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 bool valid_pat_type(unsigned t)
 708 {
 709         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 710 }
 711
 712 static bool valid_mtrr_type(unsigned t)
 713 {
 714         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 715 }
 716
 717 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 718 {
 719         int i;
 720
 721         if (!msr_mtrr_valid(msr))
 722                 return false;
 723
 724         if (msr == MSR_IA32_CR_PAT) {
 725                 for (i = 0; i < 8; i++)
 726                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 727                                 return false;
 728                 return true;
 729         } else if (msr == MSR_MTRRdefType) {
 730                 if (data & ~0xcff)
 731                         return false;
 732                 return valid_mtrr_type(data & 0xff);
 733         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 734                 for (i = 0; i < 8 ; i++)
 735                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 736                                 return false;
 737                 return true;
 738         }
 739
 740         /* variable MTRRs */
 741         return valid_mtrr_type(data & 0xff);
 742 }
 743
 744 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 745 {
 746         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 747
 748         if (!mtrr_valid(vcpu, msr, data))
 749                 return 1;
 750
 751         if (msr == MSR_MTRRdefType) {
 752                 vcpu->arch.mtrr_state.def_type = data;
 753                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 754         } else if (msr == MSR_MTRRfix64K_00000)
 755                 p[0] = data;
 756         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 757                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 758         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 759                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 760         else if (msr == MSR_IA32_CR_PAT)
 761                 vcpu->arch.pat = data;
 762         else {  /* Variable MTRRs */
 763                 int idx, is_mtrr_mask;
 764                 u64 *pt;
 765
 766                 idx = (msr - 0x200) / 2;
 767                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 768                 if (!is_mtrr_mask)
 769                         pt =
 770                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 771                 else
 772                         pt =
 773                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 774                 *pt = data;
 775         }
 776
 777         kvm_mmu_reset_context(vcpu);
 778         return 0;
 779 }
 780
 781 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 782 {
 783         switch (msr) {
 784         case MSR_EFER:
 785                 set_efer(vcpu, data);
 786                 break;
 787         case MSR_IA32_MC0_STATUS:
 788                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 789                        __func__, data);
 790                 break;
 791         case MSR_IA32_MCG_STATUS:
 792                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 793                         __func__, data);
 794                 break;
 795         case MSR_IA32_MCG_CTL:
 796                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 797                         __func__, data);
 798                 break;
 799         case MSR_IA32_DEBUGCTLMSR:
 800                 if (!data) {
 801                         /* We support the non-activated case already */
 802                         break;
 803                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 804                         /* Values other than LBR and BTF are vendor-specific,
 805                            thus reserved and should throw a #GP */
 806                         return 1;
 807                 }
 808                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 809                         __func__, data);
 810                 break;
 811         case MSR_IA32_UCODE_REV:
 812         case MSR_IA32_UCODE_WRITE:
 813         case MSR_VM_HSAVE_PA:
 814                 break;
 815         case 0x200 ... 0x2ff:
 816                 return set_msr_mtrr(vcpu, msr, data);
 817         case MSR_IA32_APICBASE:
 818                 kvm_set_apic_base(vcpu, data);
 819                 break;
 820         case MSR_IA32_MISC_ENABLE:
 821                 vcpu->arch.ia32_misc_enable_msr = data;
 822                 break;
 823         case MSR_KVM_WALL_CLOCK:
 824                 vcpu->kvm->arch.wall_clock = data;
 825                 kvm_write_wall_clock(vcpu->kvm, data);
 826                 break;
 827         case MSR_KVM_SYSTEM_TIME: {
 828                 if (vcpu->arch.time_page) {
 829                         kvm_release_page_dirty(vcpu->arch.time_page);
 830                         vcpu->arch.time_page = NULL;
 831                 }
 832
 833                 vcpu->arch.time = data;
 834
 835                 /* we verify if the enable bit is set... */
 836                 if (!(data & 1))
 837                         break;
 838
 839                 /* ...but clean it before doing the actual write */
 840                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 841
 842                 vcpu->arch.time_page =
 843                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 844
 845                 if (is_error_page(vcpu->arch.time_page)) {
 846                         kvm_release_page_clean(vcpu->arch.time_page);
 847                         vcpu->arch.time_page = NULL;
 848                 }
 849
 850                 kvm_request_guest_time_update(vcpu);
 851                 break;
 852         }
 853         default:
 854                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 855                 return 1;
 856         }
 857         return 0;
 858 }
 859 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 860
 861
 862 /*
 863  * Reads an msr value (of 'msr_index') into 'pdata'.
 864  * Returns 0 on success, non-0 otherwise.
 865  * Assumes vcpu_load() was already called.
 866  */
 867 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 868 {
 869         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 870 }
 871
 872 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 873 {
 874         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 875
 876         if (!msr_mtrr_valid(msr))
 877                 return 1;
 878
 879         if (msr == MSR_MTRRdefType)
 880                 *pdata = vcpu->arch.mtrr_state.def_type +
 881                          (vcpu->arch.mtrr_state.enabled << 10);
 882         else if (msr == MSR_MTRRfix64K_00000)
 883                 *pdata = p[0];
 884         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 885                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 886         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 887                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 888         else if (msr == MSR_IA32_CR_PAT)
 889                 *pdata = vcpu->arch.pat;
 890         else {  /* Variable MTRRs */
 891                 int idx, is_mtrr_mask;
 892                 u64 *pt;
 893
 894                 idx = (msr - 0x200) / 2;
 895                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 896                 if (!is_mtrr_mask)
 897                         pt =
 898                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 899                 else
 900                         pt =
 901                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 902                 *pdata = *pt;
 903         }
 904
 905         return 0;
 906 }
 907
 908 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 909 {
 910         u64 data;
 911
 912         switch (msr) {
 913         case 0xc0010010: /* SYSCFG */
 914         case 0xc0010015: /* HWCR */
 915         case MSR_IA32_PLATFORM_ID:
 916         case MSR_IA32_P5_MC_ADDR:
 917         case MSR_IA32_P5_MC_TYPE:
 918         case MSR_IA32_MC0_CTL:
 919         case MSR_IA32_MCG_STATUS:
 920         case MSR_IA32_MCG_CAP:
 921         case MSR_IA32_MCG_CTL:
 922         case MSR_IA32_MC0_MISC:
 923         case MSR_IA32_MC0_MISC+4:
 924         case MSR_IA32_MC0_MISC+8:
 925         case MSR_IA32_MC0_MISC+12:
 926         case MSR_IA32_MC0_MISC+16:
 927         case MSR_IA32_MC0_MISC+20:
 928         case MSR_IA32_UCODE_REV:
 929         case MSR_IA32_EBL_CR_POWERON:
 930         case MSR_IA32_DEBUGCTLMSR:
 931         case MSR_IA32_LASTBRANCHFROMIP:
 932         case MSR_IA32_LASTBRANCHTOIP:
 933         case MSR_IA32_LASTINTFROMIP:
 934         case MSR_IA32_LASTINTTOIP:
 935         case MSR_VM_HSAVE_PA:
 936         case MSR_P6_EVNTSEL0:
 937         case MSR_P6_EVNTSEL1:
 938         case MSR_K7_EVNTSEL0:
 939                 data = 0;
 940                 break;
 941         case MSR_MTRRcap:
 942                 data = 0x500 | KVM_NR_VAR_MTRR;
 943                 break;
 944         case 0x200 ... 0x2ff:
 945                 return get_msr_mtrr(vcpu, msr, pdata);
 946         case 0xcd: /* fsb frequency */
 947                 data = 3;
 948                 break;
 949         case MSR_IA32_APICBASE:
 950                 data = kvm_get_apic_base(vcpu);
 951                 break;
 952         case MSR_IA32_MISC_ENABLE:
 953                 data = vcpu->arch.ia32_misc_enable_msr;
 954                 break;
 955         case MSR_IA32_PERF_STATUS:
 956                 /* TSC increment by tick */
 957                 data = 1000ULL;
 958                 /* CPU multiplier */
 959                 data |= (((uint64_t)4ULL) << 40);
 960                 break;
 961         case MSR_EFER:
 962                 data = vcpu->arch.shadow_efer;
 963                 break;
 964         case MSR_KVM_WALL_CLOCK:
 965                 data = vcpu->kvm->arch.wall_clock;
 966                 break;
 967         case MSR_KVM_SYSTEM_TIME:
 968                 data = vcpu->arch.time;
 969                 break;
 970         default:
 971                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 972                 return 1;
 973         }
 974         *pdata = data;
 975         return 0;
 976 }
 977 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 978
 979 /*
 980  * Read or write a bunch of msrs. All parameters are kernel addresses.
 981  *
 982  * @return number of msrs set successfully.
 983  */
 984 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 985                     struct kvm_msr_entry *entries,
 986                     int (*do_msr)(struct kvm_vcpu *vcpu,
 987                                   unsigned index, u64 *data))
 988 {
 989         int i;
 990
 991         vcpu_load(vcpu);
 992
 993         down_read(&vcpu->kvm->slots_lock);
 994         for (i = 0; i < msrs->nmsrs; ++i)
 995                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 996                         break;
 997         up_read(&vcpu->kvm->slots_lock);
 998
 999         vcpu_put(vcpu);
1000
1001         return i;
1002 }
1003
1004 /*
1005  * Read or write a bunch of msrs. Parameters are user addresses.
1006  *
1007  * @return number of msrs set successfully.
1008  */
1009 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1010                   int (*do_msr)(struct kvm_vcpu *vcpu,
1011                                 unsigned index, u64 *data),
1012                   int writeback)
1013 {
1014         struct kvm_msrs msrs;
1015         struct kvm_msr_entry *entries;
1016         int r, n;
1017         unsigned size;
1018
1019         r = -EFAULT;
1020         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1021                 goto out;
1022
1023         r = -E2BIG;
1024         if (msrs.nmsrs >= MAX_IO_MSRS)
1025                 goto out;
1026
1027         r = -ENOMEM;
1028         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1029         entries = vmalloc(size);
1030         if (!entries)
1031                 goto out;
1032
1033         r = -EFAULT;
1034         if (copy_from_user(entries, user_msrs->entries, size))
1035                 goto out_free;
1036
1037         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1038         if (r < 0)
1039                 goto out_free;
1040
1041         r = -EFAULT;
1042         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1043                 goto out_free;
1044
1045         r = n;
1046
1047 out_free:
1048         vfree(entries);
1049 out:
1050         return r;
1051 }
1052
1053 int kvm_dev_ioctl_check_extension(long ext)
1054 {
1055         int r;
1056
1057         switch (ext) {
1058         case KVM_CAP_IRQCHIP:
1059         case KVM_CAP_HLT:
1060         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1061         case KVM_CAP_SET_TSS_ADDR:
1062         case KVM_CAP_EXT_CPUID:
1063         case KVM_CAP_CLOCKSOURCE:
1064         case KVM_CAP_PIT:
1065         case KVM_CAP_NOP_IO_DELAY:
1066         case KVM_CAP_MP_STATE:
1067         case KVM_CAP_SYNC_MMU:
1068         case KVM_CAP_REINJECT_CONTROL:
1069         case KVM_CAP_IRQ_INJECT_STATUS:
1070         case KVM_CAP_ASSIGN_DEV_IRQ:
1071                 r = 1;
1072                 break;
1073         case KVM_CAP_COALESCED_MMIO:
1074                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1075                 break;
1076         case KVM_CAP_VAPIC:
1077                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1078                 break;
1079         case KVM_CAP_NR_VCPUS:
1080                 r = KVM_MAX_VCPUS;
1081                 break;
1082         case KVM_CAP_NR_MEMSLOTS:
1083                 r = KVM_MEMORY_SLOTS;
1084                 break;
1085         case KVM_CAP_PV_MMU:
1086                 r = !tdp_enabled;
1087                 break;
1088         case KVM_CAP_IOMMU:
1089                 r = iommu_found();
1090                 break;
1091         default:
1092                 r = 0;
1093                 break;
1094         }
1095         return r;
1096
1097 }
1098
1099 long kvm_arch_dev_ioctl(struct file *filp,
1100                         unsigned int ioctl, unsigned long arg)
1101 {
1102         void __user *argp = (void __user *)arg;
1103         long r;
1104
1105         switch (ioctl) {
1106         case KVM_GET_MSR_INDEX_LIST: {
1107                 struct kvm_msr_list __user *user_msr_list = argp;
1108                 struct kvm_msr_list msr_list;
1109                 unsigned n;
1110
1111                 r = -EFAULT;
1112                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1113                         goto out;
1114                 n = msr_list.nmsrs;
1115                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1116                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1117                         goto out;
1118                 r = -E2BIG;
1119                 if (n < msr_list.nmsrs)
1120                         goto out;
1121                 r = -EFAULT;
1122                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1123                                  num_msrs_to_save * sizeof(u32)))
1124                         goto out;
1125                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1126                                  &emulated_msrs,
1127                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1128                         goto out;
1129                 r = 0;
1130                 break;
1131         }
1132         case KVM_GET_SUPPORTED_CPUID: {
1133                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1134                 struct kvm_cpuid2 cpuid;
1135
1136                 r = -EFAULT;
1137                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1138                         goto out;
1139                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1140                                                       cpuid_arg->entries);
1141                 if (r)
1142                         goto out;
1143
1144                 r = -EFAULT;
1145                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1146                         goto out;
1147                 r = 0;
1148                 break;
1149         }
1150         default:
1151                 r = -EINVAL;
1152         }
1153 out:
1154         return r;
1155 }
1156
1157 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1158 {
1159         kvm_x86_ops->vcpu_load(vcpu, cpu);
1160         kvm_request_guest_time_update(vcpu);
1161 }
1162
1163 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1164 {
1165         kvm_x86_ops->vcpu_put(vcpu);
1166         kvm_put_guest_fpu(vcpu);
1167 }
1168
1169 static int is_efer_nx(void)
1170 {
1171         unsigned long long efer = 0;
1172
1173         rdmsrl_safe(MSR_EFER, &efer);
1174         return efer & EFER_NX;
1175 }
1176
1177 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1178 {
1179         int i;
1180         struct kvm_cpuid_entry2 *e, *entry;
1181
1182         entry = NULL;
1183         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1184                 e = &vcpu->arch.cpuid_entries[i];
1185                 if (e->function == 0x80000001) {
1186                         entry = e;
1187                         break;
1188                 }
1189         }
1190         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1191                 entry->edx &= ~(1 << 20);
1192                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1193         }
1194 }
1195
1196 /* when an old userspace process fills a new kernel module */
1197 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1198                                     struct kvm_cpuid *cpuid,
1199                                     struct kvm_cpuid_entry __user *entries)
1200 {
1201         int r, i;
1202         struct kvm_cpuid_entry *cpuid_entries;
1203
1204         r = -E2BIG;
1205         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1206                 goto out;
1207         r = -ENOMEM;
1208         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1209         if (!cpuid_entries)
1210                 goto out;
1211         r = -EFAULT;
1212         if (copy_from_user(cpuid_entries, entries,
1213                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1214                 goto out_free;
1215         for (i = 0; i < cpuid->nent; i++) {
1216                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1217                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1218                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1219                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1220                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1221                 vcpu->arch.cpuid_entries[i].index = 0;
1222                 vcpu->arch.cpuid_entries[i].flags = 0;
1223                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1224                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1225                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1226         }
1227         vcpu->arch.cpuid_nent = cpuid->nent;
1228         cpuid_fix_nx_cap(vcpu);
1229         r = 0;
1230
1231 out_free:
1232         vfree(cpuid_entries);
1233 out:
1234         return r;
1235 }
1236
1237 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1238                                      struct kvm_cpuid2 *cpuid,
1239                                      struct kvm_cpuid_entry2 __user *entries)
1240 {
1241         int r;
1242
1243         r = -E2BIG;
1244         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1245                 goto out;
1246         r = -EFAULT;
1247         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1248                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1249                 goto out;
1250         vcpu->arch.cpuid_nent = cpuid->nent;
1251         return 0;
1252
1253 out:
1254         return r;
1255 }
1256
1257 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1258                                      struct kvm_cpuid2 *cpuid,
1259                                      struct kvm_cpuid_entry2 __user *entries)
1260 {
1261         int r;
1262
1263         r = -E2BIG;
1264         if (cpuid->nent < vcpu->arch.cpuid_nent)
1265                 goto out;
1266         r = -EFAULT;
1267         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1268                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1269                 goto out;
1270         return 0;
1271
1272 out:
1273         cpuid->nent = vcpu->arch.cpuid_nent;
1274         return r;
1275 }
1276
1277 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1278                            u32 index)
1279 {
1280         entry->function = function;
1281         entry->index = index;
1282         cpuid_count(entry->function, entry->index,
1283                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1284         entry->flags = 0;
1285 }
1286
1287 #define F(x) bit(X86_FEATURE_##x)
1288
1289 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1290                          u32 index, int *nent, int maxnent)
1291 {
1292         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1293 #ifdef CONFIG_X86_64
1294         unsigned f_lm = F(LM);
1295 #else
1296         unsigned f_lm = 0;
1297 #endif
1298
1299         /* cpuid 1.edx */
1300         const u32 kvm_supported_word0_x86_features =
1301                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1302                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1303                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1304                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1305                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1306                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1307                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1308                 0 /* HTT, TM, Reserved, PBE */;
1309         /* cpuid 0x80000001.edx */
1310         const u32 kvm_supported_word1_x86_features =
1311                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1312                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1313                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1314                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1315                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1316                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1317                 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1318                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1319         /* cpuid 1.ecx */
1320         const u32 kvm_supported_word4_x86_features =
1321                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1322                 0 /* DS-CPL, VMX, SMX, EST */ |
1323                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1324                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1325                 0 /* Reserved, DCA */ | F(XMM4_1) |
1326                 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1327                 0 /* Reserved, XSAVE, OSXSAVE */;
1328         /* cpuid 0x80000001.ecx */
1329         const u32 kvm_supported_word6_x86_features =
1330                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1331                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1332                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1333                 0 /* SKINIT */ | 0 /* WDT */;
1334
1335         /* all calls to cpuid_count() should be made on the same cpu */
1336         get_cpu();
1337         do_cpuid_1_ent(entry, function, index);
1338         ++*nent;
1339
1340         switch (function) {
1341         case 0:
1342                 entry->eax = min(entry->eax, (u32)0xb);
1343                 break;
1344         case 1:
1345                 entry->edx &= kvm_supported_word0_x86_features;
1346                 entry->ecx &= kvm_supported_word4_x86_features;
1347                 break;
1348         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1349          * may return different values. This forces us to get_cpu() before
1350          * issuing the first command, and also to emulate this annoying behavior
1351          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1352         case 2: {
1353                 int t, times = entry->eax & 0xff;
1354
1355                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1356                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1357                 for (t = 1; t < times && *nent < maxnent; ++t) {
1358                         do_cpuid_1_ent(&entry[t], function, 0);
1359                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1360                         ++*nent;
1361                 }
1362                 break;
1363         }
1364         /* function 4 and 0xb have additional index. */
1365         case 4: {
1366                 int i, cache_type;
1367
1368                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1369                 /* read more entries until cache_type is zero */
1370                 for (i = 1; *nent < maxnent; ++i) {
1371                         cache_type = entry[i - 1].eax & 0x1f;
1372                         if (!cache_type)
1373                                 break;
1374                         do_cpuid_1_ent(&entry[i], function, i);
1375                         entry[i].flags |=
1376                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1377                         ++*nent;
1378                 }
1379                 break;
1380         }
1381         case 0xb: {
1382                 int i, level_type;
1383
1384                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1385                 /* read more entries until level_type is zero */
1386                 for (i = 1; *nent < maxnent; ++i) {
1387                         level_type = entry[i - 1].ecx & 0xff00;
1388                         if (!level_type)
1389                                 break;
1390                         do_cpuid_1_ent(&entry[i], function, i);
1391                         entry[i].flags |=
1392                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1393                         ++*nent;
1394                 }
1395                 break;
1396         }
1397         case 0x80000000:
1398                 entry->eax = min(entry->eax, 0x8000001a);
1399                 break;
1400         case 0x80000001:
1401                 entry->edx &= kvm_supported_word1_x86_features;
1402                 entry->ecx &= kvm_supported_word6_x86_features;
1403                 break;
1404         }
1405         put_cpu();
1406 }
1407
1408 #undef F
1409
1410 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1411                                      struct kvm_cpuid_entry2 __user *entries)
1412 {
1413         struct kvm_cpuid_entry2 *cpuid_entries;
1414         int limit, nent = 0, r = -E2BIG;
1415         u32 func;
1416
1417         if (cpuid->nent < 1)
1418                 goto out;
1419         r = -ENOMEM;
1420         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1421         if (!cpuid_entries)
1422                 goto out;
1423
1424         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1425         limit = cpuid_entries[0].eax;
1426         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1427                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1428                              &nent, cpuid->nent);
1429         r = -E2BIG;
1430         if (nent >= cpuid->nent)
1431                 goto out_free;
1432
1433         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1434         limit = cpuid_entries[nent - 1].eax;
1435         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1436                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1437                              &nent, cpuid->nent);
1438         r = -EFAULT;
1439         if (copy_to_user(entries, cpuid_entries,
1440                          nent * sizeof(struct kvm_cpuid_entry2)))
1441                 goto out_free;
1442         cpuid->nent = nent;
1443         r = 0;
1444
1445 out_free:
1446         vfree(cpuid_entries);
1447 out:
1448         return r;
1449 }
1450
1451 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1452                                     struct kvm_lapic_state *s)
1453 {
1454         vcpu_load(vcpu);
1455         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1456         vcpu_put(vcpu);
1457
1458         return 0;
1459 }
1460
1461 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1462                                     struct kvm_lapic_state *s)
1463 {
1464         vcpu_load(vcpu);
1465         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1466         kvm_apic_post_state_restore(vcpu);
1467         vcpu_put(vcpu);
1468
1469         return 0;
1470 }
1471
1472 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1473                                     struct kvm_interrupt *irq)
1474 {
1475         if (irq->irq < 0 || irq->irq >= 256)
1476                 return -EINVAL;
1477         if (irqchip_in_kernel(vcpu->kvm))
1478                 return -ENXIO;
1479         vcpu_load(vcpu);
1480
1481         kvm_queue_interrupt(vcpu, irq->irq, false);
1482
1483         vcpu_put(vcpu);
1484
1485         return 0;
1486 }
1487
1488 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1489 {
1490         vcpu_load(vcpu);
1491         kvm_inject_nmi(vcpu);
1492         vcpu_put(vcpu);
1493
1494         return 0;
1495 }
1496
1497 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1498                                            struct kvm_tpr_access_ctl *tac)
1499 {
1500         if (tac->flags)
1501                 return -EINVAL;
1502         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1503         return 0;
1504 }
1505
1506 long kvm_arch_vcpu_ioctl(struct file *filp,
1507                          unsigned int ioctl, unsigned long arg)
1508 {
1509         struct kvm_vcpu *vcpu = filp->private_data;
1510         void __user *argp = (void __user *)arg;
1511         int r;
1512         struct kvm_lapic_state *lapic = NULL;
1513
1514         switch (ioctl) {
1515         case KVM_GET_LAPIC: {
1516                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1517
1518                 r = -ENOMEM;
1519                 if (!lapic)
1520                         goto out;
1521                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1522                 if (r)
1523                         goto out;
1524                 r = -EFAULT;
1525                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1526                         goto out;
1527                 r = 0;
1528                 break;
1529         }
1530         case KVM_SET_LAPIC: {
1531                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1532                 r = -ENOMEM;
1533                 if (!lapic)
1534                         goto out;
1535                 r = -EFAULT;
1536                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1537                         goto out;
1538                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1539                 if (r)
1540                         goto out;
1541                 r = 0;
1542                 break;
1543         }
1544         case KVM_INTERRUPT: {
1545                 struct kvm_interrupt irq;
1546
1547                 r = -EFAULT;
1548                 if (copy_from_user(&irq, argp, sizeof irq))
1549                         goto out;
1550                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1551                 if (r)
1552                         goto out;
1553                 r = 0;
1554                 break;
1555         }
1556         case KVM_NMI: {
1557                 r = kvm_vcpu_ioctl_nmi(vcpu);
1558                 if (r)
1559                         goto out;
1560                 r = 0;
1561                 break;
1562         }
1563         case KVM_SET_CPUID: {
1564                 struct kvm_cpuid __user *cpuid_arg = argp;
1565                 struct kvm_cpuid cpuid;
1566
1567                 r = -EFAULT;
1568                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1569                         goto out;
1570                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1571                 if (r)
1572                         goto out;
1573                 break;
1574         }
1575         case KVM_SET_CPUID2: {
1576                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1577                 struct kvm_cpuid2 cpuid;
1578
1579                 r = -EFAULT;
1580                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1581                         goto out;
1582                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1583                                               cpuid_arg->entries);
1584                 if (r)
1585                         goto out;
1586                 break;
1587         }
1588         case KVM_GET_CPUID2: {
1589                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1590                 struct kvm_cpuid2 cpuid;
1591
1592                 r = -EFAULT;
1593                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1594                         goto out;
1595                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1596                                               cpuid_arg->entries);
1597                 if (r)
1598                         goto out;
1599                 r = -EFAULT;
1600                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1601                         goto out;
1602                 r = 0;
1603                 break;
1604         }
1605         case KVM_GET_MSRS:
1606                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1607                 break;
1608         case KVM_SET_MSRS:
1609                 r = msr_io(vcpu, argp, do_set_msr, 0);
1610                 break;
1611         case KVM_TPR_ACCESS_REPORTING: {
1612                 struct kvm_tpr_access_ctl tac;
1613
1614                 r = -EFAULT;
1615                 if (copy_from_user(&tac, argp, sizeof tac))
1616                         goto out;
1617                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1618                 if (r)
1619                         goto out;
1620                 r = -EFAULT;
1621                 if (copy_to_user(argp, &tac, sizeof tac))
1622                         goto out;
1623                 r = 0;
1624                 break;
1625         };
1626         case KVM_SET_VAPIC_ADDR: {
1627                 struct kvm_vapic_addr va;
1628
1629                 r = -EINVAL;
1630                 if (!irqchip_in_kernel(vcpu->kvm))
1631                         goto out;
1632                 r = -EFAULT;
1633                 if (copy_from_user(&va, argp, sizeof va))
1634                         goto out;
1635                 r = 0;
1636                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1637                 break;
1638         }
1639         default:
1640                 r = -EINVAL;
1641         }
1642 out:
1643         kfree(lapic);
1644         return r;
1645 }
1646
1647 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1648 {
1649         int ret;
1650
1651         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1652                 return -1;
1653         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1654         return ret;
1655 }
1656
1657 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1658                                           u32 kvm_nr_mmu_pages)
1659 {
1660         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1661                 return -EINVAL;
1662
1663         down_write(&kvm->slots_lock);
1664         spin_lock(&kvm->mmu_lock);
1665
1666         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1667         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1668
1669         spin_unlock(&kvm->mmu_lock);
1670         up_write(&kvm->slots_lock);
1671         return 0;
1672 }
1673
1674 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1675 {
1676         return kvm->arch.n_alloc_mmu_pages;
1677 }
1678
1679 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1680 {
1681         int i;
1682         struct kvm_mem_alias *alias;
1683
1684         for (i = 0; i < kvm->arch.naliases; ++i) {
1685                 alias = &kvm->arch.aliases[i];
1686                 if (gfn >= alias->base_gfn
1687                     && gfn < alias->base_gfn + alias->npages)
1688                         return alias->target_gfn + gfn - alias->base_gfn;
1689         }
1690         return gfn;
1691 }
1692
1693 /*
1694  * Set a new alias region.  Aliases map a portion of physical memory into
1695  * another portion.  This is useful for memory windows, for example the PC
1696  * VGA region.
1697  */
1698 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1699                                          struct kvm_memory_alias *alias)
1700 {
1701         int r, n;
1702         struct kvm_mem_alias *p;
1703
1704         r = -EINVAL;
1705         /* General sanity checks */
1706         if (alias->memory_size & (PAGE_SIZE - 1))
1707                 goto out;
1708         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1709                 goto out;
1710         if (alias->slot >= KVM_ALIAS_SLOTS)
1711                 goto out;
1712         if (alias->guest_phys_addr + alias->memory_size
1713             < alias->guest_phys_addr)
1714                 goto out;
1715         if (alias->target_phys_addr + alias->memory_size
1716             < alias->target_phys_addr)
1717                 goto out;
1718
1719         down_write(&kvm->slots_lock);
1720         spin_lock(&kvm->mmu_lock);
1721
1722         p = &kvm->arch.aliases[alias->slot];
1723         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1724         p->npages = alias->memory_size >> PAGE_SHIFT;
1725         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1726
1727         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1728                 if (kvm->arch.aliases[n - 1].npages)
1729                         break;
1730         kvm->arch.naliases = n;
1731
1732         spin_unlock(&kvm->mmu_lock);
1733         kvm_mmu_zap_all(kvm);
1734
1735         up_write(&kvm->slots_lock);
1736
1737         return 0;
1738
1739 out:
1740         return r;
1741 }
1742
1743 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1744 {
1745         int r;
1746
1747         r = 0;
1748         switch (chip->chip_id) {
1749         case KVM_IRQCHIP_PIC_MASTER:
1750                 memcpy(&chip->chip.pic,
1751                         &pic_irqchip(kvm)->pics[0],
1752                         sizeof(struct kvm_pic_state));
1753                 break;
1754         case KVM_IRQCHIP_PIC_SLAVE:
1755                 memcpy(&chip->chip.pic,
1756                         &pic_irqchip(kvm)->pics[1],
1757                         sizeof(struct kvm_pic_state));
1758                 break;
1759         case KVM_IRQCHIP_IOAPIC:
1760                 memcpy(&chip->chip.ioapic,
1761                         ioapic_irqchip(kvm),
1762                         sizeof(struct kvm_ioapic_state));
1763                 break;
1764         default:
1765                 r = -EINVAL;
1766                 break;
1767         }
1768         return r;
1769 }
1770
1771 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1772 {
1773         int r;
1774
1775         r = 0;
1776         switch (chip->chip_id) {
1777         case KVM_IRQCHIP_PIC_MASTER:
1778                 memcpy(&pic_irqchip(kvm)->pics[0],
1779                         &chip->chip.pic,
1780                         sizeof(struct kvm_pic_state));
1781                 break;
1782         case KVM_IRQCHIP_PIC_SLAVE:
1783                 memcpy(&pic_irqchip(kvm)->pics[1],
1784                         &chip->chip.pic,
1785                         sizeof(struct kvm_pic_state));
1786                 break;
1787         case KVM_IRQCHIP_IOAPIC:
1788                 memcpy(ioapic_irqchip(kvm),
1789                         &chip->chip.ioapic,
1790                         sizeof(struct kvm_ioapic_state));
1791                 break;
1792         default:
1793                 r = -EINVAL;
1794                 break;
1795         }
1796         kvm_pic_update_irq(pic_irqchip(kvm));
1797         return r;
1798 }
1799
1800 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1801 {
1802         int r = 0;
1803
1804         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1805         return r;
1806 }
1807
1808 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1809 {
1810         int r = 0;
1811
1812         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1813         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1814         return r;
1815 }
1816
1817 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1818                                  struct kvm_reinject_control *control)
1819 {
1820         if (!kvm->arch.vpit)
1821                 return -ENXIO;
1822         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1823         return 0;
1824 }
1825
1826 /*
1827  * Get (and clear) the dirty memory log for a memory slot.
1828  */
1829 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1830                                       struct kvm_dirty_log *log)
1831 {
1832         int r;
1833         int n;
1834         struct kvm_memory_slot *memslot;
1835         int is_dirty = 0;
1836
1837         down_write(&kvm->slots_lock);
1838
1839         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1840         if (r)
1841                 goto out;
1842
1843         /* If nothing is dirty, don't bother messing with page tables. */
1844         if (is_dirty) {
1845                 spin_lock(&kvm->mmu_lock);
1846                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1847                 spin_unlock(&kvm->mmu_lock);
1848                 kvm_flush_remote_tlbs(kvm);
1849                 memslot = &kvm->memslots[log->slot];
1850                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1851                 memset(memslot->dirty_bitmap, 0, n);
1852         }
1853         r = 0;
1854 out:
1855         up_write(&kvm->slots_lock);
1856         return r;
1857 }
1858
1859 long kvm_arch_vm_ioctl(struct file *filp,
1860                        unsigned int ioctl, unsigned long arg)
1861 {
1862         struct kvm *kvm = filp->private_data;
1863         void __user *argp = (void __user *)arg;
1864         int r = -EINVAL;
1865         /*
1866          * This union makes it completely explicit to gcc-3.x
1867          * that these two variables' stack usage should be
1868          * combined, not added together.
1869          */
1870         union {
1871                 struct kvm_pit_state ps;
1872                 struct kvm_memory_alias alias;
1873         } u;
1874
1875         switch (ioctl) {
1876         case KVM_SET_TSS_ADDR:
1877                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1878                 if (r < 0)
1879                         goto out;
1880                 break;
1881         case KVM_SET_MEMORY_REGION: {
1882                 struct kvm_memory_region kvm_mem;
1883                 struct kvm_userspace_memory_region kvm_userspace_mem;
1884
1885                 r = -EFAULT;
1886                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1887                         goto out;
1888                 kvm_userspace_mem.slot = kvm_mem.slot;
1889                 kvm_userspace_mem.flags = kvm_mem.flags;
1890                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1891                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1892                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1893                 if (r)
1894                         goto out;
1895                 break;
1896         }
1897         case KVM_SET_NR_MMU_PAGES:
1898                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1899                 if (r)
1900                         goto out;
1901                 break;
1902         case KVM_GET_NR_MMU_PAGES:
1903                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1904                 break;
1905         case KVM_SET_MEMORY_ALIAS:
1906                 r = -EFAULT;
1907                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1908                         goto out;
1909                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1910                 if (r)
1911                         goto out;
1912                 break;
1913         case KVM_CREATE_IRQCHIP:
1914                 r = -ENOMEM;
1915                 kvm->arch.vpic = kvm_create_pic(kvm);
1916                 if (kvm->arch.vpic) {
1917                         r = kvm_ioapic_init(kvm);
1918                         if (r) {
1919                                 kfree(kvm->arch.vpic);
1920                                 kvm->arch.vpic = NULL;
1921                                 goto out;
1922                         }
1923                 } else
1924                         goto out;
1925                 r = kvm_setup_default_irq_routing(kvm);
1926                 if (r) {
1927                         kfree(kvm->arch.vpic);
1928                         kfree(kvm->arch.vioapic);
1929                         goto out;
1930                 }
1931                 break;
1932         case KVM_CREATE_PIT:
1933                 mutex_lock(&kvm->lock);
1934                 r = -EEXIST;
1935                 if (kvm->arch.vpit)
1936                         goto create_pit_unlock;
1937                 r = -ENOMEM;
1938                 kvm->arch.vpit = kvm_create_pit(kvm);
1939                 if (kvm->arch.vpit)
1940                         r = 0;
1941         create_pit_unlock:
1942                 mutex_unlock(&kvm->lock);
1943                 break;
1944         case KVM_IRQ_LINE_STATUS:
1945         case KVM_IRQ_LINE: {
1946                 struct kvm_irq_level irq_event;
1947
1948                 r = -EFAULT;
1949                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1950                         goto out;
1951                 if (irqchip_in_kernel(kvm)) {
1952                         __s32 status;
1953                         mutex_lock(&kvm->lock);
1954                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1955                                         irq_event.irq, irq_event.level);
1956                         mutex_unlock(&kvm->lock);
1957                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1958                                 irq_event.status = status;
1959                                 if (copy_to_user(argp, &irq_event,
1960                                                         sizeof irq_event))
1961                                         goto out;
1962                         }
1963                         r = 0;
1964                 }
1965                 break;
1966         }
1967         case KVM_GET_IRQCHIP: {
1968                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1969                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1970
1971                 r = -ENOMEM;
1972                 if (!chip)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_from_user(chip, argp, sizeof *chip))
1976                         goto get_irqchip_out;
1977                 r = -ENXIO;
1978                 if (!irqchip_in_kernel(kvm))
1979                         goto get_irqchip_out;
1980                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1981                 if (r)
1982                         goto get_irqchip_out;
1983                 r = -EFAULT;
1984                 if (copy_to_user(argp, chip, sizeof *chip))
1985                         goto get_irqchip_out;
1986                 r = 0;
1987         get_irqchip_out:
1988                 kfree(chip);
1989                 if (r)
1990                         goto out;
1991                 break;
1992         }
1993         case KVM_SET_IRQCHIP: {
1994                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1995                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1996
1997                 r = -ENOMEM;
1998                 if (!chip)
1999                         goto out;
2000                 r = -EFAULT;
2001                 if (copy_from_user(chip, argp, sizeof *chip))
2002                         goto set_irqchip_out;
2003                 r = -ENXIO;
2004                 if (!irqchip_in_kernel(kvm))
2005                         goto set_irqchip_out;
2006                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2007                 if (r)
2008                         goto set_irqchip_out;
2009                 r = 0;
2010         set_irqchip_out:
2011                 kfree(chip);
2012                 if (r)
2013                         goto out;
2014                 break;
2015         }
2016         case KVM_GET_PIT: {
2017                 r = -EFAULT;
2018                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2019                         goto out;
2020                 r = -ENXIO;
2021                 if (!kvm->arch.vpit)
2022                         goto out;
2023                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2024                 if (r)
2025                         goto out;
2026                 r = -EFAULT;
2027                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2028                         goto out;
2029                 r = 0;
2030                 break;
2031         }
2032         case KVM_SET_PIT: {
2033                 r = -EFAULT;
2034                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2035                         goto out;
2036                 r = -ENXIO;
2037                 if (!kvm->arch.vpit)
2038                         goto out;
2039                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2040                 if (r)
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         case KVM_REINJECT_CONTROL: {
2046                 struct kvm_reinject_control control;
2047                 r =  -EFAULT;
2048                 if (copy_from_user(&control, argp, sizeof(control)))
2049                         goto out;
2050                 r = kvm_vm_ioctl_reinject(kvm, &control);
2051                 if (r)
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056         default:
2057                 ;
2058         }
2059 out:
2060         return r;
2061 }
2062
2063 static void kvm_init_msr_list(void)
2064 {
2065         u32 dummy[2];
2066         unsigned i, j;
2067
2068         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2069                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2070                         continue;
2071                 if (j < i)
2072                         msrs_to_save[j] = msrs_to_save[i];
2073                 j++;
2074         }
2075         num_msrs_to_save = j;
2076 }
2077
2078 /*
2079  * Only apic need an MMIO device hook, so shortcut now..
2080  */
2081 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2082                                                 gpa_t addr, int len,
2083                                                 int is_write)
2084 {
2085         struct kvm_io_device *dev;
2086
2087         if (vcpu->arch.apic) {
2088                 dev = &vcpu->arch.apic->dev;
2089                 if (dev->in_range(dev, addr, len, is_write))
2090                         return dev;
2091         }
2092         return NULL;
2093 }
2094
2095
2096 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2097                                                 gpa_t addr, int len,
2098                                                 int is_write)
2099 {
2100         struct kvm_io_device *dev;
2101
2102         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2103         if (dev == NULL)
2104                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2105                                           is_write);
2106         return dev;
2107 }
2108
2109 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2110                                struct kvm_vcpu *vcpu)
2111 {
2112         void *data = val;
2113         int r = X86EMUL_CONTINUE;
2114
2115         while (bytes) {
2116                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2117                 unsigned offset = addr & (PAGE_SIZE-1);
2118                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2119                 int ret;
2120
2121                 if (gpa == UNMAPPED_GVA) {
2122                         r = X86EMUL_PROPAGATE_FAULT;
2123                         goto out;
2124                 }
2125                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2126                 if (ret < 0) {
2127                         r = X86EMUL_UNHANDLEABLE;
2128                         goto out;
2129                 }
2130
2131                 bytes -= toread;
2132                 data += toread;
2133                 addr += toread;
2134         }
2135 out:
2136         return r;
2137 }
2138
2139 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2140                                 struct kvm_vcpu *vcpu)
2141 {
2142         void *data = val;
2143         int r = X86EMUL_CONTINUE;
2144
2145         while (bytes) {
2146                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2147                 unsigned offset = addr & (PAGE_SIZE-1);
2148                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2149                 int ret;
2150
2151                 if (gpa == UNMAPPED_GVA) {
2152                         r = X86EMUL_PROPAGATE_FAULT;
2153                         goto out;
2154                 }
2155                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2156                 if (ret < 0) {
2157                         r = X86EMUL_UNHANDLEABLE;
2158                         goto out;
2159                 }
2160
2161                 bytes -= towrite;
2162                 data += towrite;
2163                 addr += towrite;
2164         }
2165 out:
2166         return r;
2167 }
2168
2169
2170 static int emulator_read_emulated(unsigned long addr,
2171                                   void *val,
2172                                   unsigned int bytes,
2173                                   struct kvm_vcpu *vcpu)
2174 {
2175         struct kvm_io_device *mmio_dev;
2176         gpa_t                 gpa;
2177
2178         if (vcpu->mmio_read_completed) {
2179                 memcpy(val, vcpu->mmio_data, bytes);
2180                 vcpu->mmio_read_completed = 0;
2181                 return X86EMUL_CONTINUE;
2182         }
2183
2184         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2185
2186         /* For APIC access vmexit */
2187         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2188                 goto mmio;
2189
2190         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2191                                 == X86EMUL_CONTINUE)
2192                 return X86EMUL_CONTINUE;
2193         if (gpa == UNMAPPED_GVA)
2194                 return X86EMUL_PROPAGATE_FAULT;
2195
2196 mmio:
2197         /*
2198          * Is this MMIO handled locally?
2199          */
2200         mutex_lock(&vcpu->kvm->lock);
2201         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2202         if (mmio_dev) {
2203                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2204                 mutex_unlock(&vcpu->kvm->lock);
2205                 return X86EMUL_CONTINUE;
2206         }
2207         mutex_unlock(&vcpu->kvm->lock);
2208
2209         vcpu->mmio_needed = 1;
2210         vcpu->mmio_phys_addr = gpa;
2211         vcpu->mmio_size = bytes;
2212         vcpu->mmio_is_write = 0;
2213
2214         return X86EMUL_UNHANDLEABLE;
2215 }
2216
2217 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2218                           const void *val, int bytes)
2219 {
2220         int ret;
2221
2222         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2223         if (ret < 0)
2224                 return 0;
2225         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2226         return 1;
2227 }
2228
2229 static int emulator_write_emulated_onepage(unsigned long addr,
2230                                            const void *val,
2231                                            unsigned int bytes,
2232                                            struct kvm_vcpu *vcpu)
2233 {
2234         struct kvm_io_device *mmio_dev;
2235         gpa_t                 gpa;
2236
2237         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2238
2239         if (gpa == UNMAPPED_GVA) {
2240                 kvm_inject_page_fault(vcpu, addr, 2);
2241                 return X86EMUL_PROPAGATE_FAULT;
2242         }
2243
2244         /* For APIC access vmexit */
2245         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2246                 goto mmio;
2247
2248         if (emulator_write_phys(vcpu, gpa, val, bytes))
2249                 return X86EMUL_CONTINUE;
2250
2251 mmio:
2252         /*
2253          * Is this MMIO handled locally?
2254          */
2255         mutex_lock(&vcpu->kvm->lock);
2256         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2257         if (mmio_dev) {
2258                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2259                 mutex_unlock(&vcpu->kvm->lock);
2260                 return X86EMUL_CONTINUE;
2261         }
2262         mutex_unlock(&vcpu->kvm->lock);
2263
2264         vcpu->mmio_needed = 1;
2265         vcpu->mmio_phys_addr = gpa;
2266         vcpu->mmio_size = bytes;
2267         vcpu->mmio_is_write = 1;
2268         memcpy(vcpu->mmio_data, val, bytes);
2269
2270         return X86EMUL_CONTINUE;
2271 }
2272
2273 int emulator_write_emulated(unsigned long addr,
2274                                    const void *val,
2275                                    unsigned int bytes,
2276                                    struct kvm_vcpu *vcpu)
2277 {
2278         /* Crossing a page boundary? */
2279         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2280                 int rc, now;
2281
2282                 now = -addr & ~PAGE_MASK;
2283                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2284                 if (rc != X86EMUL_CONTINUE)
2285                         return rc;
2286                 addr += now;
2287                 val += now;
2288                 bytes -= now;
2289         }
2290         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2291 }
2292 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2293
2294 static int emulator_cmpxchg_emulated(unsigned long addr,
2295                                      const void *old,
2296                                      const void *new,
2297                                      unsigned int bytes,
2298                                      struct kvm_vcpu *vcpu)
2299 {
2300         static int reported;
2301
2302         if (!reported) {
2303                 reported = 1;
2304                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2305         }
2306 #ifndef CONFIG_X86_64
2307         /* guests cmpxchg8b have to be emulated atomically */
2308         if (bytes == 8) {
2309                 gpa_t gpa;
2310                 struct page *page;
2311                 char *kaddr;
2312                 u64 val;
2313
2314                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2315
2316                 if (gpa == UNMAPPED_GVA ||
2317                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2318                         goto emul_write;
2319
2320                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2321                         goto emul_write;
2322
2323                 val = *(u64 *)new;
2324
2325                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2326
2327                 kaddr = kmap_atomic(page, KM_USER0);
2328                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2329                 kunmap_atomic(kaddr, KM_USER0);
2330                 kvm_release_page_dirty(page);
2331         }
2332 emul_write:
2333 #endif
2334
2335         return emulator_write_emulated(addr, new, bytes, vcpu);
2336 }
2337
2338 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2339 {
2340         return kvm_x86_ops->get_segment_base(vcpu, seg);
2341 }
2342
2343 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2344 {
2345         kvm_mmu_invlpg(vcpu, address);
2346         return X86EMUL_CONTINUE;
2347 }
2348
2349 int emulate_clts(struct kvm_vcpu *vcpu)
2350 {
2351         KVMTRACE_0D(CLTS, vcpu, handler);
2352         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2353         return X86EMUL_CONTINUE;
2354 }
2355
2356 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2357 {
2358         struct kvm_vcpu *vcpu = ctxt->vcpu;
2359
2360         switch (dr) {
2361         case 0 ... 3:
2362                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2363                 return X86EMUL_CONTINUE;
2364         default:
2365                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2366                 return X86EMUL_UNHANDLEABLE;
2367         }
2368 }
2369
2370 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2371 {
2372         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2373         int exception;
2374
2375         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2376         if (exception) {
2377                 /* FIXME: better handling */
2378                 return X86EMUL_UNHANDLEABLE;
2379         }
2380         return X86EMUL_CONTINUE;
2381 }
2382
2383 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2384 {
2385         u8 opcodes[4];
2386         unsigned long rip = kvm_rip_read(vcpu);
2387         unsigned long rip_linear;
2388
2389         if (!printk_ratelimit())
2390                 return;
2391
2392         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2393
2394         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2395
2396         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2397                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2398 }
2399 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2400
2401 static struct x86_emulate_ops emulate_ops = {
2402         .read_std            = kvm_read_guest_virt,
2403         .read_emulated       = emulator_read_emulated,
2404         .write_emulated      = emulator_write_emulated,
2405         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2406 };
2407
2408 static void cache_all_regs(struct kvm_vcpu *vcpu)
2409 {
2410         kvm_register_read(vcpu, VCPU_REGS_RAX);
2411         kvm_register_read(vcpu, VCPU_REGS_RSP);
2412         kvm_register_read(vcpu, VCPU_REGS_RIP);
2413         vcpu->arch.regs_dirty = ~0;
2414 }
2415
2416 int emulate_instruction(struct kvm_vcpu *vcpu,
2417                         struct kvm_run *run,
2418                         unsigned long cr2,
2419                         u16 error_code,
2420                         int emulation_type)
2421 {
2422         int r, shadow_mask;
2423         struct decode_cache *c;
2424
2425         kvm_clear_exception_queue(vcpu);
2426         vcpu->arch.mmio_fault_cr2 = cr2;
2427         /*
2428          * TODO: fix x86_emulate.c to use guest_read/write_register
2429          * instead of direct ->regs accesses, can save hundred cycles
2430          * on Intel for instructions that don't read/change RSP, for
2431          * for example.
2432          */
2433         cache_all_regs(vcpu);
2434
2435         vcpu->mmio_is_write = 0;
2436         vcpu->arch.pio.string = 0;
2437
2438         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2439                 int cs_db, cs_l;
2440                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2441
2442                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2443                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2444                 vcpu->arch.emulate_ctxt.mode =
2445                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2446                         ? X86EMUL_MODE_REAL : cs_l
2447                         ? X86EMUL_MODE_PROT64 : cs_db
2448                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2449
2450                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2451
2452                 /* Reject the instructions other than VMCALL/VMMCALL when
2453                  * try to emulate invalid opcode */
2454                 c = &vcpu->arch.emulate_ctxt.decode;
2455                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2456                     (!(c->twobyte && c->b == 0x01 &&
2457                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2458                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2459                         return EMULATE_FAIL;
2460
2461                 ++vcpu->stat.insn_emulation;
2462                 if (r)  {
2463                         ++vcpu->stat.insn_emulation_fail;
2464                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2465                                 return EMULATE_DONE;
2466                         return EMULATE_FAIL;
2467                 }
2468         }
2469
2470         if (emulation_type & EMULTYPE_SKIP) {
2471                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2472                 return EMULATE_DONE;
2473         }
2474
2475         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2476         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2477
2478         if (r == 0)
2479                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2480
2481         if (vcpu->arch.pio.string)
2482                 return EMULATE_DO_MMIO;
2483
2484         if ((r || vcpu->mmio_is_write) && run) {
2485                 run->exit_reason = KVM_EXIT_MMIO;
2486                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2487                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2488                 run->mmio.len = vcpu->mmio_size;
2489                 run->mmio.is_write = vcpu->mmio_is_write;
2490         }
2491
2492         if (r) {
2493                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2494                         return EMULATE_DONE;
2495                 if (!vcpu->mmio_needed) {
2496                         kvm_report_emulation_failure(vcpu, "mmio");
2497                         return EMULATE_FAIL;
2498                 }
2499                 return EMULATE_DO_MMIO;
2500         }
2501
2502         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2503
2504         if (vcpu->mmio_is_write) {
2505                 vcpu->mmio_needed = 0;
2506                 return EMULATE_DO_MMIO;
2507         }
2508
2509         return EMULATE_DONE;
2510 }
2511 EXPORT_SYMBOL_GPL(emulate_instruction);
2512
2513 static int pio_copy_data(struct kvm_vcpu *vcpu)
2514 {
2515         void *p = vcpu->arch.pio_data;
2516         gva_t q = vcpu->arch.pio.guest_gva;
2517         unsigned bytes;
2518         int ret;
2519
2520         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2521         if (vcpu->arch.pio.in)
2522                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2523         else
2524                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2525         return ret;
2526 }
2527
2528 int complete_pio(struct kvm_vcpu *vcpu)
2529 {
2530         struct kvm_pio_request *io = &vcpu->arch.pio;
2531         long delta;
2532         int r;
2533         unsigned long val;
2534
2535         if (!io->string) {
2536                 if (io->in) {
2537                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2538                         memcpy(&val, vcpu->arch.pio_data, io->size);
2539                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2540                 }
2541         } else {
2542                 if (io->in) {
2543                         r = pio_copy_data(vcpu);
2544                         if (r)
2545                                 return r;
2546                 }
2547
2548                 delta = 1;
2549                 if (io->rep) {
2550                         delta *= io->cur_count;
2551                         /*
2552                          * The size of the register should really depend on
2553                          * current address size.
2554                          */
2555                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2556                         val -= delta;
2557                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2558                 }
2559                 if (io->down)
2560                         delta = -delta;
2561                 delta *= io->size;
2562                 if (io->in) {
2563                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2564                         val += delta;
2565                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2566                 } else {
2567                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2568                         val += delta;
2569                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2570                 }
2571         }
2572
2573         io->count -= io->cur_count;
2574         io->cur_count = 0;
2575
2576         return 0;
2577 }
2578
2579 static void kernel_pio(struct kvm_io_device *pio_dev,
2580                        struct kvm_vcpu *vcpu,
2581                        void *pd)
2582 {
2583         /* TODO: String I/O for in kernel device */
2584
2585         mutex_lock(&vcpu->kvm->lock);
2586         if (vcpu->arch.pio.in)
2587                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2588                                   vcpu->arch.pio.size,
2589                                   pd);
2590         else
2591                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2592                                    vcpu->arch.pio.size,
2593                                    pd);
2594         mutex_unlock(&vcpu->kvm->lock);
2595 }
2596
2597 static void pio_string_write(struct kvm_io_device *pio_dev,
2598                              struct kvm_vcpu *vcpu)
2599 {
2600         struct kvm_pio_request *io = &vcpu->arch.pio;
2601         void *pd = vcpu->arch.pio_data;
2602         int i;
2603
2604         mutex_lock(&vcpu->kvm->lock);
2605         for (i = 0; i < io->cur_count; i++) {
2606                 kvm_iodevice_write(pio_dev, io->port,
2607                                    io->size,
2608                                    pd);
2609                 pd += io->size;
2610         }
2611         mutex_unlock(&vcpu->kvm->lock);
2612 }
2613
2614 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2615                                                gpa_t addr, int len,
2616                                                int is_write)
2617 {
2618         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2619 }
2620
2621 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2622                   int size, unsigned port)
2623 {
2624         struct kvm_io_device *pio_dev;
2625         unsigned long val;
2626
2627         vcpu->run->exit_reason = KVM_EXIT_IO;
2628         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2629         vcpu->run->io.size = vcpu->arch.pio.size = size;
2630         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2631         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2632         vcpu->run->io.port = vcpu->arch.pio.port = port;
2633         vcpu->arch.pio.in = in;
2634         vcpu->arch.pio.string = 0;
2635         vcpu->arch.pio.down = 0;
2636         vcpu->arch.pio.rep = 0;
2637
2638         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2639                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2640                             handler);
2641         else
2642                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2643                             handler);
2644
2645         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2646         memcpy(vcpu->arch.pio_data, &val, 4);
2647
2648         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2649         if (pio_dev) {
2650                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2651                 complete_pio(vcpu);
2652                 return 1;
2653         }
2654         return 0;
2655 }
2656 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2657
2658 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2659                   int size, unsigned long count, int down,
2660                   gva_t address, int rep, unsigned port)
2661 {
2662         unsigned now, in_page;
2663         int ret = 0;
2664         struct kvm_io_device *pio_dev;
2665
2666         vcpu->run->exit_reason = KVM_EXIT_IO;
2667         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2668         vcpu->run->io.size = vcpu->arch.pio.size = size;
2669         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2670         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2671         vcpu->run->io.port = vcpu->arch.pio.port = port;
2672         vcpu->arch.pio.in = in;
2673         vcpu->arch.pio.string = 1;
2674         vcpu->arch.pio.down = down;
2675         vcpu->arch.pio.rep = rep;
2676
2677         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2678                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2679                             handler);
2680         else
2681                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2682                             handler);
2683
2684         if (!count) {
2685                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2686                 return 1;
2687         }
2688
2689         if (!down)
2690                 in_page = PAGE_SIZE - offset_in_page(address);
2691         else
2692                 in_page = offset_in_page(address) + size;
2693         now = min(count, (unsigned long)in_page / size);
2694         if (!now)
2695                 now = 1;
2696         if (down) {
2697                 /*
2698                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2699                  */
2700                 pr_unimpl(vcpu, "guest string pio down\n");
2701                 kvm_inject_gp(vcpu, 0);
2702                 return 1;
2703         }
2704         vcpu->run->io.count = now;
2705         vcpu->arch.pio.cur_count = now;
2706
2707         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2708                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2709
2710         vcpu->arch.pio.guest_gva = address;
2711
2712         pio_dev = vcpu_find_pio_dev(vcpu, port,
2713                                     vcpu->arch.pio.cur_count,
2714                                     !vcpu->arch.pio.in);
2715         if (!vcpu->arch.pio.in) {
2716                 /* string PIO write */
2717                 ret = pio_copy_data(vcpu);
2718                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2719                         kvm_inject_gp(vcpu, 0);
2720                         return 1;
2721                 }
2722                 if (ret == 0 && pio_dev) {
2723                         pio_string_write(pio_dev, vcpu);
2724                         complete_pio(vcpu);
2725                         if (vcpu->arch.pio.count == 0)
2726                                 ret = 1;
2727                 }
2728         } else if (pio_dev)
2729                 pr_unimpl(vcpu, "no string pio read support yet, "
2730                        "port %x size %d count %ld\n",
2731                         port, size, count);
2732
2733         return ret;
2734 }
2735 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2736
2737 static void bounce_off(void *info)
2738 {
2739         /* nothing */
2740 }
2741
2742 static unsigned int  ref_freq;
2743 static unsigned long tsc_khz_ref;
2744
2745 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2746                                      void *data)
2747 {
2748         struct cpufreq_freqs *freq = data;
2749         struct kvm *kvm;
2750         struct kvm_vcpu *vcpu;
2751         int i, send_ipi = 0;
2752
2753         if (!ref_freq)
2754                 ref_freq = freq->old;
2755
2756         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2757                 return 0;
2758         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2759                 return 0;
2760         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2761
2762         spin_lock(&kvm_lock);
2763         list_for_each_entry(kvm, &vm_list, vm_list) {
2764                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2765                         vcpu = kvm->vcpus[i];
2766                         if (!vcpu)
2767                                 continue;
2768                         if (vcpu->cpu != freq->cpu)
2769                                 continue;
2770                         if (!kvm_request_guest_time_update(vcpu))
2771                                 continue;
2772                         if (vcpu->cpu != smp_processor_id())
2773                                 send_ipi++;
2774                 }
2775         }
2776         spin_unlock(&kvm_lock);
2777
2778         if (freq->old < freq->new && send_ipi) {
2779                 /*
2780                  * We upscale the frequency.  Must make the guest
2781                  * doesn't see old kvmclock values while running with
2782                  * the new frequency, otherwise we risk the guest sees
2783                  * time go backwards.
2784                  *
2785                  * In case we update the frequency for another cpu
2786                  * (which might be in guest context) send an interrupt
2787                  * to kick the cpu out of guest context.  Next time
2788                  * guest context is entered kvmclock will be updated,
2789                  * so the guest will not see stale values.
2790                  */
2791                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2792         }
2793         return 0;
2794 }
2795
2796 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2797         .notifier_call  = kvmclock_cpufreq_notifier
2798 };
2799
2800 int kvm_arch_init(void *opaque)
2801 {
2802         int r, cpu;
2803         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2804
2805         if (kvm_x86_ops) {
2806                 printk(KERN_ERR "kvm: already loaded the other module\n");
2807                 r = -EEXIST;
2808                 goto out;
2809         }
2810
2811         if (!ops->cpu_has_kvm_support()) {
2812                 printk(KERN_ERR "kvm: no hardware support\n");
2813                 r = -EOPNOTSUPP;
2814                 goto out;
2815         }
2816         if (ops->disabled_by_bios()) {
2817                 printk(KERN_ERR "kvm: disabled by bios\n");
2818                 r = -EOPNOTSUPP;
2819                 goto out;
2820         }
2821
2822         r = kvm_mmu_module_init();
2823         if (r)
2824                 goto out;
2825
2826         kvm_init_msr_list();
2827
2828         kvm_x86_ops = ops;
2829         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2830         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2831         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2832                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2833
2834         for_each_possible_cpu(cpu)
2835                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2836         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2837                 tsc_khz_ref = tsc_khz;
2838                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2839                                           CPUFREQ_TRANSITION_NOTIFIER);
2840         }
2841
2842         return 0;
2843
2844 out:
2845         return r;
2846 }
2847
2848 void kvm_arch_exit(void)
2849 {
2850         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2851                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2852                                             CPUFREQ_TRANSITION_NOTIFIER);
2853         kvm_x86_ops = NULL;
2854         kvm_mmu_module_exit();
2855 }
2856
2857 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2858 {
2859         ++vcpu->stat.halt_exits;
2860         KVMTRACE_0D(HLT, vcpu, handler);
2861         if (irqchip_in_kernel(vcpu->kvm)) {
2862                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2863                 return 1;
2864         } else {
2865                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2866                 return 0;
2867         }
2868 }
2869 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2870
2871 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2872                            unsigned long a1)
2873 {
2874         if (is_long_mode(vcpu))
2875                 return a0;
2876         else
2877                 return a0 | ((gpa_t)a1 << 32);
2878 }
2879
2880 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2881 {
2882         unsigned long nr, a0, a1, a2, a3, ret;
2883         int r = 1;
2884
2885         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2886         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2887         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2888         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2889         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2890
2891         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2892
2893         if (!is_long_mode(vcpu)) {
2894                 nr &= 0xFFFFFFFF;
2895                 a0 &= 0xFFFFFFFF;
2896                 a1 &= 0xFFFFFFFF;
2897                 a2 &= 0xFFFFFFFF;
2898                 a3 &= 0xFFFFFFFF;
2899         }
2900
2901         switch (nr) {
2902         case KVM_HC_VAPIC_POLL_IRQ:
2903                 ret = 0;
2904                 break;
2905         case KVM_HC_MMU_OP:
2906                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2907                 break;
2908         default:
2909                 ret = -KVM_ENOSYS;
2910                 break;
2911         }
2912         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2913         ++vcpu->stat.hypercalls;
2914         return r;
2915 }
2916 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2917
2918 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2919 {
2920         char instruction[3];
2921         int ret = 0;
2922         unsigned long rip = kvm_rip_read(vcpu);
2923
2924
2925         /*
2926          * Blow out the MMU to ensure that no other VCPU has an active mapping
2927          * to ensure that the updated hypercall appears atomically across all
2928          * VCPUs.
2929          */
2930         kvm_mmu_zap_all(vcpu->kvm);
2931
2932         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2933         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2934             != X86EMUL_CONTINUE)
2935                 ret = -EFAULT;
2936
2937         return ret;
2938 }
2939
2940 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2941 {
2942         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2943 }
2944
2945 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2946 {
2947         struct descriptor_table dt = { limit, base };
2948
2949         kvm_x86_ops->set_gdt(vcpu, &dt);
2950 }
2951
2952 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2953 {
2954         struct descriptor_table dt = { limit, base };
2955
2956         kvm_x86_ops->set_idt(vcpu, &dt);
2957 }
2958
2959 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2960                    unsigned long *rflags)
2961 {
2962         kvm_lmsw(vcpu, msw);
2963         *rflags = kvm_x86_ops->get_rflags(vcpu);
2964 }
2965
2966 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2967 {
2968         unsigned long value;
2969
2970         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2971         switch (cr) {
2972         case 0:
2973                 value = vcpu->arch.cr0;
2974                 break;
2975         case 2:
2976                 value = vcpu->arch.cr2;
2977                 break;
2978         case 3:
2979                 value = vcpu->arch.cr3;
2980                 break;
2981         case 4:
2982                 value = vcpu->arch.cr4;
2983                 break;
2984         case 8:
2985                 value = kvm_get_cr8(vcpu);
2986                 break;
2987         default:
2988                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2989                 return 0;
2990         }
2991         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2992                     (u32)((u64)value >> 32), handler);
2993
2994         return value;
2995 }
2996
2997 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2998                      unsigned long *rflags)
2999 {
3000         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
3001                     (u32)((u64)val >> 32), handler);
3002
3003         switch (cr) {
3004         case 0:
3005                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3006                 *rflags = kvm_x86_ops->get_rflags(vcpu);
3007                 break;
3008         case 2:
3009                 vcpu->arch.cr2 = val;
3010                 break;
3011         case 3:
3012                 kvm_set_cr3(vcpu, val);
3013                 break;
3014         case 4:
3015                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3016                 break;
3017         case 8:
3018                 kvm_set_cr8(vcpu, val & 0xfUL);
3019                 break;
3020         default:
3021                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3022         }
3023 }
3024
3025 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3026 {
3027         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3028         int j, nent = vcpu->arch.cpuid_nent;
3029
3030         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3031         /* when no next entry is found, the current entry[i] is reselected */
3032         for (j = i + 1; ; j = (j + 1) % nent) {
3033                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3034                 if (ej->function == e->function) {
3035                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3036                         return j;
3037                 }
3038         }
3039         return 0; /* silence gcc, even though control never reaches here */
3040 }
3041
3042 /* find an entry with matching function, matching index (if needed), and that
3043  * should be read next (if it's stateful) */
3044 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3045         u32 function, u32 index)
3046 {
3047         if (e->function != function)
3048                 return 0;
3049         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3050                 return 0;
3051         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3052             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3053                 return 0;
3054         return 1;
3055 }
3056
3057 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3058                                               u32 function, u32 index)
3059 {
3060         int i;
3061         struct kvm_cpuid_entry2 *best = NULL;
3062
3063         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3064                 struct kvm_cpuid_entry2 *e;
3065
3066                 e = &vcpu->arch.cpuid_entries[i];
3067                 if (is_matching_cpuid_entry(e, function, index)) {
3068                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3069                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3070                         best = e;
3071                         break;
3072                 }
3073                 /*
3074                  * Both basic or both extended?
3075                  */
3076                 if (((e->function ^ function) & 0x80000000) == 0)
3077                         if (!best || e->function > best->function)
3078                                 best = e;
3079         }
3080         return best;
3081 }
3082
3083 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3084 {
3085         struct kvm_cpuid_entry2 *best;
3086
3087         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3088         if (best)
3089                 return best->eax & 0xff;
3090         return 36;
3091 }
3092
3093 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3094 {
3095         u32 function, index;
3096         struct kvm_cpuid_entry2 *best;
3097
3098         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3099         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3100         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3101         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3102         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3103         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3104         best = kvm_find_cpuid_entry(vcpu, function, index);
3105         if (best) {
3106                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3107                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3108                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3109                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3110         }
3111         kvm_x86_ops->skip_emulated_instruction(vcpu);
3112         KVMTRACE_5D(CPUID, vcpu, function,
3113                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3114                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3115                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3116                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3117 }
3118 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3119
3120 /*
3121  * Check if userspace requested an interrupt window, and that the
3122  * interrupt window is open.
3123  *
3124  * No need to exit to userspace if we already have an interrupt queued.
3125  */
3126 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3127                                           struct kvm_run *kvm_run)
3128 {
3129         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3130                 kvm_run->request_interrupt_window &&
3131                 kvm_arch_interrupt_allowed(vcpu));
3132 }
3133
3134 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3135                               struct kvm_run *kvm_run)
3136 {
3137         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3138         kvm_run->cr8 = kvm_get_cr8(vcpu);
3139         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3140         if (irqchip_in_kernel(vcpu->kvm))
3141                 kvm_run->ready_for_interrupt_injection = 1;
3142         else
3143                 kvm_run->ready_for_interrupt_injection =
3144                         kvm_arch_interrupt_allowed(vcpu) &&
3145                         !kvm_cpu_has_interrupt(vcpu) &&
3146                         !kvm_event_needs_reinjection(vcpu);
3147 }
3148
3149 static void vapic_enter(struct kvm_vcpu *vcpu)
3150 {
3151         struct kvm_lapic *apic = vcpu->arch.apic;
3152         struct page *page;
3153
3154         if (!apic || !apic->vapic_addr)
3155                 return;
3156
3157         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3158
3159         vcpu->arch.apic->vapic_page = page;
3160 }
3161
3162 static void vapic_exit(struct kvm_vcpu *vcpu)
3163 {
3164         struct kvm_lapic *apic = vcpu->arch.apic;
3165
3166         if (!apic || !apic->vapic_addr)
3167                 return;
3168
3169         down_read(&vcpu->kvm->slots_lock);
3170         kvm_release_page_dirty(apic->vapic_page);
3171         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3172         up_read(&vcpu->kvm->slots_lock);
3173 }
3174
3175 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3176 {
3177         int max_irr, tpr;
3178
3179         if (!kvm_x86_ops->update_cr8_intercept)
3180                 return;
3181
3182         if (!vcpu->arch.apic->vapic_addr)
3183                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3184         else
3185                 max_irr = -1;
3186
3187         if (max_irr != -1)
3188                 max_irr >>= 4;
3189
3190         tpr = kvm_lapic_get_cr8(vcpu);
3191
3192         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3193 }
3194
3195 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3196 {
3197         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3198                 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
3199
3200         /* try to reinject previous events if any */
3201         if (vcpu->arch.nmi_injected) {
3202                 kvm_x86_ops->set_nmi(vcpu);
3203                 return;
3204         }
3205
3206         if (vcpu->arch.interrupt.pending) {
3207                 kvm_x86_ops->set_irq(vcpu);
3208                 return;
3209         }
3210
3211         /* try to inject new event if pending */
3212         if (vcpu->arch.nmi_pending) {
3213                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3214                         vcpu->arch.nmi_pending = false;
3215                         vcpu->arch.nmi_injected = true;
3216                         kvm_x86_ops->set_nmi(vcpu);
3217                 }
3218         } else if (kvm_cpu_has_interrupt(vcpu)) {
3219                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3220                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3221                                             false);
3222                         kvm_x86_ops->set_irq(vcpu);
3223                 }
3224         }
3225 }
3226
3227 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3228 {
3229         int r;
3230         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3231                 kvm_run->request_interrupt_window;
3232
3233         if (vcpu->requests)
3234                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3235                         kvm_mmu_unload(vcpu);
3236
3237         r = kvm_mmu_reload(vcpu);
3238         if (unlikely(r))
3239                 goto out;
3240
3241         if (vcpu->requests) {
3242                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3243                         __kvm_migrate_timers(vcpu);
3244                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3245                         kvm_write_guest_time(vcpu);
3246                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3247                         kvm_mmu_sync_roots(vcpu);
3248                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3249                         kvm_x86_ops->tlb_flush(vcpu);
3250                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3251                                        &vcpu->requests)) {
3252                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3253                         r = 0;
3254                         goto out;
3255                 }
3256                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3257                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3258                         r = 0;
3259                         goto out;
3260                 }
3261         }
3262
3263         preempt_disable();
3264
3265         kvm_x86_ops->prepare_guest_switch(vcpu);
3266         kvm_load_guest_fpu(vcpu);
3267
3268         local_irq_disable();
3269
3270         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3271         smp_mb__after_clear_bit();
3272
3273         if (vcpu->requests || need_resched() || signal_pending(current)) {
3274                 local_irq_enable();
3275                 preempt_enable();
3276                 r = 1;
3277                 goto out;
3278         }
3279
3280         if (vcpu->arch.exception.pending)
3281                 __queue_exception(vcpu);
3282         else
3283                 inject_pending_irq(vcpu, kvm_run);
3284
3285         /* enable NMI/IRQ window open exits if needed */
3286         if (vcpu->arch.nmi_pending)
3287                 kvm_x86_ops->enable_nmi_window(vcpu);
3288         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3289                 kvm_x86_ops->enable_irq_window(vcpu);
3290
3291         if (kvm_lapic_enabled(vcpu)) {
3292                 update_cr8_intercept(vcpu);
3293                 kvm_lapic_sync_to_vapic(vcpu);
3294         }
3295
3296         up_read(&vcpu->kvm->slots_lock);
3297
3298         kvm_guest_enter();
3299
3300         get_debugreg(vcpu->arch.host_dr6, 6);
3301         get_debugreg(vcpu->arch.host_dr7, 7);
3302         if (unlikely(vcpu->arch.switch_db_regs)) {
3303                 get_debugreg(vcpu->arch.host_db[0], 0);
3304                 get_debugreg(vcpu->arch.host_db[1], 1);
3305                 get_debugreg(vcpu->arch.host_db[2], 2);
3306                 get_debugreg(vcpu->arch.host_db[3], 3);
3307
3308                 set_debugreg(0, 7);
3309                 set_debugreg(vcpu->arch.eff_db[0], 0);
3310                 set_debugreg(vcpu->arch.eff_db[1], 1);
3311                 set_debugreg(vcpu->arch.eff_db[2], 2);
3312                 set_debugreg(vcpu->arch.eff_db[3], 3);
3313         }
3314
3315         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3316         kvm_x86_ops->run(vcpu, kvm_run);
3317
3318         if (unlikely(vcpu->arch.switch_db_regs)) {
3319                 set_debugreg(0, 7);
3320                 set_debugreg(vcpu->arch.host_db[0], 0);
3321                 set_debugreg(vcpu->arch.host_db[1], 1);
3322                 set_debugreg(vcpu->arch.host_db[2], 2);
3323                 set_debugreg(vcpu->arch.host_db[3], 3);
3324         }
3325         set_debugreg(vcpu->arch.host_dr6, 6);
3326         set_debugreg(vcpu->arch.host_dr7, 7);
3327
3328         set_bit(KVM_REQ_KICK, &vcpu->requests);
3329         local_irq_enable();
3330
3331         ++vcpu->stat.exits;
3332
3333         /*
3334          * We must have an instruction between local_irq_enable() and
3335          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3336          * the interrupt shadow.  The stat.exits increment will do nicely.
3337          * But we need to prevent reordering, hence this barrier():
3338          */
3339         barrier();
3340
3341         kvm_guest_exit();
3342
3343         preempt_enable();
3344
3345         down_read(&vcpu->kvm->slots_lock);
3346
3347         /*
3348          * Profile KVM exit RIPs:
3349          */
3350         if (unlikely(prof_on == KVM_PROFILING)) {
3351                 unsigned long rip = kvm_rip_read(vcpu);
3352                 profile_hit(KVM_PROFILING, (void *)rip);
3353         }
3354
3355
3356         kvm_lapic_sync_from_vapic(vcpu);
3357
3358         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3359 out:
3360         return r;
3361 }
3362
3363
3364 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3365 {
3366         int r;
3367
3368         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3369                 pr_debug("vcpu %d received sipi with vector # %x\n",
3370                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3371                 kvm_lapic_reset(vcpu);
3372                 r = kvm_arch_vcpu_reset(vcpu);
3373                 if (r)
3374                         return r;
3375                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3376         }
3377
3378         down_read(&vcpu->kvm->slots_lock);
3379         vapic_enter(vcpu);
3380
3381         r = 1;
3382         while (r > 0) {
3383                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3384                         r = vcpu_enter_guest(vcpu, kvm_run);
3385                 else {
3386                         up_read(&vcpu->kvm->slots_lock);
3387                         kvm_vcpu_block(vcpu);
3388                         down_read(&vcpu->kvm->slots_lock);
3389                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3390                         {
3391                                 switch(vcpu->arch.mp_state) {
3392                                 case KVM_MP_STATE_HALTED:
3393                                         vcpu->arch.mp_state =
3394                                                 KVM_MP_STATE_RUNNABLE;
3395                                 case KVM_MP_STATE_RUNNABLE:
3396                                         break;
3397                                 case KVM_MP_STATE_SIPI_RECEIVED:
3398                                 default:
3399                                         r = -EINTR;
3400                                         break;
3401                                 }
3402                         }
3403                 }
3404
3405                 if (r <= 0)
3406                         break;
3407
3408                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3409                 if (kvm_cpu_has_pending_timer(vcpu))
3410                         kvm_inject_pending_timer_irqs(vcpu);
3411
3412                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3413                         r = -EINTR;
3414                         kvm_run->exit_reason = KVM_EXIT_INTR;
3415                         ++vcpu->stat.request_irq_exits;
3416                 }
3417                 if (signal_pending(current)) {
3418                         r = -EINTR;
3419                         kvm_run->exit_reason = KVM_EXIT_INTR;
3420                         ++vcpu->stat.signal_exits;
3421                 }
3422                 if (need_resched()) {
3423                         up_read(&vcpu->kvm->slots_lock);
3424                         kvm_resched(vcpu);
3425                         down_read(&vcpu->kvm->slots_lock);
3426                 }
3427         }
3428
3429         up_read(&vcpu->kvm->slots_lock);
3430         post_kvm_run_save(vcpu, kvm_run);
3431
3432         vapic_exit(vcpu);
3433
3434         return r;
3435 }
3436
3437 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3438 {
3439         int r;
3440         sigset_t sigsaved;
3441
3442         vcpu_load(vcpu);
3443
3444         if (vcpu->sigset_active)
3445                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3446
3447         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3448                 kvm_vcpu_block(vcpu);
3449                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3450                 r = -EAGAIN;
3451                 goto out;
3452         }
3453
3454         /* re-sync apic's tpr */
3455         if (!irqchip_in_kernel(vcpu->kvm))
3456                 kvm_set_cr8(vcpu, kvm_run->cr8);
3457
3458         if (vcpu->arch.pio.cur_count) {
3459                 r = complete_pio(vcpu);
3460                 if (r)
3461                         goto out;
3462         }
3463 #if CONFIG_HAS_IOMEM
3464         if (vcpu->mmio_needed) {
3465                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3466                 vcpu->mmio_read_completed = 1;
3467                 vcpu->mmio_needed = 0;
3468
3469                 down_read(&vcpu->kvm->slots_lock);
3470                 r = emulate_instruction(vcpu, kvm_run,
3471                                         vcpu->arch.mmio_fault_cr2, 0,
3472                                         EMULTYPE_NO_DECODE);
3473                 up_read(&vcpu->kvm->slots_lock);
3474                 if (r == EMULATE_DO_MMIO) {
3475                         /*
3476                          * Read-modify-write.  Back to userspace.
3477                          */
3478                         r = 0;
3479                         goto out;
3480                 }
3481         }
3482 #endif
3483         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3484                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3485                                      kvm_run->hypercall.ret);
3486
3487         r = __vcpu_run(vcpu, kvm_run);
3488
3489 out:
3490         if (vcpu->sigset_active)
3491                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3492
3493         vcpu_put(vcpu);
3494         return r;
3495 }
3496
3497 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3498 {
3499         vcpu_load(vcpu);
3500
3501         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3502         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3503         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3504         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3505         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3506         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3507         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3508         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3509 #ifdef CONFIG_X86_64
3510         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3511         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3512         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3513         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3514         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3515         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3516         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3517         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3518 #endif
3519
3520         regs->rip = kvm_rip_read(vcpu);
3521         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3522
3523         /*
3524          * Don't leak debug flags in case they were set for guest debugging
3525          */
3526         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3527                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3528
3529         vcpu_put(vcpu);
3530
3531         return 0;
3532 }
3533
3534 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3535 {
3536         vcpu_load(vcpu);
3537
3538         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3539         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3540         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3541         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3542         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3543         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3544         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3545         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3546 #ifdef CONFIG_X86_64
3547         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3548         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3549         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3550         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3551         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3552         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3553         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3554         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3555
3556 #endif
3557
3558         kvm_rip_write(vcpu, regs->rip);
3559         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3560
3561
3562         vcpu->arch.exception.pending = false;
3563
3564         vcpu_put(vcpu);
3565
3566         return 0;
3567 }
3568
3569 void kvm_get_segment(struct kvm_vcpu *vcpu,
3570                      struct kvm_segment *var, int seg)
3571 {
3572         kvm_x86_ops->get_segment(vcpu, var, seg);
3573 }
3574
3575 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3576 {
3577         struct kvm_segment cs;
3578
3579         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3580         *db = cs.db;
3581         *l = cs.l;
3582 }
3583 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3584
3585 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3586                                   struct kvm_sregs *sregs)
3587 {
3588         struct descriptor_table dt;
3589
3590         vcpu_load(vcpu);
3591
3592         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3593         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3594         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3595         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3596         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3597         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3598
3599         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3600         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3601
3602         kvm_x86_ops->get_idt(vcpu, &dt);
3603         sregs->idt.limit = dt.limit;
3604         sregs->idt.base = dt.base;
3605         kvm_x86_ops->get_gdt(vcpu, &dt);
3606         sregs->gdt.limit = dt.limit;
3607         sregs->gdt.base = dt.base;
3608
3609         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3610         sregs->cr0 = vcpu->arch.cr0;
3611         sregs->cr2 = vcpu->arch.cr2;
3612         sregs->cr3 = vcpu->arch.cr3;
3613         sregs->cr4 = vcpu->arch.cr4;
3614         sregs->cr8 = kvm_get_cr8(vcpu);
3615         sregs->efer = vcpu->arch.shadow_efer;
3616         sregs->apic_base = kvm_get_apic_base(vcpu);
3617
3618         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3619
3620         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3621                 set_bit(vcpu->arch.interrupt.nr,
3622                         (unsigned long *)sregs->interrupt_bitmap);
3623
3624         vcpu_put(vcpu);
3625
3626         return 0;
3627 }
3628
3629 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3630                                     struct kvm_mp_state *mp_state)
3631 {
3632         vcpu_load(vcpu);
3633         mp_state->mp_state = vcpu->arch.mp_state;
3634         vcpu_put(vcpu);
3635         return 0;
3636 }
3637
3638 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3639                                     struct kvm_mp_state *mp_state)
3640 {
3641         vcpu_load(vcpu);
3642         vcpu->arch.mp_state = mp_state->mp_state;
3643         vcpu_put(vcpu);
3644         return 0;
3645 }
3646
3647 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3648                         struct kvm_segment *var, int seg)
3649 {
3650         kvm_x86_ops->set_segment(vcpu, var, seg);
3651 }
3652
3653 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3654                                    struct kvm_segment *kvm_desct)
3655 {
3656         kvm_desct->base = seg_desc->base0;
3657         kvm_desct->base |= seg_desc->base1 << 16;
3658         kvm_desct->base |= seg_desc->base2 << 24;
3659         kvm_desct->limit = seg_desc->limit0;
3660         kvm_desct->limit |= seg_desc->limit << 16;
3661         if (seg_desc->g) {
3662                 kvm_desct->limit <<= 12;
3663                 kvm_desct->limit |= 0xfff;
3664         }
3665         kvm_desct->selector = selector;
3666         kvm_desct->type = seg_desc->type;
3667         kvm_desct->present = seg_desc->p;
3668         kvm_desct->dpl = seg_desc->dpl;
3669         kvm_desct->db = seg_desc->d;
3670         kvm_desct->s = seg_desc->s;
3671         kvm_desct->l = seg_desc->l;
3672         kvm_desct->g = seg_desc->g;
3673         kvm_desct->avl = seg_desc->avl;
3674         if (!selector)
3675                 kvm_desct->unusable = 1;
3676         else
3677                 kvm_desct->unusable = 0;
3678         kvm_desct->padding = 0;
3679 }
3680
3681 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3682                                           u16 selector,
3683                                           struct descriptor_table *dtable)
3684 {
3685         if (selector & 1 << 2) {
3686                 struct kvm_segment kvm_seg;
3687
3688                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3689
3690                 if (kvm_seg.unusable)
3691                         dtable->limit = 0;
3692                 else
3693                         dtable->limit = kvm_seg.limit;
3694                 dtable->base = kvm_seg.base;
3695         }
3696         else
3697                 kvm_x86_ops->get_gdt(vcpu, dtable);
3698 }
3699
3700 /* allowed just for 8 bytes segments */
3701 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3702                                          struct desc_struct *seg_desc)
3703 {
3704         gpa_t gpa;
3705         struct descriptor_table dtable;
3706         u16 index = selector >> 3;
3707
3708         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3709
3710         if (dtable.limit < index * 8 + 7) {
3711                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3712                 return 1;
3713         }
3714         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3715         gpa += index * 8;
3716         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3717 }
3718
3719 /* allowed just for 8 bytes segments */
3720 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3721                                          struct desc_struct *seg_desc)
3722 {
3723         gpa_t gpa;
3724         struct descriptor_table dtable;
3725         u16 index = selector >> 3;
3726
3727         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3728
3729         if (dtable.limit < index * 8 + 7)
3730                 return 1;
3731         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3732         gpa += index * 8;
3733         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3734 }
3735
3736 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3737                              struct desc_struct *seg_desc)
3738 {
3739         u32 base_addr;
3740
3741         base_addr = seg_desc->base0;
3742         base_addr |= (seg_desc->base1 << 16);
3743         base_addr |= (seg_desc->base2 << 24);
3744
3745         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3746 }
3747
3748 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3749 {
3750         struct kvm_segment kvm_seg;
3751
3752         kvm_get_segment(vcpu, &kvm_seg, seg);
3753         return kvm_seg.selector;
3754 }
3755
3756 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3757                                                 u16 selector,
3758                                                 struct kvm_segment *kvm_seg)
3759 {
3760         struct desc_struct seg_desc;
3761
3762         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3763                 return 1;
3764         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3765         return 0;
3766 }
3767
3768 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3769 {
3770         struct kvm_segment segvar = {
3771                 .base = selector << 4,
3772                 .limit = 0xffff,
3773                 .selector = selector,
3774                 .type = 3,
3775                 .present = 1,
3776                 .dpl = 3,
3777                 .db = 0,
3778                 .s = 1,
3779                 .l = 0,
3780                 .g = 0,
3781                 .avl = 0,
3782                 .unusable = 0,
3783         };
3784         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3785         return 0;
3786 }
3787
3788 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3789                                 int type_bits, int seg)
3790 {
3791         struct kvm_segment kvm_seg;
3792
3793         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3794                 return kvm_load_realmode_segment(vcpu, selector, seg);
3795         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3796                 return 1;
3797         kvm_seg.type |= type_bits;
3798
3799         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3800             seg != VCPU_SREG_LDTR)
3801                 if (!kvm_seg.s)
3802                         kvm_seg.unusable = 1;
3803
3804         kvm_set_segment(vcpu, &kvm_seg, seg);
3805         return 0;
3806 }
3807
3808 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3809                                 struct tss_segment_32 *tss)
3810 {
3811         tss->cr3 = vcpu->arch.cr3;
3812         tss->eip = kvm_rip_read(vcpu);
3813         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3814         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3815         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3816         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3817         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3818         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3819         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3820         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3821         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3822         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3823         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3824         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3825         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3826         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3827         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3828         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3829 }
3830
3831 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3832                                   struct tss_segment_32 *tss)
3833 {
3834         kvm_set_cr3(vcpu, tss->cr3);
3835
3836         kvm_rip_write(vcpu, tss->eip);
3837         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3838
3839         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3840         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3841         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3842         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3843         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3844         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3845         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3846         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3847
3848         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3849                 return 1;
3850
3851         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3852                 return 1;
3853
3854         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3855                 return 1;
3856
3857         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3858                 return 1;
3859
3860         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3861                 return 1;
3862
3863         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3864                 return 1;
3865
3866         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3867                 return 1;
3868         return 0;
3869 }
3870
3871 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3872                                 struct tss_segment_16 *tss)
3873 {
3874         tss->ip = kvm_rip_read(vcpu);
3875         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3876         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3877         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3878         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3879         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3880         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3881         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3882         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3883         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3884
3885         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3886         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3887         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3888         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3889         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3890         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3891 }
3892
3893 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3894                                  struct tss_segment_16 *tss)
3895 {
3896         kvm_rip_write(vcpu, tss->ip);
3897         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3898         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3899         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3900         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3901         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3902         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3903         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3904         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3905         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3906
3907         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3908                 return 1;
3909
3910         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3911                 return 1;
3912
3913         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3914                 return 1;
3915
3916         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3917                 return 1;
3918
3919         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3920                 return 1;
3921         return 0;
3922 }
3923
3924 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3925                               u16 old_tss_sel, u32 old_tss_base,
3926                               struct desc_struct *nseg_desc)
3927 {
3928         struct tss_segment_16 tss_segment_16;
3929         int ret = 0;
3930
3931         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3932                            sizeof tss_segment_16))
3933                 goto out;
3934
3935         save_state_to_tss16(vcpu, &tss_segment_16);
3936
3937         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3938                             sizeof tss_segment_16))
3939                 goto out;
3940
3941         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3942                            &tss_segment_16, sizeof tss_segment_16))
3943                 goto out;
3944
3945         if (old_tss_sel != 0xffff) {
3946                 tss_segment_16.prev_task_link = old_tss_sel;
3947
3948                 if (kvm_write_guest(vcpu->kvm,
3949                                     get_tss_base_addr(vcpu, nseg_desc),
3950                                     &tss_segment_16.prev_task_link,
3951                                     sizeof tss_segment_16.prev_task_link))
3952                         goto out;
3953         }
3954
3955         if (load_state_from_tss16(vcpu, &tss_segment_16))
3956                 goto out;
3957
3958         ret = 1;
3959 out:
3960         return ret;
3961 }
3962
3963 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3964                        u16 old_tss_sel, u32 old_tss_base,
3965                        struct desc_struct *nseg_desc)
3966 {
3967         struct tss_segment_32 tss_segment_32;
3968         int ret = 0;
3969
3970         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3971                            sizeof tss_segment_32))
3972                 goto out;
3973
3974         save_state_to_tss32(vcpu, &tss_segment_32);
3975
3976         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3977                             sizeof tss_segment_32))
3978                 goto out;
3979
3980         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3981                            &tss_segment_32, sizeof tss_segment_32))
3982                 goto out;
3983
3984         if (old_tss_sel != 0xffff) {
3985                 tss_segment_32.prev_task_link = old_tss_sel;
3986
3987                 if (kvm_write_guest(vcpu->kvm,
3988                                     get_tss_base_addr(vcpu, nseg_desc),
3989                                     &tss_segment_32.prev_task_link,
3990                                     sizeof tss_segment_32.prev_task_link))
3991                         goto out;
3992         }
3993
3994         if (load_state_from_tss32(vcpu, &tss_segment_32))
3995                 goto out;
3996
3997         ret = 1;
3998 out:
3999         return ret;
4000 }
4001
4002 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4003 {
4004         struct kvm_segment tr_seg;
4005         struct desc_struct cseg_desc;
4006         struct desc_struct nseg_desc;
4007         int ret = 0;
4008         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4009         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4010
4011         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4012
4013         /* FIXME: Handle errors. Failure to read either TSS or their
4014          * descriptors should generate a pagefault.
4015          */
4016         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4017                 goto out;
4018
4019         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4020                 goto out;
4021
4022         if (reason != TASK_SWITCH_IRET) {
4023                 int cpl;
4024
4025                 cpl = kvm_x86_ops->get_cpl(vcpu);
4026                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4027                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4028                         return 1;
4029                 }
4030         }
4031
4032         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4033                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4034                 return 1;
4035         }
4036
4037         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4038                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4039                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4040         }
4041
4042         if (reason == TASK_SWITCH_IRET) {
4043                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4044                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4045         }
4046
4047         /* set back link to prev task only if NT bit is set in eflags
4048            note that old_tss_sel is not used afetr this point */
4049         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4050                 old_tss_sel = 0xffff;
4051
4052         /* set back link to prev task only if NT bit is set in eflags
4053            note that old_tss_sel is not used afetr this point */
4054         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4055                 old_tss_sel = 0xffff;
4056
4057         if (nseg_desc.type & 8)
4058                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4059                                          old_tss_base, &nseg_desc);
4060         else
4061                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4062                                          old_tss_base, &nseg_desc);
4063
4064         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4065                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4066                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4067         }
4068
4069         if (reason != TASK_SWITCH_IRET) {
4070                 nseg_desc.type |= (1 << 1);
4071                 save_guest_segment_descriptor(vcpu, tss_selector,
4072                                               &nseg_desc);
4073         }
4074
4075         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4076         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4077         tr_seg.type = 11;
4078         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4079 out:
4080         return ret;
4081 }
4082 EXPORT_SYMBOL_GPL(kvm_task_switch);
4083
4084 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4085                                   struct kvm_sregs *sregs)
4086 {
4087         int mmu_reset_needed = 0;
4088         int pending_vec, max_bits;
4089         struct descriptor_table dt;
4090
4091         vcpu_load(vcpu);
4092
4093         dt.limit = sregs->idt.limit;
4094         dt.base = sregs->idt.base;
4095         kvm_x86_ops->set_idt(vcpu, &dt);
4096         dt.limit = sregs->gdt.limit;
4097         dt.base = sregs->gdt.base;
4098         kvm_x86_ops->set_gdt(vcpu, &dt);
4099
4100         vcpu->arch.cr2 = sregs->cr2;
4101         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4102
4103         down_read(&vcpu->kvm->slots_lock);
4104         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4105                 vcpu->arch.cr3 = sregs->cr3;
4106         else
4107                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4108         up_read(&vcpu->kvm->slots_lock);
4109
4110         kvm_set_cr8(vcpu, sregs->cr8);
4111
4112         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4113         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4114         kvm_set_apic_base(vcpu, sregs->apic_base);
4115
4116         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4117
4118         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4119         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4120         vcpu->arch.cr0 = sregs->cr0;
4121
4122         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4123         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4124         if (!is_long_mode(vcpu) && is_pae(vcpu))
4125                 load_pdptrs(vcpu, vcpu->arch.cr3);
4126
4127         if (mmu_reset_needed)
4128                 kvm_mmu_reset_context(vcpu);
4129
4130         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4131         pending_vec = find_first_bit(
4132                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4133         if (pending_vec < max_bits) {
4134                 kvm_queue_interrupt(vcpu, pending_vec, false);
4135                 pr_debug("Set back pending irq %d\n", pending_vec);
4136                 if (irqchip_in_kernel(vcpu->kvm))
4137                         kvm_pic_clear_isr_ack(vcpu->kvm);
4138         }
4139
4140         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4141         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4142         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4143         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4144         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4145         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4146
4147         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4148         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4149
4150         /* Older userspace won't unhalt the vcpu on reset. */
4151         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4152             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4153             !(vcpu->arch.cr0 & X86_CR0_PE))
4154                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4155
4156         vcpu_put(vcpu);
4157
4158         return 0;
4159 }
4160
4161 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4162                                         struct kvm_guest_debug *dbg)
4163 {
4164         int i, r;
4165
4166         vcpu_load(vcpu);
4167
4168         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4169             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4170                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4171                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4172                 vcpu->arch.switch_db_regs =
4173                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4174         } else {
4175                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4176                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4177                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4178         }
4179
4180         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4181
4182         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4183                 kvm_queue_exception(vcpu, DB_VECTOR);
4184         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4185                 kvm_queue_exception(vcpu, BP_VECTOR);
4186
4187         vcpu_put(vcpu);
4188
4189         return r;
4190 }
4191
4192 /*
4193  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4194  * we have asm/x86/processor.h
4195  */
4196 struct fxsave {
4197         u16     cwd;
4198         u16     swd;
4199         u16     twd;
4200         u16     fop;
4201         u64     rip;
4202         u64     rdp;
4203         u32     mxcsr;
4204         u32     mxcsr_mask;
4205         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4206 #ifdef CONFIG_X86_64
4207         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4208 #else
4209         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4210 #endif
4211 };
4212
4213 /*
4214  * Translate a guest virtual address to a guest physical address.
4215  */
4216 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4217                                     struct kvm_translation *tr)
4218 {
4219         unsigned long vaddr = tr->linear_address;
4220         gpa_t gpa;
4221
4222         vcpu_load(vcpu);
4223         down_read(&vcpu->kvm->slots_lock);
4224         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4225         up_read(&vcpu->kvm->slots_lock);
4226         tr->physical_address = gpa;
4227         tr->valid = gpa != UNMAPPED_GVA;
4228         tr->writeable = 1;
4229         tr->usermode = 0;
4230         vcpu_put(vcpu);
4231
4232         return 0;
4233 }
4234
4235 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4236 {
4237         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4238
4239         vcpu_load(vcpu);
4240
4241         memcpy(fpu->fpr, fxsave->st_space, 128);
4242         fpu->fcw = fxsave->cwd;
4243         fpu->fsw = fxsave->swd;
4244         fpu->ftwx = fxsave->twd;
4245         fpu->last_opcode = fxsave->fop;
4246         fpu->last_ip = fxsave->rip;
4247         fpu->last_dp = fxsave->rdp;
4248         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4249
4250         vcpu_put(vcpu);
4251
4252         return 0;
4253 }
4254
4255 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4256 {
4257         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4258
4259         vcpu_load(vcpu);
4260
4261         memcpy(fxsave->st_space, fpu->fpr, 128);
4262         fxsave->cwd = fpu->fcw;
4263         fxsave->swd = fpu->fsw;
4264         fxsave->twd = fpu->ftwx;
4265         fxsave->fop = fpu->last_opcode;
4266         fxsave->rip = fpu->last_ip;
4267         fxsave->rdp = fpu->last_dp;
4268         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4269
4270         vcpu_put(vcpu);
4271
4272         return 0;
4273 }
4274
4275 void fx_init(struct kvm_vcpu *vcpu)
4276 {
4277         unsigned after_mxcsr_mask;
4278
4279         /*
4280          * Touch the fpu the first time in non atomic context as if
4281          * this is the first fpu instruction the exception handler
4282          * will fire before the instruction returns and it'll have to
4283          * allocate ram with GFP_KERNEL.
4284          */
4285         if (!used_math())
4286                 kvm_fx_save(&vcpu->arch.host_fx_image);
4287
4288         /* Initialize guest FPU by resetting ours and saving into guest's */
4289         preempt_disable();
4290         kvm_fx_save(&vcpu->arch.host_fx_image);
4291         kvm_fx_finit();
4292         kvm_fx_save(&vcpu->arch.guest_fx_image);
4293         kvm_fx_restore(&vcpu->arch.host_fx_image);
4294         preempt_enable();
4295
4296         vcpu->arch.cr0 |= X86_CR0_ET;
4297         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4298         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4299         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4300                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4301 }
4302 EXPORT_SYMBOL_GPL(fx_init);
4303
4304 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4305 {
4306         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4307                 return;
4308
4309         vcpu->guest_fpu_loaded = 1;
4310         kvm_fx_save(&vcpu->arch.host_fx_image);
4311         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4312 }
4313 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4314
4315 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4316 {
4317         if (!vcpu->guest_fpu_loaded)
4318                 return;
4319
4320         vcpu->guest_fpu_loaded = 0;
4321         kvm_fx_save(&vcpu->arch.guest_fx_image);
4322         kvm_fx_restore(&vcpu->arch.host_fx_image);
4323         ++vcpu->stat.fpu_reload;
4324 }
4325 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4326
4327 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4328 {
4329         if (vcpu->arch.time_page) {
4330                 kvm_release_page_dirty(vcpu->arch.time_page);
4331                 vcpu->arch.time_page = NULL;
4332         }
4333
4334         kvm_x86_ops->vcpu_free(vcpu);
4335 }
4336
4337 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4338                                                 unsigned int id)
4339 {
4340         return kvm_x86_ops->vcpu_create(kvm, id);
4341 }
4342
4343 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4344 {
4345         int r;
4346
4347         /* We do fxsave: this must be aligned. */
4348         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4349
4350         vcpu->arch.mtrr_state.have_fixed = 1;
4351         vcpu_load(vcpu);
4352         r = kvm_arch_vcpu_reset(vcpu);
4353         if (r == 0)
4354                 r = kvm_mmu_setup(vcpu);
4355         vcpu_put(vcpu);
4356         if (r < 0)
4357                 goto free_vcpu;
4358
4359         return 0;
4360 free_vcpu:
4361         kvm_x86_ops->vcpu_free(vcpu);
4362         return r;
4363 }
4364
4365 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4366 {
4367         vcpu_load(vcpu);
4368         kvm_mmu_unload(vcpu);
4369         vcpu_put(vcpu);
4370
4371         kvm_x86_ops->vcpu_free(vcpu);
4372 }
4373
4374 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4375 {
4376         vcpu->arch.nmi_pending = false;
4377         vcpu->arch.nmi_injected = false;
4378
4379         vcpu->arch.switch_db_regs = 0;
4380         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4381         vcpu->arch.dr6 = DR6_FIXED_1;
4382         vcpu->arch.dr7 = DR7_FIXED_1;
4383
4384         return kvm_x86_ops->vcpu_reset(vcpu);
4385 }
4386
4387 void kvm_arch_hardware_enable(void *garbage)
4388 {
4389         kvm_x86_ops->hardware_enable(garbage);
4390 }
4391
4392 void kvm_arch_hardware_disable(void *garbage)
4393 {
4394         kvm_x86_ops->hardware_disable(garbage);
4395 }
4396
4397 int kvm_arch_hardware_setup(void)
4398 {
4399         return kvm_x86_ops->hardware_setup();
4400 }
4401
4402 void kvm_arch_hardware_unsetup(void)
4403 {
4404         kvm_x86_ops->hardware_unsetup();
4405 }
4406
4407 void kvm_arch_check_processor_compat(void *rtn)
4408 {
4409         kvm_x86_ops->check_processor_compatibility(rtn);
4410 }
4411
4412 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4413 {
4414         struct page *page;
4415         struct kvm *kvm;
4416         int r;
4417
4418         BUG_ON(vcpu->kvm == NULL);
4419         kvm = vcpu->kvm;
4420
4421         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4422         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4423                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4424         else
4425                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4426
4427         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4428         if (!page) {
4429                 r = -ENOMEM;
4430                 goto fail;
4431         }
4432         vcpu->arch.pio_data = page_address(page);
4433
4434         r = kvm_mmu_create(vcpu);
4435         if (r < 0)
4436                 goto fail_free_pio_data;
4437
4438         if (irqchip_in_kernel(kvm)) {
4439                 r = kvm_create_lapic(vcpu);
4440                 if (r < 0)
4441                         goto fail_mmu_destroy;
4442         }
4443
4444         return 0;
4445
4446 fail_mmu_destroy:
4447         kvm_mmu_destroy(vcpu);
4448 fail_free_pio_data:
4449         free_page((unsigned long)vcpu->arch.pio_data);
4450 fail:
4451         return r;
4452 }
4453
4454 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4455 {
4456         kvm_free_lapic(vcpu);
4457         down_read(&vcpu->kvm->slots_lock);
4458         kvm_mmu_destroy(vcpu);
4459         up_read(&vcpu->kvm->slots_lock);
4460         free_page((unsigned long)vcpu->arch.pio_data);
4461 }
4462
4463 struct  kvm *kvm_arch_create_vm(void)
4464 {
4465         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4466
4467         if (!kvm)
4468                 return ERR_PTR(-ENOMEM);
4469
4470         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4471         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4472
4473         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4474         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4475
4476         rdtscll(kvm->arch.vm_init_tsc);
4477
4478         return kvm;
4479 }
4480
4481 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4482 {
4483         vcpu_load(vcpu);
4484         kvm_mmu_unload(vcpu);
4485         vcpu_put(vcpu);
4486 }
4487
4488 static void kvm_free_vcpus(struct kvm *kvm)
4489 {
4490         unsigned int i;
4491
4492         /*
4493          * Unpin any mmu pages first.
4494          */
4495         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4496                 if (kvm->vcpus[i])
4497                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4498         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4499                 if (kvm->vcpus[i]) {
4500                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4501                         kvm->vcpus[i] = NULL;
4502                 }
4503         }
4504
4505 }
4506
4507 void kvm_arch_sync_events(struct kvm *kvm)
4508 {
4509         kvm_free_all_assigned_devices(kvm);
4510 }
4511
4512 void kvm_arch_destroy_vm(struct kvm *kvm)
4513 {
4514         kvm_iommu_unmap_guest(kvm);
4515         kvm_free_pit(kvm);
4516         kfree(kvm->arch.vpic);
4517         kfree(kvm->arch.vioapic);
4518         kvm_free_vcpus(kvm);
4519         kvm_free_physmem(kvm);
4520         if (kvm->arch.apic_access_page)
4521                 put_page(kvm->arch.apic_access_page);
4522         if (kvm->arch.ept_identity_pagetable)
4523                 put_page(kvm->arch.ept_identity_pagetable);
4524         kfree(kvm);
4525 }
4526
4527 int kvm_arch_set_memory_region(struct kvm *kvm,
4528                                 struct kvm_userspace_memory_region *mem,
4529                                 struct kvm_memory_slot old,
4530                                 int user_alloc)
4531 {
4532         int npages = mem->memory_size >> PAGE_SHIFT;
4533         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4534
4535         /*To keep backward compatibility with older userspace,
4536          *x86 needs to hanlde !user_alloc case.
4537          */
4538         if (!user_alloc) {
4539                 if (npages && !old.rmap) {
4540                         unsigned long userspace_addr;
4541
4542                         down_write(&current->mm->mmap_sem);
4543                         userspace_addr = do_mmap(NULL, 0,
4544                                                  npages * PAGE_SIZE,
4545                                                  PROT_READ | PROT_WRITE,
4546                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4547                                                  0);
4548                         up_write(&current->mm->mmap_sem);
4549
4550                         if (IS_ERR((void *)userspace_addr))
4551                                 return PTR_ERR((void *)userspace_addr);
4552
4553                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4554                         spin_lock(&kvm->mmu_lock);
4555                         memslot->userspace_addr = userspace_addr;
4556                         spin_unlock(&kvm->mmu_lock);
4557                 } else {
4558                         if (!old.user_alloc && old.rmap) {
4559                                 int ret;
4560
4561                                 down_write(&current->mm->mmap_sem);
4562                                 ret = do_munmap(current->mm, old.userspace_addr,
4563                                                 old.npages * PAGE_SIZE);
4564                                 up_write(&current->mm->mmap_sem);
4565                                 if (ret < 0)
4566                                         printk(KERN_WARNING
4567                                        "kvm_vm_ioctl_set_memory_region: "
4568                                        "failed to munmap memory\n");
4569                         }
4570                 }
4571         }
4572
4573         spin_lock(&kvm->mmu_lock);
4574         if (!kvm->arch.n_requested_mmu_pages) {
4575                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4576                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4577         }
4578
4579         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4580         spin_unlock(&kvm->mmu_lock);
4581         kvm_flush_remote_tlbs(kvm);
4582
4583         return 0;
4584 }
4585
4586 void kvm_arch_flush_shadow(struct kvm *kvm)
4587 {
4588         kvm_mmu_zap_all(kvm);
4589         kvm_reload_remote_mmus(kvm);
4590 }
4591
4592 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4593 {
4594         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4595                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4596                || vcpu->arch.nmi_pending;
4597 }
4598
4599 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4600 {
4601         int me;
4602         int cpu = vcpu->cpu;
4603
4604         if (waitqueue_active(&vcpu->wq)) {
4605                 wake_up_interruptible(&vcpu->wq);
4606                 ++vcpu->stat.halt_wakeup;
4607         }
4608
4609         me = get_cpu();
4610         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4611                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4612                         smp_send_reschedule(cpu);
4613         put_cpu();
4614 }
4615
4616 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4617 {
4618         return kvm_x86_ops->interrupt_allowed(vcpu);
4619 }