SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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         case MSR_P6_EVNTSEL0:
 899         case MSR_P6_EVNTSEL1:
 900                 data = 0;
 901                 break;
 902         case MSR_MTRRcap:
 903                 data = 0x500 | KVM_NR_VAR_MTRR;
 904                 break;
 905         case 0x200 ... 0x2ff:
 906                 return get_msr_mtrr(vcpu, msr, pdata);
 907         case 0xcd: /* fsb frequency */
 908                 data = 3;
 909                 break;
 910         case MSR_IA32_APICBASE:
 911                 data = kvm_get_apic_base(vcpu);
 912                 break;
 913         case MSR_IA32_MISC_ENABLE:
 914                 data = vcpu->arch.ia32_misc_enable_msr;
 915                 break;
 916         case MSR_IA32_PERF_STATUS:
 917                 /* TSC increment by tick */
 918                 data = 1000ULL;
 919                 /* CPU multiplier */
 920                 data |= (((uint64_t)4ULL) << 40);
 921                 break;
 922         case MSR_EFER:
 923                 data = vcpu->arch.shadow_efer;
 924                 break;
 925         case MSR_KVM_WALL_CLOCK:
 926                 data = vcpu->kvm->arch.wall_clock;
 927                 break;
 928         case MSR_KVM_SYSTEM_TIME:
 929                 data = vcpu->arch.time;
 930                 break;
 931         default:
 932                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 933                 return 1;
 934         }
 935         *pdata = data;
 936         return 0;
 937 }
 938 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 939
 940 /*
 941  * Read or write a bunch of msrs. All parameters are kernel addresses.
 942  *
 943  * @return number of msrs set successfully.
 944  */
 945 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 946                     struct kvm_msr_entry *entries,
 947                     int (*do_msr)(struct kvm_vcpu *vcpu,
 948                                   unsigned index, u64 *data))
 949 {
 950         int i;
 951
 952         vcpu_load(vcpu);
 953
 954         down_read(&vcpu->kvm->slots_lock);
 955         for (i = 0; i < msrs->nmsrs; ++i)
 956                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 957                         break;
 958         up_read(&vcpu->kvm->slots_lock);
 959
 960         vcpu_put(vcpu);
 961
 962         return i;
 963 }
 964
 965 /*
 966  * Read or write a bunch of msrs. Parameters are user addresses.
 967  *
 968  * @return number of msrs set successfully.
 969  */
 970 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 971                   int (*do_msr)(struct kvm_vcpu *vcpu,
 972                                 unsigned index, u64 *data),
 973                   int writeback)
 974 {
 975         struct kvm_msrs msrs;
 976         struct kvm_msr_entry *entries;
 977         int r, n;
 978         unsigned size;
 979
 980         r = -EFAULT;
 981         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 982                 goto out;
 983
 984         r = -E2BIG;
 985         if (msrs.nmsrs >= MAX_IO_MSRS)
 986                 goto out;
 987
 988         r = -ENOMEM;
 989         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 990         entries = vmalloc(size);
 991         if (!entries)
 992                 goto out;
 993
 994         r = -EFAULT;
 995         if (copy_from_user(entries, user_msrs->entries, size))
 996                 goto out_free;
 997
 998         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 999         if (r < 0)
1000                 goto out_free;
1001
1002         r = -EFAULT;
1003         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1004                 goto out_free;
1005
1006         r = n;
1007
1008 out_free:
1009         vfree(entries);
1010 out:
1011         return r;
1012 }
1013
1014 int kvm_dev_ioctl_check_extension(long ext)
1015 {
1016         int r;
1017
1018         switch (ext) {
1019         case KVM_CAP_IRQCHIP:
1020         case KVM_CAP_HLT:
1021         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1022         case KVM_CAP_SET_TSS_ADDR:
1023         case KVM_CAP_EXT_CPUID:
1024         case KVM_CAP_CLOCKSOURCE:
1025         case KVM_CAP_PIT:
1026         case KVM_CAP_NOP_IO_DELAY:
1027         case KVM_CAP_MP_STATE:
1028         case KVM_CAP_SYNC_MMU:
1029         case KVM_CAP_REINJECT_CONTROL:
1030         case KVM_CAP_IRQ_INJECT_STATUS:
1031         case KVM_CAP_ASSIGN_DEV_IRQ:
1032                 r = 1;
1033                 break;
1034         case KVM_CAP_COALESCED_MMIO:
1035                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1036                 break;
1037         case KVM_CAP_VAPIC:
1038                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1039                 break;
1040         case KVM_CAP_NR_VCPUS:
1041                 r = KVM_MAX_VCPUS;
1042                 break;
1043         case KVM_CAP_NR_MEMSLOTS:
1044                 r = KVM_MEMORY_SLOTS;
1045                 break;
1046         case KVM_CAP_PV_MMU:
1047                 r = !tdp_enabled;
1048                 break;
1049         case KVM_CAP_IOMMU:
1050                 r = iommu_found();
1051                 break;
1052         default:
1053                 r = 0;
1054                 break;
1055         }
1056         return r;
1057
1058 }
1059
1060 long kvm_arch_dev_ioctl(struct file *filp,
1061                         unsigned int ioctl, unsigned long arg)
1062 {
1063         void __user *argp = (void __user *)arg;
1064         long r;
1065
1066         switch (ioctl) {
1067         case KVM_GET_MSR_INDEX_LIST: {
1068                 struct kvm_msr_list __user *user_msr_list = argp;
1069                 struct kvm_msr_list msr_list;
1070                 unsigned n;
1071
1072                 r = -EFAULT;
1073                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1074                         goto out;
1075                 n = msr_list.nmsrs;
1076                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1077                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1078                         goto out;
1079                 r = -E2BIG;
1080                 if (n < num_msrs_to_save)
1081                         goto out;
1082                 r = -EFAULT;
1083                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1084                                  num_msrs_to_save * sizeof(u32)))
1085                         goto out;
1086                 if (copy_to_user(user_msr_list->indices
1087                                  + num_msrs_to_save * sizeof(u32),
1088                                  &emulated_msrs,
1089                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1090                         goto out;
1091                 r = 0;
1092                 break;
1093         }
1094         case KVM_GET_SUPPORTED_CPUID: {
1095                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1096                 struct kvm_cpuid2 cpuid;
1097
1098                 r = -EFAULT;
1099                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1100                         goto out;
1101                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1102                                                       cpuid_arg->entries);
1103                 if (r)
1104                         goto out;
1105
1106                 r = -EFAULT;
1107                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1108                         goto out;
1109                 r = 0;
1110                 break;
1111         }
1112         default:
1113                 r = -EINVAL;
1114         }
1115 out:
1116         return r;
1117 }
1118
1119 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1120 {
1121         kvm_x86_ops->vcpu_load(vcpu, cpu);
1122         kvm_request_guest_time_update(vcpu);
1123 }
1124
1125 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1126 {
1127         kvm_x86_ops->vcpu_put(vcpu);
1128         kvm_put_guest_fpu(vcpu);
1129 }
1130
1131 static int is_efer_nx(void)
1132 {
1133         unsigned long long efer = 0;
1134
1135         rdmsrl_safe(MSR_EFER, &efer);
1136         return efer & EFER_NX;
1137 }
1138
1139 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1140 {
1141         int i;
1142         struct kvm_cpuid_entry2 *e, *entry;
1143
1144         entry = NULL;
1145         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1146                 e = &vcpu->arch.cpuid_entries[i];
1147                 if (e->function == 0x80000001) {
1148                         entry = e;
1149                         break;
1150                 }
1151         }
1152         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1153                 entry->edx &= ~(1 << 20);
1154                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1155         }
1156 }
1157
1158 /* when an old userspace process fills a new kernel module */
1159 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1160                                     struct kvm_cpuid *cpuid,
1161                                     struct kvm_cpuid_entry __user *entries)
1162 {
1163         int r, i;
1164         struct kvm_cpuid_entry *cpuid_entries;
1165
1166         r = -E2BIG;
1167         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1168                 goto out;
1169         r = -ENOMEM;
1170         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1171         if (!cpuid_entries)
1172                 goto out;
1173         r = -EFAULT;
1174         if (copy_from_user(cpuid_entries, entries,
1175                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1176                 goto out_free;
1177         for (i = 0; i < cpuid->nent; i++) {
1178                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1179                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1180                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1181                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1182                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1183                 vcpu->arch.cpuid_entries[i].index = 0;
1184                 vcpu->arch.cpuid_entries[i].flags = 0;
1185                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1186                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1187                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1188         }
1189         vcpu->arch.cpuid_nent = cpuid->nent;
1190         cpuid_fix_nx_cap(vcpu);
1191         r = 0;
1192
1193 out_free:
1194         vfree(cpuid_entries);
1195 out:
1196         return r;
1197 }
1198
1199 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1200                                      struct kvm_cpuid2 *cpuid,
1201                                      struct kvm_cpuid_entry2 __user *entries)
1202 {
1203         int r;
1204
1205         r = -E2BIG;
1206         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1207                 goto out;
1208         r = -EFAULT;
1209         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1210                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1211                 goto out;
1212         vcpu->arch.cpuid_nent = cpuid->nent;
1213         return 0;
1214
1215 out:
1216         return r;
1217 }
1218
1219 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1220                                      struct kvm_cpuid2 *cpuid,
1221                                      struct kvm_cpuid_entry2 __user *entries)
1222 {
1223         int r;
1224
1225         r = -E2BIG;
1226         if (cpuid->nent < vcpu->arch.cpuid_nent)
1227                 goto out;
1228         r = -EFAULT;
1229         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1230                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1231                 goto out;
1232         return 0;
1233
1234 out:
1235         cpuid->nent = vcpu->arch.cpuid_nent;
1236         return r;
1237 }
1238
1239 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1240                            u32 index)
1241 {
1242         entry->function = function;
1243         entry->index = index;
1244         cpuid_count(entry->function, entry->index,
1245                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1246         entry->flags = 0;
1247 }
1248
1249 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1250                          u32 index, int *nent, int maxnent)
1251 {
1252         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1253                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1254                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1255                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1256                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1257                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1258                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1259                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1260                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1261                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1262         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1263                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1264                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1265                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1266                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1267                 bit(X86_FEATURE_PGE) |
1268                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1269                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1270                 bit(X86_FEATURE_SYSCALL) |
1271                 (is_efer_nx() ? bit(X86_FEATURE_NX) : 0) |
1272 #ifdef CONFIG_X86_64
1273                 bit(X86_FEATURE_LM) |
1274 #endif
1275                 bit(X86_FEATURE_FXSR_OPT) |
1276                 bit(X86_FEATURE_MMXEXT) |
1277                 bit(X86_FEATURE_3DNOWEXT) |
1278                 bit(X86_FEATURE_3DNOW);
1279         const u32 kvm_supported_word3_x86_features =
1280                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1281         const u32 kvm_supported_word6_x86_features =
1282                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1283                 bit(X86_FEATURE_SVM);
1284
1285         /* all calls to cpuid_count() should be made on the same cpu */
1286         get_cpu();
1287         do_cpuid_1_ent(entry, function, index);
1288         ++*nent;
1289
1290         switch (function) {
1291         case 0:
1292                 entry->eax = min(entry->eax, (u32)0xb);
1293                 break;
1294         case 1:
1295                 entry->edx &= kvm_supported_word0_x86_features;
1296                 entry->ecx &= kvm_supported_word3_x86_features;
1297                 break;
1298         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1299          * may return different values. This forces us to get_cpu() before
1300          * issuing the first command, and also to emulate this annoying behavior
1301          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1302         case 2: {
1303                 int t, times = entry->eax & 0xff;
1304
1305                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1306                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1307                 for (t = 1; t < times && *nent < maxnent; ++t) {
1308                         do_cpuid_1_ent(&entry[t], function, 0);
1309                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1310                         ++*nent;
1311                 }
1312                 break;
1313         }
1314         /* function 4 and 0xb have additional index. */
1315         case 4: {
1316                 int i, cache_type;
1317
1318                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1319                 /* read more entries until cache_type is zero */
1320                 for (i = 1; *nent < maxnent; ++i) {
1321                         cache_type = entry[i - 1].eax & 0x1f;
1322                         if (!cache_type)
1323                                 break;
1324                         do_cpuid_1_ent(&entry[i], function, i);
1325                         entry[i].flags |=
1326                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1327                         ++*nent;
1328                 }
1329                 break;
1330         }
1331         case 0xb: {
1332                 int i, level_type;
1333
1334                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1335                 /* read more entries until level_type is zero */
1336                 for (i = 1; *nent < maxnent; ++i) {
1337                         level_type = entry[i - 1].ecx & 0xff00;
1338                         if (!level_type)
1339                                 break;
1340                         do_cpuid_1_ent(&entry[i], function, i);
1341                         entry[i].flags |=
1342                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1343                         ++*nent;
1344                 }
1345                 break;
1346         }
1347         case 0x80000000:
1348                 entry->eax = min(entry->eax, 0x8000001a);
1349                 break;
1350         case 0x80000001:
1351                 entry->edx &= kvm_supported_word1_x86_features;
1352                 entry->ecx &= kvm_supported_word6_x86_features;
1353                 break;
1354         }
1355         put_cpu();
1356 }
1357
1358 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1359                                      struct kvm_cpuid_entry2 __user *entries)
1360 {
1361         struct kvm_cpuid_entry2 *cpuid_entries;
1362         int limit, nent = 0, r = -E2BIG;
1363         u32 func;
1364
1365         if (cpuid->nent < 1)
1366                 goto out;
1367         r = -ENOMEM;
1368         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1369         if (!cpuid_entries)
1370                 goto out;
1371
1372         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1373         limit = cpuid_entries[0].eax;
1374         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1375                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1376                              &nent, cpuid->nent);
1377         r = -E2BIG;
1378         if (nent >= cpuid->nent)
1379                 goto out_free;
1380
1381         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1382         limit = cpuid_entries[nent - 1].eax;
1383         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1384                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1385                              &nent, cpuid->nent);
1386         r = -EFAULT;
1387         if (copy_to_user(entries, cpuid_entries,
1388                          nent * sizeof(struct kvm_cpuid_entry2)))
1389                 goto out_free;
1390         cpuid->nent = nent;
1391         r = 0;
1392
1393 out_free:
1394         vfree(cpuid_entries);
1395 out:
1396         return r;
1397 }
1398
1399 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1400                                     struct kvm_lapic_state *s)
1401 {
1402         vcpu_load(vcpu);
1403         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1404         vcpu_put(vcpu);
1405
1406         return 0;
1407 }
1408
1409 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1410                                     struct kvm_lapic_state *s)
1411 {
1412         vcpu_load(vcpu);
1413         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1414         kvm_apic_post_state_restore(vcpu);
1415         vcpu_put(vcpu);
1416
1417         return 0;
1418 }
1419
1420 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1421                                     struct kvm_interrupt *irq)
1422 {
1423         if (irq->irq < 0 || irq->irq >= 256)
1424                 return -EINVAL;
1425         if (irqchip_in_kernel(vcpu->kvm))
1426                 return -ENXIO;
1427         vcpu_load(vcpu);
1428
1429         set_bit(irq->irq, vcpu->arch.irq_pending);
1430         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1431
1432         vcpu_put(vcpu);
1433
1434         return 0;
1435 }
1436
1437 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1438 {
1439         vcpu_load(vcpu);
1440         kvm_inject_nmi(vcpu);
1441         vcpu_put(vcpu);
1442
1443         return 0;
1444 }
1445
1446 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1447                                            struct kvm_tpr_access_ctl *tac)
1448 {
1449         if (tac->flags)
1450                 return -EINVAL;
1451         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1452         return 0;
1453 }
1454
1455 long kvm_arch_vcpu_ioctl(struct file *filp,
1456                          unsigned int ioctl, unsigned long arg)
1457 {
1458         struct kvm_vcpu *vcpu = filp->private_data;
1459         void __user *argp = (void __user *)arg;
1460         int r;
1461         struct kvm_lapic_state *lapic = NULL;
1462
1463         switch (ioctl) {
1464         case KVM_GET_LAPIC: {
1465                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1466
1467                 r = -ENOMEM;
1468                 if (!lapic)
1469                         goto out;
1470                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1471                 if (r)
1472                         goto out;
1473                 r = -EFAULT;
1474                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1475                         goto out;
1476                 r = 0;
1477                 break;
1478         }
1479         case KVM_SET_LAPIC: {
1480                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1481                 r = -ENOMEM;
1482                 if (!lapic)
1483                         goto out;
1484                 r = -EFAULT;
1485                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1486                         goto out;
1487                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1488                 if (r)
1489                         goto out;
1490                 r = 0;
1491                 break;
1492         }
1493         case KVM_INTERRUPT: {
1494                 struct kvm_interrupt irq;
1495
1496                 r = -EFAULT;
1497                 if (copy_from_user(&irq, argp, sizeof irq))
1498                         goto out;
1499                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1500                 if (r)
1501                         goto out;
1502                 r = 0;
1503                 break;
1504         }
1505         case KVM_NMI: {
1506                 r = kvm_vcpu_ioctl_nmi(vcpu);
1507                 if (r)
1508                         goto out;
1509                 r = 0;
1510                 break;
1511         }
1512         case KVM_SET_CPUID: {
1513                 struct kvm_cpuid __user *cpuid_arg = argp;
1514                 struct kvm_cpuid cpuid;
1515
1516                 r = -EFAULT;
1517                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1518                         goto out;
1519                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1520                 if (r)
1521                         goto out;
1522                 break;
1523         }
1524         case KVM_SET_CPUID2: {
1525                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1526                 struct kvm_cpuid2 cpuid;
1527
1528                 r = -EFAULT;
1529                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1530                         goto out;
1531                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1532                                               cpuid_arg->entries);
1533                 if (r)
1534                         goto out;
1535                 break;
1536         }
1537         case KVM_GET_CPUID2: {
1538                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1539                 struct kvm_cpuid2 cpuid;
1540
1541                 r = -EFAULT;
1542                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1543                         goto out;
1544                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1545                                               cpuid_arg->entries);
1546                 if (r)
1547                         goto out;
1548                 r = -EFAULT;
1549                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1550                         goto out;
1551                 r = 0;
1552                 break;
1553         }
1554         case KVM_GET_MSRS:
1555                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1556                 break;
1557         case KVM_SET_MSRS:
1558                 r = msr_io(vcpu, argp, do_set_msr, 0);
1559                 break;
1560         case KVM_TPR_ACCESS_REPORTING: {
1561                 struct kvm_tpr_access_ctl tac;
1562
1563                 r = -EFAULT;
1564                 if (copy_from_user(&tac, argp, sizeof tac))
1565                         goto out;
1566                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1567                 if (r)
1568                         goto out;
1569                 r = -EFAULT;
1570                 if (copy_to_user(argp, &tac, sizeof tac))
1571                         goto out;
1572                 r = 0;
1573                 break;
1574         };
1575         case KVM_SET_VAPIC_ADDR: {
1576                 struct kvm_vapic_addr va;
1577
1578                 r = -EINVAL;
1579                 if (!irqchip_in_kernel(vcpu->kvm))
1580                         goto out;
1581                 r = -EFAULT;
1582                 if (copy_from_user(&va, argp, sizeof va))
1583                         goto out;
1584                 r = 0;
1585                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1586                 break;
1587         }
1588         default:
1589                 r = -EINVAL;
1590         }
1591 out:
1592         if (lapic)
1593                 kfree(lapic);
1594         return r;
1595 }
1596
1597 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1598 {
1599         int ret;
1600
1601         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1602                 return -1;
1603         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1604         return ret;
1605 }
1606
1607 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1608                                           u32 kvm_nr_mmu_pages)
1609 {
1610         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1611                 return -EINVAL;
1612
1613         down_write(&kvm->slots_lock);
1614
1615         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1616         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1617
1618         up_write(&kvm->slots_lock);
1619         return 0;
1620 }
1621
1622 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1623 {
1624         return kvm->arch.n_alloc_mmu_pages;
1625 }
1626
1627 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1628 {
1629         int i;
1630         struct kvm_mem_alias *alias;
1631
1632         for (i = 0; i < kvm->arch.naliases; ++i) {
1633                 alias = &kvm->arch.aliases[i];
1634                 if (gfn >= alias->base_gfn
1635                     && gfn < alias->base_gfn + alias->npages)
1636                         return alias->target_gfn + gfn - alias->base_gfn;
1637         }
1638         return gfn;
1639 }
1640
1641 /*
1642  * Set a new alias region.  Aliases map a portion of physical memory into
1643  * another portion.  This is useful for memory windows, for example the PC
1644  * VGA region.
1645  */
1646 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1647                                          struct kvm_memory_alias *alias)
1648 {
1649         int r, n;
1650         struct kvm_mem_alias *p;
1651
1652         r = -EINVAL;
1653         /* General sanity checks */
1654         if (alias->memory_size & (PAGE_SIZE - 1))
1655                 goto out;
1656         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1657                 goto out;
1658         if (alias->slot >= KVM_ALIAS_SLOTS)
1659                 goto out;
1660         if (alias->guest_phys_addr + alias->memory_size
1661             < alias->guest_phys_addr)
1662                 goto out;
1663         if (alias->target_phys_addr + alias->memory_size
1664             < alias->target_phys_addr)
1665                 goto out;
1666
1667         down_write(&kvm->slots_lock);
1668         spin_lock(&kvm->mmu_lock);
1669
1670         p = &kvm->arch.aliases[alias->slot];
1671         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1672         p->npages = alias->memory_size >> PAGE_SHIFT;
1673         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1674
1675         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1676                 if (kvm->arch.aliases[n - 1].npages)
1677                         break;
1678         kvm->arch.naliases = n;
1679
1680         spin_unlock(&kvm->mmu_lock);
1681         kvm_mmu_zap_all(kvm);
1682
1683         up_write(&kvm->slots_lock);
1684
1685         return 0;
1686
1687 out:
1688         return r;
1689 }
1690
1691 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1692 {
1693         int r;
1694
1695         r = 0;
1696         switch (chip->chip_id) {
1697         case KVM_IRQCHIP_PIC_MASTER:
1698                 memcpy(&chip->chip.pic,
1699                         &pic_irqchip(kvm)->pics[0],
1700                         sizeof(struct kvm_pic_state));
1701                 break;
1702         case KVM_IRQCHIP_PIC_SLAVE:
1703                 memcpy(&chip->chip.pic,
1704                         &pic_irqchip(kvm)->pics[1],
1705                         sizeof(struct kvm_pic_state));
1706                 break;
1707         case KVM_IRQCHIP_IOAPIC:
1708                 memcpy(&chip->chip.ioapic,
1709                         ioapic_irqchip(kvm),
1710                         sizeof(struct kvm_ioapic_state));
1711                 break;
1712         default:
1713                 r = -EINVAL;
1714                 break;
1715         }
1716         return r;
1717 }
1718
1719 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1720 {
1721         int r;
1722
1723         r = 0;
1724         switch (chip->chip_id) {
1725         case KVM_IRQCHIP_PIC_MASTER:
1726                 memcpy(&pic_irqchip(kvm)->pics[0],
1727                         &chip->chip.pic,
1728                         sizeof(struct kvm_pic_state));
1729                 break;
1730         case KVM_IRQCHIP_PIC_SLAVE:
1731                 memcpy(&pic_irqchip(kvm)->pics[1],
1732                         &chip->chip.pic,
1733                         sizeof(struct kvm_pic_state));
1734                 break;
1735         case KVM_IRQCHIP_IOAPIC:
1736                 memcpy(ioapic_irqchip(kvm),
1737                         &chip->chip.ioapic,
1738                         sizeof(struct kvm_ioapic_state));
1739                 break;
1740         default:
1741                 r = -EINVAL;
1742                 break;
1743         }
1744         kvm_pic_update_irq(pic_irqchip(kvm));
1745         return r;
1746 }
1747
1748 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1749 {
1750         int r = 0;
1751
1752         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1753         return r;
1754 }
1755
1756 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1757 {
1758         int r = 0;
1759
1760         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1761         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1762         return r;
1763 }
1764
1765 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1766                                  struct kvm_reinject_control *control)
1767 {
1768         if (!kvm->arch.vpit)
1769                 return -ENXIO;
1770         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1771         return 0;
1772 }
1773
1774 /*
1775  * Get (and clear) the dirty memory log for a memory slot.
1776  */
1777 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1778                                       struct kvm_dirty_log *log)
1779 {
1780         int r;
1781         int n;
1782         struct kvm_memory_slot *memslot;
1783         int is_dirty = 0;
1784
1785         down_write(&kvm->slots_lock);
1786
1787         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1788         if (r)
1789                 goto out;
1790
1791         /* If nothing is dirty, don't bother messing with page tables. */
1792         if (is_dirty) {
1793                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1794                 kvm_flush_remote_tlbs(kvm);
1795                 memslot = &kvm->memslots[log->slot];
1796                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1797                 memset(memslot->dirty_bitmap, 0, n);
1798         }
1799         r = 0;
1800 out:
1801         up_write(&kvm->slots_lock);
1802         return r;
1803 }
1804
1805 long kvm_arch_vm_ioctl(struct file *filp,
1806                        unsigned int ioctl, unsigned long arg)
1807 {
1808         struct kvm *kvm = filp->private_data;
1809         void __user *argp = (void __user *)arg;
1810         int r = -EINVAL;
1811         /*
1812          * This union makes it completely explicit to gcc-3.x
1813          * that these two variables' stack usage should be
1814          * combined, not added together.
1815          */
1816         union {
1817                 struct kvm_pit_state ps;
1818                 struct kvm_memory_alias alias;
1819         } u;
1820
1821         switch (ioctl) {
1822         case KVM_SET_TSS_ADDR:
1823                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1824                 if (r < 0)
1825                         goto out;
1826                 break;
1827         case KVM_SET_MEMORY_REGION: {
1828                 struct kvm_memory_region kvm_mem;
1829                 struct kvm_userspace_memory_region kvm_userspace_mem;
1830
1831                 r = -EFAULT;
1832                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1833                         goto out;
1834                 kvm_userspace_mem.slot = kvm_mem.slot;
1835                 kvm_userspace_mem.flags = kvm_mem.flags;
1836                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1837                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1838                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1839                 if (r)
1840                         goto out;
1841                 break;
1842         }
1843         case KVM_SET_NR_MMU_PAGES:
1844                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1845                 if (r)
1846                         goto out;
1847                 break;
1848         case KVM_GET_NR_MMU_PAGES:
1849                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1850                 break;
1851         case KVM_SET_MEMORY_ALIAS:
1852                 r = -EFAULT;
1853                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1854                         goto out;
1855                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1856                 if (r)
1857                         goto out;
1858                 break;
1859         case KVM_CREATE_IRQCHIP:
1860                 r = -ENOMEM;
1861                 kvm->arch.vpic = kvm_create_pic(kvm);
1862                 if (kvm->arch.vpic) {
1863                         r = kvm_ioapic_init(kvm);
1864                         if (r) {
1865                                 kfree(kvm->arch.vpic);
1866                                 kvm->arch.vpic = NULL;
1867                                 goto out;
1868                         }
1869                 } else
1870                         goto out;
1871                 r = kvm_setup_default_irq_routing(kvm);
1872                 if (r) {
1873                         kfree(kvm->arch.vpic);
1874                         kfree(kvm->arch.vioapic);
1875                         goto out;
1876                 }
1877                 break;
1878         case KVM_CREATE_PIT:
1879                 mutex_lock(&kvm->lock);
1880                 r = -EEXIST;
1881                 if (kvm->arch.vpit)
1882                         goto create_pit_unlock;
1883                 r = -ENOMEM;
1884                 kvm->arch.vpit = kvm_create_pit(kvm);
1885                 if (kvm->arch.vpit)
1886                         r = 0;
1887         create_pit_unlock:
1888                 mutex_unlock(&kvm->lock);
1889                 break;
1890         case KVM_IRQ_LINE_STATUS:
1891         case KVM_IRQ_LINE: {
1892                 struct kvm_irq_level irq_event;
1893
1894                 r = -EFAULT;
1895                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1896                         goto out;
1897                 if (irqchip_in_kernel(kvm)) {
1898                         __s32 status;
1899                         mutex_lock(&kvm->lock);
1900                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1901                                         irq_event.irq, irq_event.level);
1902                         mutex_unlock(&kvm->lock);
1903                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1904                                 irq_event.status = status;
1905                                 if (copy_to_user(argp, &irq_event,
1906                                                         sizeof irq_event))
1907                                         goto out;
1908                         }
1909                         r = 0;
1910                 }
1911                 break;
1912         }
1913         case KVM_GET_IRQCHIP: {
1914                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1915                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1916
1917                 r = -ENOMEM;
1918                 if (!chip)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_from_user(chip, argp, sizeof *chip))
1922                         goto get_irqchip_out;
1923                 r = -ENXIO;
1924                 if (!irqchip_in_kernel(kvm))
1925                         goto get_irqchip_out;
1926                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1927                 if (r)
1928                         goto get_irqchip_out;
1929                 r = -EFAULT;
1930                 if (copy_to_user(argp, chip, sizeof *chip))
1931                         goto get_irqchip_out;
1932                 r = 0;
1933         get_irqchip_out:
1934                 kfree(chip);
1935                 if (r)
1936                         goto out;
1937                 break;
1938         }
1939         case KVM_SET_IRQCHIP: {
1940                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1941                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1942
1943                 r = -ENOMEM;
1944                 if (!chip)
1945                         goto out;
1946                 r = -EFAULT;
1947                 if (copy_from_user(chip, argp, sizeof *chip))
1948                         goto set_irqchip_out;
1949                 r = -ENXIO;
1950                 if (!irqchip_in_kernel(kvm))
1951                         goto set_irqchip_out;
1952                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1953                 if (r)
1954                         goto set_irqchip_out;
1955                 r = 0;
1956         set_irqchip_out:
1957                 kfree(chip);
1958                 if (r)
1959                         goto out;
1960                 break;
1961         }
1962         case KVM_GET_PIT: {
1963                 r = -EFAULT;
1964                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1965                         goto out;
1966                 r = -ENXIO;
1967                 if (!kvm->arch.vpit)
1968                         goto out;
1969                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1970                 if (r)
1971                         goto out;
1972                 r = -EFAULT;
1973                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1974                         goto out;
1975                 r = 0;
1976                 break;
1977         }
1978         case KVM_SET_PIT: {
1979                 r = -EFAULT;
1980                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1981                         goto out;
1982                 r = -ENXIO;
1983                 if (!kvm->arch.vpit)
1984                         goto out;
1985                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1986                 if (r)
1987                         goto out;
1988                 r = 0;
1989                 break;
1990         }
1991         case KVM_REINJECT_CONTROL: {
1992                 struct kvm_reinject_control control;
1993                 r =  -EFAULT;
1994                 if (copy_from_user(&control, argp, sizeof(control)))
1995                         goto out;
1996                 r = kvm_vm_ioctl_reinject(kvm, &control);
1997                 if (r)
1998                         goto out;
1999                 r = 0;
2000                 break;
2001         }
2002         default:
2003                 ;
2004         }
2005 out:
2006         return r;
2007 }
2008
2009 static void kvm_init_msr_list(void)
2010 {
2011         u32 dummy[2];
2012         unsigned i, j;
2013
2014         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2015                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2016                         continue;
2017                 if (j < i)
2018                         msrs_to_save[j] = msrs_to_save[i];
2019                 j++;
2020         }
2021         num_msrs_to_save = j;
2022 }
2023
2024 /*
2025  * Only apic need an MMIO device hook, so shortcut now..
2026  */
2027 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2028                                                 gpa_t addr, int len,
2029                                                 int is_write)
2030 {
2031         struct kvm_io_device *dev;
2032
2033         if (vcpu->arch.apic) {
2034                 dev = &vcpu->arch.apic->dev;
2035                 if (dev->in_range(dev, addr, len, is_write))
2036                         return dev;
2037         }
2038         return NULL;
2039 }
2040
2041
2042 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2043                                                 gpa_t addr, int len,
2044                                                 int is_write)
2045 {
2046         struct kvm_io_device *dev;
2047
2048         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2049         if (dev == NULL)
2050                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2051                                           is_write);
2052         return dev;
2053 }
2054
2055 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2056                                struct kvm_vcpu *vcpu)
2057 {
2058         void *data = val;
2059         int r = X86EMUL_CONTINUE;
2060
2061         while (bytes) {
2062                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2063                 unsigned offset = addr & (PAGE_SIZE-1);
2064                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2065                 int ret;
2066
2067                 if (gpa == UNMAPPED_GVA) {
2068                         r = X86EMUL_PROPAGATE_FAULT;
2069                         goto out;
2070                 }
2071                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2072                 if (ret < 0) {
2073                         r = X86EMUL_UNHANDLEABLE;
2074                         goto out;
2075                 }
2076
2077                 bytes -= toread;
2078                 data += toread;
2079                 addr += toread;
2080         }
2081 out:
2082         return r;
2083 }
2084
2085 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2086                                 struct kvm_vcpu *vcpu)
2087 {
2088         void *data = val;
2089         int r = X86EMUL_CONTINUE;
2090
2091         while (bytes) {
2092                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2093                 unsigned offset = addr & (PAGE_SIZE-1);
2094                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2095                 int ret;
2096
2097                 if (gpa == UNMAPPED_GVA) {
2098                         r = X86EMUL_PROPAGATE_FAULT;
2099                         goto out;
2100                 }
2101                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2102                 if (ret < 0) {
2103                         r = X86EMUL_UNHANDLEABLE;
2104                         goto out;
2105                 }
2106
2107                 bytes -= towrite;
2108                 data += towrite;
2109                 addr += towrite;
2110         }
2111 out:
2112         return r;
2113 }
2114
2115
2116 static int emulator_read_emulated(unsigned long addr,
2117                                   void *val,
2118                                   unsigned int bytes,
2119                                   struct kvm_vcpu *vcpu)
2120 {
2121         struct kvm_io_device *mmio_dev;
2122         gpa_t                 gpa;
2123
2124         if (vcpu->mmio_read_completed) {
2125                 memcpy(val, vcpu->mmio_data, bytes);
2126                 vcpu->mmio_read_completed = 0;
2127                 return X86EMUL_CONTINUE;
2128         }
2129
2130         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2131
2132         /* For APIC access vmexit */
2133         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2134                 goto mmio;
2135
2136         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2137                                 == X86EMUL_CONTINUE)
2138                 return X86EMUL_CONTINUE;
2139         if (gpa == UNMAPPED_GVA)
2140                 return X86EMUL_PROPAGATE_FAULT;
2141
2142 mmio:
2143         /*
2144          * Is this MMIO handled locally?
2145          */
2146         mutex_lock(&vcpu->kvm->lock);
2147         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2148         if (mmio_dev) {
2149                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2150                 mutex_unlock(&vcpu->kvm->lock);
2151                 return X86EMUL_CONTINUE;
2152         }
2153         mutex_unlock(&vcpu->kvm->lock);
2154
2155         vcpu->mmio_needed = 1;
2156         vcpu->mmio_phys_addr = gpa;
2157         vcpu->mmio_size = bytes;
2158         vcpu->mmio_is_write = 0;
2159
2160         return X86EMUL_UNHANDLEABLE;
2161 }
2162
2163 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2164                           const void *val, int bytes)
2165 {
2166         int ret;
2167
2168         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2169         if (ret < 0)
2170                 return 0;
2171         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2172         return 1;
2173 }
2174
2175 static int emulator_write_emulated_onepage(unsigned long addr,
2176                                            const void *val,
2177                                            unsigned int bytes,
2178                                            struct kvm_vcpu *vcpu)
2179 {
2180         struct kvm_io_device *mmio_dev;
2181         gpa_t                 gpa;
2182
2183         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2184
2185         if (gpa == UNMAPPED_GVA) {
2186                 kvm_inject_page_fault(vcpu, addr, 2);
2187                 return X86EMUL_PROPAGATE_FAULT;
2188         }
2189
2190         /* For APIC access vmexit */
2191         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2192                 goto mmio;
2193
2194         if (emulator_write_phys(vcpu, gpa, val, bytes))
2195                 return X86EMUL_CONTINUE;
2196
2197 mmio:
2198         /*
2199          * Is this MMIO handled locally?
2200          */
2201         mutex_lock(&vcpu->kvm->lock);
2202         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2203         if (mmio_dev) {
2204                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2205                 mutex_unlock(&vcpu->kvm->lock);
2206                 return X86EMUL_CONTINUE;
2207         }
2208         mutex_unlock(&vcpu->kvm->lock);
2209
2210         vcpu->mmio_needed = 1;
2211         vcpu->mmio_phys_addr = gpa;
2212         vcpu->mmio_size = bytes;
2213         vcpu->mmio_is_write = 1;
2214         memcpy(vcpu->mmio_data, val, bytes);
2215
2216         return X86EMUL_CONTINUE;
2217 }
2218
2219 int emulator_write_emulated(unsigned long addr,
2220                                    const void *val,
2221                                    unsigned int bytes,
2222                                    struct kvm_vcpu *vcpu)
2223 {
2224         /* Crossing a page boundary? */
2225         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2226                 int rc, now;
2227
2228                 now = -addr & ~PAGE_MASK;
2229                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2230                 if (rc != X86EMUL_CONTINUE)
2231                         return rc;
2232                 addr += now;
2233                 val += now;
2234                 bytes -= now;
2235         }
2236         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2237 }
2238 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2239
2240 static int emulator_cmpxchg_emulated(unsigned long addr,
2241                                      const void *old,
2242                                      const void *new,
2243                                      unsigned int bytes,
2244                                      struct kvm_vcpu *vcpu)
2245 {
2246         static int reported;
2247
2248         if (!reported) {
2249                 reported = 1;
2250                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2251         }
2252 #ifndef CONFIG_X86_64
2253         /* guests cmpxchg8b have to be emulated atomically */
2254         if (bytes == 8) {
2255                 gpa_t gpa;
2256                 struct page *page;
2257                 char *kaddr;
2258                 u64 val;
2259
2260                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2261
2262                 if (gpa == UNMAPPED_GVA ||
2263                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2264                         goto emul_write;
2265
2266                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2267                         goto emul_write;
2268
2269                 val = *(u64 *)new;
2270
2271                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2272
2273                 kaddr = kmap_atomic(page, KM_USER0);
2274                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2275                 kunmap_atomic(kaddr, KM_USER0);
2276                 kvm_release_page_dirty(page);
2277         }
2278 emul_write:
2279 #endif
2280
2281         return emulator_write_emulated(addr, new, bytes, vcpu);
2282 }
2283
2284 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2285 {
2286         return kvm_x86_ops->get_segment_base(vcpu, seg);
2287 }
2288
2289 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2290 {
2291         kvm_mmu_invlpg(vcpu, address);
2292         return X86EMUL_CONTINUE;
2293 }
2294
2295 int emulate_clts(struct kvm_vcpu *vcpu)
2296 {
2297         KVMTRACE_0D(CLTS, vcpu, handler);
2298         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2299         return X86EMUL_CONTINUE;
2300 }
2301
2302 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2303 {
2304         struct kvm_vcpu *vcpu = ctxt->vcpu;
2305
2306         switch (dr) {
2307         case 0 ... 3:
2308                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2309                 return X86EMUL_CONTINUE;
2310         default:
2311                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2312                 return X86EMUL_UNHANDLEABLE;
2313         }
2314 }
2315
2316 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2317 {
2318         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2319         int exception;
2320
2321         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2322         if (exception) {
2323                 /* FIXME: better handling */
2324                 return X86EMUL_UNHANDLEABLE;
2325         }
2326         return X86EMUL_CONTINUE;
2327 }
2328
2329 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2330 {
2331         u8 opcodes[4];
2332         unsigned long rip = kvm_rip_read(vcpu);
2333         unsigned long rip_linear;
2334
2335         if (!printk_ratelimit())
2336                 return;
2337
2338         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2339
2340         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2341
2342         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2343                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2344 }
2345 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2346
2347 static struct x86_emulate_ops emulate_ops = {
2348         .read_std            = kvm_read_guest_virt,
2349         .read_emulated       = emulator_read_emulated,
2350         .write_emulated      = emulator_write_emulated,
2351         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2352 };
2353
2354 static void cache_all_regs(struct kvm_vcpu *vcpu)
2355 {
2356         kvm_register_read(vcpu, VCPU_REGS_RAX);
2357         kvm_register_read(vcpu, VCPU_REGS_RSP);
2358         kvm_register_read(vcpu, VCPU_REGS_RIP);
2359         vcpu->arch.regs_dirty = ~0;
2360 }
2361
2362 int emulate_instruction(struct kvm_vcpu *vcpu,
2363                         struct kvm_run *run,
2364                         unsigned long cr2,
2365                         u16 error_code,
2366                         int emulation_type)
2367 {
2368         int r;
2369         struct decode_cache *c;
2370
2371         kvm_clear_exception_queue(vcpu);
2372         vcpu->arch.mmio_fault_cr2 = cr2;
2373         /*
2374          * TODO: fix x86_emulate.c to use guest_read/write_register
2375          * instead of direct ->regs accesses, can save hundred cycles
2376          * on Intel for instructions that don't read/change RSP, for
2377          * for example.
2378          */
2379         cache_all_regs(vcpu);
2380
2381         vcpu->mmio_is_write = 0;
2382         vcpu->arch.pio.string = 0;
2383
2384         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2385                 int cs_db, cs_l;
2386                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2387
2388                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2389                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2390                 vcpu->arch.emulate_ctxt.mode =
2391                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2392                         ? X86EMUL_MODE_REAL : cs_l
2393                         ? X86EMUL_MODE_PROT64 : cs_db
2394                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2395
2396                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2397
2398                 /* Reject the instructions other than VMCALL/VMMCALL when
2399                  * try to emulate invalid opcode */
2400                 c = &vcpu->arch.emulate_ctxt.decode;
2401                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2402                     (!(c->twobyte && c->b == 0x01 &&
2403                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2404                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2405                         return EMULATE_FAIL;
2406
2407                 ++vcpu->stat.insn_emulation;
2408                 if (r)  {
2409                         ++vcpu->stat.insn_emulation_fail;
2410                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2411                                 return EMULATE_DONE;
2412                         return EMULATE_FAIL;
2413                 }
2414         }
2415
2416         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2417
2418         if (vcpu->arch.pio.string)
2419                 return EMULATE_DO_MMIO;
2420
2421         if ((r || vcpu->mmio_is_write) && run) {
2422                 run->exit_reason = KVM_EXIT_MMIO;
2423                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2424                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2425                 run->mmio.len = vcpu->mmio_size;
2426                 run->mmio.is_write = vcpu->mmio_is_write;
2427         }
2428
2429         if (r) {
2430                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2431                         return EMULATE_DONE;
2432                 if (!vcpu->mmio_needed) {
2433                         kvm_report_emulation_failure(vcpu, "mmio");
2434                         return EMULATE_FAIL;
2435                 }
2436                 return EMULATE_DO_MMIO;
2437         }
2438
2439         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2440
2441         if (vcpu->mmio_is_write) {
2442                 vcpu->mmio_needed = 0;
2443                 return EMULATE_DO_MMIO;
2444         }
2445
2446         return EMULATE_DONE;
2447 }
2448 EXPORT_SYMBOL_GPL(emulate_instruction);
2449
2450 static int pio_copy_data(struct kvm_vcpu *vcpu)
2451 {
2452         void *p = vcpu->arch.pio_data;
2453         gva_t q = vcpu->arch.pio.guest_gva;
2454         unsigned bytes;
2455         int ret;
2456
2457         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2458         if (vcpu->arch.pio.in)
2459                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2460         else
2461                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2462         return ret;
2463 }
2464
2465 int complete_pio(struct kvm_vcpu *vcpu)
2466 {
2467         struct kvm_pio_request *io = &vcpu->arch.pio;
2468         long delta;
2469         int r;
2470         unsigned long val;
2471
2472         if (!io->string) {
2473                 if (io->in) {
2474                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2475                         memcpy(&val, vcpu->arch.pio_data, io->size);
2476                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2477                 }
2478         } else {
2479                 if (io->in) {
2480                         r = pio_copy_data(vcpu);
2481                         if (r)
2482                                 return r;
2483                 }
2484
2485                 delta = 1;
2486                 if (io->rep) {
2487                         delta *= io->cur_count;
2488                         /*
2489                          * The size of the register should really depend on
2490                          * current address size.
2491                          */
2492                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2493                         val -= delta;
2494                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2495                 }
2496                 if (io->down)
2497                         delta = -delta;
2498                 delta *= io->size;
2499                 if (io->in) {
2500                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2501                         val += delta;
2502                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2503                 } else {
2504                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2505                         val += delta;
2506                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2507                 }
2508         }
2509
2510         io->count -= io->cur_count;
2511         io->cur_count = 0;
2512
2513         return 0;
2514 }
2515
2516 static void kernel_pio(struct kvm_io_device *pio_dev,
2517                        struct kvm_vcpu *vcpu,
2518                        void *pd)
2519 {
2520         /* TODO: String I/O for in kernel device */
2521
2522         mutex_lock(&vcpu->kvm->lock);
2523         if (vcpu->arch.pio.in)
2524                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2525                                   vcpu->arch.pio.size,
2526                                   pd);
2527         else
2528                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2529                                    vcpu->arch.pio.size,
2530                                    pd);
2531         mutex_unlock(&vcpu->kvm->lock);
2532 }
2533
2534 static void pio_string_write(struct kvm_io_device *pio_dev,
2535                              struct kvm_vcpu *vcpu)
2536 {
2537         struct kvm_pio_request *io = &vcpu->arch.pio;
2538         void *pd = vcpu->arch.pio_data;
2539         int i;
2540
2541         mutex_lock(&vcpu->kvm->lock);
2542         for (i = 0; i < io->cur_count; i++) {
2543                 kvm_iodevice_write(pio_dev, io->port,
2544                                    io->size,
2545                                    pd);
2546                 pd += io->size;
2547         }
2548         mutex_unlock(&vcpu->kvm->lock);
2549 }
2550
2551 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2552                                                gpa_t addr, int len,
2553                                                int is_write)
2554 {
2555         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2556 }
2557
2558 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2559                   int size, unsigned port)
2560 {
2561         struct kvm_io_device *pio_dev;
2562         unsigned long val;
2563
2564         vcpu->run->exit_reason = KVM_EXIT_IO;
2565         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2566         vcpu->run->io.size = vcpu->arch.pio.size = size;
2567         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2568         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2569         vcpu->run->io.port = vcpu->arch.pio.port = port;
2570         vcpu->arch.pio.in = in;
2571         vcpu->arch.pio.string = 0;
2572         vcpu->arch.pio.down = 0;
2573         vcpu->arch.pio.rep = 0;
2574
2575         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2576                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2577                             handler);
2578         else
2579                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2580                             handler);
2581
2582         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2583         memcpy(vcpu->arch.pio_data, &val, 4);
2584
2585         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2586         if (pio_dev) {
2587                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2588                 complete_pio(vcpu);
2589                 return 1;
2590         }
2591         return 0;
2592 }
2593 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2594
2595 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2596                   int size, unsigned long count, int down,
2597                   gva_t address, int rep, unsigned port)
2598 {
2599         unsigned now, in_page;
2600         int ret = 0;
2601         struct kvm_io_device *pio_dev;
2602
2603         vcpu->run->exit_reason = KVM_EXIT_IO;
2604         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2605         vcpu->run->io.size = vcpu->arch.pio.size = size;
2606         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2607         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2608         vcpu->run->io.port = vcpu->arch.pio.port = port;
2609         vcpu->arch.pio.in = in;
2610         vcpu->arch.pio.string = 1;
2611         vcpu->arch.pio.down = down;
2612         vcpu->arch.pio.rep = rep;
2613
2614         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2615                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2616                             handler);
2617         else
2618                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2619                             handler);
2620
2621         if (!count) {
2622                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2623                 return 1;
2624         }
2625
2626         if (!down)
2627                 in_page = PAGE_SIZE - offset_in_page(address);
2628         else
2629                 in_page = offset_in_page(address) + size;
2630         now = min(count, (unsigned long)in_page / size);
2631         if (!now)
2632                 now = 1;
2633         if (down) {
2634                 /*
2635                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2636                  */
2637                 pr_unimpl(vcpu, "guest string pio down\n");
2638                 kvm_inject_gp(vcpu, 0);
2639                 return 1;
2640         }
2641         vcpu->run->io.count = now;
2642         vcpu->arch.pio.cur_count = now;
2643
2644         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2645                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2646
2647         vcpu->arch.pio.guest_gva = address;
2648
2649         pio_dev = vcpu_find_pio_dev(vcpu, port,
2650                                     vcpu->arch.pio.cur_count,
2651                                     !vcpu->arch.pio.in);
2652         if (!vcpu->arch.pio.in) {
2653                 /* string PIO write */
2654                 ret = pio_copy_data(vcpu);
2655                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2656                         kvm_inject_gp(vcpu, 0);
2657                         return 1;
2658                 }
2659                 if (ret == 0 && pio_dev) {
2660                         pio_string_write(pio_dev, vcpu);
2661                         complete_pio(vcpu);
2662                         if (vcpu->arch.pio.count == 0)
2663                                 ret = 1;
2664                 }
2665         } else if (pio_dev)
2666                 pr_unimpl(vcpu, "no string pio read support yet, "
2667                        "port %x size %d count %ld\n",
2668                         port, size, count);
2669
2670         return ret;
2671 }
2672 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2673
2674 static void bounce_off(void *info)
2675 {
2676         /* nothing */
2677 }
2678
2679 static unsigned int  ref_freq;
2680 static unsigned long tsc_khz_ref;
2681
2682 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2683                                      void *data)
2684 {
2685         struct cpufreq_freqs *freq = data;
2686         struct kvm *kvm;
2687         struct kvm_vcpu *vcpu;
2688         int i, send_ipi = 0;
2689
2690         if (!ref_freq)
2691                 ref_freq = freq->old;
2692
2693         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2694                 return 0;
2695         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2696                 return 0;
2697         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2698
2699         spin_lock(&kvm_lock);
2700         list_for_each_entry(kvm, &vm_list, vm_list) {
2701                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2702                         vcpu = kvm->vcpus[i];
2703                         if (!vcpu)
2704                                 continue;
2705                         if (vcpu->cpu != freq->cpu)
2706                                 continue;
2707                         if (!kvm_request_guest_time_update(vcpu))
2708                                 continue;
2709                         if (vcpu->cpu != smp_processor_id())
2710                                 send_ipi++;
2711                 }
2712         }
2713         spin_unlock(&kvm_lock);
2714
2715         if (freq->old < freq->new && send_ipi) {
2716                 /*
2717                  * We upscale the frequency.  Must make the guest
2718                  * doesn't see old kvmclock values while running with
2719                  * the new frequency, otherwise we risk the guest sees
2720                  * time go backwards.
2721                  *
2722                  * In case we update the frequency for another cpu
2723                  * (which might be in guest context) send an interrupt
2724                  * to kick the cpu out of guest context.  Next time
2725                  * guest context is entered kvmclock will be updated,
2726                  * so the guest will not see stale values.
2727                  */
2728                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2729         }
2730         return 0;
2731 }
2732
2733 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2734         .notifier_call  = kvmclock_cpufreq_notifier
2735 };
2736
2737 int kvm_arch_init(void *opaque)
2738 {
2739         int r, cpu;
2740         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2741
2742         if (kvm_x86_ops) {
2743                 printk(KERN_ERR "kvm: already loaded the other module\n");
2744                 r = -EEXIST;
2745                 goto out;
2746         }
2747
2748         if (!ops->cpu_has_kvm_support()) {
2749                 printk(KERN_ERR "kvm: no hardware support\n");
2750                 r = -EOPNOTSUPP;
2751                 goto out;
2752         }
2753         if (ops->disabled_by_bios()) {
2754                 printk(KERN_ERR "kvm: disabled by bios\n");
2755                 r = -EOPNOTSUPP;
2756                 goto out;
2757         }
2758
2759         r = kvm_mmu_module_init();
2760         if (r)
2761                 goto out;
2762
2763         kvm_init_msr_list();
2764
2765         kvm_x86_ops = ops;
2766         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2767         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2768         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2769                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2770
2771         for_each_possible_cpu(cpu)
2772                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2773         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2774                 tsc_khz_ref = tsc_khz;
2775                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2776                                           CPUFREQ_TRANSITION_NOTIFIER);
2777         }
2778
2779         return 0;
2780
2781 out:
2782         return r;
2783 }
2784
2785 void kvm_arch_exit(void)
2786 {
2787         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2788                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2789                                             CPUFREQ_TRANSITION_NOTIFIER);
2790         kvm_x86_ops = NULL;
2791         kvm_mmu_module_exit();
2792 }
2793
2794 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2795 {
2796         ++vcpu->stat.halt_exits;
2797         KVMTRACE_0D(HLT, vcpu, handler);
2798         if (irqchip_in_kernel(vcpu->kvm)) {
2799                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2800                 return 1;
2801         } else {
2802                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2803                 return 0;
2804         }
2805 }
2806 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2807
2808 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2809                            unsigned long a1)
2810 {
2811         if (is_long_mode(vcpu))
2812                 return a0;
2813         else
2814                 return a0 | ((gpa_t)a1 << 32);
2815 }
2816
2817 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2818 {
2819         unsigned long nr, a0, a1, a2, a3, ret;
2820         int r = 1;
2821
2822         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2823         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2824         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2825         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2826         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2827
2828         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2829
2830         if (!is_long_mode(vcpu)) {
2831                 nr &= 0xFFFFFFFF;
2832                 a0 &= 0xFFFFFFFF;
2833                 a1 &= 0xFFFFFFFF;
2834                 a2 &= 0xFFFFFFFF;
2835                 a3 &= 0xFFFFFFFF;
2836         }
2837
2838         switch (nr) {
2839         case KVM_HC_VAPIC_POLL_IRQ:
2840                 ret = 0;
2841                 break;
2842         case KVM_HC_MMU_OP:
2843                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2844                 break;
2845         default:
2846                 ret = -KVM_ENOSYS;
2847                 break;
2848         }
2849         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2850         ++vcpu->stat.hypercalls;
2851         return r;
2852 }
2853 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2854
2855 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2856 {
2857         char instruction[3];
2858         int ret = 0;
2859         unsigned long rip = kvm_rip_read(vcpu);
2860
2861
2862         /*
2863          * Blow out the MMU to ensure that no other VCPU has an active mapping
2864          * to ensure that the updated hypercall appears atomically across all
2865          * VCPUs.
2866          */
2867         kvm_mmu_zap_all(vcpu->kvm);
2868
2869         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2870         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2871             != X86EMUL_CONTINUE)
2872                 ret = -EFAULT;
2873
2874         return ret;
2875 }
2876
2877 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2878 {
2879         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2880 }
2881
2882 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2883 {
2884         struct descriptor_table dt = { limit, base };
2885
2886         kvm_x86_ops->set_gdt(vcpu, &dt);
2887 }
2888
2889 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2890 {
2891         struct descriptor_table dt = { limit, base };
2892
2893         kvm_x86_ops->set_idt(vcpu, &dt);
2894 }
2895
2896 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2897                    unsigned long *rflags)
2898 {
2899         kvm_lmsw(vcpu, msw);
2900         *rflags = kvm_x86_ops->get_rflags(vcpu);
2901 }
2902
2903 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2904 {
2905         unsigned long value;
2906
2907         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2908         switch (cr) {
2909         case 0:
2910                 value = vcpu->arch.cr0;
2911                 break;
2912         case 2:
2913                 value = vcpu->arch.cr2;
2914                 break;
2915         case 3:
2916                 value = vcpu->arch.cr3;
2917                 break;
2918         case 4:
2919                 value = vcpu->arch.cr4;
2920                 break;
2921         case 8:
2922                 value = kvm_get_cr8(vcpu);
2923                 break;
2924         default:
2925                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2926                 return 0;
2927         }
2928         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2929                     (u32)((u64)value >> 32), handler);
2930
2931         return value;
2932 }
2933
2934 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2935                      unsigned long *rflags)
2936 {
2937         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2938                     (u32)((u64)val >> 32), handler);
2939
2940         switch (cr) {
2941         case 0:
2942                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2943                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2944                 break;
2945         case 2:
2946                 vcpu->arch.cr2 = val;
2947                 break;
2948         case 3:
2949                 kvm_set_cr3(vcpu, val);
2950                 break;
2951         case 4:
2952                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2953                 break;
2954         case 8:
2955                 kvm_set_cr8(vcpu, val & 0xfUL);
2956                 break;
2957         default:
2958                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2959         }
2960 }
2961
2962 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2963 {
2964         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2965         int j, nent = vcpu->arch.cpuid_nent;
2966
2967         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2968         /* when no next entry is found, the current entry[i] is reselected */
2969         for (j = i + 1; ; j = (j + 1) % nent) {
2970                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2971                 if (ej->function == e->function) {
2972                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2973                         return j;
2974                 }
2975         }
2976         return 0; /* silence gcc, even though control never reaches here */
2977 }
2978
2979 /* find an entry with matching function, matching index (if needed), and that
2980  * should be read next (if it's stateful) */
2981 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2982         u32 function, u32 index)
2983 {
2984         if (e->function != function)
2985                 return 0;
2986         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2987                 return 0;
2988         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2989             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2990                 return 0;
2991         return 1;
2992 }
2993
2994 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2995                                               u32 function, u32 index)
2996 {
2997         int i;
2998         struct kvm_cpuid_entry2 *best = NULL;
2999
3000         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3001                 struct kvm_cpuid_entry2 *e;
3002
3003                 e = &vcpu->arch.cpuid_entries[i];
3004                 if (is_matching_cpuid_entry(e, function, index)) {
3005                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3006                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3007                         best = e;
3008                         break;
3009                 }
3010                 /*
3011                  * Both basic or both extended?
3012                  */
3013                 if (((e->function ^ function) & 0x80000000) == 0)
3014                         if (!best || e->function > best->function)
3015                                 best = e;
3016         }
3017         return best;
3018 }
3019
3020 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3021 {
3022         u32 function, index;
3023         struct kvm_cpuid_entry2 *best;
3024
3025         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3026         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3027         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3028         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3029         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3030         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3031         best = kvm_find_cpuid_entry(vcpu, function, index);
3032         if (best) {
3033                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3034                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3035                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3036                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3037         }
3038         kvm_x86_ops->skip_emulated_instruction(vcpu);
3039         KVMTRACE_5D(CPUID, vcpu, function,
3040                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3041                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3042                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3043                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3044 }
3045 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3046
3047 /*
3048  * Check if userspace requested an interrupt window, and that the
3049  * interrupt window is open.
3050  *
3051  * No need to exit to userspace if we already have an interrupt queued.
3052  */
3053 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3054                                           struct kvm_run *kvm_run)
3055 {
3056         return (!vcpu->arch.irq_summary &&
3057                 kvm_run->request_interrupt_window &&
3058                 vcpu->arch.interrupt_window_open &&
3059                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3060 }
3061
3062 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3063                               struct kvm_run *kvm_run)
3064 {
3065         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3066         kvm_run->cr8 = kvm_get_cr8(vcpu);
3067         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3068         if (irqchip_in_kernel(vcpu->kvm))
3069                 kvm_run->ready_for_interrupt_injection = 1;
3070         else
3071                 kvm_run->ready_for_interrupt_injection =
3072                                         (vcpu->arch.interrupt_window_open &&
3073                                          vcpu->arch.irq_summary == 0);
3074 }
3075
3076 static void vapic_enter(struct kvm_vcpu *vcpu)
3077 {
3078         struct kvm_lapic *apic = vcpu->arch.apic;
3079         struct page *page;
3080
3081         if (!apic || !apic->vapic_addr)
3082                 return;
3083
3084         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3085
3086         vcpu->arch.apic->vapic_page = page;
3087 }
3088
3089 static void vapic_exit(struct kvm_vcpu *vcpu)
3090 {
3091         struct kvm_lapic *apic = vcpu->arch.apic;
3092
3093         if (!apic || !apic->vapic_addr)
3094                 return;
3095
3096         down_read(&vcpu->kvm->slots_lock);
3097         kvm_release_page_dirty(apic->vapic_page);
3098         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3099         up_read(&vcpu->kvm->slots_lock);
3100 }
3101
3102 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3103 {
3104         int r;
3105
3106         if (vcpu->requests)
3107                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3108                         kvm_mmu_unload(vcpu);
3109
3110         r = kvm_mmu_reload(vcpu);
3111         if (unlikely(r))
3112                 goto out;
3113
3114         if (vcpu->requests) {
3115                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3116                         __kvm_migrate_timers(vcpu);
3117                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3118                         kvm_write_guest_time(vcpu);
3119                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3120                         kvm_mmu_sync_roots(vcpu);
3121                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3122                         kvm_x86_ops->tlb_flush(vcpu);
3123                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3124                                        &vcpu->requests)) {
3125                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3126                         r = 0;
3127                         goto out;
3128                 }
3129                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3130                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3131                         r = 0;
3132                         goto out;
3133                 }
3134         }
3135
3136         preempt_disable();
3137
3138         kvm_x86_ops->prepare_guest_switch(vcpu);
3139         kvm_load_guest_fpu(vcpu);
3140
3141         local_irq_disable();
3142
3143         if (vcpu->requests || need_resched() || signal_pending(current)) {
3144                 local_irq_enable();
3145                 preempt_enable();
3146                 r = 1;
3147                 goto out;
3148         }
3149
3150         vcpu->guest_mode = 1;
3151         /*
3152          * Make sure that guest_mode assignment won't happen after
3153          * testing the pending IRQ vector bitmap.
3154          */
3155         smp_wmb();
3156
3157         if (vcpu->arch.exception.pending)
3158                 __queue_exception(vcpu);
3159         else if (irqchip_in_kernel(vcpu->kvm))
3160                 kvm_x86_ops->inject_pending_irq(vcpu);
3161         else
3162                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3163
3164         kvm_lapic_sync_to_vapic(vcpu);
3165
3166         up_read(&vcpu->kvm->slots_lock);
3167
3168         kvm_guest_enter();
3169
3170         get_debugreg(vcpu->arch.host_dr6, 6);
3171         get_debugreg(vcpu->arch.host_dr7, 7);
3172         if (unlikely(vcpu->arch.switch_db_regs)) {
3173                 get_debugreg(vcpu->arch.host_db[0], 0);
3174                 get_debugreg(vcpu->arch.host_db[1], 1);
3175                 get_debugreg(vcpu->arch.host_db[2], 2);
3176                 get_debugreg(vcpu->arch.host_db[3], 3);
3177
3178                 set_debugreg(0, 7);
3179                 set_debugreg(vcpu->arch.eff_db[0], 0);
3180                 set_debugreg(vcpu->arch.eff_db[1], 1);
3181                 set_debugreg(vcpu->arch.eff_db[2], 2);
3182                 set_debugreg(vcpu->arch.eff_db[3], 3);
3183         }
3184
3185         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3186         kvm_x86_ops->run(vcpu, kvm_run);
3187
3188         if (unlikely(vcpu->arch.switch_db_regs)) {
3189                 set_debugreg(0, 7);
3190                 set_debugreg(vcpu->arch.host_db[0], 0);
3191                 set_debugreg(vcpu->arch.host_db[1], 1);
3192                 set_debugreg(vcpu->arch.host_db[2], 2);
3193                 set_debugreg(vcpu->arch.host_db[3], 3);
3194         }
3195         set_debugreg(vcpu->arch.host_dr6, 6);
3196         set_debugreg(vcpu->arch.host_dr7, 7);
3197
3198         vcpu->guest_mode = 0;
3199         local_irq_enable();
3200
3201         ++vcpu->stat.exits;
3202
3203         /*
3204          * We must have an instruction between local_irq_enable() and
3205          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3206          * the interrupt shadow.  The stat.exits increment will do nicely.
3207          * But we need to prevent reordering, hence this barrier():
3208          */
3209         barrier();
3210
3211         kvm_guest_exit();
3212
3213         preempt_enable();
3214
3215         down_read(&vcpu->kvm->slots_lock);
3216
3217         /*
3218          * Profile KVM exit RIPs:
3219          */
3220         if (unlikely(prof_on == KVM_PROFILING)) {
3221                 unsigned long rip = kvm_rip_read(vcpu);
3222                 profile_hit(KVM_PROFILING, (void *)rip);
3223         }
3224
3225         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3226                 vcpu->arch.exception.pending = false;
3227
3228         kvm_lapic_sync_from_vapic(vcpu);
3229
3230         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3231 out:
3232         return r;
3233 }
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                         {
3263                                 switch(vcpu->arch.mp_state) {
3264                                 case KVM_MP_STATE_HALTED:
3265                                         vcpu->arch.mp_state =
3266                                                 KVM_MP_STATE_RUNNABLE;
3267                                 case KVM_MP_STATE_RUNNABLE:
3268                                         break;
3269                                 case KVM_MP_STATE_SIPI_RECEIVED:
3270                                 default:
3271                                         r = -EINTR;
3272                                         break;
3273                                 }
3274                         }
3275                 }
3276
3277                 if (r <= 0)
3278                         break;
3279
3280                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3281                 if (kvm_cpu_has_pending_timer(vcpu))
3282                         kvm_inject_pending_timer_irqs(vcpu);
3283
3284                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3285                         r = -EINTR;
3286                         kvm_run->exit_reason = KVM_EXIT_INTR;
3287                         ++vcpu->stat.request_irq_exits;
3288                 }
3289                 if (signal_pending(current)) {
3290                         r = -EINTR;
3291                         kvm_run->exit_reason = KVM_EXIT_INTR;
3292                         ++vcpu->stat.signal_exits;
3293                 }
3294                 if (need_resched()) {
3295                         up_read(&vcpu->kvm->slots_lock);
3296                         kvm_resched(vcpu);
3297                         down_read(&vcpu->kvm->slots_lock);
3298                 }
3299         }
3300
3301         up_read(&vcpu->kvm->slots_lock);
3302         post_kvm_run_save(vcpu, kvm_run);
3303
3304         vapic_exit(vcpu);
3305
3306         return r;
3307 }
3308
3309 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3310 {
3311         int r;
3312         sigset_t sigsaved;
3313
3314         vcpu_load(vcpu);
3315
3316         if (vcpu->sigset_active)
3317                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3318
3319         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3320                 kvm_vcpu_block(vcpu);
3321                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3322                 r = -EAGAIN;
3323                 goto out;
3324         }
3325
3326         /* re-sync apic's tpr */
3327         if (!irqchip_in_kernel(vcpu->kvm))
3328                 kvm_set_cr8(vcpu, kvm_run->cr8);
3329
3330         if (vcpu->arch.pio.cur_count) {
3331                 r = complete_pio(vcpu);
3332                 if (r)
3333                         goto out;
3334         }
3335 #if CONFIG_HAS_IOMEM
3336         if (vcpu->mmio_needed) {
3337                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3338                 vcpu->mmio_read_completed = 1;
3339                 vcpu->mmio_needed = 0;
3340
3341                 down_read(&vcpu->kvm->slots_lock);
3342                 r = emulate_instruction(vcpu, kvm_run,
3343                                         vcpu->arch.mmio_fault_cr2, 0,
3344                                         EMULTYPE_NO_DECODE);
3345                 up_read(&vcpu->kvm->slots_lock);
3346                 if (r == EMULATE_DO_MMIO) {
3347                         /*
3348                          * Read-modify-write.  Back to userspace.
3349                          */
3350                         r = 0;
3351                         goto out;
3352                 }
3353         }
3354 #endif
3355         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3356                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3357                                      kvm_run->hypercall.ret);
3358
3359         r = __vcpu_run(vcpu, kvm_run);
3360
3361 out:
3362         if (vcpu->sigset_active)
3363                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3364
3365         vcpu_put(vcpu);
3366         return r;
3367 }
3368
3369 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3370 {
3371         vcpu_load(vcpu);
3372
3373         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3374         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3375         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3376         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3377         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3378         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3379         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3380         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3381 #ifdef CONFIG_X86_64
3382         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3383         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3384         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3385         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3386         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3387         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3388         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3389         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3390 #endif
3391
3392         regs->rip = kvm_rip_read(vcpu);
3393         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3394
3395         /*
3396          * Don't leak debug flags in case they were set for guest debugging
3397          */
3398         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3399                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3400
3401         vcpu_put(vcpu);
3402
3403         return 0;
3404 }
3405
3406 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3407 {
3408         vcpu_load(vcpu);
3409
3410         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3411         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3412         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3413         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3414         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3415         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3416         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3417         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3418 #ifdef CONFIG_X86_64
3419         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3420         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3421         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3422         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3423         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3424         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3425         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3426         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3427
3428 #endif
3429
3430         kvm_rip_write(vcpu, regs->rip);
3431         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3432
3433
3434         vcpu->arch.exception.pending = false;
3435
3436         vcpu_put(vcpu);
3437
3438         return 0;
3439 }
3440
3441 void kvm_get_segment(struct kvm_vcpu *vcpu,
3442                      struct kvm_segment *var, int seg)
3443 {
3444         kvm_x86_ops->get_segment(vcpu, var, seg);
3445 }
3446
3447 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3448 {
3449         struct kvm_segment cs;
3450
3451         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3452         *db = cs.db;
3453         *l = cs.l;
3454 }
3455 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3456
3457 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3458                                   struct kvm_sregs *sregs)
3459 {
3460         struct descriptor_table dt;
3461         int pending_vec;
3462
3463         vcpu_load(vcpu);
3464
3465         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3466         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3467         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3468         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3469         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3470         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3471
3472         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3473         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3474
3475         kvm_x86_ops->get_idt(vcpu, &dt);
3476         sregs->idt.limit = dt.limit;
3477         sregs->idt.base = dt.base;
3478         kvm_x86_ops->get_gdt(vcpu, &dt);
3479         sregs->gdt.limit = dt.limit;
3480         sregs->gdt.base = dt.base;
3481
3482         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3483         sregs->cr0 = vcpu->arch.cr0;
3484         sregs->cr2 = vcpu->arch.cr2;
3485         sregs->cr3 = vcpu->arch.cr3;
3486         sregs->cr4 = vcpu->arch.cr4;
3487         sregs->cr8 = kvm_get_cr8(vcpu);
3488         sregs->efer = vcpu->arch.shadow_efer;
3489         sregs->apic_base = kvm_get_apic_base(vcpu);
3490
3491         if (irqchip_in_kernel(vcpu->kvm)) {
3492                 memset(sregs->interrupt_bitmap, 0,
3493                        sizeof sregs->interrupt_bitmap);
3494                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3495                 if (pending_vec >= 0)
3496                         set_bit(pending_vec,
3497                                 (unsigned long *)sregs->interrupt_bitmap);
3498         } else
3499                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3500                        sizeof sregs->interrupt_bitmap);
3501
3502         vcpu_put(vcpu);
3503
3504         return 0;
3505 }
3506
3507 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3508                                     struct kvm_mp_state *mp_state)
3509 {
3510         vcpu_load(vcpu);
3511         mp_state->mp_state = vcpu->arch.mp_state;
3512         vcpu_put(vcpu);
3513         return 0;
3514 }
3515
3516 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3517                                     struct kvm_mp_state *mp_state)
3518 {
3519         vcpu_load(vcpu);
3520         vcpu->arch.mp_state = mp_state->mp_state;
3521         vcpu_put(vcpu);
3522         return 0;
3523 }
3524
3525 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3526                         struct kvm_segment *var, int seg)
3527 {
3528         kvm_x86_ops->set_segment(vcpu, var, seg);
3529 }
3530
3531 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3532                                    struct kvm_segment *kvm_desct)
3533 {
3534         kvm_desct->base = seg_desc->base0;
3535         kvm_desct->base |= seg_desc->base1 << 16;
3536         kvm_desct->base |= seg_desc->base2 << 24;
3537         kvm_desct->limit = seg_desc->limit0;
3538         kvm_desct->limit |= seg_desc->limit << 16;
3539         if (seg_desc->g) {
3540                 kvm_desct->limit <<= 12;
3541                 kvm_desct->limit |= 0xfff;
3542         }
3543         kvm_desct->selector = selector;
3544         kvm_desct->type = seg_desc->type;
3545         kvm_desct->present = seg_desc->p;
3546         kvm_desct->dpl = seg_desc->dpl;
3547         kvm_desct->db = seg_desc->d;
3548         kvm_desct->s = seg_desc->s;
3549         kvm_desct->l = seg_desc->l;
3550         kvm_desct->g = seg_desc->g;
3551         kvm_desct->avl = seg_desc->avl;
3552         if (!selector)
3553                 kvm_desct->unusable = 1;
3554         else
3555                 kvm_desct->unusable = 0;
3556         kvm_desct->padding = 0;
3557 }
3558
3559 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3560                                           u16 selector,
3561                                           struct descriptor_table *dtable)
3562 {
3563         if (selector & 1 << 2) {
3564                 struct kvm_segment kvm_seg;
3565
3566                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3567
3568                 if (kvm_seg.unusable)
3569                         dtable->limit = 0;
3570                 else
3571                         dtable->limit = kvm_seg.limit;
3572                 dtable->base = kvm_seg.base;
3573         }
3574         else
3575                 kvm_x86_ops->get_gdt(vcpu, dtable);
3576 }
3577
3578 /* allowed just for 8 bytes segments */
3579 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3580                                          struct desc_struct *seg_desc)
3581 {
3582         gpa_t gpa;
3583         struct descriptor_table dtable;
3584         u16 index = selector >> 3;
3585
3586         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3587
3588         if (dtable.limit < index * 8 + 7) {
3589                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3590                 return 1;
3591         }
3592         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3593         gpa += index * 8;
3594         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3595 }
3596
3597 /* allowed just for 8 bytes segments */
3598 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3599                                          struct desc_struct *seg_desc)
3600 {
3601         gpa_t gpa;
3602         struct descriptor_table dtable;
3603         u16 index = selector >> 3;
3604
3605         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3606
3607         if (dtable.limit < index * 8 + 7)
3608                 return 1;
3609         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3610         gpa += index * 8;
3611         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3612 }
3613
3614 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3615                              struct desc_struct *seg_desc)
3616 {
3617         u32 base_addr;
3618
3619         base_addr = seg_desc->base0;
3620         base_addr |= (seg_desc->base1 << 16);
3621         base_addr |= (seg_desc->base2 << 24);
3622
3623         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3624 }
3625
3626 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3627 {
3628         struct kvm_segment kvm_seg;
3629
3630         kvm_get_segment(vcpu, &kvm_seg, seg);
3631         return kvm_seg.selector;
3632 }
3633
3634 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3635                                                 u16 selector,
3636                                                 struct kvm_segment *kvm_seg)
3637 {
3638         struct desc_struct seg_desc;
3639
3640         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3641                 return 1;
3642         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3643         return 0;
3644 }
3645
3646 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3647 {
3648         struct kvm_segment segvar = {
3649                 .base = selector << 4,
3650                 .limit = 0xffff,
3651                 .selector = selector,
3652                 .type = 3,
3653                 .present = 1,
3654                 .dpl = 3,
3655                 .db = 0,
3656                 .s = 1,
3657                 .l = 0,
3658                 .g = 0,
3659                 .avl = 0,
3660                 .unusable = 0,
3661         };
3662         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3663         return 0;
3664 }
3665
3666 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3667                                 int type_bits, int seg)
3668 {
3669         struct kvm_segment kvm_seg;
3670
3671         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3672                 return kvm_load_realmode_segment(vcpu, selector, seg);
3673         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3674                 return 1;
3675         kvm_seg.type |= type_bits;
3676
3677         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3678             seg != VCPU_SREG_LDTR)
3679                 if (!kvm_seg.s)
3680                         kvm_seg.unusable = 1;
3681
3682         kvm_set_segment(vcpu, &kvm_seg, seg);
3683         return 0;
3684 }
3685
3686 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3687                                 struct tss_segment_32 *tss)
3688 {
3689         tss->cr3 = vcpu->arch.cr3;
3690         tss->eip = kvm_rip_read(vcpu);
3691         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3692         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3693         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3694         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3695         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3696         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3697         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3698         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3699         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3700         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3701         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3702         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3703         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3704         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3705         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3706         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3707         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3708 }
3709
3710 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3711                                   struct tss_segment_32 *tss)
3712 {
3713         kvm_set_cr3(vcpu, tss->cr3);
3714
3715         kvm_rip_write(vcpu, tss->eip);
3716         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3717
3718         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3719         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3720         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3721         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3722         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3723         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3724         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3725         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3726
3727         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3728                 return 1;
3729
3730         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3731                 return 1;
3732
3733         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3734                 return 1;
3735
3736         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3737                 return 1;
3738
3739         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3740                 return 1;
3741
3742         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3743                 return 1;
3744
3745         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3746                 return 1;
3747         return 0;
3748 }
3749
3750 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3751                                 struct tss_segment_16 *tss)
3752 {
3753         tss->ip = kvm_rip_read(vcpu);
3754         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3755         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3756         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3757         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3758         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3759         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3760         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3761         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3762         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3763
3764         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3765         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3766         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3767         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3768         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3769         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3770 }
3771
3772 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3773                                  struct tss_segment_16 *tss)
3774 {
3775         kvm_rip_write(vcpu, tss->ip);
3776         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3777         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3778         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3779         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3780         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3781         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3782         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3783         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3784         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3785
3786         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3787                 return 1;
3788
3789         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3790                 return 1;
3791
3792         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3793                 return 1;
3794
3795         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3796                 return 1;
3797
3798         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3799                 return 1;
3800         return 0;
3801 }
3802
3803 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3804                        u32 old_tss_base,
3805                        struct desc_struct *nseg_desc)
3806 {
3807         struct tss_segment_16 tss_segment_16;
3808         int ret = 0;
3809
3810         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3811                            sizeof tss_segment_16))
3812                 goto out;
3813
3814         save_state_to_tss16(vcpu, &tss_segment_16);
3815
3816         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3817                             sizeof tss_segment_16))
3818                 goto out;
3819
3820         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3821                            &tss_segment_16, sizeof tss_segment_16))
3822                 goto out;
3823
3824         if (load_state_from_tss16(vcpu, &tss_segment_16))
3825                 goto out;
3826
3827         ret = 1;
3828 out:
3829         return ret;
3830 }
3831
3832 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3833                        u32 old_tss_base,
3834                        struct desc_struct *nseg_desc)
3835 {
3836         struct tss_segment_32 tss_segment_32;
3837         int ret = 0;
3838
3839         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3840                            sizeof tss_segment_32))
3841                 goto out;
3842
3843         save_state_to_tss32(vcpu, &tss_segment_32);
3844
3845         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3846                             sizeof tss_segment_32))
3847                 goto out;
3848
3849         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3850                            &tss_segment_32, sizeof tss_segment_32))
3851                 goto out;
3852
3853         if (load_state_from_tss32(vcpu, &tss_segment_32))
3854                 goto out;
3855
3856         ret = 1;
3857 out:
3858         return ret;
3859 }
3860
3861 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3862 {
3863         struct kvm_segment tr_seg;
3864         struct desc_struct cseg_desc;
3865         struct desc_struct nseg_desc;
3866         int ret = 0;
3867         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3868         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3869
3870         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3871
3872         /* FIXME: Handle errors. Failure to read either TSS or their
3873          * descriptors should generate a pagefault.
3874          */
3875         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3876                 goto out;
3877
3878         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3879                 goto out;
3880
3881         if (reason != TASK_SWITCH_IRET) {
3882                 int cpl;
3883
3884                 cpl = kvm_x86_ops->get_cpl(vcpu);
3885                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3886                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3887                         return 1;
3888                 }
3889         }
3890
3891         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3892                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3893                 return 1;
3894         }
3895
3896         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3897                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3898                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3899         }
3900
3901         if (reason == TASK_SWITCH_IRET) {
3902                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3903                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3904         }
3905
3906         kvm_x86_ops->skip_emulated_instruction(vcpu);
3907
3908         if (nseg_desc.type & 8)
3909                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3910                                          &nseg_desc);
3911         else
3912                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3913                                          &nseg_desc);
3914
3915         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3916                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3917                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3918         }
3919
3920         if (reason != TASK_SWITCH_IRET) {
3921                 nseg_desc.type |= (1 << 1);
3922                 save_guest_segment_descriptor(vcpu, tss_selector,
3923                                               &nseg_desc);
3924         }
3925
3926         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3927         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3928         tr_seg.type = 11;
3929         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3930 out:
3931         return ret;
3932 }
3933 EXPORT_SYMBOL_GPL(kvm_task_switch);
3934
3935 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3936                                   struct kvm_sregs *sregs)
3937 {
3938         int mmu_reset_needed = 0;
3939         int i, pending_vec, max_bits;
3940         struct descriptor_table dt;
3941
3942         vcpu_load(vcpu);
3943
3944         dt.limit = sregs->idt.limit;
3945         dt.base = sregs->idt.base;
3946         kvm_x86_ops->set_idt(vcpu, &dt);
3947         dt.limit = sregs->gdt.limit;
3948         dt.base = sregs->gdt.base;
3949         kvm_x86_ops->set_gdt(vcpu, &dt);
3950
3951         vcpu->arch.cr2 = sregs->cr2;
3952         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3953         vcpu->arch.cr3 = sregs->cr3;
3954
3955         kvm_set_cr8(vcpu, sregs->cr8);
3956
3957         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3958         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3959         kvm_set_apic_base(vcpu, sregs->apic_base);
3960
3961         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3962
3963         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3964         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3965         vcpu->arch.cr0 = sregs->cr0;
3966
3967         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3968         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3969         if (!is_long_mode(vcpu) && is_pae(vcpu))
3970                 load_pdptrs(vcpu, vcpu->arch.cr3);
3971
3972         if (mmu_reset_needed)
3973                 kvm_mmu_reset_context(vcpu);
3974
3975         if (!irqchip_in_kernel(vcpu->kvm)) {
3976                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3977                        sizeof vcpu->arch.irq_pending);
3978                 vcpu->arch.irq_summary = 0;
3979                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3980                         if (vcpu->arch.irq_pending[i])
3981                                 __set_bit(i, &vcpu->arch.irq_summary);
3982         } else {
3983                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3984                 pending_vec = find_first_bit(
3985                         (const unsigned long *)sregs->interrupt_bitmap,
3986                         max_bits);
3987                 /* Only pending external irq is handled here */
3988                 if (pending_vec < max_bits) {
3989                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3990                         pr_debug("Set back pending irq %d\n",
3991                                  pending_vec);
3992                 }
3993                 kvm_pic_clear_isr_ack(vcpu->kvm);
3994         }
3995
3996         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3997         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3998         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3999         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4000         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4001         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4002
4003         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4004         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4005
4006         /* Older userspace won't unhalt the vcpu on reset. */
4007         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4008             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4009             !(vcpu->arch.cr0 & X86_CR0_PE))
4010                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4011
4012         vcpu_put(vcpu);
4013
4014         return 0;
4015 }
4016
4017 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4018                                         struct kvm_guest_debug *dbg)
4019 {
4020         int i, r;
4021
4022         vcpu_load(vcpu);
4023
4024         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4025             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4026                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4027                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4028                 vcpu->arch.switch_db_regs =
4029                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4030         } else {
4031                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4032                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4033                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4034         }
4035
4036         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4037
4038         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4039                 kvm_queue_exception(vcpu, DB_VECTOR);
4040         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4041                 kvm_queue_exception(vcpu, BP_VECTOR);
4042
4043         vcpu_put(vcpu);
4044
4045         return r;
4046 }
4047
4048 /*
4049  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4050  * we have asm/x86/processor.h
4051  */
4052 struct fxsave {
4053         u16     cwd;
4054         u16     swd;
4055         u16     twd;
4056         u16     fop;
4057         u64     rip;
4058         u64     rdp;
4059         u32     mxcsr;
4060         u32     mxcsr_mask;
4061         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4062 #ifdef CONFIG_X86_64
4063         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4064 #else
4065         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4066 #endif
4067 };
4068
4069 /*
4070  * Translate a guest virtual address to a guest physical address.
4071  */
4072 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4073                                     struct kvm_translation *tr)
4074 {
4075         unsigned long vaddr = tr->linear_address;
4076         gpa_t gpa;
4077
4078         vcpu_load(vcpu);
4079         down_read(&vcpu->kvm->slots_lock);
4080         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4081         up_read(&vcpu->kvm->slots_lock);
4082         tr->physical_address = gpa;
4083         tr->valid = gpa != UNMAPPED_GVA;
4084         tr->writeable = 1;
4085         tr->usermode = 0;
4086         vcpu_put(vcpu);
4087
4088         return 0;
4089 }
4090
4091 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4092 {
4093         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4094
4095         vcpu_load(vcpu);
4096
4097         memcpy(fpu->fpr, fxsave->st_space, 128);
4098         fpu->fcw = fxsave->cwd;
4099         fpu->fsw = fxsave->swd;
4100         fpu->ftwx = fxsave->twd;
4101         fpu->last_opcode = fxsave->fop;
4102         fpu->last_ip = fxsave->rip;
4103         fpu->last_dp = fxsave->rdp;
4104         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4105
4106         vcpu_put(vcpu);
4107
4108         return 0;
4109 }
4110
4111 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4112 {
4113         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4114
4115         vcpu_load(vcpu);
4116
4117         memcpy(fxsave->st_space, fpu->fpr, 128);
4118         fxsave->cwd = fpu->fcw;
4119         fxsave->swd = fpu->fsw;
4120         fxsave->twd = fpu->ftwx;
4121         fxsave->fop = fpu->last_opcode;
4122         fxsave->rip = fpu->last_ip;
4123         fxsave->rdp = fpu->last_dp;
4124         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4125
4126         vcpu_put(vcpu);
4127
4128         return 0;
4129 }
4130
4131 void fx_init(struct kvm_vcpu *vcpu)
4132 {
4133         unsigned after_mxcsr_mask;
4134
4135         /*
4136          * Touch the fpu the first time in non atomic context as if
4137          * this is the first fpu instruction the exception handler
4138          * will fire before the instruction returns and it'll have to
4139          * allocate ram with GFP_KERNEL.
4140          */
4141         if (!used_math())
4142                 kvm_fx_save(&vcpu->arch.host_fx_image);
4143
4144         /* Initialize guest FPU by resetting ours and saving into guest's */
4145         preempt_disable();
4146         kvm_fx_save(&vcpu->arch.host_fx_image);
4147         kvm_fx_finit();
4148         kvm_fx_save(&vcpu->arch.guest_fx_image);
4149         kvm_fx_restore(&vcpu->arch.host_fx_image);
4150         preempt_enable();
4151
4152         vcpu->arch.cr0 |= X86_CR0_ET;
4153         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4154         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4155         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4156                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4157 }
4158 EXPORT_SYMBOL_GPL(fx_init);
4159
4160 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4161 {
4162         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4163                 return;
4164
4165         vcpu->guest_fpu_loaded = 1;
4166         kvm_fx_save(&vcpu->arch.host_fx_image);
4167         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4168 }
4169 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4170
4171 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4172 {
4173         if (!vcpu->guest_fpu_loaded)
4174                 return;
4175
4176         vcpu->guest_fpu_loaded = 0;
4177         kvm_fx_save(&vcpu->arch.guest_fx_image);
4178         kvm_fx_restore(&vcpu->arch.host_fx_image);
4179         ++vcpu->stat.fpu_reload;
4180 }
4181 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4182
4183 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4184 {
4185         if (vcpu->arch.time_page) {
4186                 kvm_release_page_dirty(vcpu->arch.time_page);
4187                 vcpu->arch.time_page = NULL;
4188         }
4189
4190         kvm_x86_ops->vcpu_free(vcpu);
4191 }
4192
4193 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4194                                                 unsigned int id)
4195 {
4196         return kvm_x86_ops->vcpu_create(kvm, id);
4197 }
4198
4199 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4200 {
4201         int r;
4202
4203         /* We do fxsave: this must be aligned. */
4204         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4205
4206         vcpu->arch.mtrr_state.have_fixed = 1;
4207         vcpu_load(vcpu);
4208         r = kvm_arch_vcpu_reset(vcpu);
4209         if (r == 0)
4210                 r = kvm_mmu_setup(vcpu);
4211         vcpu_put(vcpu);
4212         if (r < 0)
4213                 goto free_vcpu;
4214
4215         return 0;
4216 free_vcpu:
4217         kvm_x86_ops->vcpu_free(vcpu);
4218         return r;
4219 }
4220
4221 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4222 {
4223         vcpu_load(vcpu);
4224         kvm_mmu_unload(vcpu);
4225         vcpu_put(vcpu);
4226
4227         kvm_x86_ops->vcpu_free(vcpu);
4228 }
4229
4230 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4231 {
4232         vcpu->arch.nmi_pending = false;
4233         vcpu->arch.nmi_injected = false;
4234
4235         vcpu->arch.switch_db_regs = 0;
4236         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4237         vcpu->arch.dr6 = DR6_FIXED_1;
4238         vcpu->arch.dr7 = DR7_FIXED_1;
4239
4240         return kvm_x86_ops->vcpu_reset(vcpu);
4241 }
4242
4243 void kvm_arch_hardware_enable(void *garbage)
4244 {
4245         kvm_x86_ops->hardware_enable(garbage);
4246 }
4247
4248 void kvm_arch_hardware_disable(void *garbage)
4249 {
4250         kvm_x86_ops->hardware_disable(garbage);
4251 }
4252
4253 int kvm_arch_hardware_setup(void)
4254 {
4255         return kvm_x86_ops->hardware_setup();
4256 }
4257
4258 void kvm_arch_hardware_unsetup(void)
4259 {
4260         kvm_x86_ops->hardware_unsetup();
4261 }
4262
4263 void kvm_arch_check_processor_compat(void *rtn)
4264 {
4265         kvm_x86_ops->check_processor_compatibility(rtn);
4266 }
4267
4268 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4269 {
4270         struct page *page;
4271         struct kvm *kvm;
4272         int r;
4273
4274         BUG_ON(vcpu->kvm == NULL);
4275         kvm = vcpu->kvm;
4276
4277         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4278         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4279                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4280         else
4281                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4282
4283         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4284         if (!page) {
4285                 r = -ENOMEM;
4286                 goto fail;
4287         }
4288         vcpu->arch.pio_data = page_address(page);
4289
4290         r = kvm_mmu_create(vcpu);
4291         if (r < 0)
4292                 goto fail_free_pio_data;
4293
4294         if (irqchip_in_kernel(kvm)) {
4295                 r = kvm_create_lapic(vcpu);
4296                 if (r < 0)
4297                         goto fail_mmu_destroy;
4298         }
4299
4300         return 0;
4301
4302 fail_mmu_destroy:
4303         kvm_mmu_destroy(vcpu);
4304 fail_free_pio_data:
4305         free_page((unsigned long)vcpu->arch.pio_data);
4306 fail:
4307         return r;
4308 }
4309
4310 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4311 {
4312         kvm_free_lapic(vcpu);
4313         down_read(&vcpu->kvm->slots_lock);
4314         kvm_mmu_destroy(vcpu);
4315         up_read(&vcpu->kvm->slots_lock);
4316         free_page((unsigned long)vcpu->arch.pio_data);
4317 }
4318
4319 struct  kvm *kvm_arch_create_vm(void)
4320 {
4321         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4322
4323         if (!kvm)
4324                 return ERR_PTR(-ENOMEM);
4325
4326         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4327         INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4328         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4329
4330         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4331         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4332
4333         rdtscll(kvm->arch.vm_init_tsc);
4334
4335         return kvm;
4336 }
4337
4338 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4339 {
4340         vcpu_load(vcpu);
4341         kvm_mmu_unload(vcpu);
4342         vcpu_put(vcpu);
4343 }
4344
4345 static void kvm_free_vcpus(struct kvm *kvm)
4346 {
4347         unsigned int i;
4348
4349         /*
4350          * Unpin any mmu pages first.
4351          */
4352         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4353                 if (kvm->vcpus[i])
4354                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4355         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4356                 if (kvm->vcpus[i]) {
4357                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4358                         kvm->vcpus[i] = NULL;
4359                 }
4360         }
4361
4362 }
4363
4364 void kvm_arch_sync_events(struct kvm *kvm)
4365 {
4366         kvm_free_all_assigned_devices(kvm);
4367 }
4368
4369 void kvm_arch_destroy_vm(struct kvm *kvm)
4370 {
4371         kvm_iommu_unmap_guest(kvm);
4372         kvm_free_pit(kvm);
4373         kfree(kvm->arch.vpic);
4374         kfree(kvm->arch.vioapic);
4375         kvm_free_vcpus(kvm);
4376         kvm_free_physmem(kvm);
4377         if (kvm->arch.apic_access_page)
4378                 put_page(kvm->arch.apic_access_page);
4379         if (kvm->arch.ept_identity_pagetable)
4380                 put_page(kvm->arch.ept_identity_pagetable);
4381         kfree(kvm);
4382 }
4383
4384 int kvm_arch_set_memory_region(struct kvm *kvm,
4385                                 struct kvm_userspace_memory_region *mem,
4386                                 struct kvm_memory_slot old,
4387                                 int user_alloc)
4388 {
4389         int npages = mem->memory_size >> PAGE_SHIFT;
4390         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4391
4392         /*To keep backward compatibility with older userspace,
4393          *x86 needs to hanlde !user_alloc case.
4394          */
4395         if (!user_alloc) {
4396                 if (npages && !old.rmap) {
4397                         unsigned long userspace_addr;
4398
4399                         down_write(&current->mm->mmap_sem);
4400                         userspace_addr = do_mmap(NULL, 0,
4401                                                  npages * PAGE_SIZE,
4402                                                  PROT_READ | PROT_WRITE,
4403                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4404                                                  0);
4405                         up_write(&current->mm->mmap_sem);
4406
4407                         if (IS_ERR((void *)userspace_addr))
4408                                 return PTR_ERR((void *)userspace_addr);
4409
4410                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4411                         spin_lock(&kvm->mmu_lock);
4412                         memslot->userspace_addr = userspace_addr;
4413                         spin_unlock(&kvm->mmu_lock);
4414                 } else {
4415                         if (!old.user_alloc && old.rmap) {
4416                                 int ret;
4417
4418                                 down_write(&current->mm->mmap_sem);
4419                                 ret = do_munmap(current->mm, old.userspace_addr,
4420                                                 old.npages * PAGE_SIZE);
4421                                 up_write(&current->mm->mmap_sem);
4422                                 if (ret < 0)
4423                                         printk(KERN_WARNING
4424                                        "kvm_vm_ioctl_set_memory_region: "
4425                                        "failed to munmap memory\n");
4426                         }
4427                 }
4428         }
4429
4430         if (!kvm->arch.n_requested_mmu_pages) {
4431                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4432                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4433         }
4434
4435         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4436         kvm_flush_remote_tlbs(kvm);
4437
4438         return 0;
4439 }
4440
4441 void kvm_arch_flush_shadow(struct kvm *kvm)
4442 {
4443         kvm_mmu_zap_all(kvm);
4444 }
4445
4446 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4447 {
4448         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4449                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4450                || vcpu->arch.nmi_pending;
4451 }
4452
4453 static void vcpu_kick_intr(void *info)
4454 {
4455 #ifdef DEBUG
4456         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4457         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4458 #endif
4459 }
4460
4461 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4462 {
4463         int ipi_pcpu = vcpu->cpu;
4464         int cpu = get_cpu();
4465
4466         if (waitqueue_active(&vcpu->wq)) {
4467                 wake_up_interruptible(&vcpu->wq);
4468                 ++vcpu->stat.halt_wakeup;
4469         }
4470         /*
4471          * We may be called synchronously with irqs disabled in guest mode,
4472          * So need not to call smp_call_function_single() in that case.
4473          */
4474         if (vcpu->guest_mode && vcpu->cpu != cpu)
4475                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4476         put_cpu();
4477 }