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         { "request_nmi", VCPU_STAT(request_nmi_exits) },
  95         { "irq_exits", VCPU_STAT(irq_exits) },
  96         { "host_state_reload", VCPU_STAT(host_state_reload) },
  97         { "efer_reload", VCPU_STAT(efer_reload) },
  98         { "fpu_reload", VCPU_STAT(fpu_reload) },
  99         { "insn_emulation", VCPU_STAT(insn_emulation) },
 100         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 101         { "irq_injections", VCPU_STAT(irq_injections) },
 102         { "nmi_injections", VCPU_STAT(nmi_injections) },
 103         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 104         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 105         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 106         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 107         { "mmu_flooded", VM_STAT(mmu_flooded) },
 108         { "mmu_recycled", VM_STAT(mmu_recycled) },
 109         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 110         { "mmu_unsync", VM_STAT(mmu_unsync) },
 111         { "mmu_unsync_global", VM_STAT(mmu_unsync_global) },
 112         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 113         { "largepages", VM_STAT(lpages) },
 114         { NULL }
 115 };
 116
 117 unsigned long segment_base(u16 selector)
 118 {
 119         struct descriptor_table gdt;
 120         struct desc_struct *d;
 121         unsigned long table_base;
 122         unsigned long v;
 123
 124         if (selector == 0)
 125                 return 0;
 126
 127         asm("sgdt %0" : "=m"(gdt));
 128         table_base = gdt.base;
 129
 130         if (selector & 4) {           /* from ldt */
 131                 u16 ldt_selector;
 132
 133                 asm("sldt %0" : "=g"(ldt_selector));
 134                 table_base = segment_base(ldt_selector);
 135         }
 136         d = (struct desc_struct *)(table_base + (selector & ~7));
 137         v = d->base0 | ((unsigned long)d->base1 << 16) |
 138                 ((unsigned long)d->base2 << 24);
 139 #ifdef CONFIG_X86_64
 140         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 141                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 142 #endif
 143         return v;
 144 }
 145 EXPORT_SYMBOL_GPL(segment_base);
 146
 147 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 148 {
 149         if (irqchip_in_kernel(vcpu->kvm))
 150                 return vcpu->arch.apic_base;
 151         else
 152                 return vcpu->arch.apic_base;
 153 }
 154 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 155
 156 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 157 {
 158         /* TODO: reserve bits check */
 159         if (irqchip_in_kernel(vcpu->kvm))
 160                 kvm_lapic_set_base(vcpu, data);
 161         else
 162                 vcpu->arch.apic_base = data;
 163 }
 164 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 165
 166 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 167 {
 168         WARN_ON(vcpu->arch.exception.pending);
 169         vcpu->arch.exception.pending = true;
 170         vcpu->arch.exception.has_error_code = false;
 171         vcpu->arch.exception.nr = nr;
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 174
 175 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 176                            u32 error_code)
 177 {
 178         ++vcpu->stat.pf_guest;
 179
 180         if (vcpu->arch.exception.pending) {
 181                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 182                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 183                                         " double fault 0x%lx\n", addr);
 184                         vcpu->arch.exception.nr = DF_VECTOR;
 185                         vcpu->arch.exception.error_code = 0;
 186                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 187                         /* triple fault -> shutdown */
 188                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 189                 }
 190                 return;
 191         }
 192         vcpu->arch.cr2 = addr;
 193         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 194 }
 195
 196 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 197 {
 198         vcpu->arch.nmi_pending = 1;
 199 }
 200 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 201
 202 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 203 {
 204         WARN_ON(vcpu->arch.exception.pending);
 205         vcpu->arch.exception.pending = true;
 206         vcpu->arch.exception.has_error_code = true;
 207         vcpu->arch.exception.nr = nr;
 208         vcpu->arch.exception.error_code = error_code;
 209 }
 210 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 211
 212 static void __queue_exception(struct kvm_vcpu *vcpu)
 213 {
 214         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 215                                      vcpu->arch.exception.has_error_code,
 216                                      vcpu->arch.exception.error_code);
 217 }
 218
 219 /*
 220  * Load the pae pdptrs.  Return true is they are all valid.
 221  */
 222 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 223 {
 224         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 225         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 226         int i;
 227         int ret;
 228         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 229
 230         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 231                                   offset * sizeof(u64), sizeof(pdpte));
 232         if (ret < 0) {
 233                 ret = 0;
 234                 goto out;
 235         }
 236         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 237                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 238                         ret = 0;
 239                         goto out;
 240                 }
 241         }
 242         ret = 1;
 243
 244         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 245 out:
 246
 247         return ret;
 248 }
 249 EXPORT_SYMBOL_GPL(load_pdptrs);
 250
 251 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 252 {
 253         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 254         bool changed = true;
 255         int r;
 256
 257         if (is_long_mode(vcpu) || !is_pae(vcpu))
 258                 return false;
 259
 260         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 261         if (r < 0)
 262                 goto out;
 263         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 264 out:
 265
 266         return changed;
 267 }
 268
 269 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 270 {
 271         if (cr0 & CR0_RESERVED_BITS) {
 272                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 273                        cr0, vcpu->arch.cr0);
 274                 kvm_inject_gp(vcpu, 0);
 275                 return;
 276         }
 277
 278         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 279                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 280                 kvm_inject_gp(vcpu, 0);
 281                 return;
 282         }
 283
 284         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 285                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 286                        "and a clear PE flag\n");
 287                 kvm_inject_gp(vcpu, 0);
 288                 return;
 289         }
 290
 291         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 292 #ifdef CONFIG_X86_64
 293                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 294                         int cs_db, cs_l;
 295
 296                         if (!is_pae(vcpu)) {
 297                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 298                                        "in long mode while PAE is disabled\n");
 299                                 kvm_inject_gp(vcpu, 0);
 300                                 return;
 301                         }
 302                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 303                         if (cs_l) {
 304                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 305                                        "in long mode while CS.L == 1\n");
 306                                 kvm_inject_gp(vcpu, 0);
 307                                 return;
 308
 309                         }
 310                 } else
 311 #endif
 312                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 313                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 314                                "reserved bits\n");
 315                         kvm_inject_gp(vcpu, 0);
 316                         return;
 317                 }
 318
 319         }
 320
 321         kvm_x86_ops->set_cr0(vcpu, cr0);
 322         vcpu->arch.cr0 = cr0;
 323
 324         kvm_mmu_sync_global(vcpu);
 325         kvm_mmu_reset_context(vcpu);
 326         return;
 327 }
 328 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 329
 330 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 331 {
 332         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 333         KVMTRACE_1D(LMSW, vcpu,
 334                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 335                     handler);
 336 }
 337 EXPORT_SYMBOL_GPL(kvm_lmsw);
 338
 339 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 340 {
 341         unsigned long old_cr4 = vcpu->arch.cr4;
 342         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 343
 344         if (cr4 & CR4_RESERVED_BITS) {
 345                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 346                 kvm_inject_gp(vcpu, 0);
 347                 return;
 348         }
 349
 350         if (is_long_mode(vcpu)) {
 351                 if (!(cr4 & X86_CR4_PAE)) {
 352                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 353                                "in long mode\n");
 354                         kvm_inject_gp(vcpu, 0);
 355                         return;
 356                 }
 357         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 358                    && ((cr4 ^ old_cr4) & pdptr_bits)
 359                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 360                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 361                 kvm_inject_gp(vcpu, 0);
 362                 return;
 363         }
 364
 365         if (cr4 & X86_CR4_VMXE) {
 366                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 367                 kvm_inject_gp(vcpu, 0);
 368                 return;
 369         }
 370         kvm_x86_ops->set_cr4(vcpu, cr4);
 371         vcpu->arch.cr4 = cr4;
 372         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 373         kvm_mmu_sync_global(vcpu);
 374         kvm_mmu_reset_context(vcpu);
 375 }
 376 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 377
 378 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 379 {
 380         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 381                 kvm_mmu_sync_roots(vcpu);
 382                 kvm_mmu_flush_tlb(vcpu);
 383                 return;
 384         }
 385
 386         if (is_long_mode(vcpu)) {
 387                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 388                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 389                         kvm_inject_gp(vcpu, 0);
 390                         return;
 391                 }
 392         } else {
 393                 if (is_pae(vcpu)) {
 394                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 395                                 printk(KERN_DEBUG
 396                                        "set_cr3: #GP, reserved bits\n");
 397                                 kvm_inject_gp(vcpu, 0);
 398                                 return;
 399                         }
 400                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 401                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 402                                        "reserved bits\n");
 403                                 kvm_inject_gp(vcpu, 0);
 404                                 return;
 405                         }
 406                 }
 407                 /*
 408                  * We don't check reserved bits in nonpae mode, because
 409                  * this isn't enforced, and VMware depends on this.
 410                  */
 411         }
 412
 413         /*
 414          * Does the new cr3 value map to physical memory? (Note, we
 415          * catch an invalid cr3 even in real-mode, because it would
 416          * cause trouble later on when we turn on paging anyway.)
 417          *
 418          * A real CPU would silently accept an invalid cr3 and would
 419          * attempt to use it - with largely undefined (and often hard
 420          * to debug) behavior on the guest side.
 421          */
 422         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 423                 kvm_inject_gp(vcpu, 0);
 424         else {
 425                 vcpu->arch.cr3 = cr3;
 426                 vcpu->arch.mmu.new_cr3(vcpu);
 427         }
 428 }
 429 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 430
 431 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 432 {
 433         if (cr8 & CR8_RESERVED_BITS) {
 434                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 435                 kvm_inject_gp(vcpu, 0);
 436                 return;
 437         }
 438         if (irqchip_in_kernel(vcpu->kvm))
 439                 kvm_lapic_set_tpr(vcpu, cr8);
 440         else
 441                 vcpu->arch.cr8 = cr8;
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 444
 445 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 446 {
 447         if (irqchip_in_kernel(vcpu->kvm))
 448                 return kvm_lapic_get_cr8(vcpu);
 449         else
 450                 return vcpu->arch.cr8;
 451 }
 452 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 453
 454 static inline u32 bit(int bitno)
 455 {
 456         return 1 << (bitno & 31);
 457 }
 458
 459 /*
 460  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 461  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 462  *
 463  * This list is modified at module load time to reflect the
 464  * capabilities of the host cpu.
 465  */
 466 static u32 msrs_to_save[] = {
 467         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 468         MSR_K6_STAR,
 469 #ifdef CONFIG_X86_64
 470         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 471 #endif
 472         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 473         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 474 };
 475
 476 static unsigned num_msrs_to_save;
 477
 478 static u32 emulated_msrs[] = {
 479         MSR_IA32_MISC_ENABLE,
 480 };
 481
 482 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 483 {
 484         if (efer & efer_reserved_bits) {
 485                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 486                        efer);
 487                 kvm_inject_gp(vcpu, 0);
 488                 return;
 489         }
 490
 491         if (is_paging(vcpu)
 492             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 493                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 494                 kvm_inject_gp(vcpu, 0);
 495                 return;
 496         }
 497
 498         if (efer & EFER_FFXSR) {
 499                 struct kvm_cpuid_entry2 *feat;
 500
 501                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 502                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 503                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 504                         kvm_inject_gp(vcpu, 0);
 505                         return;
 506                 }
 507         }
 508
 509         if (efer & EFER_SVME) {
 510                 struct kvm_cpuid_entry2 *feat;
 511
 512                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 513                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 514                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 515                         kvm_inject_gp(vcpu, 0);
 516                         return;
 517                 }
 518         }
 519
 520         kvm_x86_ops->set_efer(vcpu, efer);
 521
 522         efer &= ~EFER_LMA;
 523         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 524
 525         vcpu->arch.shadow_efer = efer;
 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
 634         if ((!vcpu->time_page))
 635                 return;
 636
 637         preempt_disable();
 638         if (unlikely(vcpu->hv_clock_tsc_khz != __get_cpu_var(cpu_tsc_khz))) {
 639                 kvm_set_time_scale(__get_cpu_var(cpu_tsc_khz), &vcpu->hv_clock);
 640                 vcpu->hv_clock_tsc_khz = __get_cpu_var(cpu_tsc_khz);
 641         }
 642         preempt_enable();
 643
 644         /* Keep irq disabled to prevent changes to the clock */
 645         local_irq_save(flags);
 646         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 647                           &vcpu->hv_clock.tsc_timestamp);
 648         ktime_get_ts(&ts);
 649         local_irq_restore(flags);
 650
 651         /* With all the info we got, fill in the values */
 652
 653         vcpu->hv_clock.system_time = ts.tv_nsec +
 654                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 655         /*
 656          * The interface expects us to write an even number signaling that the
 657          * update is finished. Since the guest won't see the intermediate
 658          * state, we just increase by 2 at the end.
 659          */
 660         vcpu->hv_clock.version += 2;
 661
 662         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 663
 664         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 665                sizeof(vcpu->hv_clock));
 666
 667         kunmap_atomic(shared_kaddr, KM_USER0);
 668
 669         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 670 }
 671
 672 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 673 {
 674         struct kvm_vcpu_arch *vcpu = &v->arch;
 675
 676         if (!vcpu->time_page)
 677                 return 0;
 678         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 679         return 1;
 680 }
 681
 682 static bool msr_mtrr_valid(unsigned msr)
 683 {
 684         switch (msr) {
 685         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 686         case MSR_MTRRfix64K_00000:
 687         case MSR_MTRRfix16K_80000:
 688         case MSR_MTRRfix16K_A0000:
 689         case MSR_MTRRfix4K_C0000:
 690         case MSR_MTRRfix4K_C8000:
 691         case MSR_MTRRfix4K_D0000:
 692         case MSR_MTRRfix4K_D8000:
 693         case MSR_MTRRfix4K_E0000:
 694         case MSR_MTRRfix4K_E8000:
 695         case MSR_MTRRfix4K_F0000:
 696         case MSR_MTRRfix4K_F8000:
 697         case MSR_MTRRdefType:
 698         case MSR_IA32_CR_PAT:
 699                 return true;
 700         case 0x2f8:
 701                 return true;
 702         }
 703         return false;
 704 }
 705
 706 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 707 {
 708         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 709
 710         if (!msr_mtrr_valid(msr))
 711                 return 1;
 712
 713         if (msr == MSR_MTRRdefType) {
 714                 vcpu->arch.mtrr_state.def_type = data;
 715                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 716         } else if (msr == MSR_MTRRfix64K_00000)
 717                 p[0] = data;
 718         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 719                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 720         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 721                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 722         else if (msr == MSR_IA32_CR_PAT)
 723                 vcpu->arch.pat = data;
 724         else {  /* Variable MTRRs */
 725                 int idx, is_mtrr_mask;
 726                 u64 *pt;
 727
 728                 idx = (msr - 0x200) / 2;
 729                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 730                 if (!is_mtrr_mask)
 731                         pt =
 732                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 733                 else
 734                         pt =
 735                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 736                 *pt = data;
 737         }
 738
 739         kvm_mmu_reset_context(vcpu);
 740         return 0;
 741 }
 742
 743 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 744 {
 745         switch (msr) {
 746         case MSR_EFER:
 747                 set_efer(vcpu, data);
 748                 break;
 749         case MSR_IA32_MC0_STATUS:
 750                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 751                        __func__, data);
 752                 break;
 753         case MSR_IA32_MCG_STATUS:
 754                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 755                         __func__, data);
 756                 break;
 757         case MSR_IA32_MCG_CTL:
 758                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 759                         __func__, data);
 760                 break;
 761         case MSR_IA32_DEBUGCTLMSR:
 762                 if (!data) {
 763                         /* We support the non-activated case already */
 764                         break;
 765                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 766                         /* Values other than LBR and BTF are vendor-specific,
 767                            thus reserved and should throw a #GP */
 768                         return 1;
 769                 }
 770                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 771                         __func__, data);
 772                 break;
 773         case MSR_IA32_UCODE_REV:
 774         case MSR_IA32_UCODE_WRITE:
 775         case MSR_VM_HSAVE_PA:
 776                 break;
 777         case 0x200 ... 0x2ff:
 778                 return set_msr_mtrr(vcpu, msr, data);
 779         case MSR_IA32_APICBASE:
 780                 kvm_set_apic_base(vcpu, data);
 781                 break;
 782         case MSR_IA32_MISC_ENABLE:
 783                 vcpu->arch.ia32_misc_enable_msr = data;
 784                 break;
 785         case MSR_KVM_WALL_CLOCK:
 786                 vcpu->kvm->arch.wall_clock = data;
 787                 kvm_write_wall_clock(vcpu->kvm, data);
 788                 break;
 789         case MSR_KVM_SYSTEM_TIME: {
 790                 if (vcpu->arch.time_page) {
 791                         kvm_release_page_dirty(vcpu->arch.time_page);
 792                         vcpu->arch.time_page = NULL;
 793                 }
 794
 795                 vcpu->arch.time = data;
 796
 797                 /* we verify if the enable bit is set... */
 798                 if (!(data & 1))
 799                         break;
 800
 801                 /* ...but clean it before doing the actual write */
 802                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 803
 804                 vcpu->arch.time_page =
 805                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 806
 807                 if (is_error_page(vcpu->arch.time_page)) {
 808                         kvm_release_page_clean(vcpu->arch.time_page);
 809                         vcpu->arch.time_page = NULL;
 810                 }
 811
 812                 kvm_request_guest_time_update(vcpu);
 813                 break;
 814         }
 815         default:
 816                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 817                 return 1;
 818         }
 819         return 0;
 820 }
 821 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 822
 823
 824 /*
 825  * Reads an msr value (of 'msr_index') into 'pdata'.
 826  * Returns 0 on success, non-0 otherwise.
 827  * Assumes vcpu_load() was already called.
 828  */
 829 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 830 {
 831         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 832 }
 833
 834 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 835 {
 836         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 837
 838         if (!msr_mtrr_valid(msr))
 839                 return 1;
 840
 841         if (msr == MSR_MTRRdefType)
 842                 *pdata = vcpu->arch.mtrr_state.def_type +
 843                          (vcpu->arch.mtrr_state.enabled << 10);
 844         else if (msr == MSR_MTRRfix64K_00000)
 845                 *pdata = p[0];
 846         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 847                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 848         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 849                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 850         else if (msr == MSR_IA32_CR_PAT)
 851                 *pdata = vcpu->arch.pat;
 852         else {  /* Variable MTRRs */
 853                 int idx, is_mtrr_mask;
 854                 u64 *pt;
 855
 856                 idx = (msr - 0x200) / 2;
 857                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 858                 if (!is_mtrr_mask)
 859                         pt =
 860                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 861                 else
 862                         pt =
 863                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 864                 *pdata = *pt;
 865         }
 866
 867         return 0;
 868 }
 869
 870 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 871 {
 872         u64 data;
 873
 874         switch (msr) {
 875         case 0xc0010010: /* SYSCFG */
 876         case 0xc0010015: /* HWCR */
 877         case MSR_IA32_PLATFORM_ID:
 878         case MSR_IA32_P5_MC_ADDR:
 879         case MSR_IA32_P5_MC_TYPE:
 880         case MSR_IA32_MC0_CTL:
 881         case MSR_IA32_MCG_STATUS:
 882         case MSR_IA32_MCG_CAP:
 883         case MSR_IA32_MCG_CTL:
 884         case MSR_IA32_MC0_MISC:
 885         case MSR_IA32_MC0_MISC+4:
 886         case MSR_IA32_MC0_MISC+8:
 887         case MSR_IA32_MC0_MISC+12:
 888         case MSR_IA32_MC0_MISC+16:
 889         case MSR_IA32_MC0_MISC+20:
 890         case MSR_IA32_UCODE_REV:
 891         case MSR_IA32_EBL_CR_POWERON:
 892         case MSR_IA32_DEBUGCTLMSR:
 893         case MSR_IA32_LASTBRANCHFROMIP:
 894         case MSR_IA32_LASTBRANCHTOIP:
 895         case MSR_IA32_LASTINTFROMIP:
 896         case MSR_IA32_LASTINTTOIP:
 897         case MSR_VM_HSAVE_PA:
 898                 data = 0;
 899                 break;
 900         case MSR_MTRRcap:
 901                 data = 0x500 | KVM_NR_VAR_MTRR;
 902                 break;
 903         case 0x200 ... 0x2ff:
 904                 return get_msr_mtrr(vcpu, msr, pdata);
 905         case 0xcd: /* fsb frequency */
 906                 data = 3;
 907                 break;
 908         case MSR_IA32_APICBASE:
 909                 data = kvm_get_apic_base(vcpu);
 910                 break;
 911         case MSR_IA32_MISC_ENABLE:
 912                 data = vcpu->arch.ia32_misc_enable_msr;
 913                 break;
 914         case MSR_IA32_PERF_STATUS:
 915                 /* TSC increment by tick */
 916                 data = 1000ULL;
 917                 /* CPU multiplier */
 918                 data |= (((uint64_t)4ULL) << 40);
 919                 break;
 920         case MSR_EFER:
 921                 data = vcpu->arch.shadow_efer;
 922                 break;
 923         case MSR_KVM_WALL_CLOCK:
 924                 data = vcpu->kvm->arch.wall_clock;
 925                 break;
 926         case MSR_KVM_SYSTEM_TIME:
 927                 data = vcpu->arch.time;
 928                 break;
 929         default:
 930                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 931                 return 1;
 932         }
 933         *pdata = data;
 934         return 0;
 935 }
 936 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 937
 938 /*
 939  * Read or write a bunch of msrs. All parameters are kernel addresses.
 940  *
 941  * @return number of msrs set successfully.
 942  */
 943 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 944                     struct kvm_msr_entry *entries,
 945                     int (*do_msr)(struct kvm_vcpu *vcpu,
 946                                   unsigned index, u64 *data))
 947 {
 948         int i;
 949
 950         vcpu_load(vcpu);
 951
 952         down_read(&vcpu->kvm->slots_lock);
 953         for (i = 0; i < msrs->nmsrs; ++i)
 954                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 955                         break;
 956         up_read(&vcpu->kvm->slots_lock);
 957
 958         vcpu_put(vcpu);
 959
 960         return i;
 961 }
 962
 963 /*
 964  * Read or write a bunch of msrs. Parameters are user addresses.
 965  *
 966  * @return number of msrs set successfully.
 967  */
 968 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 969                   int (*do_msr)(struct kvm_vcpu *vcpu,
 970                                 unsigned index, u64 *data),
 971                   int writeback)
 972 {
 973         struct kvm_msrs msrs;
 974         struct kvm_msr_entry *entries;
 975         int r, n;
 976         unsigned size;
 977
 978         r = -EFAULT;
 979         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 980                 goto out;
 981
 982         r = -E2BIG;
 983         if (msrs.nmsrs >= MAX_IO_MSRS)
 984                 goto out;
 985
 986         r = -ENOMEM;
 987         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 988         entries = vmalloc(size);
 989         if (!entries)
 990                 goto out;
 991
 992         r = -EFAULT;
 993         if (copy_from_user(entries, user_msrs->entries, size))
 994                 goto out_free;
 995
 996         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 997         if (r < 0)
 998                 goto out_free;
 999
1000         r = -EFAULT;
1001         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1002                 goto out_free;
1003
1004         r = n;
1005
1006 out_free:
1007         vfree(entries);
1008 out:
1009         return r;
1010 }
1011
1012 int kvm_dev_ioctl_check_extension(long ext)
1013 {
1014         int r;
1015
1016         switch (ext) {
1017         case KVM_CAP_IRQCHIP:
1018         case KVM_CAP_HLT:
1019         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1020         case KVM_CAP_SET_TSS_ADDR:
1021         case KVM_CAP_EXT_CPUID:
1022         case KVM_CAP_CLOCKSOURCE:
1023         case KVM_CAP_PIT:
1024         case KVM_CAP_NOP_IO_DELAY:
1025         case KVM_CAP_MP_STATE:
1026         case KVM_CAP_SYNC_MMU:
1027         case KVM_CAP_REINJECT_CONTROL:
1028         case KVM_CAP_IRQ_INJECT_STATUS:
1029                 r = 1;
1030                 break;
1031         case KVM_CAP_COALESCED_MMIO:
1032                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1033                 break;
1034         case KVM_CAP_VAPIC:
1035                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1036                 break;
1037         case KVM_CAP_NR_VCPUS:
1038                 r = KVM_MAX_VCPUS;
1039                 break;
1040         case KVM_CAP_NR_MEMSLOTS:
1041                 r = KVM_MEMORY_SLOTS;
1042                 break;
1043         case KVM_CAP_PV_MMU:
1044                 r = !tdp_enabled;
1045                 break;
1046         case KVM_CAP_IOMMU:
1047                 r = iommu_found();
1048                 break;
1049         default:
1050                 r = 0;
1051                 break;
1052         }
1053         return r;
1054
1055 }
1056
1057 long kvm_arch_dev_ioctl(struct file *filp,
1058                         unsigned int ioctl, unsigned long arg)
1059 {
1060         void __user *argp = (void __user *)arg;
1061         long r;
1062
1063         switch (ioctl) {
1064         case KVM_GET_MSR_INDEX_LIST: {
1065                 struct kvm_msr_list __user *user_msr_list = argp;
1066                 struct kvm_msr_list msr_list;
1067                 unsigned n;
1068
1069                 r = -EFAULT;
1070                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1071                         goto out;
1072                 n = msr_list.nmsrs;
1073                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1074                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1075                         goto out;
1076                 r = -E2BIG;
1077                 if (n < num_msrs_to_save)
1078                         goto out;
1079                 r = -EFAULT;
1080                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1081                                  num_msrs_to_save * sizeof(u32)))
1082                         goto out;
1083                 if (copy_to_user(user_msr_list->indices
1084                                  + num_msrs_to_save * sizeof(u32),
1085                                  &emulated_msrs,
1086                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1087                         goto out;
1088                 r = 0;
1089                 break;
1090         }
1091         case KVM_GET_SUPPORTED_CPUID: {
1092                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1093                 struct kvm_cpuid2 cpuid;
1094
1095                 r = -EFAULT;
1096                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1097                         goto out;
1098                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1099                                                       cpuid_arg->entries);
1100                 if (r)
1101                         goto out;
1102
1103                 r = -EFAULT;
1104                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1105                         goto out;
1106                 r = 0;
1107                 break;
1108         }
1109         default:
1110                 r = -EINVAL;
1111         }
1112 out:
1113         return r;
1114 }
1115
1116 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1117 {
1118         kvm_x86_ops->vcpu_load(vcpu, cpu);
1119         kvm_request_guest_time_update(vcpu);
1120 }
1121
1122 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1123 {
1124         kvm_x86_ops->vcpu_put(vcpu);
1125         kvm_put_guest_fpu(vcpu);
1126 }
1127
1128 static int is_efer_nx(void)
1129 {
1130         unsigned long long efer = 0;
1131
1132         rdmsrl_safe(MSR_EFER, &efer);
1133         return efer & EFER_NX;
1134 }
1135
1136 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1137 {
1138         int i;
1139         struct kvm_cpuid_entry2 *e, *entry;
1140
1141         entry = NULL;
1142         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1143                 e = &vcpu->arch.cpuid_entries[i];
1144                 if (e->function == 0x80000001) {
1145                         entry = e;
1146                         break;
1147                 }
1148         }
1149         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1150                 entry->edx &= ~(1 << 20);
1151                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1152         }
1153 }
1154
1155 /* when an old userspace process fills a new kernel module */
1156 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1157                                     struct kvm_cpuid *cpuid,
1158                                     struct kvm_cpuid_entry __user *entries)
1159 {
1160         int r, i;
1161         struct kvm_cpuid_entry *cpuid_entries;
1162
1163         r = -E2BIG;
1164         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1165                 goto out;
1166         r = -ENOMEM;
1167         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1168         if (!cpuid_entries)
1169                 goto out;
1170         r = -EFAULT;
1171         if (copy_from_user(cpuid_entries, entries,
1172                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1173                 goto out_free;
1174         for (i = 0; i < cpuid->nent; i++) {
1175                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1176                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1177                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1178                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1179                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1180                 vcpu->arch.cpuid_entries[i].index = 0;
1181                 vcpu->arch.cpuid_entries[i].flags = 0;
1182                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1183                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1184                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1185         }
1186         vcpu->arch.cpuid_nent = cpuid->nent;
1187         cpuid_fix_nx_cap(vcpu);
1188         r = 0;
1189
1190 out_free:
1191         vfree(cpuid_entries);
1192 out:
1193         return r;
1194 }
1195
1196 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1197                                      struct kvm_cpuid2 *cpuid,
1198                                      struct kvm_cpuid_entry2 __user *entries)
1199 {
1200         int r;
1201
1202         r = -E2BIG;
1203         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1204                 goto out;
1205         r = -EFAULT;
1206         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1207                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1208                 goto out;
1209         vcpu->arch.cpuid_nent = cpuid->nent;
1210         return 0;
1211
1212 out:
1213         return r;
1214 }
1215
1216 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1217                                      struct kvm_cpuid2 *cpuid,
1218                                      struct kvm_cpuid_entry2 __user *entries)
1219 {
1220         int r;
1221
1222         r = -E2BIG;
1223         if (cpuid->nent < vcpu->arch.cpuid_nent)
1224                 goto out;
1225         r = -EFAULT;
1226         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1227                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1228                 goto out;
1229         return 0;
1230
1231 out:
1232         cpuid->nent = vcpu->arch.cpuid_nent;
1233         return r;
1234 }
1235
1236 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1237                            u32 index)
1238 {
1239         entry->function = function;
1240         entry->index = index;
1241         cpuid_count(entry->function, entry->index,
1242                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1243         entry->flags = 0;
1244 }
1245
1246 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1247                          u32 index, int *nent, int maxnent)
1248 {
1249         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1250                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1251                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1252                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1253                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1254                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1255                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1256                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1257                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1258                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1259         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1260                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1261                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1262                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1263                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1264                 bit(X86_FEATURE_PGE) |
1265                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1266                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1267                 bit(X86_FEATURE_SYSCALL) |
1268                 (is_efer_nx() ? bit(X86_FEATURE_NX) : 0) |
1269 #ifdef CONFIG_X86_64
1270                 bit(X86_FEATURE_LM) |
1271 #endif
1272                 bit(X86_FEATURE_FXSR_OPT) |
1273                 bit(X86_FEATURE_MMXEXT) |
1274                 bit(X86_FEATURE_3DNOWEXT) |
1275                 bit(X86_FEATURE_3DNOW);
1276         const u32 kvm_supported_word3_x86_features =
1277                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1278         const u32 kvm_supported_word6_x86_features =
1279                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1280                 bit(X86_FEATURE_SVM);
1281
1282         /* all calls to cpuid_count() should be made on the same cpu */
1283         get_cpu();
1284         do_cpuid_1_ent(entry, function, index);
1285         ++*nent;
1286
1287         switch (function) {
1288         case 0:
1289                 entry->eax = min(entry->eax, (u32)0xb);
1290                 break;
1291         case 1:
1292                 entry->edx &= kvm_supported_word0_x86_features;
1293                 entry->ecx &= kvm_supported_word3_x86_features;
1294                 break;
1295         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1296          * may return different values. This forces us to get_cpu() before
1297          * issuing the first command, and also to emulate this annoying behavior
1298          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1299         case 2: {
1300                 int t, times = entry->eax & 0xff;
1301
1302                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1303                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1304                 for (t = 1; t < times && *nent < maxnent; ++t) {
1305                         do_cpuid_1_ent(&entry[t], function, 0);
1306                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1307                         ++*nent;
1308                 }
1309                 break;
1310         }
1311         /* function 4 and 0xb have additional index. */
1312         case 4: {
1313                 int i, cache_type;
1314
1315                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1316                 /* read more entries until cache_type is zero */
1317                 for (i = 1; *nent < maxnent; ++i) {
1318                         cache_type = entry[i - 1].eax & 0x1f;
1319                         if (!cache_type)
1320                                 break;
1321                         do_cpuid_1_ent(&entry[i], function, i);
1322                         entry[i].flags |=
1323                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1324                         ++*nent;
1325                 }
1326                 break;
1327         }
1328         case 0xb: {
1329                 int i, level_type;
1330
1331                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1332                 /* read more entries until level_type is zero */
1333                 for (i = 1; *nent < maxnent; ++i) {
1334                         level_type = entry[i - 1].ecx & 0xff00;
1335                         if (!level_type)
1336                                 break;
1337                         do_cpuid_1_ent(&entry[i], function, i);
1338                         entry[i].flags |=
1339                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1340                         ++*nent;
1341                 }
1342                 break;
1343         }
1344         case 0x80000000:
1345                 entry->eax = min(entry->eax, 0x8000001a);
1346                 break;
1347         case 0x80000001:
1348                 entry->edx &= kvm_supported_word1_x86_features;
1349                 entry->ecx &= kvm_supported_word6_x86_features;
1350                 break;
1351         }
1352         put_cpu();
1353 }
1354
1355 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1356                                      struct kvm_cpuid_entry2 __user *entries)
1357 {
1358         struct kvm_cpuid_entry2 *cpuid_entries;
1359         int limit, nent = 0, r = -E2BIG;
1360         u32 func;
1361
1362         if (cpuid->nent < 1)
1363                 goto out;
1364         r = -ENOMEM;
1365         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1366         if (!cpuid_entries)
1367                 goto out;
1368
1369         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1370         limit = cpuid_entries[0].eax;
1371         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1372                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1373                              &nent, cpuid->nent);
1374         r = -E2BIG;
1375         if (nent >= cpuid->nent)
1376                 goto out_free;
1377
1378         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1379         limit = cpuid_entries[nent - 1].eax;
1380         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1381                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1382                              &nent, cpuid->nent);
1383         r = -EFAULT;
1384         if (copy_to_user(entries, cpuid_entries,
1385                          nent * sizeof(struct kvm_cpuid_entry2)))
1386                 goto out_free;
1387         cpuid->nent = nent;
1388         r = 0;
1389
1390 out_free:
1391         vfree(cpuid_entries);
1392 out:
1393         return r;
1394 }
1395
1396 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1397                                     struct kvm_lapic_state *s)
1398 {
1399         vcpu_load(vcpu);
1400         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1401         vcpu_put(vcpu);
1402
1403         return 0;
1404 }
1405
1406 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1407                                     struct kvm_lapic_state *s)
1408 {
1409         vcpu_load(vcpu);
1410         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1411         kvm_apic_post_state_restore(vcpu);
1412         vcpu_put(vcpu);
1413
1414         return 0;
1415 }
1416
1417 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1418                                     struct kvm_interrupt *irq)
1419 {
1420         if (irq->irq < 0 || irq->irq >= 256)
1421                 return -EINVAL;
1422         if (irqchip_in_kernel(vcpu->kvm))
1423                 return -ENXIO;
1424         vcpu_load(vcpu);
1425
1426         set_bit(irq->irq, vcpu->arch.irq_pending);
1427         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1428
1429         vcpu_put(vcpu);
1430
1431         return 0;
1432 }
1433
1434 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1435 {
1436         vcpu_load(vcpu);
1437         kvm_inject_nmi(vcpu);
1438         vcpu_put(vcpu);
1439
1440         return 0;
1441 }
1442
1443 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1444                                            struct kvm_tpr_access_ctl *tac)
1445 {
1446         if (tac->flags)
1447                 return -EINVAL;
1448         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1449         return 0;
1450 }
1451
1452 long kvm_arch_vcpu_ioctl(struct file *filp,
1453                          unsigned int ioctl, unsigned long arg)
1454 {
1455         struct kvm_vcpu *vcpu = filp->private_data;
1456         void __user *argp = (void __user *)arg;
1457         int r;
1458         struct kvm_lapic_state *lapic = NULL;
1459
1460         switch (ioctl) {
1461         case KVM_GET_LAPIC: {
1462                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1463
1464                 r = -ENOMEM;
1465                 if (!lapic)
1466                         goto out;
1467                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1468                 if (r)
1469                         goto out;
1470                 r = -EFAULT;
1471                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1472                         goto out;
1473                 r = 0;
1474                 break;
1475         }
1476         case KVM_SET_LAPIC: {
1477                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1478                 r = -ENOMEM;
1479                 if (!lapic)
1480                         goto out;
1481                 r = -EFAULT;
1482                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1483                         goto out;
1484                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1485                 if (r)
1486                         goto out;
1487                 r = 0;
1488                 break;
1489         }
1490         case KVM_INTERRUPT: {
1491                 struct kvm_interrupt irq;
1492
1493                 r = -EFAULT;
1494                 if (copy_from_user(&irq, argp, sizeof irq))
1495                         goto out;
1496                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1497                 if (r)
1498                         goto out;
1499                 r = 0;
1500                 break;
1501         }
1502         case KVM_NMI: {
1503                 r = kvm_vcpu_ioctl_nmi(vcpu);
1504                 if (r)
1505                         goto out;
1506                 r = 0;
1507                 break;
1508         }
1509         case KVM_SET_CPUID: {
1510                 struct kvm_cpuid __user *cpuid_arg = argp;
1511                 struct kvm_cpuid cpuid;
1512
1513                 r = -EFAULT;
1514                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1515                         goto out;
1516                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1517                 if (r)
1518                         goto out;
1519                 break;
1520         }
1521         case KVM_SET_CPUID2: {
1522                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1523                 struct kvm_cpuid2 cpuid;
1524
1525                 r = -EFAULT;
1526                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1527                         goto out;
1528                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1529                                               cpuid_arg->entries);
1530                 if (r)
1531                         goto out;
1532                 break;
1533         }
1534         case KVM_GET_CPUID2: {
1535                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1536                 struct kvm_cpuid2 cpuid;
1537
1538                 r = -EFAULT;
1539                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1540                         goto out;
1541                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1542                                               cpuid_arg->entries);
1543                 if (r)
1544                         goto out;
1545                 r = -EFAULT;
1546                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1547                         goto out;
1548                 r = 0;
1549                 break;
1550         }
1551         case KVM_GET_MSRS:
1552                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1553                 break;
1554         case KVM_SET_MSRS:
1555                 r = msr_io(vcpu, argp, do_set_msr, 0);
1556                 break;
1557         case KVM_TPR_ACCESS_REPORTING: {
1558                 struct kvm_tpr_access_ctl tac;
1559
1560                 r = -EFAULT;
1561                 if (copy_from_user(&tac, argp, sizeof tac))
1562                         goto out;
1563                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1564                 if (r)
1565                         goto out;
1566                 r = -EFAULT;
1567                 if (copy_to_user(argp, &tac, sizeof tac))
1568                         goto out;
1569                 r = 0;
1570                 break;
1571         };
1572         case KVM_SET_VAPIC_ADDR: {
1573                 struct kvm_vapic_addr va;
1574
1575                 r = -EINVAL;
1576                 if (!irqchip_in_kernel(vcpu->kvm))
1577                         goto out;
1578                 r = -EFAULT;
1579                 if (copy_from_user(&va, argp, sizeof va))
1580                         goto out;
1581                 r = 0;
1582                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1583                 break;
1584         }
1585         default:
1586                 r = -EINVAL;
1587         }
1588 out:
1589         if (lapic)
1590                 kfree(lapic);
1591         return r;
1592 }
1593
1594 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1595 {
1596         int ret;
1597
1598         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1599                 return -1;
1600         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1601         return ret;
1602 }
1603
1604 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1605                                           u32 kvm_nr_mmu_pages)
1606 {
1607         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1608                 return -EINVAL;
1609
1610         down_write(&kvm->slots_lock);
1611         spin_lock(&kvm->mmu_lock);
1612
1613         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1614         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1615
1616         spin_unlock(&kvm->mmu_lock);
1617         up_write(&kvm->slots_lock);
1618         return 0;
1619 }
1620
1621 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1622 {
1623         return kvm->arch.n_alloc_mmu_pages;
1624 }
1625
1626 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1627 {
1628         int i;
1629         struct kvm_mem_alias *alias;
1630
1631         for (i = 0; i < kvm->arch.naliases; ++i) {
1632                 alias = &kvm->arch.aliases[i];
1633                 if (gfn >= alias->base_gfn
1634                     && gfn < alias->base_gfn + alias->npages)
1635                         return alias->target_gfn + gfn - alias->base_gfn;
1636         }
1637         return gfn;
1638 }
1639
1640 /*
1641  * Set a new alias region.  Aliases map a portion of physical memory into
1642  * another portion.  This is useful for memory windows, for example the PC
1643  * VGA region.
1644  */
1645 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1646                                          struct kvm_memory_alias *alias)
1647 {
1648         int r, n;
1649         struct kvm_mem_alias *p;
1650
1651         r = -EINVAL;
1652         /* General sanity checks */
1653         if (alias->memory_size & (PAGE_SIZE - 1))
1654                 goto out;
1655         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1656                 goto out;
1657         if (alias->slot >= KVM_ALIAS_SLOTS)
1658                 goto out;
1659         if (alias->guest_phys_addr + alias->memory_size
1660             < alias->guest_phys_addr)
1661                 goto out;
1662         if (alias->target_phys_addr + alias->memory_size
1663             < alias->target_phys_addr)
1664                 goto out;
1665
1666         down_write(&kvm->slots_lock);
1667         spin_lock(&kvm->mmu_lock);
1668
1669         p = &kvm->arch.aliases[alias->slot];
1670         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1671         p->npages = alias->memory_size >> PAGE_SHIFT;
1672         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1673
1674         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1675                 if (kvm->arch.aliases[n - 1].npages)
1676                         break;
1677         kvm->arch.naliases = n;
1678
1679         spin_unlock(&kvm->mmu_lock);
1680         kvm_mmu_zap_all(kvm);
1681
1682         up_write(&kvm->slots_lock);
1683
1684         return 0;
1685
1686 out:
1687         return r;
1688 }
1689
1690 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1691 {
1692         int r;
1693
1694         r = 0;
1695         switch (chip->chip_id) {
1696         case KVM_IRQCHIP_PIC_MASTER:
1697                 memcpy(&chip->chip.pic,
1698                         &pic_irqchip(kvm)->pics[0],
1699                         sizeof(struct kvm_pic_state));
1700                 break;
1701         case KVM_IRQCHIP_PIC_SLAVE:
1702                 memcpy(&chip->chip.pic,
1703                         &pic_irqchip(kvm)->pics[1],
1704                         sizeof(struct kvm_pic_state));
1705                 break;
1706         case KVM_IRQCHIP_IOAPIC:
1707                 memcpy(&chip->chip.ioapic,
1708                         ioapic_irqchip(kvm),
1709                         sizeof(struct kvm_ioapic_state));
1710                 break;
1711         default:
1712                 r = -EINVAL;
1713                 break;
1714         }
1715         return r;
1716 }
1717
1718 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1719 {
1720         int r;
1721
1722         r = 0;
1723         switch (chip->chip_id) {
1724         case KVM_IRQCHIP_PIC_MASTER:
1725                 memcpy(&pic_irqchip(kvm)->pics[0],
1726                         &chip->chip.pic,
1727                         sizeof(struct kvm_pic_state));
1728                 break;
1729         case KVM_IRQCHIP_PIC_SLAVE:
1730                 memcpy(&pic_irqchip(kvm)->pics[1],
1731                         &chip->chip.pic,
1732                         sizeof(struct kvm_pic_state));
1733                 break;
1734         case KVM_IRQCHIP_IOAPIC:
1735                 memcpy(ioapic_irqchip(kvm),
1736                         &chip->chip.ioapic,
1737                         sizeof(struct kvm_ioapic_state));
1738                 break;
1739         default:
1740                 r = -EINVAL;
1741                 break;
1742         }
1743         kvm_pic_update_irq(pic_irqchip(kvm));
1744         return r;
1745 }
1746
1747 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1748 {
1749         int r = 0;
1750
1751         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1752         return r;
1753 }
1754
1755 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1756 {
1757         int r = 0;
1758
1759         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1760         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1761         return r;
1762 }
1763
1764 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1765                                  struct kvm_reinject_control *control)
1766 {
1767         if (!kvm->arch.vpit)
1768                 return -ENXIO;
1769         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1770         return 0;
1771 }
1772
1773 /*
1774  * Get (and clear) the dirty memory log for a memory slot.
1775  */
1776 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1777                                       struct kvm_dirty_log *log)
1778 {
1779         int r;
1780         int n;
1781         struct kvm_memory_slot *memslot;
1782         int is_dirty = 0;
1783
1784         down_write(&kvm->slots_lock);
1785
1786         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1787         if (r)
1788                 goto out;
1789
1790         /* If nothing is dirty, don't bother messing with page tables. */
1791         if (is_dirty) {
1792                 spin_lock(&kvm->mmu_lock);
1793                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1794                 spin_unlock(&kvm->mmu_lock);
1795                 kvm_flush_remote_tlbs(kvm);
1796                 memslot = &kvm->memslots[log->slot];
1797                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1798                 memset(memslot->dirty_bitmap, 0, n);
1799         }
1800         r = 0;
1801 out:
1802         up_write(&kvm->slots_lock);
1803         return r;
1804 }
1805
1806 long kvm_arch_vm_ioctl(struct file *filp,
1807                        unsigned int ioctl, unsigned long arg)
1808 {
1809         struct kvm *kvm = filp->private_data;
1810         void __user *argp = (void __user *)arg;
1811         int r = -EINVAL;
1812         /*
1813          * This union makes it completely explicit to gcc-3.x
1814          * that these two variables' stack usage should be
1815          * combined, not added together.
1816          */
1817         union {
1818                 struct kvm_pit_state ps;
1819                 struct kvm_memory_alias alias;
1820         } u;
1821
1822         switch (ioctl) {
1823         case KVM_SET_TSS_ADDR:
1824                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1825                 if (r < 0)
1826                         goto out;
1827                 break;
1828         case KVM_SET_MEMORY_REGION: {
1829                 struct kvm_memory_region kvm_mem;
1830                 struct kvm_userspace_memory_region kvm_userspace_mem;
1831
1832                 r = -EFAULT;
1833                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1834                         goto out;
1835                 kvm_userspace_mem.slot = kvm_mem.slot;
1836                 kvm_userspace_mem.flags = kvm_mem.flags;
1837                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1838                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1839                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1840                 if (r)
1841                         goto out;
1842                 break;
1843         }
1844         case KVM_SET_NR_MMU_PAGES:
1845                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1846                 if (r)
1847                         goto out;
1848                 break;
1849         case KVM_GET_NR_MMU_PAGES:
1850                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1851                 break;
1852         case KVM_SET_MEMORY_ALIAS:
1853                 r = -EFAULT;
1854                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1855                         goto out;
1856                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1857                 if (r)
1858                         goto out;
1859                 break;
1860         case KVM_CREATE_IRQCHIP:
1861                 r = -ENOMEM;
1862                 kvm->arch.vpic = kvm_create_pic(kvm);
1863                 if (kvm->arch.vpic) {
1864                         r = kvm_ioapic_init(kvm);
1865                         if (r) {
1866                                 kfree(kvm->arch.vpic);
1867                                 kvm->arch.vpic = NULL;
1868                                 goto out;
1869                         }
1870                 } else
1871                         goto out;
1872                 r = kvm_setup_default_irq_routing(kvm);
1873                 if (r) {
1874                         kfree(kvm->arch.vpic);
1875                         kfree(kvm->arch.vioapic);
1876                         goto out;
1877                 }
1878                 break;
1879         case KVM_CREATE_PIT:
1880                 mutex_lock(&kvm->lock);
1881                 r = -EEXIST;
1882                 if (kvm->arch.vpit)
1883                         goto create_pit_unlock;
1884                 r = -ENOMEM;
1885                 kvm->arch.vpit = kvm_create_pit(kvm);
1886                 if (kvm->arch.vpit)
1887                         r = 0;
1888         create_pit_unlock:
1889                 mutex_unlock(&kvm->lock);
1890                 break;
1891         case KVM_IRQ_LINE_STATUS:
1892         case KVM_IRQ_LINE: {
1893                 struct kvm_irq_level irq_event;
1894
1895                 r = -EFAULT;
1896                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1897                         goto out;
1898                 if (irqchip_in_kernel(kvm)) {
1899                         __s32 status;
1900                         mutex_lock(&kvm->lock);
1901                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1902                                         irq_event.irq, irq_event.level);
1903                         mutex_unlock(&kvm->lock);
1904                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1905                                 irq_event.status = status;
1906                                 if (copy_to_user(argp, &irq_event,
1907                                                         sizeof irq_event))
1908                                         goto out;
1909                         }
1910                         r = 0;
1911                 }
1912                 break;
1913         }
1914         case KVM_GET_IRQCHIP: {
1915                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1916                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1917
1918                 r = -ENOMEM;
1919                 if (!chip)
1920                         goto out;
1921                 r = -EFAULT;
1922                 if (copy_from_user(chip, argp, sizeof *chip))
1923                         goto get_irqchip_out;
1924                 r = -ENXIO;
1925                 if (!irqchip_in_kernel(kvm))
1926                         goto get_irqchip_out;
1927                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1928                 if (r)
1929                         goto get_irqchip_out;
1930                 r = -EFAULT;
1931                 if (copy_to_user(argp, chip, sizeof *chip))
1932                         goto get_irqchip_out;
1933                 r = 0;
1934         get_irqchip_out:
1935                 kfree(chip);
1936                 if (r)
1937                         goto out;
1938                 break;
1939         }
1940         case KVM_SET_IRQCHIP: {
1941                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1942                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1943
1944                 r = -ENOMEM;
1945                 if (!chip)
1946                         goto out;
1947                 r = -EFAULT;
1948                 if (copy_from_user(chip, argp, sizeof *chip))
1949                         goto set_irqchip_out;
1950                 r = -ENXIO;
1951                 if (!irqchip_in_kernel(kvm))
1952                         goto set_irqchip_out;
1953                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1954                 if (r)
1955                         goto set_irqchip_out;
1956                 r = 0;
1957         set_irqchip_out:
1958                 kfree(chip);
1959                 if (r)
1960                         goto out;
1961                 break;
1962         }
1963         case KVM_GET_PIT: {
1964                 r = -EFAULT;
1965                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1966                         goto out;
1967                 r = -ENXIO;
1968                 if (!kvm->arch.vpit)
1969                         goto out;
1970                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1971                 if (r)
1972                         goto out;
1973                 r = -EFAULT;
1974                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1975                         goto out;
1976                 r = 0;
1977                 break;
1978         }
1979         case KVM_SET_PIT: {
1980                 r = -EFAULT;
1981                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1982                         goto out;
1983                 r = -ENXIO;
1984                 if (!kvm->arch.vpit)
1985                         goto out;
1986                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1987                 if (r)
1988                         goto out;
1989                 r = 0;
1990                 break;
1991         }
1992         case KVM_REINJECT_CONTROL: {
1993                 struct kvm_reinject_control control;
1994                 r =  -EFAULT;
1995                 if (copy_from_user(&control, argp, sizeof(control)))
1996                         goto out;
1997                 r = kvm_vm_ioctl_reinject(kvm, &control);
1998                 if (r)
1999                         goto out;
2000                 r = 0;
2001                 break;
2002         }
2003         default:
2004                 ;
2005         }
2006 out:
2007         return r;
2008 }
2009
2010 static void kvm_init_msr_list(void)
2011 {
2012         u32 dummy[2];
2013         unsigned i, j;
2014
2015         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2016                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2017                         continue;
2018                 if (j < i)
2019                         msrs_to_save[j] = msrs_to_save[i];
2020                 j++;
2021         }
2022         num_msrs_to_save = j;
2023 }
2024
2025 /*
2026  * Only apic need an MMIO device hook, so shortcut now..
2027  */
2028 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2029                                                 gpa_t addr, int len,
2030                                                 int is_write)
2031 {
2032         struct kvm_io_device *dev;
2033
2034         if (vcpu->arch.apic) {
2035                 dev = &vcpu->arch.apic->dev;
2036                 if (dev->in_range(dev, addr, len, is_write))
2037                         return dev;
2038         }
2039         return NULL;
2040 }
2041
2042
2043 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2044                                                 gpa_t addr, int len,
2045                                                 int is_write)
2046 {
2047         struct kvm_io_device *dev;
2048
2049         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2050         if (dev == NULL)
2051                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2052                                           is_write);
2053         return dev;
2054 }
2055
2056 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2057                                struct kvm_vcpu *vcpu)
2058 {
2059         void *data = val;
2060         int r = X86EMUL_CONTINUE;
2061
2062         while (bytes) {
2063                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2064                 unsigned offset = addr & (PAGE_SIZE-1);
2065                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2066                 int ret;
2067
2068                 if (gpa == UNMAPPED_GVA) {
2069                         r = X86EMUL_PROPAGATE_FAULT;
2070                         goto out;
2071                 }
2072                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2073                 if (ret < 0) {
2074                         r = X86EMUL_UNHANDLEABLE;
2075                         goto out;
2076                 }
2077
2078                 bytes -= toread;
2079                 data += toread;
2080                 addr += toread;
2081         }
2082 out:
2083         return r;
2084 }
2085
2086 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2087                                 struct kvm_vcpu *vcpu)
2088 {
2089         void *data = val;
2090         int r = X86EMUL_CONTINUE;
2091
2092         while (bytes) {
2093                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2094                 unsigned offset = addr & (PAGE_SIZE-1);
2095                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2096                 int ret;
2097
2098                 if (gpa == UNMAPPED_GVA) {
2099                         r = X86EMUL_PROPAGATE_FAULT;
2100                         goto out;
2101                 }
2102                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2103                 if (ret < 0) {
2104                         r = X86EMUL_UNHANDLEABLE;
2105                         goto out;
2106                 }
2107
2108                 bytes -= towrite;
2109                 data += towrite;
2110                 addr += towrite;
2111         }
2112 out:
2113         return r;
2114 }
2115
2116
2117 static int emulator_read_emulated(unsigned long addr,
2118                                   void *val,
2119                                   unsigned int bytes,
2120                                   struct kvm_vcpu *vcpu)
2121 {
2122         struct kvm_io_device *mmio_dev;
2123         gpa_t                 gpa;
2124
2125         if (vcpu->mmio_read_completed) {
2126                 memcpy(val, vcpu->mmio_data, bytes);
2127                 vcpu->mmio_read_completed = 0;
2128                 return X86EMUL_CONTINUE;
2129         }
2130
2131         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2132
2133         /* For APIC access vmexit */
2134         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2135                 goto mmio;
2136
2137         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2138                                 == X86EMUL_CONTINUE)
2139                 return X86EMUL_CONTINUE;
2140         if (gpa == UNMAPPED_GVA)
2141                 return X86EMUL_PROPAGATE_FAULT;
2142
2143 mmio:
2144         /*
2145          * Is this MMIO handled locally?
2146          */
2147         mutex_lock(&vcpu->kvm->lock);
2148         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2149         if (mmio_dev) {
2150                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2151                 mutex_unlock(&vcpu->kvm->lock);
2152                 return X86EMUL_CONTINUE;
2153         }
2154         mutex_unlock(&vcpu->kvm->lock);
2155
2156         vcpu->mmio_needed = 1;
2157         vcpu->mmio_phys_addr = gpa;
2158         vcpu->mmio_size = bytes;
2159         vcpu->mmio_is_write = 0;
2160
2161         return X86EMUL_UNHANDLEABLE;
2162 }
2163
2164 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2165                           const void *val, int bytes)
2166 {
2167         int ret;
2168
2169         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2170         if (ret < 0)
2171                 return 0;
2172         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2173         return 1;
2174 }
2175
2176 static int emulator_write_emulated_onepage(unsigned long addr,
2177                                            const void *val,
2178                                            unsigned int bytes,
2179                                            struct kvm_vcpu *vcpu)
2180 {
2181         struct kvm_io_device *mmio_dev;
2182         gpa_t                 gpa;
2183
2184         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2185
2186         if (gpa == UNMAPPED_GVA) {
2187                 kvm_inject_page_fault(vcpu, addr, 2);
2188                 return X86EMUL_PROPAGATE_FAULT;
2189         }
2190
2191         /* For APIC access vmexit */
2192         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2193                 goto mmio;
2194
2195         if (emulator_write_phys(vcpu, gpa, val, bytes))
2196                 return X86EMUL_CONTINUE;
2197
2198 mmio:
2199         /*
2200          * Is this MMIO handled locally?
2201          */
2202         mutex_lock(&vcpu->kvm->lock);
2203         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2204         if (mmio_dev) {
2205                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2206                 mutex_unlock(&vcpu->kvm->lock);
2207                 return X86EMUL_CONTINUE;
2208         }
2209         mutex_unlock(&vcpu->kvm->lock);
2210
2211         vcpu->mmio_needed = 1;
2212         vcpu->mmio_phys_addr = gpa;
2213         vcpu->mmio_size = bytes;
2214         vcpu->mmio_is_write = 1;
2215         memcpy(vcpu->mmio_data, val, bytes);
2216
2217         return X86EMUL_CONTINUE;
2218 }
2219
2220 int emulator_write_emulated(unsigned long addr,
2221                                    const void *val,
2222                                    unsigned int bytes,
2223                                    struct kvm_vcpu *vcpu)
2224 {
2225         /* Crossing a page boundary? */
2226         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2227                 int rc, now;
2228
2229                 now = -addr & ~PAGE_MASK;
2230                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2231                 if (rc != X86EMUL_CONTINUE)
2232                         return rc;
2233                 addr += now;
2234                 val += now;
2235                 bytes -= now;
2236         }
2237         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2238 }
2239 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2240
2241 static int emulator_cmpxchg_emulated(unsigned long addr,
2242                                      const void *old,
2243                                      const void *new,
2244                                      unsigned int bytes,
2245                                      struct kvm_vcpu *vcpu)
2246 {
2247         static int reported;
2248
2249         if (!reported) {
2250                 reported = 1;
2251                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2252         }
2253 #ifndef CONFIG_X86_64
2254         /* guests cmpxchg8b have to be emulated atomically */
2255         if (bytes == 8) {
2256                 gpa_t gpa;
2257                 struct page *page;
2258                 char *kaddr;
2259                 u64 val;
2260
2261                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2262
2263                 if (gpa == UNMAPPED_GVA ||
2264                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2265                         goto emul_write;
2266
2267                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2268                         goto emul_write;
2269
2270                 val = *(u64 *)new;
2271
2272                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2273
2274                 kaddr = kmap_atomic(page, KM_USER0);
2275                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2276                 kunmap_atomic(kaddr, KM_USER0);
2277                 kvm_release_page_dirty(page);
2278         }
2279 emul_write:
2280 #endif
2281
2282         return emulator_write_emulated(addr, new, bytes, vcpu);
2283 }
2284
2285 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2286 {
2287         return kvm_x86_ops->get_segment_base(vcpu, seg);
2288 }
2289
2290 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2291 {
2292         kvm_mmu_invlpg(vcpu, address);
2293         return X86EMUL_CONTINUE;
2294 }
2295
2296 int emulate_clts(struct kvm_vcpu *vcpu)
2297 {
2298         KVMTRACE_0D(CLTS, vcpu, handler);
2299         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2300         return X86EMUL_CONTINUE;
2301 }
2302
2303 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2304 {
2305         struct kvm_vcpu *vcpu = ctxt->vcpu;
2306
2307         switch (dr) {
2308         case 0 ... 3:
2309                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2310                 return X86EMUL_CONTINUE;
2311         default:
2312                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2313                 return X86EMUL_UNHANDLEABLE;
2314         }
2315 }
2316
2317 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2318 {
2319         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2320         int exception;
2321
2322         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2323         if (exception) {
2324                 /* FIXME: better handling */
2325                 return X86EMUL_UNHANDLEABLE;
2326         }
2327         return X86EMUL_CONTINUE;
2328 }
2329
2330 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2331 {
2332         u8 opcodes[4];
2333         unsigned long rip = kvm_rip_read(vcpu);
2334         unsigned long rip_linear;
2335
2336         if (!printk_ratelimit())
2337                 return;
2338
2339         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2340
2341         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2342
2343         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2344                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2345 }
2346 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2347
2348 static struct x86_emulate_ops emulate_ops = {
2349         .read_std            = kvm_read_guest_virt,
2350         .read_emulated       = emulator_read_emulated,
2351         .write_emulated      = emulator_write_emulated,
2352         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2353 };
2354
2355 static void cache_all_regs(struct kvm_vcpu *vcpu)
2356 {
2357         kvm_register_read(vcpu, VCPU_REGS_RAX);
2358         kvm_register_read(vcpu, VCPU_REGS_RSP);
2359         kvm_register_read(vcpu, VCPU_REGS_RIP);
2360         vcpu->arch.regs_dirty = ~0;
2361 }
2362
2363 int emulate_instruction(struct kvm_vcpu *vcpu,
2364                         struct kvm_run *run,
2365                         unsigned long cr2,
2366                         u16 error_code,
2367                         int emulation_type)
2368 {
2369         int r;
2370         struct decode_cache *c;
2371
2372         kvm_clear_exception_queue(vcpu);
2373         vcpu->arch.mmio_fault_cr2 = cr2;
2374         /*
2375          * TODO: fix x86_emulate.c to use guest_read/write_register
2376          * instead of direct ->regs accesses, can save hundred cycles
2377          * on Intel for instructions that don't read/change RSP, for
2378          * for example.
2379          */
2380         cache_all_regs(vcpu);
2381
2382         vcpu->mmio_is_write = 0;
2383         vcpu->arch.pio.string = 0;
2384
2385         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2386                 int cs_db, cs_l;
2387                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2388
2389                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2390                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2391                 vcpu->arch.emulate_ctxt.mode =
2392                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2393                         ? X86EMUL_MODE_REAL : cs_l
2394                         ? X86EMUL_MODE_PROT64 : cs_db
2395                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2396
2397                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2398
2399                 /* Reject the instructions other than VMCALL/VMMCALL when
2400                  * try to emulate invalid opcode */
2401                 c = &vcpu->arch.emulate_ctxt.decode;
2402                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2403                     (!(c->twobyte && c->b == 0x01 &&
2404                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2405                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2406                         return EMULATE_FAIL;
2407
2408                 ++vcpu->stat.insn_emulation;
2409                 if (r)  {
2410                         ++vcpu->stat.insn_emulation_fail;
2411                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2412                                 return EMULATE_DONE;
2413                         return EMULATE_FAIL;
2414                 }
2415         }
2416
2417         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2418
2419         if (vcpu->arch.pio.string)
2420                 return EMULATE_DO_MMIO;
2421
2422         if ((r || vcpu->mmio_is_write) && run) {
2423                 run->exit_reason = KVM_EXIT_MMIO;
2424                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2425                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2426                 run->mmio.len = vcpu->mmio_size;
2427                 run->mmio.is_write = vcpu->mmio_is_write;
2428         }
2429
2430         if (r) {
2431                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2432                         return EMULATE_DONE;
2433                 if (!vcpu->mmio_needed) {
2434                         kvm_report_emulation_failure(vcpu, "mmio");
2435                         return EMULATE_FAIL;
2436                 }
2437                 return EMULATE_DO_MMIO;
2438         }
2439
2440         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2441
2442         if (vcpu->mmio_is_write) {
2443                 vcpu->mmio_needed = 0;
2444                 return EMULATE_DO_MMIO;
2445         }
2446
2447         return EMULATE_DONE;
2448 }
2449 EXPORT_SYMBOL_GPL(emulate_instruction);
2450
2451 static int pio_copy_data(struct kvm_vcpu *vcpu)
2452 {
2453         void *p = vcpu->arch.pio_data;
2454         gva_t q = vcpu->arch.pio.guest_gva;
2455         unsigned bytes;
2456         int ret;
2457
2458         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2459         if (vcpu->arch.pio.in)
2460                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2461         else
2462                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2463         return ret;
2464 }
2465
2466 int complete_pio(struct kvm_vcpu *vcpu)
2467 {
2468         struct kvm_pio_request *io = &vcpu->arch.pio;
2469         long delta;
2470         int r;
2471         unsigned long val;
2472
2473         if (!io->string) {
2474                 if (io->in) {
2475                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2476                         memcpy(&val, vcpu->arch.pio_data, io->size);
2477                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2478                 }
2479         } else {
2480                 if (io->in) {
2481                         r = pio_copy_data(vcpu);
2482                         if (r)
2483                                 return r;
2484                 }
2485
2486                 delta = 1;
2487                 if (io->rep) {
2488                         delta *= io->cur_count;
2489                         /*
2490                          * The size of the register should really depend on
2491                          * current address size.
2492                          */
2493                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2494                         val -= delta;
2495                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2496                 }
2497                 if (io->down)
2498                         delta = -delta;
2499                 delta *= io->size;
2500                 if (io->in) {
2501                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2502                         val += delta;
2503                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2504                 } else {
2505                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2506                         val += delta;
2507                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2508                 }
2509         }
2510
2511         io->count -= io->cur_count;
2512         io->cur_count = 0;
2513
2514         return 0;
2515 }
2516
2517 static void kernel_pio(struct kvm_io_device *pio_dev,
2518                        struct kvm_vcpu *vcpu,
2519                        void *pd)
2520 {
2521         /* TODO: String I/O for in kernel device */
2522
2523         mutex_lock(&vcpu->kvm->lock);
2524         if (vcpu->arch.pio.in)
2525                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2526                                   vcpu->arch.pio.size,
2527                                   pd);
2528         else
2529                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2530                                    vcpu->arch.pio.size,
2531                                    pd);
2532         mutex_unlock(&vcpu->kvm->lock);
2533 }
2534
2535 static void pio_string_write(struct kvm_io_device *pio_dev,
2536                              struct kvm_vcpu *vcpu)
2537 {
2538         struct kvm_pio_request *io = &vcpu->arch.pio;
2539         void *pd = vcpu->arch.pio_data;
2540         int i;
2541
2542         mutex_lock(&vcpu->kvm->lock);
2543         for (i = 0; i < io->cur_count; i++) {
2544                 kvm_iodevice_write(pio_dev, io->port,
2545                                    io->size,
2546                                    pd);
2547                 pd += io->size;
2548         }
2549         mutex_unlock(&vcpu->kvm->lock);
2550 }
2551
2552 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2553                                                gpa_t addr, int len,
2554                                                int is_write)
2555 {
2556         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2557 }
2558
2559 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2560                   int size, unsigned port)
2561 {
2562         struct kvm_io_device *pio_dev;
2563         unsigned long val;
2564
2565         vcpu->run->exit_reason = KVM_EXIT_IO;
2566         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2567         vcpu->run->io.size = vcpu->arch.pio.size = size;
2568         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2569         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2570         vcpu->run->io.port = vcpu->arch.pio.port = port;
2571         vcpu->arch.pio.in = in;
2572         vcpu->arch.pio.string = 0;
2573         vcpu->arch.pio.down = 0;
2574         vcpu->arch.pio.rep = 0;
2575
2576         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2577                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2578                             handler);
2579         else
2580                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2581                             handler);
2582
2583         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2584         memcpy(vcpu->arch.pio_data, &val, 4);
2585
2586         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2587         if (pio_dev) {
2588                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2589                 complete_pio(vcpu);
2590                 return 1;
2591         }
2592         return 0;
2593 }
2594 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2595
2596 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2597                   int size, unsigned long count, int down,
2598                   gva_t address, int rep, unsigned port)
2599 {
2600         unsigned now, in_page;
2601         int ret = 0;
2602         struct kvm_io_device *pio_dev;
2603
2604         vcpu->run->exit_reason = KVM_EXIT_IO;
2605         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2606         vcpu->run->io.size = vcpu->arch.pio.size = size;
2607         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2608         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2609         vcpu->run->io.port = vcpu->arch.pio.port = port;
2610         vcpu->arch.pio.in = in;
2611         vcpu->arch.pio.string = 1;
2612         vcpu->arch.pio.down = down;
2613         vcpu->arch.pio.rep = rep;
2614
2615         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2616                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2617                             handler);
2618         else
2619                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2620                             handler);
2621
2622         if (!count) {
2623                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2624                 return 1;
2625         }
2626
2627         if (!down)
2628                 in_page = PAGE_SIZE - offset_in_page(address);
2629         else
2630                 in_page = offset_in_page(address) + size;
2631         now = min(count, (unsigned long)in_page / size);
2632         if (!now)
2633                 now = 1;
2634         if (down) {
2635                 /*
2636                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2637                  */
2638                 pr_unimpl(vcpu, "guest string pio down\n");
2639                 kvm_inject_gp(vcpu, 0);
2640                 return 1;
2641         }
2642         vcpu->run->io.count = now;
2643         vcpu->arch.pio.cur_count = now;
2644
2645         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2646                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2647
2648         vcpu->arch.pio.guest_gva = address;
2649
2650         pio_dev = vcpu_find_pio_dev(vcpu, port,
2651                                     vcpu->arch.pio.cur_count,
2652                                     !vcpu->arch.pio.in);
2653         if (!vcpu->arch.pio.in) {
2654                 /* string PIO write */
2655                 ret = pio_copy_data(vcpu);
2656                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2657                         kvm_inject_gp(vcpu, 0);
2658                         return 1;
2659                 }
2660                 if (ret == 0 && pio_dev) {
2661                         pio_string_write(pio_dev, vcpu);
2662                         complete_pio(vcpu);
2663                         if (vcpu->arch.pio.count == 0)
2664                                 ret = 1;
2665                 }
2666         } else if (pio_dev)
2667                 pr_unimpl(vcpu, "no string pio read support yet, "
2668                        "port %x size %d count %ld\n",
2669                         port, size, count);
2670
2671         return ret;
2672 }
2673 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2674
2675 static void bounce_off(void *info)
2676 {
2677         /* nothing */
2678 }
2679
2680 static unsigned int  ref_freq;
2681 static unsigned long tsc_khz_ref;
2682
2683 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2684                                      void *data)
2685 {
2686         struct cpufreq_freqs *freq = data;
2687         struct kvm *kvm;
2688         struct kvm_vcpu *vcpu;
2689         int i, send_ipi = 0;
2690
2691         if (!ref_freq)
2692                 ref_freq = freq->old;
2693
2694         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2695                 return 0;
2696         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2697                 return 0;
2698         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2699
2700         spin_lock(&kvm_lock);
2701         list_for_each_entry(kvm, &vm_list, vm_list) {
2702                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2703                         vcpu = kvm->vcpus[i];
2704                         if (!vcpu)
2705                                 continue;
2706                         if (vcpu->cpu != freq->cpu)
2707                                 continue;
2708                         if (!kvm_request_guest_time_update(vcpu))
2709                                 continue;
2710                         if (vcpu->cpu != smp_processor_id())
2711                                 send_ipi++;
2712                 }
2713         }
2714         spin_unlock(&kvm_lock);
2715
2716         if (freq->old < freq->new && send_ipi) {
2717                 /*
2718                  * We upscale the frequency.  Must make the guest
2719                  * doesn't see old kvmclock values while running with
2720                  * the new frequency, otherwise we risk the guest sees
2721                  * time go backwards.
2722                  *
2723                  * In case we update the frequency for another cpu
2724                  * (which might be in guest context) send an interrupt
2725                  * to kick the cpu out of guest context.  Next time
2726                  * guest context is entered kvmclock will be updated,
2727                  * so the guest will not see stale values.
2728                  */
2729                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2730         }
2731         return 0;
2732 }
2733
2734 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2735         .notifier_call  = kvmclock_cpufreq_notifier
2736 };
2737
2738 int kvm_arch_init(void *opaque)
2739 {
2740         int r, cpu;
2741         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2742
2743         if (kvm_x86_ops) {
2744                 printk(KERN_ERR "kvm: already loaded the other module\n");
2745                 r = -EEXIST;
2746                 goto out;
2747         }
2748
2749         if (!ops->cpu_has_kvm_support()) {
2750                 printk(KERN_ERR "kvm: no hardware support\n");
2751                 r = -EOPNOTSUPP;
2752                 goto out;
2753         }
2754         if (ops->disabled_by_bios()) {
2755                 printk(KERN_ERR "kvm: disabled by bios\n");
2756                 r = -EOPNOTSUPP;
2757                 goto out;
2758         }
2759
2760         r = kvm_mmu_module_init();
2761         if (r)
2762                 goto out;
2763
2764         kvm_init_msr_list();
2765
2766         kvm_x86_ops = ops;
2767         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2768         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2769         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2770                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2771
2772         for_each_possible_cpu(cpu)
2773                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2774         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2775                 tsc_khz_ref = tsc_khz;
2776                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2777                                           CPUFREQ_TRANSITION_NOTIFIER);
2778         }
2779
2780         return 0;
2781
2782 out:
2783         return r;
2784 }
2785
2786 void kvm_arch_exit(void)
2787 {
2788         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2789                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2790                                             CPUFREQ_TRANSITION_NOTIFIER);
2791         kvm_x86_ops = NULL;
2792         kvm_mmu_module_exit();
2793 }
2794
2795 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2796 {
2797         ++vcpu->stat.halt_exits;
2798         KVMTRACE_0D(HLT, vcpu, handler);
2799         if (irqchip_in_kernel(vcpu->kvm)) {
2800                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2801                 return 1;
2802         } else {
2803                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2804                 return 0;
2805         }
2806 }
2807 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2808
2809 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2810                            unsigned long a1)
2811 {
2812         if (is_long_mode(vcpu))
2813                 return a0;
2814         else
2815                 return a0 | ((gpa_t)a1 << 32);
2816 }
2817
2818 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2819 {
2820         unsigned long nr, a0, a1, a2, a3, ret;
2821         int r = 1;
2822
2823         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2824         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2825         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2826         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2827         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2828
2829         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2830
2831         if (!is_long_mode(vcpu)) {
2832                 nr &= 0xFFFFFFFF;
2833                 a0 &= 0xFFFFFFFF;
2834                 a1 &= 0xFFFFFFFF;
2835                 a2 &= 0xFFFFFFFF;
2836                 a3 &= 0xFFFFFFFF;
2837         }
2838
2839         switch (nr) {
2840         case KVM_HC_VAPIC_POLL_IRQ:
2841                 ret = 0;
2842                 break;
2843         case KVM_HC_MMU_OP:
2844                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2845                 break;
2846         default:
2847                 ret = -KVM_ENOSYS;
2848                 break;
2849         }
2850         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2851         ++vcpu->stat.hypercalls;
2852         return r;
2853 }
2854 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2855
2856 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2857 {
2858         char instruction[3];
2859         int ret = 0;
2860         unsigned long rip = kvm_rip_read(vcpu);
2861
2862
2863         /*
2864          * Blow out the MMU to ensure that no other VCPU has an active mapping
2865          * to ensure that the updated hypercall appears atomically across all
2866          * VCPUs.
2867          */
2868         kvm_mmu_zap_all(vcpu->kvm);
2869
2870         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2871         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2872             != X86EMUL_CONTINUE)
2873                 ret = -EFAULT;
2874
2875         return ret;
2876 }
2877
2878 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2879 {
2880         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2881 }
2882
2883 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2884 {
2885         struct descriptor_table dt = { limit, base };
2886
2887         kvm_x86_ops->set_gdt(vcpu, &dt);
2888 }
2889
2890 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2891 {
2892         struct descriptor_table dt = { limit, base };
2893
2894         kvm_x86_ops->set_idt(vcpu, &dt);
2895 }
2896
2897 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2898                    unsigned long *rflags)
2899 {
2900         kvm_lmsw(vcpu, msw);
2901         *rflags = kvm_x86_ops->get_rflags(vcpu);
2902 }
2903
2904 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2905 {
2906         unsigned long value;
2907
2908         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2909         switch (cr) {
2910         case 0:
2911                 value = vcpu->arch.cr0;
2912                 break;
2913         case 2:
2914                 value = vcpu->arch.cr2;
2915                 break;
2916         case 3:
2917                 value = vcpu->arch.cr3;
2918                 break;
2919         case 4:
2920                 value = vcpu->arch.cr4;
2921                 break;
2922         case 8:
2923                 value = kvm_get_cr8(vcpu);
2924                 break;
2925         default:
2926                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2927                 return 0;
2928         }
2929         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2930                     (u32)((u64)value >> 32), handler);
2931
2932         return value;
2933 }
2934
2935 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2936                      unsigned long *rflags)
2937 {
2938         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2939                     (u32)((u64)val >> 32), handler);
2940
2941         switch (cr) {
2942         case 0:
2943                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2944                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2945                 break;
2946         case 2:
2947                 vcpu->arch.cr2 = val;
2948                 break;
2949         case 3:
2950                 kvm_set_cr3(vcpu, val);
2951                 break;
2952         case 4:
2953                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2954                 break;
2955         case 8:
2956                 kvm_set_cr8(vcpu, val & 0xfUL);
2957                 break;
2958         default:
2959                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2960         }
2961 }
2962
2963 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2964 {
2965         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2966         int j, nent = vcpu->arch.cpuid_nent;
2967
2968         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2969         /* when no next entry is found, the current entry[i] is reselected */
2970         for (j = i + 1; ; j = (j + 1) % nent) {
2971                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2972                 if (ej->function == e->function) {
2973                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2974                         return j;
2975                 }
2976         }
2977         return 0; /* silence gcc, even though control never reaches here */
2978 }
2979
2980 /* find an entry with matching function, matching index (if needed), and that
2981  * should be read next (if it's stateful) */
2982 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2983         u32 function, u32 index)
2984 {
2985         if (e->function != function)
2986                 return 0;
2987         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2988                 return 0;
2989         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2990             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2991                 return 0;
2992         return 1;
2993 }
2994
2995 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2996                                               u32 function, u32 index)
2997 {
2998         int i;
2999         struct kvm_cpuid_entry2 *best = NULL;
3000
3001         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3002                 struct kvm_cpuid_entry2 *e;
3003
3004                 e = &vcpu->arch.cpuid_entries[i];
3005                 if (is_matching_cpuid_entry(e, function, index)) {
3006                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3007                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3008                         best = e;
3009                         break;
3010                 }
3011                 /*
3012                  * Both basic or both extended?
3013                  */
3014                 if (((e->function ^ function) & 0x80000000) == 0)
3015                         if (!best || e->function > best->function)
3016                                 best = e;
3017         }
3018         return best;
3019 }
3020
3021 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3022 {
3023         u32 function, index;
3024         struct kvm_cpuid_entry2 *best;
3025
3026         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3027         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3028         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3029         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3030         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3031         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3032         best = kvm_find_cpuid_entry(vcpu, function, index);
3033         if (best) {
3034                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3035                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3036                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3037                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3038         }
3039         kvm_x86_ops->skip_emulated_instruction(vcpu);
3040         KVMTRACE_5D(CPUID, vcpu, function,
3041                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3042                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3043                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3044                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3045 }
3046 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3047
3048 /*
3049  * Check if userspace requested an interrupt window, and that the
3050  * interrupt window is open.
3051  *
3052  * No need to exit to userspace if we already have an interrupt queued.
3053  */
3054 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3055                                           struct kvm_run *kvm_run)
3056 {
3057         return (!vcpu->arch.irq_summary &&
3058                 kvm_run->request_interrupt_window &&
3059                 vcpu->arch.interrupt_window_open &&
3060                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3061 }
3062
3063 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3064                               struct kvm_run *kvm_run)
3065 {
3066         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3067         kvm_run->cr8 = kvm_get_cr8(vcpu);
3068         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3069         if (irqchip_in_kernel(vcpu->kvm))
3070                 kvm_run->ready_for_interrupt_injection = 1;
3071         else
3072                 kvm_run->ready_for_interrupt_injection =
3073                                         (vcpu->arch.interrupt_window_open &&
3074                                          vcpu->arch.irq_summary == 0);
3075 }
3076
3077 static void vapic_enter(struct kvm_vcpu *vcpu)
3078 {
3079         struct kvm_lapic *apic = vcpu->arch.apic;
3080         struct page *page;
3081
3082         if (!apic || !apic->vapic_addr)
3083                 return;
3084
3085         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3086
3087         vcpu->arch.apic->vapic_page = page;
3088 }
3089
3090 static void vapic_exit(struct kvm_vcpu *vcpu)
3091 {
3092         struct kvm_lapic *apic = vcpu->arch.apic;
3093
3094         if (!apic || !apic->vapic_addr)
3095                 return;
3096
3097         down_read(&vcpu->kvm->slots_lock);
3098         kvm_release_page_dirty(apic->vapic_page);
3099         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3100         up_read(&vcpu->kvm->slots_lock);
3101 }
3102
3103 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3104 {
3105         int r;
3106
3107         if (vcpu->requests)
3108                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3109                         kvm_mmu_unload(vcpu);
3110
3111         r = kvm_mmu_reload(vcpu);
3112         if (unlikely(r))
3113                 goto out;
3114
3115         if (vcpu->requests) {
3116                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3117                         __kvm_migrate_timers(vcpu);
3118                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3119                         kvm_write_guest_time(vcpu);
3120                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3121                         kvm_mmu_sync_roots(vcpu);
3122                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3123                         kvm_x86_ops->tlb_flush(vcpu);
3124                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3125                                        &vcpu->requests)) {
3126                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3127                         r = 0;
3128                         goto out;
3129                 }
3130                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3131                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3132                         r = 0;
3133                         goto out;
3134                 }
3135         }
3136
3137         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3138         kvm_inject_pending_timer_irqs(vcpu);
3139
3140         preempt_disable();
3141
3142         kvm_x86_ops->prepare_guest_switch(vcpu);
3143         kvm_load_guest_fpu(vcpu);
3144
3145         local_irq_disable();
3146
3147         if (vcpu->requests || need_resched() || signal_pending(current)) {
3148                 local_irq_enable();
3149                 preempt_enable();
3150                 r = 1;
3151                 goto out;
3152         }
3153
3154         vcpu->guest_mode = 1;
3155         /*
3156          * Make sure that guest_mode assignment won't happen after
3157          * testing the pending IRQ vector bitmap.
3158          */
3159         smp_wmb();
3160
3161         if (vcpu->arch.exception.pending)
3162                 __queue_exception(vcpu);
3163         else if (irqchip_in_kernel(vcpu->kvm))
3164                 kvm_x86_ops->inject_pending_irq(vcpu);
3165         else
3166                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3167
3168         kvm_lapic_sync_to_vapic(vcpu);
3169
3170         up_read(&vcpu->kvm->slots_lock);
3171
3172         kvm_guest_enter();
3173
3174         get_debugreg(vcpu->arch.host_dr6, 6);
3175         get_debugreg(vcpu->arch.host_dr7, 7);
3176         if (unlikely(vcpu->arch.switch_db_regs)) {
3177                 get_debugreg(vcpu->arch.host_db[0], 0);
3178                 get_debugreg(vcpu->arch.host_db[1], 1);
3179                 get_debugreg(vcpu->arch.host_db[2], 2);
3180                 get_debugreg(vcpu->arch.host_db[3], 3);
3181
3182                 set_debugreg(0, 7);
3183                 set_debugreg(vcpu->arch.eff_db[0], 0);
3184                 set_debugreg(vcpu->arch.eff_db[1], 1);
3185                 set_debugreg(vcpu->arch.eff_db[2], 2);
3186                 set_debugreg(vcpu->arch.eff_db[3], 3);
3187         }
3188
3189         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3190         kvm_x86_ops->run(vcpu, kvm_run);
3191
3192         if (unlikely(vcpu->arch.switch_db_regs)) {
3193                 set_debugreg(0, 7);
3194                 set_debugreg(vcpu->arch.host_db[0], 0);
3195                 set_debugreg(vcpu->arch.host_db[1], 1);
3196                 set_debugreg(vcpu->arch.host_db[2], 2);
3197                 set_debugreg(vcpu->arch.host_db[3], 3);
3198         }
3199         set_debugreg(vcpu->arch.host_dr6, 6);
3200         set_debugreg(vcpu->arch.host_dr7, 7);
3201
3202         vcpu->guest_mode = 0;
3203         local_irq_enable();
3204
3205         ++vcpu->stat.exits;
3206
3207         /*
3208          * We must have an instruction between local_irq_enable() and
3209          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3210          * the interrupt shadow.  The stat.exits increment will do nicely.
3211          * But we need to prevent reordering, hence this barrier():
3212          */
3213         barrier();
3214
3215         kvm_guest_exit();
3216
3217         preempt_enable();
3218
3219         down_read(&vcpu->kvm->slots_lock);
3220
3221         /*
3222          * Profile KVM exit RIPs:
3223          */
3224         if (unlikely(prof_on == KVM_PROFILING)) {
3225                 unsigned long rip = kvm_rip_read(vcpu);
3226                 profile_hit(KVM_PROFILING, (void *)rip);
3227         }
3228
3229         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3230                 vcpu->arch.exception.pending = false;
3231
3232         kvm_lapic_sync_from_vapic(vcpu);
3233
3234         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3235 out:
3236         return r;
3237 }
3238
3239 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3240 {
3241         int r;
3242
3243         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3244                 pr_debug("vcpu %d received sipi with vector # %x\n",
3245                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3246                 kvm_lapic_reset(vcpu);
3247                 r = kvm_arch_vcpu_reset(vcpu);
3248                 if (r)
3249                         return r;
3250                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3251         }
3252
3253         down_read(&vcpu->kvm->slots_lock);
3254         vapic_enter(vcpu);
3255
3256         r = 1;
3257         while (r > 0) {
3258                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3259                         r = vcpu_enter_guest(vcpu, kvm_run);
3260                 else {
3261                         up_read(&vcpu->kvm->slots_lock);
3262                         kvm_vcpu_block(vcpu);
3263                         down_read(&vcpu->kvm->slots_lock);
3264                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3265                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3266                                         vcpu->arch.mp_state =
3267                                                         KVM_MP_STATE_RUNNABLE;
3268                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3269                                 r = -EINTR;
3270                 }
3271
3272                 if (r > 0) {
3273                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3274                                 r = -EINTR;
3275                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3276                                 ++vcpu->stat.request_irq_exits;
3277                         }
3278                         if (signal_pending(current)) {
3279                                 r = -EINTR;
3280                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3281                                 ++vcpu->stat.signal_exits;
3282                         }
3283                         if (need_resched()) {
3284                                 up_read(&vcpu->kvm->slots_lock);
3285                                 kvm_resched(vcpu);
3286                                 down_read(&vcpu->kvm->slots_lock);
3287                         }
3288                 }
3289         }
3290
3291         up_read(&vcpu->kvm->slots_lock);
3292         post_kvm_run_save(vcpu, kvm_run);
3293
3294         vapic_exit(vcpu);
3295
3296         return r;
3297 }
3298
3299 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3300 {
3301         int r;
3302         sigset_t sigsaved;
3303
3304         vcpu_load(vcpu);
3305
3306         if (vcpu->sigset_active)
3307                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3308
3309         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3310                 kvm_vcpu_block(vcpu);
3311                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3312                 r = -EAGAIN;
3313                 goto out;
3314         }
3315
3316         /* re-sync apic's tpr */
3317         if (!irqchip_in_kernel(vcpu->kvm))
3318                 kvm_set_cr8(vcpu, kvm_run->cr8);
3319
3320         if (vcpu->arch.pio.cur_count) {
3321                 r = complete_pio(vcpu);
3322                 if (r)
3323                         goto out;
3324         }
3325 #if CONFIG_HAS_IOMEM
3326         if (vcpu->mmio_needed) {
3327                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3328                 vcpu->mmio_read_completed = 1;
3329                 vcpu->mmio_needed = 0;
3330
3331                 down_read(&vcpu->kvm->slots_lock);
3332                 r = emulate_instruction(vcpu, kvm_run,
3333                                         vcpu->arch.mmio_fault_cr2, 0,
3334                                         EMULTYPE_NO_DECODE);
3335                 up_read(&vcpu->kvm->slots_lock);
3336                 if (r == EMULATE_DO_MMIO) {
3337                         /*
3338                          * Read-modify-write.  Back to userspace.
3339                          */
3340                         r = 0;
3341                         goto out;
3342                 }
3343         }
3344 #endif
3345         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3346                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3347                                      kvm_run->hypercall.ret);
3348
3349         r = __vcpu_run(vcpu, kvm_run);
3350
3351 out:
3352         if (vcpu->sigset_active)
3353                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3354
3355         vcpu_put(vcpu);
3356         return r;
3357 }
3358
3359 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3360 {
3361         vcpu_load(vcpu);
3362
3363         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3364         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3365         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3366         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3367         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3368         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3369         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3370         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3371 #ifdef CONFIG_X86_64
3372         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3373         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3374         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3375         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3376         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3377         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3378         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3379         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3380 #endif
3381
3382         regs->rip = kvm_rip_read(vcpu);
3383         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3384
3385         /*
3386          * Don't leak debug flags in case they were set for guest debugging
3387          */
3388         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3389                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3390
3391         vcpu_put(vcpu);
3392
3393         return 0;
3394 }
3395
3396 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3397 {
3398         vcpu_load(vcpu);
3399
3400         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3401         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3402         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3403         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3404         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3405         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3406         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3407         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3408 #ifdef CONFIG_X86_64
3409         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3410         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3411         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3412         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3413         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3414         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3415         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3416         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3417
3418 #endif
3419
3420         kvm_rip_write(vcpu, regs->rip);
3421         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3422
3423
3424         vcpu->arch.exception.pending = false;
3425
3426         vcpu_put(vcpu);
3427
3428         return 0;
3429 }
3430
3431 void kvm_get_segment(struct kvm_vcpu *vcpu,
3432                      struct kvm_segment *var, int seg)
3433 {
3434         kvm_x86_ops->get_segment(vcpu, var, seg);
3435 }
3436
3437 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3438 {
3439         struct kvm_segment cs;
3440
3441         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3442         *db = cs.db;
3443         *l = cs.l;
3444 }
3445 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3446
3447 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3448                                   struct kvm_sregs *sregs)
3449 {
3450         struct descriptor_table dt;
3451         int pending_vec;
3452
3453         vcpu_load(vcpu);
3454
3455         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3456         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3457         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3458         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3459         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3460         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3461
3462         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3463         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3464
3465         kvm_x86_ops->get_idt(vcpu, &dt);
3466         sregs->idt.limit = dt.limit;
3467         sregs->idt.base = dt.base;
3468         kvm_x86_ops->get_gdt(vcpu, &dt);
3469         sregs->gdt.limit = dt.limit;
3470         sregs->gdt.base = dt.base;
3471
3472         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3473         sregs->cr0 = vcpu->arch.cr0;
3474         sregs->cr2 = vcpu->arch.cr2;
3475         sregs->cr3 = vcpu->arch.cr3;
3476         sregs->cr4 = vcpu->arch.cr4;
3477         sregs->cr8 = kvm_get_cr8(vcpu);
3478         sregs->efer = vcpu->arch.shadow_efer;
3479         sregs->apic_base = kvm_get_apic_base(vcpu);
3480
3481         if (irqchip_in_kernel(vcpu->kvm)) {
3482                 memset(sregs->interrupt_bitmap, 0,
3483                        sizeof sregs->interrupt_bitmap);
3484                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3485                 if (pending_vec >= 0)
3486                         set_bit(pending_vec,
3487                                 (unsigned long *)sregs->interrupt_bitmap);
3488         } else
3489                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3490                        sizeof sregs->interrupt_bitmap);
3491
3492         vcpu_put(vcpu);
3493
3494         return 0;
3495 }
3496
3497 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3498                                     struct kvm_mp_state *mp_state)
3499 {
3500         vcpu_load(vcpu);
3501         mp_state->mp_state = vcpu->arch.mp_state;
3502         vcpu_put(vcpu);
3503         return 0;
3504 }
3505
3506 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3507                                     struct kvm_mp_state *mp_state)
3508 {
3509         vcpu_load(vcpu);
3510         vcpu->arch.mp_state = mp_state->mp_state;
3511         vcpu_put(vcpu);
3512         return 0;
3513 }
3514
3515 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3516                         struct kvm_segment *var, int seg)
3517 {
3518         kvm_x86_ops->set_segment(vcpu, var, seg);
3519 }
3520
3521 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3522                                    struct kvm_segment *kvm_desct)
3523 {
3524         kvm_desct->base = seg_desc->base0;
3525         kvm_desct->base |= seg_desc->base1 << 16;
3526         kvm_desct->base |= seg_desc->base2 << 24;
3527         kvm_desct->limit = seg_desc->limit0;
3528         kvm_desct->limit |= seg_desc->limit << 16;
3529         if (seg_desc->g) {
3530                 kvm_desct->limit <<= 12;
3531                 kvm_desct->limit |= 0xfff;
3532         }
3533         kvm_desct->selector = selector;
3534         kvm_desct->type = seg_desc->type;
3535         kvm_desct->present = seg_desc->p;
3536         kvm_desct->dpl = seg_desc->dpl;
3537         kvm_desct->db = seg_desc->d;
3538         kvm_desct->s = seg_desc->s;
3539         kvm_desct->l = seg_desc->l;
3540         kvm_desct->g = seg_desc->g;
3541         kvm_desct->avl = seg_desc->avl;
3542         if (!selector)
3543                 kvm_desct->unusable = 1;
3544         else
3545                 kvm_desct->unusable = 0;
3546         kvm_desct->padding = 0;
3547 }
3548
3549 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3550                                           u16 selector,
3551                                           struct descriptor_table *dtable)
3552 {
3553         if (selector & 1 << 2) {
3554                 struct kvm_segment kvm_seg;
3555
3556                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3557
3558                 if (kvm_seg.unusable)
3559                         dtable->limit = 0;
3560                 else
3561                         dtable->limit = kvm_seg.limit;
3562                 dtable->base = kvm_seg.base;
3563         }
3564         else
3565                 kvm_x86_ops->get_gdt(vcpu, dtable);
3566 }
3567
3568 /* allowed just for 8 bytes segments */
3569 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3570                                          struct desc_struct *seg_desc)
3571 {
3572         gpa_t gpa;
3573         struct descriptor_table dtable;
3574         u16 index = selector >> 3;
3575
3576         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3577
3578         if (dtable.limit < index * 8 + 7) {
3579                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3580                 return 1;
3581         }
3582         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3583         gpa += index * 8;
3584         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3585 }
3586
3587 /* allowed just for 8 bytes segments */
3588 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3589                                          struct desc_struct *seg_desc)
3590 {
3591         gpa_t gpa;
3592         struct descriptor_table dtable;
3593         u16 index = selector >> 3;
3594
3595         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3596
3597         if (dtable.limit < index * 8 + 7)
3598                 return 1;
3599         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3600         gpa += index * 8;
3601         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3602 }
3603
3604 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3605                              struct desc_struct *seg_desc)
3606 {
3607         u32 base_addr;
3608
3609         base_addr = seg_desc->base0;
3610         base_addr |= (seg_desc->base1 << 16);
3611         base_addr |= (seg_desc->base2 << 24);
3612
3613         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3614 }
3615
3616 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3617 {
3618         struct kvm_segment kvm_seg;
3619
3620         kvm_get_segment(vcpu, &kvm_seg, seg);
3621         return kvm_seg.selector;
3622 }
3623
3624 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3625                                                 u16 selector,
3626                                                 struct kvm_segment *kvm_seg)
3627 {
3628         struct desc_struct seg_desc;
3629
3630         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3631                 return 1;
3632         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3633         return 0;
3634 }
3635
3636 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3637 {
3638         struct kvm_segment segvar = {
3639                 .base = selector << 4,
3640                 .limit = 0xffff,
3641                 .selector = selector,
3642                 .type = 3,
3643                 .present = 1,
3644                 .dpl = 3,
3645                 .db = 0,
3646                 .s = 1,
3647                 .l = 0,
3648                 .g = 0,
3649                 .avl = 0,
3650                 .unusable = 0,
3651         };
3652         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3653         return 0;
3654 }
3655
3656 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3657                                 int type_bits, int seg)
3658 {
3659         struct kvm_segment kvm_seg;
3660
3661         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3662                 return kvm_load_realmode_segment(vcpu, selector, seg);
3663         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3664                 return 1;
3665         kvm_seg.type |= type_bits;
3666
3667         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3668             seg != VCPU_SREG_LDTR)
3669                 if (!kvm_seg.s)
3670                         kvm_seg.unusable = 1;
3671
3672         kvm_set_segment(vcpu, &kvm_seg, seg);
3673         return 0;
3674 }
3675
3676 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3677                                 struct tss_segment_32 *tss)
3678 {
3679         tss->cr3 = vcpu->arch.cr3;
3680         tss->eip = kvm_rip_read(vcpu);
3681         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3682         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3683         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3684         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3685         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3686         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3687         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3688         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3689         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3690         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3691         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3692         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3693         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3694         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3695         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3696         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3697         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3698 }
3699
3700 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3701                                   struct tss_segment_32 *tss)
3702 {
3703         kvm_set_cr3(vcpu, tss->cr3);
3704
3705         kvm_rip_write(vcpu, tss->eip);
3706         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3707
3708         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3709         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3710         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3711         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3712         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3713         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3714         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3715         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3716
3717         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3718                 return 1;
3719
3720         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3721                 return 1;
3722
3723         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3724                 return 1;
3725
3726         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3727                 return 1;
3728
3729         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3730                 return 1;
3731
3732         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3733                 return 1;
3734
3735         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3736                 return 1;
3737         return 0;
3738 }
3739
3740 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3741                                 struct tss_segment_16 *tss)
3742 {
3743         tss->ip = kvm_rip_read(vcpu);
3744         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3745         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3746         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3747         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3748         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3749         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3750         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3751         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3752         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3753
3754         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3755         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3756         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3757         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3758         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3759         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3760 }
3761
3762 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3763                                  struct tss_segment_16 *tss)
3764 {
3765         kvm_rip_write(vcpu, tss->ip);
3766         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3767         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3768         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3769         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3770         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3771         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3772         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3773         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3774         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3775
3776         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3777                 return 1;
3778
3779         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3780                 return 1;
3781
3782         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3783                 return 1;
3784
3785         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3786                 return 1;
3787
3788         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3789                 return 1;
3790         return 0;
3791 }
3792
3793 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3794                        u32 old_tss_base,
3795                        struct desc_struct *nseg_desc)
3796 {
3797         struct tss_segment_16 tss_segment_16;
3798         int ret = 0;
3799
3800         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3801                            sizeof tss_segment_16))
3802                 goto out;
3803
3804         save_state_to_tss16(vcpu, &tss_segment_16);
3805
3806         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3807                             sizeof tss_segment_16))
3808                 goto out;
3809
3810         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3811                            &tss_segment_16, sizeof tss_segment_16))
3812                 goto out;
3813
3814         if (load_state_from_tss16(vcpu, &tss_segment_16))
3815                 goto out;
3816
3817         ret = 1;
3818 out:
3819         return ret;
3820 }
3821
3822 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3823                        u32 old_tss_base,
3824                        struct desc_struct *nseg_desc)
3825 {
3826         struct tss_segment_32 tss_segment_32;
3827         int ret = 0;
3828
3829         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3830                            sizeof tss_segment_32))
3831                 goto out;
3832
3833         save_state_to_tss32(vcpu, &tss_segment_32);
3834
3835         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3836                             sizeof tss_segment_32))
3837                 goto out;
3838
3839         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3840                            &tss_segment_32, sizeof tss_segment_32))
3841                 goto out;
3842
3843         if (load_state_from_tss32(vcpu, &tss_segment_32))
3844                 goto out;
3845
3846         ret = 1;
3847 out:
3848         return ret;
3849 }
3850
3851 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3852 {
3853         struct kvm_segment tr_seg;
3854         struct desc_struct cseg_desc;
3855         struct desc_struct nseg_desc;
3856         int ret = 0;
3857         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3858         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3859
3860         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3861
3862         /* FIXME: Handle errors. Failure to read either TSS or their
3863          * descriptors should generate a pagefault.
3864          */
3865         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3866                 goto out;
3867
3868         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3869                 goto out;
3870
3871         if (reason != TASK_SWITCH_IRET) {
3872                 int cpl;
3873
3874                 cpl = kvm_x86_ops->get_cpl(vcpu);
3875                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3876                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3877                         return 1;
3878                 }
3879         }
3880
3881         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3882                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3883                 return 1;
3884         }
3885
3886         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3887                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3888                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3889         }
3890
3891         if (reason == TASK_SWITCH_IRET) {
3892                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3893                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3894         }
3895
3896         kvm_x86_ops->skip_emulated_instruction(vcpu);
3897
3898         if (nseg_desc.type & 8)
3899                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3900                                          &nseg_desc);
3901         else
3902                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3903                                          &nseg_desc);
3904
3905         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3906                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3907                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3908         }
3909
3910         if (reason != TASK_SWITCH_IRET) {
3911                 nseg_desc.type |= (1 << 1);
3912                 save_guest_segment_descriptor(vcpu, tss_selector,
3913                                               &nseg_desc);
3914         }
3915
3916         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3917         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3918         tr_seg.type = 11;
3919         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3920 out:
3921         return ret;
3922 }
3923 EXPORT_SYMBOL_GPL(kvm_task_switch);
3924
3925 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3926                                   struct kvm_sregs *sregs)
3927 {
3928         int mmu_reset_needed = 0;
3929         int i, pending_vec, max_bits;
3930         struct descriptor_table dt;
3931
3932         vcpu_load(vcpu);
3933
3934         dt.limit = sregs->idt.limit;
3935         dt.base = sregs->idt.base;
3936         kvm_x86_ops->set_idt(vcpu, &dt);
3937         dt.limit = sregs->gdt.limit;
3938         dt.base = sregs->gdt.base;
3939         kvm_x86_ops->set_gdt(vcpu, &dt);
3940
3941         vcpu->arch.cr2 = sregs->cr2;
3942         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3943
3944         down_read(&vcpu->kvm->slots_lock);
3945         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
3946                 vcpu->arch.cr3 = sregs->cr3;
3947         else
3948                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
3949         up_read(&vcpu->kvm->slots_lock);
3950
3951         kvm_set_cr8(vcpu, sregs->cr8);
3952
3953         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3954         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3955         kvm_set_apic_base(vcpu, sregs->apic_base);
3956
3957         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3958
3959         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3960         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3961         vcpu->arch.cr0 = sregs->cr0;
3962
3963         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3964         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3965         if (!is_long_mode(vcpu) && is_pae(vcpu))
3966                 load_pdptrs(vcpu, vcpu->arch.cr3);
3967
3968         if (mmu_reset_needed)
3969                 kvm_mmu_reset_context(vcpu);
3970
3971         if (!irqchip_in_kernel(vcpu->kvm)) {
3972                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3973                        sizeof vcpu->arch.irq_pending);
3974                 vcpu->arch.irq_summary = 0;
3975                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3976                         if (vcpu->arch.irq_pending[i])
3977                                 __set_bit(i, &vcpu->arch.irq_summary);
3978         } else {
3979                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3980                 pending_vec = find_first_bit(
3981                         (const unsigned long *)sregs->interrupt_bitmap,
3982                         max_bits);
3983                 /* Only pending external irq is handled here */
3984                 if (pending_vec < max_bits) {
3985                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3986                         pr_debug("Set back pending irq %d\n",
3987                                  pending_vec);
3988                 }
3989                 kvm_pic_clear_isr_ack(vcpu->kvm);
3990         }
3991
3992         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3993         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3994         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3995         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3996         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3997         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3998
3999         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4000         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4001
4002         /* Older userspace won't unhalt the vcpu on reset. */
4003         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4004             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4005             !(vcpu->arch.cr0 & X86_CR0_PE))
4006                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4007
4008         vcpu_put(vcpu);
4009
4010         return 0;
4011 }
4012
4013 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4014                                         struct kvm_guest_debug *dbg)
4015 {
4016         int i, r;
4017
4018         vcpu_load(vcpu);
4019
4020         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4021             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4022                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4023                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4024                 vcpu->arch.switch_db_regs =
4025                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4026         } else {
4027                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4028                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4029                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4030         }
4031
4032         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4033
4034         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4035                 kvm_queue_exception(vcpu, DB_VECTOR);
4036         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4037                 kvm_queue_exception(vcpu, BP_VECTOR);
4038
4039         vcpu_put(vcpu);
4040
4041         return r;
4042 }
4043
4044 /*
4045  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4046  * we have asm/x86/processor.h
4047  */
4048 struct fxsave {
4049         u16     cwd;
4050         u16     swd;
4051         u16     twd;
4052         u16     fop;
4053         u64     rip;
4054         u64     rdp;
4055         u32     mxcsr;
4056         u32     mxcsr_mask;
4057         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4058 #ifdef CONFIG_X86_64
4059         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4060 #else
4061         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4062 #endif
4063 };
4064
4065 /*
4066  * Translate a guest virtual address to a guest physical address.
4067  */
4068 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4069                                     struct kvm_translation *tr)
4070 {
4071         unsigned long vaddr = tr->linear_address;
4072         gpa_t gpa;
4073
4074         vcpu_load(vcpu);
4075         down_read(&vcpu->kvm->slots_lock);
4076         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4077         up_read(&vcpu->kvm->slots_lock);
4078         tr->physical_address = gpa;
4079         tr->valid = gpa != UNMAPPED_GVA;
4080         tr->writeable = 1;
4081         tr->usermode = 0;
4082         vcpu_put(vcpu);
4083
4084         return 0;
4085 }
4086
4087 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4088 {
4089         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4090
4091         vcpu_load(vcpu);
4092
4093         memcpy(fpu->fpr, fxsave->st_space, 128);
4094         fpu->fcw = fxsave->cwd;
4095         fpu->fsw = fxsave->swd;
4096         fpu->ftwx = fxsave->twd;
4097         fpu->last_opcode = fxsave->fop;
4098         fpu->last_ip = fxsave->rip;
4099         fpu->last_dp = fxsave->rdp;
4100         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4101
4102         vcpu_put(vcpu);
4103
4104         return 0;
4105 }
4106
4107 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4108 {
4109         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4110
4111         vcpu_load(vcpu);
4112
4113         memcpy(fxsave->st_space, fpu->fpr, 128);
4114         fxsave->cwd = fpu->fcw;
4115         fxsave->swd = fpu->fsw;
4116         fxsave->twd = fpu->ftwx;
4117         fxsave->fop = fpu->last_opcode;
4118         fxsave->rip = fpu->last_ip;
4119         fxsave->rdp = fpu->last_dp;
4120         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4121
4122         vcpu_put(vcpu);
4123
4124         return 0;
4125 }
4126
4127 void fx_init(struct kvm_vcpu *vcpu)
4128 {
4129         unsigned after_mxcsr_mask;
4130
4131         /*
4132          * Touch the fpu the first time in non atomic context as if
4133          * this is the first fpu instruction the exception handler
4134          * will fire before the instruction returns and it'll have to
4135          * allocate ram with GFP_KERNEL.
4136          */
4137         if (!used_math())
4138                 kvm_fx_save(&vcpu->arch.host_fx_image);
4139
4140         /* Initialize guest FPU by resetting ours and saving into guest's */
4141         preempt_disable();
4142         kvm_fx_save(&vcpu->arch.host_fx_image);
4143         kvm_fx_finit();
4144         kvm_fx_save(&vcpu->arch.guest_fx_image);
4145         kvm_fx_restore(&vcpu->arch.host_fx_image);
4146         preempt_enable();
4147
4148         vcpu->arch.cr0 |= X86_CR0_ET;
4149         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4150         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4151         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4152                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4153 }
4154 EXPORT_SYMBOL_GPL(fx_init);
4155
4156 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4157 {
4158         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4159                 return;
4160
4161         vcpu->guest_fpu_loaded = 1;
4162         kvm_fx_save(&vcpu->arch.host_fx_image);
4163         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4164 }
4165 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4166
4167 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4168 {
4169         if (!vcpu->guest_fpu_loaded)
4170                 return;
4171
4172         vcpu->guest_fpu_loaded = 0;
4173         kvm_fx_save(&vcpu->arch.guest_fx_image);
4174         kvm_fx_restore(&vcpu->arch.host_fx_image);
4175         ++vcpu->stat.fpu_reload;
4176 }
4177 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4178
4179 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4180 {
4181         if (vcpu->arch.time_page) {
4182                 kvm_release_page_dirty(vcpu->arch.time_page);
4183                 vcpu->arch.time_page = NULL;
4184         }
4185
4186         kvm_x86_ops->vcpu_free(vcpu);
4187 }
4188
4189 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4190                                                 unsigned int id)
4191 {
4192         return kvm_x86_ops->vcpu_create(kvm, id);
4193 }
4194
4195 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4196 {
4197         int r;
4198
4199         /* We do fxsave: this must be aligned. */
4200         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4201
4202         vcpu->arch.mtrr_state.have_fixed = 1;
4203         vcpu_load(vcpu);
4204         r = kvm_arch_vcpu_reset(vcpu);
4205         if (r == 0)
4206                 r = kvm_mmu_setup(vcpu);
4207         vcpu_put(vcpu);
4208         if (r < 0)
4209                 goto free_vcpu;
4210
4211         return 0;
4212 free_vcpu:
4213         kvm_x86_ops->vcpu_free(vcpu);
4214         return r;
4215 }
4216
4217 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4218 {
4219         vcpu_load(vcpu);
4220         kvm_mmu_unload(vcpu);
4221         vcpu_put(vcpu);
4222
4223         kvm_x86_ops->vcpu_free(vcpu);
4224 }
4225
4226 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4227 {
4228         vcpu->arch.nmi_pending = false;
4229         vcpu->arch.nmi_injected = false;
4230
4231         vcpu->arch.switch_db_regs = 0;
4232         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4233         vcpu->arch.dr6 = DR6_FIXED_1;
4234         vcpu->arch.dr7 = DR7_FIXED_1;
4235
4236         return kvm_x86_ops->vcpu_reset(vcpu);
4237 }
4238
4239 void kvm_arch_hardware_enable(void *garbage)
4240 {
4241         kvm_x86_ops->hardware_enable(garbage);
4242 }
4243
4244 void kvm_arch_hardware_disable(void *garbage)
4245 {
4246         kvm_x86_ops->hardware_disable(garbage);
4247 }
4248
4249 int kvm_arch_hardware_setup(void)
4250 {
4251         return kvm_x86_ops->hardware_setup();
4252 }
4253
4254 void kvm_arch_hardware_unsetup(void)
4255 {
4256         kvm_x86_ops->hardware_unsetup();
4257 }
4258
4259 void kvm_arch_check_processor_compat(void *rtn)
4260 {
4261         kvm_x86_ops->check_processor_compatibility(rtn);
4262 }
4263
4264 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4265 {
4266         struct page *page;
4267         struct kvm *kvm;
4268         int r;
4269
4270         BUG_ON(vcpu->kvm == NULL);
4271         kvm = vcpu->kvm;
4272
4273         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4274         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4275                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4276         else
4277                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4278
4279         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4280         if (!page) {
4281                 r = -ENOMEM;
4282                 goto fail;
4283         }
4284         vcpu->arch.pio_data = page_address(page);
4285
4286         r = kvm_mmu_create(vcpu);
4287         if (r < 0)
4288                 goto fail_free_pio_data;
4289
4290         if (irqchip_in_kernel(kvm)) {
4291                 r = kvm_create_lapic(vcpu);
4292                 if (r < 0)
4293                         goto fail_mmu_destroy;
4294         }
4295
4296         return 0;
4297
4298 fail_mmu_destroy:
4299         kvm_mmu_destroy(vcpu);
4300 fail_free_pio_data:
4301         free_page((unsigned long)vcpu->arch.pio_data);
4302 fail:
4303         return r;
4304 }
4305
4306 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4307 {
4308         kvm_free_lapic(vcpu);
4309         down_read(&vcpu->kvm->slots_lock);
4310         kvm_mmu_destroy(vcpu);
4311         up_read(&vcpu->kvm->slots_lock);
4312         free_page((unsigned long)vcpu->arch.pio_data);
4313 }
4314
4315 struct  kvm *kvm_arch_create_vm(void)
4316 {
4317         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4318
4319         if (!kvm)
4320                 return ERR_PTR(-ENOMEM);
4321
4322         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4323         INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4324         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4325
4326         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4327         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4328
4329         rdtscll(kvm->arch.vm_init_tsc);
4330
4331         return kvm;
4332 }
4333
4334 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4335 {
4336         vcpu_load(vcpu);
4337         kvm_mmu_unload(vcpu);
4338         vcpu_put(vcpu);
4339 }
4340
4341 static void kvm_free_vcpus(struct kvm *kvm)
4342 {
4343         unsigned int i;
4344
4345         /*
4346          * Unpin any mmu pages first.
4347          */
4348         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4349                 if (kvm->vcpus[i])
4350                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4351         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4352                 if (kvm->vcpus[i]) {
4353                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4354                         kvm->vcpus[i] = NULL;
4355                 }
4356         }
4357
4358 }
4359
4360 void kvm_arch_sync_events(struct kvm *kvm)
4361 {
4362         kvm_free_all_assigned_devices(kvm);
4363 }
4364
4365 void kvm_arch_destroy_vm(struct kvm *kvm)
4366 {
4367         kvm_iommu_unmap_guest(kvm);
4368         kvm_free_pit(kvm);
4369         kfree(kvm->arch.vpic);
4370         kfree(kvm->arch.vioapic);
4371         kvm_free_vcpus(kvm);
4372         kvm_free_physmem(kvm);
4373         if (kvm->arch.apic_access_page)
4374                 put_page(kvm->arch.apic_access_page);
4375         if (kvm->arch.ept_identity_pagetable)
4376                 put_page(kvm->arch.ept_identity_pagetable);
4377         kfree(kvm);
4378 }
4379
4380 int kvm_arch_set_memory_region(struct kvm *kvm,
4381                                 struct kvm_userspace_memory_region *mem,
4382                                 struct kvm_memory_slot old,
4383                                 int user_alloc)
4384 {
4385         int npages = mem->memory_size >> PAGE_SHIFT;
4386         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4387
4388         /*To keep backward compatibility with older userspace,
4389          *x86 needs to hanlde !user_alloc case.
4390          */
4391         if (!user_alloc) {
4392                 if (npages && !old.rmap) {
4393                         unsigned long userspace_addr;
4394
4395                         down_write(&current->mm->mmap_sem);
4396                         userspace_addr = do_mmap(NULL, 0,
4397                                                  npages * PAGE_SIZE,
4398                                                  PROT_READ | PROT_WRITE,
4399                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4400                                                  0);
4401                         up_write(&current->mm->mmap_sem);
4402
4403                         if (IS_ERR((void *)userspace_addr))
4404                                 return PTR_ERR((void *)userspace_addr);
4405
4406                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4407                         spin_lock(&kvm->mmu_lock);
4408                         memslot->userspace_addr = userspace_addr;
4409                         spin_unlock(&kvm->mmu_lock);
4410                 } else {
4411                         if (!old.user_alloc && old.rmap) {
4412                                 int ret;
4413
4414                                 down_write(&current->mm->mmap_sem);
4415                                 ret = do_munmap(current->mm, old.userspace_addr,
4416                                                 old.npages * PAGE_SIZE);
4417                                 up_write(&current->mm->mmap_sem);
4418                                 if (ret < 0)
4419                                         printk(KERN_WARNING
4420                                        "kvm_vm_ioctl_set_memory_region: "
4421                                        "failed to munmap memory\n");
4422                         }
4423                 }
4424         }
4425
4426         spin_lock(&kvm->mmu_lock);
4427         if (!kvm->arch.n_requested_mmu_pages) {
4428                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4429                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4430         }
4431
4432         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4433         spin_unlock(&kvm->mmu_lock);
4434         kvm_flush_remote_tlbs(kvm);
4435
4436         return 0;
4437 }
4438
4439 void kvm_arch_flush_shadow(struct kvm *kvm)
4440 {
4441         kvm_mmu_zap_all(kvm);
4442         kvm_reload_remote_mmus(kvm);
4443 }
4444
4445 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4446 {
4447         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4448                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4449                || vcpu->arch.nmi_pending;
4450 }
4451
4452 static void vcpu_kick_intr(void *info)
4453 {
4454 #ifdef DEBUG
4455         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4456         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4457 #endif
4458 }
4459
4460 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4461 {
4462         int ipi_pcpu = vcpu->cpu;
4463         int cpu = get_cpu();
4464
4465         if (waitqueue_active(&vcpu->wq)) {
4466                 wake_up_interruptible(&vcpu->wq);
4467                 ++vcpu->stat.halt_wakeup;
4468         }
4469         /*
4470          * We may be called synchronously with irqs disabled in guest mode,
4471          * So need not to call smp_call_function_single() in that case.
4472          */
4473         if (vcpu->guest_mode && vcpu->cpu != cpu)
4474                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4475         put_cpu();
4476 }