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         set_bit(irq->irq, vcpu->arch.irq_pending);
1445         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1446
1447         vcpu_put(vcpu);
1448
1449         return 0;
1450 }
1451
1452 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1453 {
1454         vcpu_load(vcpu);
1455         kvm_inject_nmi(vcpu);
1456         vcpu_put(vcpu);
1457
1458         return 0;
1459 }
1460
1461 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1462                                            struct kvm_tpr_access_ctl *tac)
1463 {
1464         if (tac->flags)
1465                 return -EINVAL;
1466         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1467         return 0;
1468 }
1469
1470 long kvm_arch_vcpu_ioctl(struct file *filp,
1471                          unsigned int ioctl, unsigned long arg)
1472 {
1473         struct kvm_vcpu *vcpu = filp->private_data;
1474         void __user *argp = (void __user *)arg;
1475         int r;
1476         struct kvm_lapic_state *lapic = NULL;
1477
1478         switch (ioctl) {
1479         case KVM_GET_LAPIC: {
1480                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1481
1482                 r = -ENOMEM;
1483                 if (!lapic)
1484                         goto out;
1485                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1486                 if (r)
1487                         goto out;
1488                 r = -EFAULT;
1489                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1490                         goto out;
1491                 r = 0;
1492                 break;
1493         }
1494         case KVM_SET_LAPIC: {
1495                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1496                 r = -ENOMEM;
1497                 if (!lapic)
1498                         goto out;
1499                 r = -EFAULT;
1500                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1501                         goto out;
1502                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1503                 if (r)
1504                         goto out;
1505                 r = 0;
1506                 break;
1507         }
1508         case KVM_INTERRUPT: {
1509                 struct kvm_interrupt irq;
1510
1511                 r = -EFAULT;
1512                 if (copy_from_user(&irq, argp, sizeof irq))
1513                         goto out;
1514                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1515                 if (r)
1516                         goto out;
1517                 r = 0;
1518                 break;
1519         }
1520         case KVM_NMI: {
1521                 r = kvm_vcpu_ioctl_nmi(vcpu);
1522                 if (r)
1523                         goto out;
1524                 r = 0;
1525                 break;
1526         }
1527         case KVM_SET_CPUID: {
1528                 struct kvm_cpuid __user *cpuid_arg = argp;
1529                 struct kvm_cpuid cpuid;
1530
1531                 r = -EFAULT;
1532                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1533                         goto out;
1534                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1535                 if (r)
1536                         goto out;
1537                 break;
1538         }
1539         case KVM_SET_CPUID2: {
1540                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1541                 struct kvm_cpuid2 cpuid;
1542
1543                 r = -EFAULT;
1544                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1545                         goto out;
1546                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1547                                               cpuid_arg->entries);
1548                 if (r)
1549                         goto out;
1550                 break;
1551         }
1552         case KVM_GET_CPUID2: {
1553                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1554                 struct kvm_cpuid2 cpuid;
1555
1556                 r = -EFAULT;
1557                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1558                         goto out;
1559                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1560                                               cpuid_arg->entries);
1561                 if (r)
1562                         goto out;
1563                 r = -EFAULT;
1564                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1565                         goto out;
1566                 r = 0;
1567                 break;
1568         }
1569         case KVM_GET_MSRS:
1570                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1571                 break;
1572         case KVM_SET_MSRS:
1573                 r = msr_io(vcpu, argp, do_set_msr, 0);
1574                 break;
1575         case KVM_TPR_ACCESS_REPORTING: {
1576                 struct kvm_tpr_access_ctl tac;
1577
1578                 r = -EFAULT;
1579                 if (copy_from_user(&tac, argp, sizeof tac))
1580                         goto out;
1581                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1582                 if (r)
1583                         goto out;
1584                 r = -EFAULT;
1585                 if (copy_to_user(argp, &tac, sizeof tac))
1586                         goto out;
1587                 r = 0;
1588                 break;
1589         };
1590         case KVM_SET_VAPIC_ADDR: {
1591                 struct kvm_vapic_addr va;
1592
1593                 r = -EINVAL;
1594                 if (!irqchip_in_kernel(vcpu->kvm))
1595                         goto out;
1596                 r = -EFAULT;
1597                 if (copy_from_user(&va, argp, sizeof va))
1598                         goto out;
1599                 r = 0;
1600                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1601                 break;
1602         }
1603         default:
1604                 r = -EINVAL;
1605         }
1606 out:
1607         kfree(lapic);
1608         return r;
1609 }
1610
1611 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1612 {
1613         int ret;
1614
1615         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1616                 return -1;
1617         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1618         return ret;
1619 }
1620
1621 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1622                                           u32 kvm_nr_mmu_pages)
1623 {
1624         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1625                 return -EINVAL;
1626
1627         down_write(&kvm->slots_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         up_write(&kvm->slots_lock);
1633         return 0;
1634 }
1635
1636 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1637 {
1638         return kvm->arch.n_alloc_mmu_pages;
1639 }
1640
1641 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1642 {
1643         int i;
1644         struct kvm_mem_alias *alias;
1645
1646         for (i = 0; i < kvm->arch.naliases; ++i) {
1647                 alias = &kvm->arch.aliases[i];
1648                 if (gfn >= alias->base_gfn
1649                     && gfn < alias->base_gfn + alias->npages)
1650                         return alias->target_gfn + gfn - alias->base_gfn;
1651         }
1652         return gfn;
1653 }
1654
1655 /*
1656  * Set a new alias region.  Aliases map a portion of physical memory into
1657  * another portion.  This is useful for memory windows, for example the PC
1658  * VGA region.
1659  */
1660 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1661                                          struct kvm_memory_alias *alias)
1662 {
1663         int r, n;
1664         struct kvm_mem_alias *p;
1665
1666         r = -EINVAL;
1667         /* General sanity checks */
1668         if (alias->memory_size & (PAGE_SIZE - 1))
1669                 goto out;
1670         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1671                 goto out;
1672         if (alias->slot >= KVM_ALIAS_SLOTS)
1673                 goto out;
1674         if (alias->guest_phys_addr + alias->memory_size
1675             < alias->guest_phys_addr)
1676                 goto out;
1677         if (alias->target_phys_addr + alias->memory_size
1678             < alias->target_phys_addr)
1679                 goto out;
1680
1681         down_write(&kvm->slots_lock);
1682         spin_lock(&kvm->mmu_lock);
1683
1684         p = &kvm->arch.aliases[alias->slot];
1685         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1686         p->npages = alias->memory_size >> PAGE_SHIFT;
1687         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1688
1689         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1690                 if (kvm->arch.aliases[n - 1].npages)
1691                         break;
1692         kvm->arch.naliases = n;
1693
1694         spin_unlock(&kvm->mmu_lock);
1695         kvm_mmu_zap_all(kvm);
1696
1697         up_write(&kvm->slots_lock);
1698
1699         return 0;
1700
1701 out:
1702         return r;
1703 }
1704
1705 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1706 {
1707         int r;
1708
1709         r = 0;
1710         switch (chip->chip_id) {
1711         case KVM_IRQCHIP_PIC_MASTER:
1712                 memcpy(&chip->chip.pic,
1713                         &pic_irqchip(kvm)->pics[0],
1714                         sizeof(struct kvm_pic_state));
1715                 break;
1716         case KVM_IRQCHIP_PIC_SLAVE:
1717                 memcpy(&chip->chip.pic,
1718                         &pic_irqchip(kvm)->pics[1],
1719                         sizeof(struct kvm_pic_state));
1720                 break;
1721         case KVM_IRQCHIP_IOAPIC:
1722                 memcpy(&chip->chip.ioapic,
1723                         ioapic_irqchip(kvm),
1724                         sizeof(struct kvm_ioapic_state));
1725                 break;
1726         default:
1727                 r = -EINVAL;
1728                 break;
1729         }
1730         return r;
1731 }
1732
1733 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1734 {
1735         int r;
1736
1737         r = 0;
1738         switch (chip->chip_id) {
1739         case KVM_IRQCHIP_PIC_MASTER:
1740                 memcpy(&pic_irqchip(kvm)->pics[0],
1741                         &chip->chip.pic,
1742                         sizeof(struct kvm_pic_state));
1743                 break;
1744         case KVM_IRQCHIP_PIC_SLAVE:
1745                 memcpy(&pic_irqchip(kvm)->pics[1],
1746                         &chip->chip.pic,
1747                         sizeof(struct kvm_pic_state));
1748                 break;
1749         case KVM_IRQCHIP_IOAPIC:
1750                 memcpy(ioapic_irqchip(kvm),
1751                         &chip->chip.ioapic,
1752                         sizeof(struct kvm_ioapic_state));
1753                 break;
1754         default:
1755                 r = -EINVAL;
1756                 break;
1757         }
1758         kvm_pic_update_irq(pic_irqchip(kvm));
1759         return r;
1760 }
1761
1762 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1763 {
1764         int r = 0;
1765
1766         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1767         return r;
1768 }
1769
1770 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1771 {
1772         int r = 0;
1773
1774         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1775         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1776         return r;
1777 }
1778
1779 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1780                                  struct kvm_reinject_control *control)
1781 {
1782         if (!kvm->arch.vpit)
1783                 return -ENXIO;
1784         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1785         return 0;
1786 }
1787
1788 /*
1789  * Get (and clear) the dirty memory log for a memory slot.
1790  */
1791 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1792                                       struct kvm_dirty_log *log)
1793 {
1794         int r;
1795         int n;
1796         struct kvm_memory_slot *memslot;
1797         int is_dirty = 0;
1798
1799         down_write(&kvm->slots_lock);
1800
1801         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1802         if (r)
1803                 goto out;
1804
1805         /* If nothing is dirty, don't bother messing with page tables. */
1806         if (is_dirty) {
1807                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1808                 kvm_flush_remote_tlbs(kvm);
1809                 memslot = &kvm->memslots[log->slot];
1810                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1811                 memset(memslot->dirty_bitmap, 0, n);
1812         }
1813         r = 0;
1814 out:
1815         up_write(&kvm->slots_lock);
1816         return r;
1817 }
1818
1819 long kvm_arch_vm_ioctl(struct file *filp,
1820                        unsigned int ioctl, unsigned long arg)
1821 {
1822         struct kvm *kvm = filp->private_data;
1823         void __user *argp = (void __user *)arg;
1824         int r = -EINVAL;
1825         /*
1826          * This union makes it completely explicit to gcc-3.x
1827          * that these two variables' stack usage should be
1828          * combined, not added together.
1829          */
1830         union {
1831                 struct kvm_pit_state ps;
1832                 struct kvm_memory_alias alias;
1833         } u;
1834
1835         switch (ioctl) {
1836         case KVM_SET_TSS_ADDR:
1837                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1838                 if (r < 0)
1839                         goto out;
1840                 break;
1841         case KVM_SET_MEMORY_REGION: {
1842                 struct kvm_memory_region kvm_mem;
1843                 struct kvm_userspace_memory_region kvm_userspace_mem;
1844
1845                 r = -EFAULT;
1846                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1847                         goto out;
1848                 kvm_userspace_mem.slot = kvm_mem.slot;
1849                 kvm_userspace_mem.flags = kvm_mem.flags;
1850                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1851                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1852                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1853                 if (r)
1854                         goto out;
1855                 break;
1856         }
1857         case KVM_SET_NR_MMU_PAGES:
1858                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1859                 if (r)
1860                         goto out;
1861                 break;
1862         case KVM_GET_NR_MMU_PAGES:
1863                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1864                 break;
1865         case KVM_SET_MEMORY_ALIAS:
1866                 r = -EFAULT;
1867                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1868                         goto out;
1869                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1870                 if (r)
1871                         goto out;
1872                 break;
1873         case KVM_CREATE_IRQCHIP:
1874                 r = -ENOMEM;
1875                 kvm->arch.vpic = kvm_create_pic(kvm);
1876                 if (kvm->arch.vpic) {
1877                         r = kvm_ioapic_init(kvm);
1878                         if (r) {
1879                                 kfree(kvm->arch.vpic);
1880                                 kvm->arch.vpic = NULL;
1881                                 goto out;
1882                         }
1883                 } else
1884                         goto out;
1885                 r = kvm_setup_default_irq_routing(kvm);
1886                 if (r) {
1887                         kfree(kvm->arch.vpic);
1888                         kfree(kvm->arch.vioapic);
1889                         goto out;
1890                 }
1891                 break;
1892         case KVM_CREATE_PIT:
1893                 mutex_lock(&kvm->lock);
1894                 r = -EEXIST;
1895                 if (kvm->arch.vpit)
1896                         goto create_pit_unlock;
1897                 r = -ENOMEM;
1898                 kvm->arch.vpit = kvm_create_pit(kvm);
1899                 if (kvm->arch.vpit)
1900                         r = 0;
1901         create_pit_unlock:
1902                 mutex_unlock(&kvm->lock);
1903                 break;
1904         case KVM_IRQ_LINE_STATUS:
1905         case KVM_IRQ_LINE: {
1906                 struct kvm_irq_level irq_event;
1907
1908                 r = -EFAULT;
1909                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1910                         goto out;
1911                 if (irqchip_in_kernel(kvm)) {
1912                         __s32 status;
1913                         mutex_lock(&kvm->lock);
1914                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1915                                         irq_event.irq, irq_event.level);
1916                         mutex_unlock(&kvm->lock);
1917                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1918                                 irq_event.status = status;
1919                                 if (copy_to_user(argp, &irq_event,
1920                                                         sizeof irq_event))
1921                                         goto out;
1922                         }
1923                         r = 0;
1924                 }
1925                 break;
1926         }
1927         case KVM_GET_IRQCHIP: {
1928                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1929                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1930
1931                 r = -ENOMEM;
1932                 if (!chip)
1933                         goto out;
1934                 r = -EFAULT;
1935                 if (copy_from_user(chip, argp, sizeof *chip))
1936                         goto get_irqchip_out;
1937                 r = -ENXIO;
1938                 if (!irqchip_in_kernel(kvm))
1939                         goto get_irqchip_out;
1940                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1941                 if (r)
1942                         goto get_irqchip_out;
1943                 r = -EFAULT;
1944                 if (copy_to_user(argp, chip, sizeof *chip))
1945                         goto get_irqchip_out;
1946                 r = 0;
1947         get_irqchip_out:
1948                 kfree(chip);
1949                 if (r)
1950                         goto out;
1951                 break;
1952         }
1953         case KVM_SET_IRQCHIP: {
1954                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1955                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1956
1957                 r = -ENOMEM;
1958                 if (!chip)
1959                         goto out;
1960                 r = -EFAULT;
1961                 if (copy_from_user(chip, argp, sizeof *chip))
1962                         goto set_irqchip_out;
1963                 r = -ENXIO;
1964                 if (!irqchip_in_kernel(kvm))
1965                         goto set_irqchip_out;
1966                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1967                 if (r)
1968                         goto set_irqchip_out;
1969                 r = 0;
1970         set_irqchip_out:
1971                 kfree(chip);
1972                 if (r)
1973                         goto out;
1974                 break;
1975         }
1976         case KVM_GET_PIT: {
1977                 r = -EFAULT;
1978                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1979                         goto out;
1980                 r = -ENXIO;
1981                 if (!kvm->arch.vpit)
1982                         goto out;
1983                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1984                 if (r)
1985                         goto out;
1986                 r = -EFAULT;
1987                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1988                         goto out;
1989                 r = 0;
1990                 break;
1991         }
1992         case KVM_SET_PIT: {
1993                 r = -EFAULT;
1994                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1995                         goto out;
1996                 r = -ENXIO;
1997                 if (!kvm->arch.vpit)
1998                         goto out;
1999                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2000                 if (r)
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_REINJECT_CONTROL: {
2006                 struct kvm_reinject_control control;
2007                 r =  -EFAULT;
2008                 if (copy_from_user(&control, argp, sizeof(control)))
2009                         goto out;
2010                 r = kvm_vm_ioctl_reinject(kvm, &control);
2011                 if (r)
2012                         goto out;
2013                 r = 0;
2014                 break;
2015         }
2016         default:
2017                 ;
2018         }
2019 out:
2020         return r;
2021 }
2022
2023 static void kvm_init_msr_list(void)
2024 {
2025         u32 dummy[2];
2026         unsigned i, j;
2027
2028         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2029                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2030                         continue;
2031                 if (j < i)
2032                         msrs_to_save[j] = msrs_to_save[i];
2033                 j++;
2034         }
2035         num_msrs_to_save = j;
2036 }
2037
2038 /*
2039  * Only apic need an MMIO device hook, so shortcut now..
2040  */
2041 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2042                                                 gpa_t addr, int len,
2043                                                 int is_write)
2044 {
2045         struct kvm_io_device *dev;
2046
2047         if (vcpu->arch.apic) {
2048                 dev = &vcpu->arch.apic->dev;
2049                 if (dev->in_range(dev, addr, len, is_write))
2050                         return dev;
2051         }
2052         return NULL;
2053 }
2054
2055
2056 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2057                                                 gpa_t addr, int len,
2058                                                 int is_write)
2059 {
2060         struct kvm_io_device *dev;
2061
2062         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2063         if (dev == NULL)
2064                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2065                                           is_write);
2066         return dev;
2067 }
2068
2069 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2070                                struct kvm_vcpu *vcpu)
2071 {
2072         void *data = val;
2073         int r = X86EMUL_CONTINUE;
2074
2075         while (bytes) {
2076                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2077                 unsigned offset = addr & (PAGE_SIZE-1);
2078                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2079                 int ret;
2080
2081                 if (gpa == UNMAPPED_GVA) {
2082                         r = X86EMUL_PROPAGATE_FAULT;
2083                         goto out;
2084                 }
2085                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2086                 if (ret < 0) {
2087                         r = X86EMUL_UNHANDLEABLE;
2088                         goto out;
2089                 }
2090
2091                 bytes -= toread;
2092                 data += toread;
2093                 addr += toread;
2094         }
2095 out:
2096         return r;
2097 }
2098
2099 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2100                                 struct kvm_vcpu *vcpu)
2101 {
2102         void *data = val;
2103         int r = X86EMUL_CONTINUE;
2104
2105         while (bytes) {
2106                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2107                 unsigned offset = addr & (PAGE_SIZE-1);
2108                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2109                 int ret;
2110
2111                 if (gpa == UNMAPPED_GVA) {
2112                         r = X86EMUL_PROPAGATE_FAULT;
2113                         goto out;
2114                 }
2115                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2116                 if (ret < 0) {
2117                         r = X86EMUL_UNHANDLEABLE;
2118                         goto out;
2119                 }
2120
2121                 bytes -= towrite;
2122                 data += towrite;
2123                 addr += towrite;
2124         }
2125 out:
2126         return r;
2127 }
2128
2129
2130 static int emulator_read_emulated(unsigned long addr,
2131                                   void *val,
2132                                   unsigned int bytes,
2133                                   struct kvm_vcpu *vcpu)
2134 {
2135         struct kvm_io_device *mmio_dev;
2136         gpa_t                 gpa;
2137
2138         if (vcpu->mmio_read_completed) {
2139                 memcpy(val, vcpu->mmio_data, bytes);
2140                 vcpu->mmio_read_completed = 0;
2141                 return X86EMUL_CONTINUE;
2142         }
2143
2144         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2145
2146         /* For APIC access vmexit */
2147         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2148                 goto mmio;
2149
2150         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2151                                 == X86EMUL_CONTINUE)
2152                 return X86EMUL_CONTINUE;
2153         if (gpa == UNMAPPED_GVA)
2154                 return X86EMUL_PROPAGATE_FAULT;
2155
2156 mmio:
2157         /*
2158          * Is this MMIO handled locally?
2159          */
2160         mutex_lock(&vcpu->kvm->lock);
2161         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2162         if (mmio_dev) {
2163                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2164                 mutex_unlock(&vcpu->kvm->lock);
2165                 return X86EMUL_CONTINUE;
2166         }
2167         mutex_unlock(&vcpu->kvm->lock);
2168
2169         vcpu->mmio_needed = 1;
2170         vcpu->mmio_phys_addr = gpa;
2171         vcpu->mmio_size = bytes;
2172         vcpu->mmio_is_write = 0;
2173
2174         return X86EMUL_UNHANDLEABLE;
2175 }
2176
2177 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2178                           const void *val, int bytes)
2179 {
2180         int ret;
2181
2182         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2183         if (ret < 0)
2184                 return 0;
2185         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2186         return 1;
2187 }
2188
2189 static int emulator_write_emulated_onepage(unsigned long addr,
2190                                            const void *val,
2191                                            unsigned int bytes,
2192                                            struct kvm_vcpu *vcpu)
2193 {
2194         struct kvm_io_device *mmio_dev;
2195         gpa_t                 gpa;
2196
2197         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2198
2199         if (gpa == UNMAPPED_GVA) {
2200                 kvm_inject_page_fault(vcpu, addr, 2);
2201                 return X86EMUL_PROPAGATE_FAULT;
2202         }
2203
2204         /* For APIC access vmexit */
2205         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2206                 goto mmio;
2207
2208         if (emulator_write_phys(vcpu, gpa, val, bytes))
2209                 return X86EMUL_CONTINUE;
2210
2211 mmio:
2212         /*
2213          * Is this MMIO handled locally?
2214          */
2215         mutex_lock(&vcpu->kvm->lock);
2216         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2217         if (mmio_dev) {
2218                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2219                 mutex_unlock(&vcpu->kvm->lock);
2220                 return X86EMUL_CONTINUE;
2221         }
2222         mutex_unlock(&vcpu->kvm->lock);
2223
2224         vcpu->mmio_needed = 1;
2225         vcpu->mmio_phys_addr = gpa;
2226         vcpu->mmio_size = bytes;
2227         vcpu->mmio_is_write = 1;
2228         memcpy(vcpu->mmio_data, val, bytes);
2229
2230         return X86EMUL_CONTINUE;
2231 }
2232
2233 int emulator_write_emulated(unsigned long addr,
2234                                    const void *val,
2235                                    unsigned int bytes,
2236                                    struct kvm_vcpu *vcpu)
2237 {
2238         /* Crossing a page boundary? */
2239         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2240                 int rc, now;
2241
2242                 now = -addr & ~PAGE_MASK;
2243                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2244                 if (rc != X86EMUL_CONTINUE)
2245                         return rc;
2246                 addr += now;
2247                 val += now;
2248                 bytes -= now;
2249         }
2250         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2251 }
2252 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2253
2254 static int emulator_cmpxchg_emulated(unsigned long addr,
2255                                      const void *old,
2256                                      const void *new,
2257                                      unsigned int bytes,
2258                                      struct kvm_vcpu *vcpu)
2259 {
2260         static int reported;
2261
2262         if (!reported) {
2263                 reported = 1;
2264                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2265         }
2266 #ifndef CONFIG_X86_64
2267         /* guests cmpxchg8b have to be emulated atomically */
2268         if (bytes == 8) {
2269                 gpa_t gpa;
2270                 struct page *page;
2271                 char *kaddr;
2272                 u64 val;
2273
2274                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2275
2276                 if (gpa == UNMAPPED_GVA ||
2277                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2278                         goto emul_write;
2279
2280                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2281                         goto emul_write;
2282
2283                 val = *(u64 *)new;
2284
2285                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2286
2287                 kaddr = kmap_atomic(page, KM_USER0);
2288                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2289                 kunmap_atomic(kaddr, KM_USER0);
2290                 kvm_release_page_dirty(page);
2291         }
2292 emul_write:
2293 #endif
2294
2295         return emulator_write_emulated(addr, new, bytes, vcpu);
2296 }
2297
2298 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2299 {
2300         return kvm_x86_ops->get_segment_base(vcpu, seg);
2301 }
2302
2303 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2304 {
2305         kvm_mmu_invlpg(vcpu, address);
2306         return X86EMUL_CONTINUE;
2307 }
2308
2309 int emulate_clts(struct kvm_vcpu *vcpu)
2310 {
2311         KVMTRACE_0D(CLTS, vcpu, handler);
2312         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2313         return X86EMUL_CONTINUE;
2314 }
2315
2316 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2317 {
2318         struct kvm_vcpu *vcpu = ctxt->vcpu;
2319
2320         switch (dr) {
2321         case 0 ... 3:
2322                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2323                 return X86EMUL_CONTINUE;
2324         default:
2325                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2326                 return X86EMUL_UNHANDLEABLE;
2327         }
2328 }
2329
2330 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2331 {
2332         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2333         int exception;
2334
2335         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2336         if (exception) {
2337                 /* FIXME: better handling */
2338                 return X86EMUL_UNHANDLEABLE;
2339         }
2340         return X86EMUL_CONTINUE;
2341 }
2342
2343 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2344 {
2345         u8 opcodes[4];
2346         unsigned long rip = kvm_rip_read(vcpu);
2347         unsigned long rip_linear;
2348
2349         if (!printk_ratelimit())
2350                 return;
2351
2352         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2353
2354         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2355
2356         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2357                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2358 }
2359 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2360
2361 static struct x86_emulate_ops emulate_ops = {
2362         .read_std            = kvm_read_guest_virt,
2363         .read_emulated       = emulator_read_emulated,
2364         .write_emulated      = emulator_write_emulated,
2365         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2366 };
2367
2368 static void cache_all_regs(struct kvm_vcpu *vcpu)
2369 {
2370         kvm_register_read(vcpu, VCPU_REGS_RAX);
2371         kvm_register_read(vcpu, VCPU_REGS_RSP);
2372         kvm_register_read(vcpu, VCPU_REGS_RIP);
2373         vcpu->arch.regs_dirty = ~0;
2374 }
2375
2376 int emulate_instruction(struct kvm_vcpu *vcpu,
2377                         struct kvm_run *run,
2378                         unsigned long cr2,
2379                         u16 error_code,
2380                         int emulation_type)
2381 {
2382         int r;
2383         struct decode_cache *c;
2384
2385         kvm_clear_exception_queue(vcpu);
2386         vcpu->arch.mmio_fault_cr2 = cr2;
2387         /*
2388          * TODO: fix x86_emulate.c to use guest_read/write_register
2389          * instead of direct ->regs accesses, can save hundred cycles
2390          * on Intel for instructions that don't read/change RSP, for
2391          * for example.
2392          */
2393         cache_all_regs(vcpu);
2394
2395         vcpu->mmio_is_write = 0;
2396         vcpu->arch.pio.string = 0;
2397
2398         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2399                 int cs_db, cs_l;
2400                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2401
2402                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2403                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2404                 vcpu->arch.emulate_ctxt.mode =
2405                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2406                         ? X86EMUL_MODE_REAL : cs_l
2407                         ? X86EMUL_MODE_PROT64 : cs_db
2408                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2409
2410                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2411
2412                 /* Reject the instructions other than VMCALL/VMMCALL when
2413                  * try to emulate invalid opcode */
2414                 c = &vcpu->arch.emulate_ctxt.decode;
2415                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2416                     (!(c->twobyte && c->b == 0x01 &&
2417                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2418                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2419                         return EMULATE_FAIL;
2420
2421                 ++vcpu->stat.insn_emulation;
2422                 if (r)  {
2423                         ++vcpu->stat.insn_emulation_fail;
2424                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2425                                 return EMULATE_DONE;
2426                         return EMULATE_FAIL;
2427                 }
2428         }
2429
2430         if (emulation_type & EMULTYPE_SKIP) {
2431                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2432                 return EMULATE_DONE;
2433         }
2434
2435         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2436
2437         if (vcpu->arch.pio.string)
2438                 return EMULATE_DO_MMIO;
2439
2440         if ((r || vcpu->mmio_is_write) && run) {
2441                 run->exit_reason = KVM_EXIT_MMIO;
2442                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2443                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2444                 run->mmio.len = vcpu->mmio_size;
2445                 run->mmio.is_write = vcpu->mmio_is_write;
2446         }
2447
2448         if (r) {
2449                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2450                         return EMULATE_DONE;
2451                 if (!vcpu->mmio_needed) {
2452                         kvm_report_emulation_failure(vcpu, "mmio");
2453                         return EMULATE_FAIL;
2454                 }
2455                 return EMULATE_DO_MMIO;
2456         }
2457
2458         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2459
2460         if (vcpu->mmio_is_write) {
2461                 vcpu->mmio_needed = 0;
2462                 return EMULATE_DO_MMIO;
2463         }
2464
2465         return EMULATE_DONE;
2466 }
2467 EXPORT_SYMBOL_GPL(emulate_instruction);
2468
2469 static int pio_copy_data(struct kvm_vcpu *vcpu)
2470 {
2471         void *p = vcpu->arch.pio_data;
2472         gva_t q = vcpu->arch.pio.guest_gva;
2473         unsigned bytes;
2474         int ret;
2475
2476         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2477         if (vcpu->arch.pio.in)
2478                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2479         else
2480                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2481         return ret;
2482 }
2483
2484 int complete_pio(struct kvm_vcpu *vcpu)
2485 {
2486         struct kvm_pio_request *io = &vcpu->arch.pio;
2487         long delta;
2488         int r;
2489         unsigned long val;
2490
2491         if (!io->string) {
2492                 if (io->in) {
2493                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2494                         memcpy(&val, vcpu->arch.pio_data, io->size);
2495                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2496                 }
2497         } else {
2498                 if (io->in) {
2499                         r = pio_copy_data(vcpu);
2500                         if (r)
2501                                 return r;
2502                 }
2503
2504                 delta = 1;
2505                 if (io->rep) {
2506                         delta *= io->cur_count;
2507                         /*
2508                          * The size of the register should really depend on
2509                          * current address size.
2510                          */
2511                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2512                         val -= delta;
2513                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2514                 }
2515                 if (io->down)
2516                         delta = -delta;
2517                 delta *= io->size;
2518                 if (io->in) {
2519                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2520                         val += delta;
2521                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2522                 } else {
2523                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2524                         val += delta;
2525                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2526                 }
2527         }
2528
2529         io->count -= io->cur_count;
2530         io->cur_count = 0;
2531
2532         return 0;
2533 }
2534
2535 static void kernel_pio(struct kvm_io_device *pio_dev,
2536                        struct kvm_vcpu *vcpu,
2537                        void *pd)
2538 {
2539         /* TODO: String I/O for in kernel device */
2540
2541         mutex_lock(&vcpu->kvm->lock);
2542         if (vcpu->arch.pio.in)
2543                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2544                                   vcpu->arch.pio.size,
2545                                   pd);
2546         else
2547                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2548                                    vcpu->arch.pio.size,
2549                                    pd);
2550         mutex_unlock(&vcpu->kvm->lock);
2551 }
2552
2553 static void pio_string_write(struct kvm_io_device *pio_dev,
2554                              struct kvm_vcpu *vcpu)
2555 {
2556         struct kvm_pio_request *io = &vcpu->arch.pio;
2557         void *pd = vcpu->arch.pio_data;
2558         int i;
2559
2560         mutex_lock(&vcpu->kvm->lock);
2561         for (i = 0; i < io->cur_count; i++) {
2562                 kvm_iodevice_write(pio_dev, io->port,
2563                                    io->size,
2564                                    pd);
2565                 pd += io->size;
2566         }
2567         mutex_unlock(&vcpu->kvm->lock);
2568 }
2569
2570 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2571                                                gpa_t addr, int len,
2572                                                int is_write)
2573 {
2574         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2575 }
2576
2577 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2578                   int size, unsigned port)
2579 {
2580         struct kvm_io_device *pio_dev;
2581         unsigned long val;
2582
2583         vcpu->run->exit_reason = KVM_EXIT_IO;
2584         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2585         vcpu->run->io.size = vcpu->arch.pio.size = size;
2586         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2587         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2588         vcpu->run->io.port = vcpu->arch.pio.port = port;
2589         vcpu->arch.pio.in = in;
2590         vcpu->arch.pio.string = 0;
2591         vcpu->arch.pio.down = 0;
2592         vcpu->arch.pio.rep = 0;
2593
2594         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2595                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2596                             handler);
2597         else
2598                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2599                             handler);
2600
2601         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2602         memcpy(vcpu->arch.pio_data, &val, 4);
2603
2604         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2605         if (pio_dev) {
2606                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2607                 complete_pio(vcpu);
2608                 return 1;
2609         }
2610         return 0;
2611 }
2612 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2613
2614 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2615                   int size, unsigned long count, int down,
2616                   gva_t address, int rep, unsigned port)
2617 {
2618         unsigned now, in_page;
2619         int ret = 0;
2620         struct kvm_io_device *pio_dev;
2621
2622         vcpu->run->exit_reason = KVM_EXIT_IO;
2623         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2624         vcpu->run->io.size = vcpu->arch.pio.size = size;
2625         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2626         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2627         vcpu->run->io.port = vcpu->arch.pio.port = port;
2628         vcpu->arch.pio.in = in;
2629         vcpu->arch.pio.string = 1;
2630         vcpu->arch.pio.down = down;
2631         vcpu->arch.pio.rep = rep;
2632
2633         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2634                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2635                             handler);
2636         else
2637                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2638                             handler);
2639
2640         if (!count) {
2641                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2642                 return 1;
2643         }
2644
2645         if (!down)
2646                 in_page = PAGE_SIZE - offset_in_page(address);
2647         else
2648                 in_page = offset_in_page(address) + size;
2649         now = min(count, (unsigned long)in_page / size);
2650         if (!now)
2651                 now = 1;
2652         if (down) {
2653                 /*
2654                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2655                  */
2656                 pr_unimpl(vcpu, "guest string pio down\n");
2657                 kvm_inject_gp(vcpu, 0);
2658                 return 1;
2659         }
2660         vcpu->run->io.count = now;
2661         vcpu->arch.pio.cur_count = now;
2662
2663         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2664                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2665
2666         vcpu->arch.pio.guest_gva = address;
2667
2668         pio_dev = vcpu_find_pio_dev(vcpu, port,
2669                                     vcpu->arch.pio.cur_count,
2670                                     !vcpu->arch.pio.in);
2671         if (!vcpu->arch.pio.in) {
2672                 /* string PIO write */
2673                 ret = pio_copy_data(vcpu);
2674                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2675                         kvm_inject_gp(vcpu, 0);
2676                         return 1;
2677                 }
2678                 if (ret == 0 && pio_dev) {
2679                         pio_string_write(pio_dev, vcpu);
2680                         complete_pio(vcpu);
2681                         if (vcpu->arch.pio.count == 0)
2682                                 ret = 1;
2683                 }
2684         } else if (pio_dev)
2685                 pr_unimpl(vcpu, "no string pio read support yet, "
2686                        "port %x size %d count %ld\n",
2687                         port, size, count);
2688
2689         return ret;
2690 }
2691 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2692
2693 static void bounce_off(void *info)
2694 {
2695         /* nothing */
2696 }
2697
2698 static unsigned int  ref_freq;
2699 static unsigned long tsc_khz_ref;
2700
2701 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2702                                      void *data)
2703 {
2704         struct cpufreq_freqs *freq = data;
2705         struct kvm *kvm;
2706         struct kvm_vcpu *vcpu;
2707         int i, send_ipi = 0;
2708
2709         if (!ref_freq)
2710                 ref_freq = freq->old;
2711
2712         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2713                 return 0;
2714         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2715                 return 0;
2716         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2717
2718         spin_lock(&kvm_lock);
2719         list_for_each_entry(kvm, &vm_list, vm_list) {
2720                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2721                         vcpu = kvm->vcpus[i];
2722                         if (!vcpu)
2723                                 continue;
2724                         if (vcpu->cpu != freq->cpu)
2725                                 continue;
2726                         if (!kvm_request_guest_time_update(vcpu))
2727                                 continue;
2728                         if (vcpu->cpu != smp_processor_id())
2729                                 send_ipi++;
2730                 }
2731         }
2732         spin_unlock(&kvm_lock);
2733
2734         if (freq->old < freq->new && send_ipi) {
2735                 /*
2736                  * We upscale the frequency.  Must make the guest
2737                  * doesn't see old kvmclock values while running with
2738                  * the new frequency, otherwise we risk the guest sees
2739                  * time go backwards.
2740                  *
2741                  * In case we update the frequency for another cpu
2742                  * (which might be in guest context) send an interrupt
2743                  * to kick the cpu out of guest context.  Next time
2744                  * guest context is entered kvmclock will be updated,
2745                  * so the guest will not see stale values.
2746                  */
2747                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2748         }
2749         return 0;
2750 }
2751
2752 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2753         .notifier_call  = kvmclock_cpufreq_notifier
2754 };
2755
2756 int kvm_arch_init(void *opaque)
2757 {
2758         int r, cpu;
2759         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2760
2761         if (kvm_x86_ops) {
2762                 printk(KERN_ERR "kvm: already loaded the other module\n");
2763                 r = -EEXIST;
2764                 goto out;
2765         }
2766
2767         if (!ops->cpu_has_kvm_support()) {
2768                 printk(KERN_ERR "kvm: no hardware support\n");
2769                 r = -EOPNOTSUPP;
2770                 goto out;
2771         }
2772         if (ops->disabled_by_bios()) {
2773                 printk(KERN_ERR "kvm: disabled by bios\n");
2774                 r = -EOPNOTSUPP;
2775                 goto out;
2776         }
2777
2778         r = kvm_mmu_module_init();
2779         if (r)
2780                 goto out;
2781
2782         kvm_init_msr_list();
2783
2784         kvm_x86_ops = ops;
2785         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2786         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2787         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2788                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2789
2790         for_each_possible_cpu(cpu)
2791                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2792         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2793                 tsc_khz_ref = tsc_khz;
2794                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2795                                           CPUFREQ_TRANSITION_NOTIFIER);
2796         }
2797
2798         return 0;
2799
2800 out:
2801         return r;
2802 }
2803
2804 void kvm_arch_exit(void)
2805 {
2806         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2807                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2808                                             CPUFREQ_TRANSITION_NOTIFIER);
2809         kvm_x86_ops = NULL;
2810         kvm_mmu_module_exit();
2811 }
2812
2813 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2814 {
2815         ++vcpu->stat.halt_exits;
2816         KVMTRACE_0D(HLT, vcpu, handler);
2817         if (irqchip_in_kernel(vcpu->kvm)) {
2818                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2819                 return 1;
2820         } else {
2821                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2822                 return 0;
2823         }
2824 }
2825 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2826
2827 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2828                            unsigned long a1)
2829 {
2830         if (is_long_mode(vcpu))
2831                 return a0;
2832         else
2833                 return a0 | ((gpa_t)a1 << 32);
2834 }
2835
2836 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2837 {
2838         unsigned long nr, a0, a1, a2, a3, ret;
2839         int r = 1;
2840
2841         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2842         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2843         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2844         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2845         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2846
2847         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2848
2849         if (!is_long_mode(vcpu)) {
2850                 nr &= 0xFFFFFFFF;
2851                 a0 &= 0xFFFFFFFF;
2852                 a1 &= 0xFFFFFFFF;
2853                 a2 &= 0xFFFFFFFF;
2854                 a3 &= 0xFFFFFFFF;
2855         }
2856
2857         switch (nr) {
2858         case KVM_HC_VAPIC_POLL_IRQ:
2859                 ret = 0;
2860                 break;
2861         case KVM_HC_MMU_OP:
2862                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2863                 break;
2864         default:
2865                 ret = -KVM_ENOSYS;
2866                 break;
2867         }
2868         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2869         ++vcpu->stat.hypercalls;
2870         return r;
2871 }
2872 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2873
2874 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2875 {
2876         char instruction[3];
2877         int ret = 0;
2878         unsigned long rip = kvm_rip_read(vcpu);
2879
2880
2881         /*
2882          * Blow out the MMU to ensure that no other VCPU has an active mapping
2883          * to ensure that the updated hypercall appears atomically across all
2884          * VCPUs.
2885          */
2886         kvm_mmu_zap_all(vcpu->kvm);
2887
2888         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2889         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2890             != X86EMUL_CONTINUE)
2891                 ret = -EFAULT;
2892
2893         return ret;
2894 }
2895
2896 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2897 {
2898         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2899 }
2900
2901 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2902 {
2903         struct descriptor_table dt = { limit, base };
2904
2905         kvm_x86_ops->set_gdt(vcpu, &dt);
2906 }
2907
2908 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2909 {
2910         struct descriptor_table dt = { limit, base };
2911
2912         kvm_x86_ops->set_idt(vcpu, &dt);
2913 }
2914
2915 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2916                    unsigned long *rflags)
2917 {
2918         kvm_lmsw(vcpu, msw);
2919         *rflags = kvm_x86_ops->get_rflags(vcpu);
2920 }
2921
2922 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2923 {
2924         unsigned long value;
2925
2926         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2927         switch (cr) {
2928         case 0:
2929                 value = vcpu->arch.cr0;
2930                 break;
2931         case 2:
2932                 value = vcpu->arch.cr2;
2933                 break;
2934         case 3:
2935                 value = vcpu->arch.cr3;
2936                 break;
2937         case 4:
2938                 value = vcpu->arch.cr4;
2939                 break;
2940         case 8:
2941                 value = kvm_get_cr8(vcpu);
2942                 break;
2943         default:
2944                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2945                 return 0;
2946         }
2947         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2948                     (u32)((u64)value >> 32), handler);
2949
2950         return value;
2951 }
2952
2953 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2954                      unsigned long *rflags)
2955 {
2956         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2957                     (u32)((u64)val >> 32), handler);
2958
2959         switch (cr) {
2960         case 0:
2961                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2962                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2963                 break;
2964         case 2:
2965                 vcpu->arch.cr2 = val;
2966                 break;
2967         case 3:
2968                 kvm_set_cr3(vcpu, val);
2969                 break;
2970         case 4:
2971                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2972                 break;
2973         case 8:
2974                 kvm_set_cr8(vcpu, val & 0xfUL);
2975                 break;
2976         default:
2977                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2978         }
2979 }
2980
2981 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2982 {
2983         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2984         int j, nent = vcpu->arch.cpuid_nent;
2985
2986         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2987         /* when no next entry is found, the current entry[i] is reselected */
2988         for (j = i + 1; ; j = (j + 1) % nent) {
2989                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2990                 if (ej->function == e->function) {
2991                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2992                         return j;
2993                 }
2994         }
2995         return 0; /* silence gcc, even though control never reaches here */
2996 }
2997
2998 /* find an entry with matching function, matching index (if needed), and that
2999  * should be read next (if it's stateful) */
3000 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3001         u32 function, u32 index)
3002 {
3003         if (e->function != function)
3004                 return 0;
3005         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3006                 return 0;
3007         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3008             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3009                 return 0;
3010         return 1;
3011 }
3012
3013 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3014                                               u32 function, u32 index)
3015 {
3016         int i;
3017         struct kvm_cpuid_entry2 *best = NULL;
3018
3019         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3020                 struct kvm_cpuid_entry2 *e;
3021
3022                 e = &vcpu->arch.cpuid_entries[i];
3023                 if (is_matching_cpuid_entry(e, function, index)) {
3024                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3025                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3026                         best = e;
3027                         break;
3028                 }
3029                 /*
3030                  * Both basic or both extended?
3031                  */
3032                 if (((e->function ^ function) & 0x80000000) == 0)
3033                         if (!best || e->function > best->function)
3034                                 best = e;
3035         }
3036         return best;
3037 }
3038
3039 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3040 {
3041         struct kvm_cpuid_entry2 *best;
3042
3043         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3044         if (best)
3045                 return best->eax & 0xff;
3046         return 36;
3047 }
3048
3049 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3050 {
3051         u32 function, index;
3052         struct kvm_cpuid_entry2 *best;
3053
3054         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3055         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3056         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3057         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3058         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3059         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3060         best = kvm_find_cpuid_entry(vcpu, function, index);
3061         if (best) {
3062                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3063                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3064                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3065                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3066         }
3067         kvm_x86_ops->skip_emulated_instruction(vcpu);
3068         KVMTRACE_5D(CPUID, vcpu, function,
3069                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3070                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3071                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3072                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3073 }
3074 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3075
3076 /*
3077  * Check if userspace requested an interrupt window, and that the
3078  * interrupt window is open.
3079  *
3080  * No need to exit to userspace if we already have an interrupt queued.
3081  */
3082 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3083                                           struct kvm_run *kvm_run)
3084 {
3085         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3086                 kvm_run->request_interrupt_window &&
3087                 kvm_arch_interrupt_allowed(vcpu));
3088 }
3089
3090 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3091                               struct kvm_run *kvm_run)
3092 {
3093         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3094         kvm_run->cr8 = kvm_get_cr8(vcpu);
3095         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3096         if (irqchip_in_kernel(vcpu->kvm))
3097                 kvm_run->ready_for_interrupt_injection = 1;
3098         else
3099                 kvm_run->ready_for_interrupt_injection =
3100                                         (kvm_arch_interrupt_allowed(vcpu) &&
3101                                          !kvm_cpu_has_interrupt(vcpu));
3102 }
3103
3104 static void vapic_enter(struct kvm_vcpu *vcpu)
3105 {
3106         struct kvm_lapic *apic = vcpu->arch.apic;
3107         struct page *page;
3108
3109         if (!apic || !apic->vapic_addr)
3110                 return;
3111
3112         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3113
3114         vcpu->arch.apic->vapic_page = page;
3115 }
3116
3117 static void vapic_exit(struct kvm_vcpu *vcpu)
3118 {
3119         struct kvm_lapic *apic = vcpu->arch.apic;
3120
3121         if (!apic || !apic->vapic_addr)
3122                 return;
3123
3124         down_read(&vcpu->kvm->slots_lock);
3125         kvm_release_page_dirty(apic->vapic_page);
3126         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3127         up_read(&vcpu->kvm->slots_lock);
3128 }
3129
3130 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3131 {
3132         int max_irr, tpr;
3133
3134         if (!kvm_x86_ops->update_cr8_intercept)
3135                 return;
3136
3137         max_irr = kvm_lapic_find_highest_irr(vcpu);
3138
3139         if (max_irr != -1)
3140                 max_irr >>= 4;
3141
3142         tpr = kvm_lapic_get_cr8(vcpu);
3143
3144         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3145 }
3146
3147 static void inject_irq(struct kvm_vcpu *vcpu)
3148 {
3149         /* try to reinject previous events if any */
3150         if (vcpu->arch.nmi_injected) {
3151                 kvm_x86_ops->set_nmi(vcpu);
3152                 return;
3153         }
3154
3155         if (vcpu->arch.interrupt.pending) {
3156                 kvm_x86_ops->set_irq(vcpu, vcpu->arch.interrupt.nr);
3157                 return;
3158         }
3159
3160         /* try to inject new event if pending */
3161         if (vcpu->arch.nmi_pending) {
3162                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3163                         vcpu->arch.nmi_pending = false;
3164                         vcpu->arch.nmi_injected = true;
3165                         kvm_x86_ops->set_nmi(vcpu);
3166                 }
3167         } else if (kvm_cpu_has_interrupt(vcpu)) {
3168                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3169                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3170                         kvm_x86_ops->set_irq(vcpu, vcpu->arch.interrupt.nr);
3171                 }
3172         }
3173 }
3174
3175 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3176 {
3177         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3178                 kvm_run->request_interrupt_window;
3179
3180         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3181                 kvm_x86_ops->drop_interrupt_shadow(vcpu);
3182
3183         inject_irq(vcpu);
3184
3185         /* enable NMI/IRQ window open exits if needed */
3186         if (vcpu->arch.nmi_pending)
3187                 kvm_x86_ops->enable_nmi_window(vcpu);
3188         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3189                 kvm_x86_ops->enable_irq_window(vcpu);
3190 }
3191
3192 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3193 {
3194         int r;
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         if (vcpu->requests || need_resched() || signal_pending(current)) {
3234                 local_irq_enable();
3235                 preempt_enable();
3236                 r = 1;
3237                 goto out;
3238         }
3239
3240         vcpu->guest_mode = 1;
3241         /*
3242          * Make sure that guest_mode assignment won't happen after
3243          * testing the pending IRQ vector bitmap.
3244          */
3245         smp_wmb();
3246
3247         if (vcpu->arch.exception.pending)
3248                 __queue_exception(vcpu);
3249         else
3250                 inject_pending_irq(vcpu, kvm_run);
3251
3252         if (kvm_lapic_enabled(vcpu)) {
3253                 if (!vcpu->arch.apic->vapic_addr)
3254                         update_cr8_intercept(vcpu);
3255                 else
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         vcpu->guest_mode = 0;
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         if (irqchip_in_kernel(vcpu->kvm))
3582                 memset(sregs->interrupt_bitmap, 0,
3583                        sizeof sregs->interrupt_bitmap);
3584         else
3585                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3586                        sizeof sregs->interrupt_bitmap);
3587
3588         if (vcpu->arch.interrupt.pending)
3589                 set_bit(vcpu->arch.interrupt.nr,
3590                         (unsigned long *)sregs->interrupt_bitmap);
3591
3592         vcpu_put(vcpu);
3593
3594         return 0;
3595 }
3596
3597 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3598                                     struct kvm_mp_state *mp_state)
3599 {
3600         vcpu_load(vcpu);
3601         mp_state->mp_state = vcpu->arch.mp_state;
3602         vcpu_put(vcpu);
3603         return 0;
3604 }
3605
3606 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3607                                     struct kvm_mp_state *mp_state)
3608 {
3609         vcpu_load(vcpu);
3610         vcpu->arch.mp_state = mp_state->mp_state;
3611         vcpu_put(vcpu);
3612         return 0;
3613 }
3614
3615 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3616                         struct kvm_segment *var, int seg)
3617 {
3618         kvm_x86_ops->set_segment(vcpu, var, seg);
3619 }
3620
3621 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3622                                    struct kvm_segment *kvm_desct)
3623 {
3624         kvm_desct->base = seg_desc->base0;
3625         kvm_desct->base |= seg_desc->base1 << 16;
3626         kvm_desct->base |= seg_desc->base2 << 24;
3627         kvm_desct->limit = seg_desc->limit0;
3628         kvm_desct->limit |= seg_desc->limit << 16;
3629         if (seg_desc->g) {
3630                 kvm_desct->limit <<= 12;
3631                 kvm_desct->limit |= 0xfff;
3632         }
3633         kvm_desct->selector = selector;
3634         kvm_desct->type = seg_desc->type;
3635         kvm_desct->present = seg_desc->p;
3636         kvm_desct->dpl = seg_desc->dpl;
3637         kvm_desct->db = seg_desc->d;
3638         kvm_desct->s = seg_desc->s;
3639         kvm_desct->l = seg_desc->l;
3640         kvm_desct->g = seg_desc->g;
3641         kvm_desct->avl = seg_desc->avl;
3642         if (!selector)
3643                 kvm_desct->unusable = 1;
3644         else
3645                 kvm_desct->unusable = 0;
3646         kvm_desct->padding = 0;
3647 }
3648
3649 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3650                                           u16 selector,
3651                                           struct descriptor_table *dtable)
3652 {
3653         if (selector & 1 << 2) {
3654                 struct kvm_segment kvm_seg;
3655
3656                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3657
3658                 if (kvm_seg.unusable)
3659                         dtable->limit = 0;
3660                 else
3661                         dtable->limit = kvm_seg.limit;
3662                 dtable->base = kvm_seg.base;
3663         }
3664         else
3665                 kvm_x86_ops->get_gdt(vcpu, dtable);
3666 }
3667
3668 /* allowed just for 8 bytes segments */
3669 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3670                                          struct desc_struct *seg_desc)
3671 {
3672         gpa_t gpa;
3673         struct descriptor_table dtable;
3674         u16 index = selector >> 3;
3675
3676         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3677
3678         if (dtable.limit < index * 8 + 7) {
3679                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3680                 return 1;
3681         }
3682         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3683         gpa += index * 8;
3684         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3685 }
3686
3687 /* allowed just for 8 bytes segments */
3688 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3689                                          struct desc_struct *seg_desc)
3690 {
3691         gpa_t gpa;
3692         struct descriptor_table dtable;
3693         u16 index = selector >> 3;
3694
3695         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3696
3697         if (dtable.limit < index * 8 + 7)
3698                 return 1;
3699         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3700         gpa += index * 8;
3701         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3702 }
3703
3704 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3705                              struct desc_struct *seg_desc)
3706 {
3707         u32 base_addr;
3708
3709         base_addr = seg_desc->base0;
3710         base_addr |= (seg_desc->base1 << 16);
3711         base_addr |= (seg_desc->base2 << 24);
3712
3713         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3714 }
3715
3716 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3717 {
3718         struct kvm_segment kvm_seg;
3719
3720         kvm_get_segment(vcpu, &kvm_seg, seg);
3721         return kvm_seg.selector;
3722 }
3723
3724 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3725                                                 u16 selector,
3726                                                 struct kvm_segment *kvm_seg)
3727 {
3728         struct desc_struct seg_desc;
3729
3730         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3731                 return 1;
3732         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3733         return 0;
3734 }
3735
3736 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3737 {
3738         struct kvm_segment segvar = {
3739                 .base = selector << 4,
3740                 .limit = 0xffff,
3741                 .selector = selector,
3742                 .type = 3,
3743                 .present = 1,
3744                 .dpl = 3,
3745                 .db = 0,
3746                 .s = 1,
3747                 .l = 0,
3748                 .g = 0,
3749                 .avl = 0,
3750                 .unusable = 0,
3751         };
3752         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3753         return 0;
3754 }
3755
3756 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3757                                 int type_bits, int seg)
3758 {
3759         struct kvm_segment kvm_seg;
3760
3761         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3762                 return kvm_load_realmode_segment(vcpu, selector, seg);
3763         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3764                 return 1;
3765         kvm_seg.type |= type_bits;
3766
3767         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3768             seg != VCPU_SREG_LDTR)
3769                 if (!kvm_seg.s)
3770                         kvm_seg.unusable = 1;
3771
3772         kvm_set_segment(vcpu, &kvm_seg, seg);
3773         return 0;
3774 }
3775
3776 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3777                                 struct tss_segment_32 *tss)
3778 {
3779         tss->cr3 = vcpu->arch.cr3;
3780         tss->eip = kvm_rip_read(vcpu);
3781         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3782         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3783         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3784         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3785         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3786         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3787         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3788         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3789         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3790         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3791         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3792         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3793         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3794         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3795         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3796         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3797 }
3798
3799 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3800                                   struct tss_segment_32 *tss)
3801 {
3802         kvm_set_cr3(vcpu, tss->cr3);
3803
3804         kvm_rip_write(vcpu, tss->eip);
3805         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3806
3807         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3808         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3809         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3810         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3811         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3812         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3813         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3814         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3815
3816         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3817                 return 1;
3818
3819         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3820                 return 1;
3821
3822         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3823                 return 1;
3824
3825         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3826                 return 1;
3827
3828         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3829                 return 1;
3830
3831         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3832                 return 1;
3833
3834         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3835                 return 1;
3836         return 0;
3837 }
3838
3839 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3840                                 struct tss_segment_16 *tss)
3841 {
3842         tss->ip = kvm_rip_read(vcpu);
3843         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3844         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3845         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3846         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3847         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3848         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3849         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3850         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3851         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3852
3853         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3854         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3855         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3856         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3857         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3858         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3859 }
3860
3861 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3862                                  struct tss_segment_16 *tss)
3863 {
3864         kvm_rip_write(vcpu, tss->ip);
3865         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3866         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3867         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3868         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3869         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3870         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3871         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3872         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3873         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3874
3875         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3876                 return 1;
3877
3878         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3879                 return 1;
3880
3881         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3882                 return 1;
3883
3884         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3885                 return 1;
3886
3887         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3888                 return 1;
3889         return 0;
3890 }
3891
3892 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3893                               u16 old_tss_sel, u32 old_tss_base,
3894                               struct desc_struct *nseg_desc)
3895 {
3896         struct tss_segment_16 tss_segment_16;
3897         int ret = 0;
3898
3899         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3900                            sizeof tss_segment_16))
3901                 goto out;
3902
3903         save_state_to_tss16(vcpu, &tss_segment_16);
3904
3905         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3906                             sizeof tss_segment_16))
3907                 goto out;
3908
3909         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3910                            &tss_segment_16, sizeof tss_segment_16))
3911                 goto out;
3912
3913         if (old_tss_sel != 0xffff) {
3914                 tss_segment_16.prev_task_link = old_tss_sel;
3915
3916                 if (kvm_write_guest(vcpu->kvm,
3917                                     get_tss_base_addr(vcpu, nseg_desc),
3918                                     &tss_segment_16.prev_task_link,
3919                                     sizeof tss_segment_16.prev_task_link))
3920                         goto out;
3921         }
3922
3923         if (load_state_from_tss16(vcpu, &tss_segment_16))
3924                 goto out;
3925
3926         ret = 1;
3927 out:
3928         return ret;
3929 }
3930
3931 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3932                        u16 old_tss_sel, u32 old_tss_base,
3933                        struct desc_struct *nseg_desc)
3934 {
3935         struct tss_segment_32 tss_segment_32;
3936         int ret = 0;
3937
3938         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3939                            sizeof tss_segment_32))
3940                 goto out;
3941
3942         save_state_to_tss32(vcpu, &tss_segment_32);
3943
3944         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3945                             sizeof tss_segment_32))
3946                 goto out;
3947
3948         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3949                            &tss_segment_32, sizeof tss_segment_32))
3950                 goto out;
3951
3952         if (old_tss_sel != 0xffff) {
3953                 tss_segment_32.prev_task_link = old_tss_sel;
3954
3955                 if (kvm_write_guest(vcpu->kvm,
3956                                     get_tss_base_addr(vcpu, nseg_desc),
3957                                     &tss_segment_32.prev_task_link,
3958                                     sizeof tss_segment_32.prev_task_link))
3959                         goto out;
3960         }
3961
3962         if (load_state_from_tss32(vcpu, &tss_segment_32))
3963                 goto out;
3964
3965         ret = 1;
3966 out:
3967         return ret;
3968 }
3969
3970 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3971 {
3972         struct kvm_segment tr_seg;
3973         struct desc_struct cseg_desc;
3974         struct desc_struct nseg_desc;
3975         int ret = 0;
3976         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3977         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3978
3979         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3980
3981         /* FIXME: Handle errors. Failure to read either TSS or their
3982          * descriptors should generate a pagefault.
3983          */
3984         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3985                 goto out;
3986
3987         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3988                 goto out;
3989
3990         if (reason != TASK_SWITCH_IRET) {
3991                 int cpl;
3992
3993                 cpl = kvm_x86_ops->get_cpl(vcpu);
3994                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3995                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3996                         return 1;
3997                 }
3998         }
3999
4000         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4001                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4002                 return 1;
4003         }
4004
4005         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4006                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4007                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4008         }
4009
4010         if (reason == TASK_SWITCH_IRET) {
4011                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4012                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4013         }
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         /* set back link to prev task only if NT bit is set in eflags
4021            note that old_tss_sel is not used afetr this point */
4022         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4023                 old_tss_sel = 0xffff;
4024
4025         if (nseg_desc.type & 8)
4026                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4027                                          old_tss_base, &nseg_desc);
4028         else
4029                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4030                                          old_tss_base, &nseg_desc);
4031
4032         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4033                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4034                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4035         }
4036
4037         if (reason != TASK_SWITCH_IRET) {
4038                 nseg_desc.type |= (1 << 1);
4039                 save_guest_segment_descriptor(vcpu, tss_selector,
4040                                               &nseg_desc);
4041         }
4042
4043         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4044         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4045         tr_seg.type = 11;
4046         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4047 out:
4048         return ret;
4049 }
4050 EXPORT_SYMBOL_GPL(kvm_task_switch);
4051
4052 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4053                                   struct kvm_sregs *sregs)
4054 {
4055         int mmu_reset_needed = 0;
4056         int i, pending_vec, max_bits;
4057         struct descriptor_table dt;
4058
4059         vcpu_load(vcpu);
4060
4061         dt.limit = sregs->idt.limit;
4062         dt.base = sregs->idt.base;
4063         kvm_x86_ops->set_idt(vcpu, &dt);
4064         dt.limit = sregs->gdt.limit;
4065         dt.base = sregs->gdt.base;
4066         kvm_x86_ops->set_gdt(vcpu, &dt);
4067
4068         vcpu->arch.cr2 = sregs->cr2;
4069         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4070
4071         down_read(&vcpu->kvm->slots_lock);
4072         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4073                 vcpu->arch.cr3 = sregs->cr3;
4074         else
4075                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4076         up_read(&vcpu->kvm->slots_lock);
4077
4078         kvm_set_cr8(vcpu, sregs->cr8);
4079
4080         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4081         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4082         kvm_set_apic_base(vcpu, sregs->apic_base);
4083
4084         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4085
4086         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4087         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4088         vcpu->arch.cr0 = sregs->cr0;
4089
4090         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4091         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4092         if (!is_long_mode(vcpu) && is_pae(vcpu))
4093                 load_pdptrs(vcpu, vcpu->arch.cr3);
4094
4095         if (mmu_reset_needed)
4096                 kvm_mmu_reset_context(vcpu);
4097
4098         if (!irqchip_in_kernel(vcpu->kvm)) {
4099                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
4100                        sizeof vcpu->arch.irq_pending);
4101                 vcpu->arch.irq_summary = 0;
4102                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
4103                         if (vcpu->arch.irq_pending[i])
4104                                 __set_bit(i, &vcpu->arch.irq_summary);
4105         } else {
4106                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4107                 pending_vec = find_first_bit(
4108                         (const unsigned long *)sregs->interrupt_bitmap,
4109                         max_bits);
4110                 /* Only pending external irq is handled here */
4111                 if (pending_vec < max_bits) {
4112                         kvm_queue_interrupt(vcpu, pending_vec);
4113                         pr_debug("Set back pending irq %d\n", pending_vec);
4114                 }
4115                 kvm_pic_clear_isr_ack(vcpu->kvm);
4116         }
4117
4118         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4119         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4120         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4121         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4122         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4123         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4124
4125         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4126         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4127
4128         /* Older userspace won't unhalt the vcpu on reset. */
4129         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4130             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4131             !(vcpu->arch.cr0 & X86_CR0_PE))
4132                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4133
4134         vcpu_put(vcpu);
4135
4136         return 0;
4137 }
4138
4139 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4140                                         struct kvm_guest_debug *dbg)
4141 {
4142         int i, r;
4143
4144         vcpu_load(vcpu);
4145
4146         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4147             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4148                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4149                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4150                 vcpu->arch.switch_db_regs =
4151                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4152         } else {
4153                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4154                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4155                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4156         }
4157
4158         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4159
4160         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4161                 kvm_queue_exception(vcpu, DB_VECTOR);
4162         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4163                 kvm_queue_exception(vcpu, BP_VECTOR);
4164
4165         vcpu_put(vcpu);
4166
4167         return r;
4168 }
4169
4170 /*
4171  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4172  * we have asm/x86/processor.h
4173  */
4174 struct fxsave {
4175         u16     cwd;
4176         u16     swd;
4177         u16     twd;
4178         u16     fop;
4179         u64     rip;
4180         u64     rdp;
4181         u32     mxcsr;
4182         u32     mxcsr_mask;
4183         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4184 #ifdef CONFIG_X86_64
4185         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4186 #else
4187         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4188 #endif
4189 };
4190
4191 /*
4192  * Translate a guest virtual address to a guest physical address.
4193  */
4194 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4195                                     struct kvm_translation *tr)
4196 {
4197         unsigned long vaddr = tr->linear_address;
4198         gpa_t gpa;
4199
4200         vcpu_load(vcpu);
4201         down_read(&vcpu->kvm->slots_lock);
4202         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4203         up_read(&vcpu->kvm->slots_lock);
4204         tr->physical_address = gpa;
4205         tr->valid = gpa != UNMAPPED_GVA;
4206         tr->writeable = 1;
4207         tr->usermode = 0;
4208         vcpu_put(vcpu);
4209
4210         return 0;
4211 }
4212
4213 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4214 {
4215         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4216
4217         vcpu_load(vcpu);
4218
4219         memcpy(fpu->fpr, fxsave->st_space, 128);
4220         fpu->fcw = fxsave->cwd;
4221         fpu->fsw = fxsave->swd;
4222         fpu->ftwx = fxsave->twd;
4223         fpu->last_opcode = fxsave->fop;
4224         fpu->last_ip = fxsave->rip;
4225         fpu->last_dp = fxsave->rdp;
4226         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4227
4228         vcpu_put(vcpu);
4229
4230         return 0;
4231 }
4232
4233 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4234 {
4235         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4236
4237         vcpu_load(vcpu);
4238
4239         memcpy(fxsave->st_space, fpu->fpr, 128);
4240         fxsave->cwd = fpu->fcw;
4241         fxsave->swd = fpu->fsw;
4242         fxsave->twd = fpu->ftwx;
4243         fxsave->fop = fpu->last_opcode;
4244         fxsave->rip = fpu->last_ip;
4245         fxsave->rdp = fpu->last_dp;
4246         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4247
4248         vcpu_put(vcpu);
4249
4250         return 0;
4251 }
4252
4253 void fx_init(struct kvm_vcpu *vcpu)
4254 {
4255         unsigned after_mxcsr_mask;
4256
4257         /*
4258          * Touch the fpu the first time in non atomic context as if
4259          * this is the first fpu instruction the exception handler
4260          * will fire before the instruction returns and it'll have to
4261          * allocate ram with GFP_KERNEL.
4262          */
4263         if (!used_math())
4264                 kvm_fx_save(&vcpu->arch.host_fx_image);
4265
4266         /* Initialize guest FPU by resetting ours and saving into guest's */
4267         preempt_disable();
4268         kvm_fx_save(&vcpu->arch.host_fx_image);
4269         kvm_fx_finit();
4270         kvm_fx_save(&vcpu->arch.guest_fx_image);
4271         kvm_fx_restore(&vcpu->arch.host_fx_image);
4272         preempt_enable();
4273
4274         vcpu->arch.cr0 |= X86_CR0_ET;
4275         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4276         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4277         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4278                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4279 }
4280 EXPORT_SYMBOL_GPL(fx_init);
4281
4282 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4283 {
4284         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4285                 return;
4286
4287         vcpu->guest_fpu_loaded = 1;
4288         kvm_fx_save(&vcpu->arch.host_fx_image);
4289         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4290 }
4291 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4292
4293 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4294 {
4295         if (!vcpu->guest_fpu_loaded)
4296                 return;
4297
4298         vcpu->guest_fpu_loaded = 0;
4299         kvm_fx_save(&vcpu->arch.guest_fx_image);
4300         kvm_fx_restore(&vcpu->arch.host_fx_image);
4301         ++vcpu->stat.fpu_reload;
4302 }
4303 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4304
4305 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4306 {
4307         if (vcpu->arch.time_page) {
4308                 kvm_release_page_dirty(vcpu->arch.time_page);
4309                 vcpu->arch.time_page = NULL;
4310         }
4311
4312         kvm_x86_ops->vcpu_free(vcpu);
4313 }
4314
4315 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4316                                                 unsigned int id)
4317 {
4318         return kvm_x86_ops->vcpu_create(kvm, id);
4319 }
4320
4321 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4322 {
4323         int r;
4324
4325         /* We do fxsave: this must be aligned. */
4326         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4327
4328         vcpu->arch.mtrr_state.have_fixed = 1;
4329         vcpu_load(vcpu);
4330         r = kvm_arch_vcpu_reset(vcpu);
4331         if (r == 0)
4332                 r = kvm_mmu_setup(vcpu);
4333         vcpu_put(vcpu);
4334         if (r < 0)
4335                 goto free_vcpu;
4336
4337         return 0;
4338 free_vcpu:
4339         kvm_x86_ops->vcpu_free(vcpu);
4340         return r;
4341 }
4342
4343 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4344 {
4345         vcpu_load(vcpu);
4346         kvm_mmu_unload(vcpu);
4347         vcpu_put(vcpu);
4348
4349         kvm_x86_ops->vcpu_free(vcpu);
4350 }
4351
4352 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4353 {
4354         vcpu->arch.nmi_pending = false;
4355         vcpu->arch.nmi_injected = false;
4356
4357         vcpu->arch.switch_db_regs = 0;
4358         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4359         vcpu->arch.dr6 = DR6_FIXED_1;
4360         vcpu->arch.dr7 = DR7_FIXED_1;
4361
4362         return kvm_x86_ops->vcpu_reset(vcpu);
4363 }
4364
4365 void kvm_arch_hardware_enable(void *garbage)
4366 {
4367         kvm_x86_ops->hardware_enable(garbage);
4368 }
4369
4370 void kvm_arch_hardware_disable(void *garbage)
4371 {
4372         kvm_x86_ops->hardware_disable(garbage);
4373 }
4374
4375 int kvm_arch_hardware_setup(void)
4376 {
4377         return kvm_x86_ops->hardware_setup();
4378 }
4379
4380 void kvm_arch_hardware_unsetup(void)
4381 {
4382         kvm_x86_ops->hardware_unsetup();
4383 }
4384
4385 void kvm_arch_check_processor_compat(void *rtn)
4386 {
4387         kvm_x86_ops->check_processor_compatibility(rtn);
4388 }
4389
4390 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4391 {
4392         struct page *page;
4393         struct kvm *kvm;
4394         int r;
4395
4396         BUG_ON(vcpu->kvm == NULL);
4397         kvm = vcpu->kvm;
4398
4399         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4400         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4401                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4402         else
4403                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4404
4405         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4406         if (!page) {
4407                 r = -ENOMEM;
4408                 goto fail;
4409         }
4410         vcpu->arch.pio_data = page_address(page);
4411
4412         r = kvm_mmu_create(vcpu);
4413         if (r < 0)
4414                 goto fail_free_pio_data;
4415
4416         if (irqchip_in_kernel(kvm)) {
4417                 r = kvm_create_lapic(vcpu);
4418                 if (r < 0)
4419                         goto fail_mmu_destroy;
4420         }
4421
4422         return 0;
4423
4424 fail_mmu_destroy:
4425         kvm_mmu_destroy(vcpu);
4426 fail_free_pio_data:
4427         free_page((unsigned long)vcpu->arch.pio_data);
4428 fail:
4429         return r;
4430 }
4431
4432 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4433 {
4434         kvm_free_lapic(vcpu);
4435         down_read(&vcpu->kvm->slots_lock);
4436         kvm_mmu_destroy(vcpu);
4437         up_read(&vcpu->kvm->slots_lock);
4438         free_page((unsigned long)vcpu->arch.pio_data);
4439 }
4440
4441 struct  kvm *kvm_arch_create_vm(void)
4442 {
4443         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4444
4445         if (!kvm)
4446                 return ERR_PTR(-ENOMEM);
4447
4448         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4449         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4450
4451         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4452         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4453
4454         rdtscll(kvm->arch.vm_init_tsc);
4455
4456         return kvm;
4457 }
4458
4459 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4460 {
4461         vcpu_load(vcpu);
4462         kvm_mmu_unload(vcpu);
4463         vcpu_put(vcpu);
4464 }
4465
4466 static void kvm_free_vcpus(struct kvm *kvm)
4467 {
4468         unsigned int i;
4469
4470         /*
4471          * Unpin any mmu pages first.
4472          */
4473         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4474                 if (kvm->vcpus[i])
4475                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4476         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4477                 if (kvm->vcpus[i]) {
4478                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4479                         kvm->vcpus[i] = NULL;
4480                 }
4481         }
4482
4483 }
4484
4485 void kvm_arch_sync_events(struct kvm *kvm)
4486 {
4487         kvm_free_all_assigned_devices(kvm);
4488 }
4489
4490 void kvm_arch_destroy_vm(struct kvm *kvm)
4491 {
4492         kvm_iommu_unmap_guest(kvm);
4493         kvm_free_pit(kvm);
4494         kfree(kvm->arch.vpic);
4495         kfree(kvm->arch.vioapic);
4496         kvm_free_vcpus(kvm);
4497         kvm_free_physmem(kvm);
4498         if (kvm->arch.apic_access_page)
4499                 put_page(kvm->arch.apic_access_page);
4500         if (kvm->arch.ept_identity_pagetable)
4501                 put_page(kvm->arch.ept_identity_pagetable);
4502         kfree(kvm);
4503 }
4504
4505 int kvm_arch_set_memory_region(struct kvm *kvm,
4506                                 struct kvm_userspace_memory_region *mem,
4507                                 struct kvm_memory_slot old,
4508                                 int user_alloc)
4509 {
4510         int npages = mem->memory_size >> PAGE_SHIFT;
4511         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4512
4513         /*To keep backward compatibility with older userspace,
4514          *x86 needs to hanlde !user_alloc case.
4515          */
4516         if (!user_alloc) {
4517                 if (npages && !old.rmap) {
4518                         unsigned long userspace_addr;
4519
4520                         down_write(&current->mm->mmap_sem);
4521                         userspace_addr = do_mmap(NULL, 0,
4522                                                  npages * PAGE_SIZE,
4523                                                  PROT_READ | PROT_WRITE,
4524                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4525                                                  0);
4526                         up_write(&current->mm->mmap_sem);
4527
4528                         if (IS_ERR((void *)userspace_addr))
4529                                 return PTR_ERR((void *)userspace_addr);
4530
4531                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4532                         spin_lock(&kvm->mmu_lock);
4533                         memslot->userspace_addr = userspace_addr;
4534                         spin_unlock(&kvm->mmu_lock);
4535                 } else {
4536                         if (!old.user_alloc && old.rmap) {
4537                                 int ret;
4538
4539                                 down_write(&current->mm->mmap_sem);
4540                                 ret = do_munmap(current->mm, old.userspace_addr,
4541                                                 old.npages * PAGE_SIZE);
4542                                 up_write(&current->mm->mmap_sem);
4543                                 if (ret < 0)
4544                                         printk(KERN_WARNING
4545                                        "kvm_vm_ioctl_set_memory_region: "
4546                                        "failed to munmap memory\n");
4547                         }
4548                 }
4549         }
4550
4551         if (!kvm->arch.n_requested_mmu_pages) {
4552                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4553                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4554         }
4555
4556         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4557         kvm_flush_remote_tlbs(kvm);
4558
4559         return 0;
4560 }
4561
4562 void kvm_arch_flush_shadow(struct kvm *kvm)
4563 {
4564         kvm_mmu_zap_all(kvm);
4565 }
4566
4567 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4568 {
4569         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4570                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4571                || vcpu->arch.nmi_pending;
4572 }
4573
4574 static void vcpu_kick_intr(void *info)
4575 {
4576 #ifdef DEBUG
4577         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4578         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4579 #endif
4580 }
4581
4582 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4583 {
4584         int ipi_pcpu = vcpu->cpu;
4585         int cpu;
4586
4587         if (waitqueue_active(&vcpu->wq)) {
4588                 wake_up_interruptible(&vcpu->wq);
4589                 ++vcpu->stat.halt_wakeup;
4590         }
4591         /*
4592          * We may be called synchronously with irqs disabled in guest mode,
4593          * So need not to call smp_call_function_single() in that case.
4594          */
4595         cpu = get_cpu();
4596         if (vcpu->guest_mode && vcpu->cpu != cpu)
4597                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4598         put_cpu();
4599 }
4600
4601 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4602 {
4603         return kvm_x86_ops->interrupt_allowed(vcpu);
4604 }