }
4561
4562 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4563 {
4564         int me;
4565         int cpu = vcpu->cpu;
4566
4567         if (waitqueue_active(&vcpu->wq)) {
4568                 wake_up_interruptible(&vcpu->wq);
4569                 ++vcpu->stat.halt_wakeup;
4570         }
4571
4572         me = get_cpu();
4573         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4574                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4575                         smp_send_reschedule(cpu);
4576         put_cpu();
4577 }
4578
4579 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4580 {
4581         return kvm_x86_ops->interrupt_allowed(vcpu);
4582 }