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