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         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 528         kvm_mmu_reset_context(vcpu);
 529 }
 530
 531 void kvm_enable_efer_bits(u64 mask)
 532 {
 533        efer_reserved_bits &= ~mask;
 534 }
 535 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 536
 537
 538 /*
 539  * Writes msr value into into the appropriate "register".
 540  * Returns 0 on success, non-0 otherwise.
 541  * Assumes vcpu_load() was already called.
 542  */
 543 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 544 {
 545         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 546 }
 547
 548 /*
 549  * Adapt set_msr() to msr_io()'s calling convention
 550  */
 551 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 552 {
 553         return kvm_set_msr(vcpu, index, *data);
 554 }
 555
 556 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 557 {
 558         static int version;
 559         struct pvclock_wall_clock wc;
 560         struct timespec now, sys, boot;
 561
 562         if (!wall_clock)
 563                 return;
 564
 565         version++;
 566
 567         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 568
 569         /*
 570          * The guest calculates current wall clock time by adding
 571          * system time (updated by kvm_write_guest_time below) to the
 572          * wall clock specified here.  guest system time equals host
 573          * system time for us, thus we must fill in host boot time here.
 574          */
 575         now = current_kernel_time();
 576         ktime_get_ts(&sys);
 577         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 578
 579         wc.sec = boot.tv_sec;
 580         wc.nsec = boot.tv_nsec;
 581         wc.version = version;
 582
 583         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 584
 585         version++;
 586         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 587 }
 588
 589 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 590 {
 591         uint32_t quotient, remainder;
 592
 593         /* Don't try to replace with do_div(), this one calculates
 594          * "(dividend << 32) / divisor" */
 595         __asm__ ( "divl %4"
 596                   : "=a" (quotient), "=d" (remainder)
 597                   : "0" (0), "1" (dividend), "r" (divisor) );
 598         return quotient;
 599 }
 600
 601 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 602 {
 603         uint64_t nsecs = 1000000000LL;
 604         int32_t  shift = 0;
 605         uint64_t tps64;
 606         uint32_t tps32;
 607
 608         tps64 = tsc_khz * 1000LL;
 609         while (tps64 > nsecs*2) {
 610                 tps64 >>= 1;
 611                 shift--;
 612         }
 613
 614         tps32 = (uint32_t)tps64;
 615         while (tps32 <= (uint32_t)nsecs) {
 616                 tps32 <<= 1;
 617                 shift++;
 618         }
 619
 620         hv_clock->tsc_shift = shift;
 621         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 622
 623         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 624                  __func__, tsc_khz, hv_clock->tsc_shift,
 625                  hv_clock->tsc_to_system_mul);
 626 }
 627
 628 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 629
 630 static void kvm_write_guest_time(struct kvm_vcpu *v)
 631 {
 632         struct timespec ts;
 633         unsigned long flags;
 634         struct kvm_vcpu_arch *vcpu = &v->arch;
 635         void *shared_kaddr;
 636
 637         if ((!vcpu->time_page))
 638                 return;
 639
 640         preempt_disable();
 641         if (unlikely(vcpu->hv_clock_tsc_khz != __get_cpu_var(cpu_tsc_khz))) {
 642                 kvm_set_time_scale(__get_cpu_var(cpu_tsc_khz), &vcpu->hv_clock);
 643                 vcpu->hv_clock_tsc_khz = __get_cpu_var(cpu_tsc_khz);
 644         }
 645         preempt_enable();
 646
 647         /* Keep irq disabled to prevent changes to the clock */
 648         local_irq_save(flags);
 649         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 650                           &vcpu->hv_clock.tsc_timestamp);
 651         ktime_get_ts(&ts);
 652         local_irq_restore(flags);
 653
 654         /* With all the info we got, fill in the values */
 655
 656         vcpu->hv_clock.system_time = ts.tv_nsec +
 657                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 658         /*
 659          * The interface expects us to write an even number signaling that the
 660          * update is finished. Since the guest won't see the intermediate
 661          * state, we just increase by 2 at the end.
 662          */
 663         vcpu->hv_clock.version += 2;
 664
 665         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 666
 667         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 668                sizeof(vcpu->hv_clock));
 669
 670         kunmap_atomic(shared_kaddr, KM_USER0);
 671
 672         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 673 }
 674
 675 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 676 {
 677         struct kvm_vcpu_arch *vcpu = &v->arch;
 678
 679         if (!vcpu->time_page)
 680                 return 0;
 681         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 682         return 1;
 683 }
 684
 685 static bool msr_mtrr_valid(unsigned msr)
 686 {
 687         switch (msr) {
 688         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 689         case MSR_MTRRfix64K_00000:
 690         case MSR_MTRRfix16K_80000:
 691         case MSR_MTRRfix16K_A0000:
 692         case MSR_MTRRfix4K_C0000:
 693         case MSR_MTRRfix4K_C8000:
 694         case MSR_MTRRfix4K_D0000:
 695         case MSR_MTRRfix4K_D8000:
 696         case MSR_MTRRfix4K_E0000:
 697         case MSR_MTRRfix4K_E8000:
 698         case MSR_MTRRfix4K_F0000:
 699         case MSR_MTRRfix4K_F8000:
 700         case MSR_MTRRdefType:
 701         case MSR_IA32_CR_PAT:
 702                 return true;
 703         case 0x2f8:
 704                 return true;
 705         }
 706         return false;
 707 }
 708
 709 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 710 {
 711         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 712
 713         if (!msr_mtrr_valid(msr))
 714                 return 1;
 715
 716         if (msr == MSR_MTRRdefType) {
 717                 vcpu->arch.mtrr_state.def_type = data;
 718                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 719         } else if (msr == MSR_MTRRfix64K_00000)
 720                 p[0] = data;
 721         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 722                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 723         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 724                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 725         else if (msr == MSR_IA32_CR_PAT)
 726                 vcpu->arch.pat = data;
 727         else {  /* Variable MTRRs */
 728                 int idx, is_mtrr_mask;
 729                 u64 *pt;
 730
 731                 idx = (msr - 0x200) / 2;
 732                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 733                 if (!is_mtrr_mask)
 734                         pt =
 735                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 736                 else
 737                         pt =
 738                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 739                 *pt = data;
 740         }
 741
 742         kvm_mmu_reset_context(vcpu);
 743         return 0;
 744 }
 745
 746 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 747 {
 748         switch (msr) {
 749         case MSR_EFER:
 750                 set_efer(vcpu, data);
 751                 break;
 752         case MSR_IA32_MC0_STATUS:
 753                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 754                        __func__, data);
 755                 break;
 756         case MSR_IA32_MCG_STATUS:
 757                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 758                         __func__, data);
 759                 break;
 760         case MSR_IA32_MCG_CTL:
 761                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 762                         __func__, data);
 763                 break;
 764         case MSR_IA32_DEBUGCTLMSR:
 765                 if (!data) {
 766                         /* We support the non-activated case already */
 767                         break;
 768                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 769                         /* Values other than LBR and BTF are vendor-specific,
 770                            thus reserved and should throw a #GP */
 771                         return 1;
 772                 }
 773                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 774                         __func__, data);
 775                 break;
 776         case MSR_IA32_UCODE_REV:
 777         case MSR_IA32_UCODE_WRITE:
 778         case MSR_VM_HSAVE_PA:
 779                 break;
 780         case 0x200 ... 0x2ff:
 781                 return set_msr_mtrr(vcpu, msr, data);
 782         case MSR_IA32_APICBASE:
 783                 kvm_set_apic_base(vcpu, data);
 784                 break;
 785         case MSR_IA32_MISC_ENABLE:
 786                 vcpu->arch.ia32_misc_enable_msr = data;
 787                 break;
 788         case MSR_KVM_WALL_CLOCK:
 789                 vcpu->kvm->arch.wall_clock = data;
 790                 kvm_write_wall_clock(vcpu->kvm, data);
 791                 break;
 792         case MSR_KVM_SYSTEM_TIME: {
 793                 if (vcpu->arch.time_page) {
 794                         kvm_release_page_dirty(vcpu->arch.time_page);
 795                         vcpu->arch.time_page = NULL;
 796                 }
 797
 798                 vcpu->arch.time = data;
 799
 800                 /* we verify if the enable bit is set... */
 801                 if (!(data & 1))
 802                         break;
 803
 804                 /* ...but clean it before doing the actual write */
 805                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 806
 807                 vcpu->arch.time_page =
 808                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 809
 810                 if (is_error_page(vcpu->arch.time_page)) {
 811                         kvm_release_page_clean(vcpu->arch.time_page);
 812                         vcpu->arch.time_page = NULL;
 813                 }
 814
 815                 kvm_request_guest_time_update(vcpu);
 816                 break;
 817         }
 818         default:
 819                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 820                 return 1;
 821         }
 822         return 0;
 823 }
 824 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 825
 826
 827 /*
 828  * Reads an msr value (of 'msr_index') into 'pdata'.
 829  * Returns 0 on success, non-0 otherwise.
 830  * Assumes vcpu_load() was already called.
 831  */
 832 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 833 {
 834         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 835 }
 836
 837 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 838 {
 839         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 840
 841         if (!msr_mtrr_valid(msr))
 842                 return 1;
 843
 844         if (msr == MSR_MTRRdefType)
 845                 *pdata = vcpu->arch.mtrr_state.def_type +
 846                          (vcpu->arch.mtrr_state.enabled << 10);
 847         else if (msr == MSR_MTRRfix64K_00000)
 848                 *pdata = p[0];
 849         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 850                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 851         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 852                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 853         else if (msr == MSR_IA32_CR_PAT)
 854                 *pdata = vcpu->arch.pat;
 855         else {  /* Variable MTRRs */
 856                 int idx, is_mtrr_mask;
 857                 u64 *pt;
 858
 859                 idx = (msr - 0x200) / 2;
 860                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 861                 if (!is_mtrr_mask)
 862                         pt =
 863                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 864                 else
 865                         pt =
 866                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 867                 *pdata = *pt;
 868         }
 869
 870         return 0;
 871 }
 872
 873 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 874 {
 875         u64 data;
 876
 877         switch (msr) {
 878         case 0xc0010010: /* SYSCFG */
 879         case 0xc0010015: /* HWCR */
 880         case MSR_IA32_PLATFORM_ID:
 881         case MSR_IA32_P5_MC_ADDR:
 882         case MSR_IA32_P5_MC_TYPE:
 883         case MSR_IA32_MC0_CTL:
 884         case MSR_IA32_MCG_STATUS:
 885         case MSR_IA32_MCG_CAP:
 886         case MSR_IA32_MCG_CTL:
 887         case MSR_IA32_MC0_MISC:
 888         case MSR_IA32_MC0_MISC+4:
 889         case MSR_IA32_MC0_MISC+8:
 890         case MSR_IA32_MC0_MISC+12:
 891         case MSR_IA32_MC0_MISC+16:
 892         case MSR_IA32_MC0_MISC+20:
 893         case MSR_IA32_UCODE_REV:
 894         case MSR_IA32_EBL_CR_POWERON:
 895         case MSR_IA32_DEBUGCTLMSR:
 896         case MSR_IA32_LASTBRANCHFROMIP:
 897         case MSR_IA32_LASTBRANCHTOIP:
 898         case MSR_IA32_LASTINTFROMIP:
 899         case MSR_IA32_LASTINTTOIP:
 900         case MSR_VM_HSAVE_PA:
 901         case MSR_P6_EVNTSEL0:
 902         case MSR_P6_EVNTSEL1:
 903                 data = 0;
 904                 break;
 905         case MSR_MTRRcap:
 906                 data = 0x500 | KVM_NR_VAR_MTRR;
 907                 break;
 908         case 0x200 ... 0x2ff:
 909                 return get_msr_mtrr(vcpu, msr, pdata);
 910         case 0xcd: /* fsb frequency */
 911                 data = 3;
 912                 break;
 913         case MSR_IA32_APICBASE:
 914                 data = kvm_get_apic_base(vcpu);
 915                 break;
 916         case MSR_IA32_MISC_ENABLE:
 917                 data = vcpu->arch.ia32_misc_enable_msr;
 918                 break;
 919         case MSR_IA32_PERF_STATUS:
 920                 /* TSC increment by tick */
 921                 data = 1000ULL;
 922                 /* CPU multiplier */
 923                 data |= (((uint64_t)4ULL) << 40);
 924                 break;
 925         case MSR_EFER:
 926                 data = vcpu->arch.shadow_efer;
 927                 break;
 928         case MSR_KVM_WALL_CLOCK:
 929                 data = vcpu->kvm->arch.wall_clock;
 930                 break;
 931         case MSR_KVM_SYSTEM_TIME:
 932                 data = vcpu->arch.time;
 933                 break;
 934         default:
 935                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 936                 return 1;
 937         }
 938         *pdata = data;
 939         return 0;
 940 }
 941 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 942
 943 /*
 944  * Read or write a bunch of msrs. All parameters are kernel addresses.
 945  *
 946  * @return number of msrs set successfully.
 947  */
 948 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 949                     struct kvm_msr_entry *entries,
 950                     int (*do_msr)(struct kvm_vcpu *vcpu,
 951                                   unsigned index, u64 *data))
 952 {
 953         int i;
 954
 955         vcpu_load(vcpu);
 956
 957         down_read(&vcpu->kvm->slots_lock);
 958         for (i = 0; i < msrs->nmsrs; ++i)
 959                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 960                         break;
 961         up_read(&vcpu->kvm->slots_lock);
 962
 963         vcpu_put(vcpu);
 964
 965         return i;
 966 }
 967
 968 /*
 969  * Read or write a bunch of msrs. Parameters are user addresses.
 970  *
 971  * @return number of msrs set successfully.
 972  */
 973 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 974                   int (*do_msr)(struct kvm_vcpu *vcpu,
 975                                 unsigned index, u64 *data),
 976                   int writeback)
 977 {
 978         struct kvm_msrs msrs;
 979         struct kvm_msr_entry *entries;
 980         int r, n;
 981         unsigned size;
 982
 983         r = -EFAULT;
 984         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 985                 goto out;
 986
 987         r = -E2BIG;
 988         if (msrs.nmsrs >= MAX_IO_MSRS)
 989                 goto out;
 990
 991         r = -ENOMEM;
 992         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 993         entries = vmalloc(size);
 994         if (!entries)
 995                 goto out;
 996
 997         r = -EFAULT;
 998         if (copy_from_user(entries, user_msrs->entries, size))
 999                 goto out_free;
1000
1001         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1002         if (r < 0)
1003                 goto out_free;
1004
1005         r = -EFAULT;
1006         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1007                 goto out_free;
1008
1009         r = n;
1010
1011 out_free:
1012         vfree(entries);
1013 out:
1014         return r;
1015 }
1016
1017 int kvm_dev_ioctl_check_extension(long ext)
1018 {
1019         int r;
1020
1021         switch (ext) {
1022         case KVM_CAP_IRQCHIP:
1023         case KVM_CAP_HLT:
1024         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1025         case KVM_CAP_SET_TSS_ADDR:
1026         case KVM_CAP_EXT_CPUID:
1027         case KVM_CAP_CLOCKSOURCE:
1028         case KVM_CAP_PIT:
1029         case KVM_CAP_NOP_IO_DELAY:
1030         case KVM_CAP_MP_STATE:
1031         case KVM_CAP_SYNC_MMU:
1032         case KVM_CAP_REINJECT_CONTROL:
1033         case KVM_CAP_IRQ_INJECT_STATUS:
1034         case KVM_CAP_ASSIGN_DEV_IRQ:
1035                 r = 1;
1036                 break;
1037         case KVM_CAP_COALESCED_MMIO:
1038                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1039                 break;
1040         case KVM_CAP_VAPIC:
1041                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1042                 break;
1043         case KVM_CAP_NR_VCPUS:
1044                 r = KVM_MAX_VCPUS;
1045                 break;
1046         case KVM_CAP_NR_MEMSLOTS:
1047                 r = KVM_MEMORY_SLOTS;
1048                 break;
1049         case KVM_CAP_PV_MMU:
1050                 r = !tdp_enabled;
1051                 break;
1052         case KVM_CAP_IOMMU:
1053                 r = iommu_found();
1054                 break;
1055         default:
1056                 r = 0;
1057                 break;
1058         }
1059         return r;
1060
1061 }
1062
1063 long kvm_arch_dev_ioctl(struct file *filp,
1064                         unsigned int ioctl, unsigned long arg)
1065 {
1066         void __user *argp = (void __user *)arg;
1067         long r;
1068
1069         switch (ioctl) {
1070         case KVM_GET_MSR_INDEX_LIST: {
1071                 struct kvm_msr_list __user *user_msr_list = argp;
1072                 struct kvm_msr_list msr_list;
1073                 unsigned n;
1074
1075                 r = -EFAULT;
1076                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1077                         goto out;
1078                 n = msr_list.nmsrs;
1079                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1080                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1081                         goto out;
1082                 r = -E2BIG;
1083                 if (n < num_msrs_to_save)
1084                         goto out;
1085                 r = -EFAULT;
1086                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1087                                  num_msrs_to_save * sizeof(u32)))
1088                         goto out;
1089                 if (copy_to_user(user_msr_list->indices
1090                                  + num_msrs_to_save * sizeof(u32),
1091                                  &emulated_msrs,
1092                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1093                         goto out;
1094                 r = 0;
1095                 break;
1096         }
1097         case KVM_GET_SUPPORTED_CPUID: {
1098                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1099                 struct kvm_cpuid2 cpuid;
1100
1101                 r = -EFAULT;
1102                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1103                         goto out;
1104                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1105                                                       cpuid_arg->entries);
1106                 if (r)
1107                         goto out;
1108
1109                 r = -EFAULT;
1110                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1111                         goto out;
1112                 r = 0;
1113                 break;
1114         }
1115         default:
1116                 r = -EINVAL;
1117         }
1118 out:
1119         return r;
1120 }
1121
1122 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1123 {
1124         kvm_x86_ops->vcpu_load(vcpu, cpu);
1125         kvm_request_guest_time_update(vcpu);
1126 }
1127
1128 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1129 {
1130         kvm_x86_ops->vcpu_put(vcpu);
1131         kvm_put_guest_fpu(vcpu);
1132 }
1133
1134 static int is_efer_nx(void)
1135 {
1136         unsigned long long efer = 0;
1137
1138         rdmsrl_safe(MSR_EFER, &efer);
1139         return efer & EFER_NX;
1140 }
1141
1142 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1143 {
1144         int i;
1145         struct kvm_cpuid_entry2 *e, *entry;
1146
1147         entry = NULL;
1148         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1149                 e = &vcpu->arch.cpuid_entries[i];
1150                 if (e->function == 0x80000001) {
1151                         entry = e;
1152                         break;
1153                 }
1154         }
1155         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1156                 entry->edx &= ~(1 << 20);
1157                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1158         }
1159 }
1160
1161 /* when an old userspace process fills a new kernel module */
1162 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1163                                     struct kvm_cpuid *cpuid,
1164                                     struct kvm_cpuid_entry __user *entries)
1165 {
1166         int r, i;
1167         struct kvm_cpuid_entry *cpuid_entries;
1168
1169         r = -E2BIG;
1170         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1171                 goto out;
1172         r = -ENOMEM;
1173         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1174         if (!cpuid_entries)
1175                 goto out;
1176         r = -EFAULT;
1177         if (copy_from_user(cpuid_entries, entries,
1178                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1179                 goto out_free;
1180         for (i = 0; i < cpuid->nent; i++) {
1181                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1182                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1183                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1184                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1185                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1186                 vcpu->arch.cpuid_entries[i].index = 0;
1187                 vcpu->arch.cpuid_entries[i].flags = 0;
1188                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1189                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1190                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1191         }
1192         vcpu->arch.cpuid_nent = cpuid->nent;
1193         cpuid_fix_nx_cap(vcpu);
1194         r = 0;
1195
1196 out_free:
1197         vfree(cpuid_entries);
1198 out:
1199         return r;
1200 }
1201
1202 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1203                                      struct kvm_cpuid2 *cpuid,
1204                                      struct kvm_cpuid_entry2 __user *entries)
1205 {
1206         int r;
1207
1208         r = -E2BIG;
1209         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1210                 goto out;
1211         r = -EFAULT;
1212         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1213                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1214                 goto out;
1215         vcpu->arch.cpuid_nent = cpuid->nent;
1216         return 0;
1217
1218 out:
1219         return r;
1220 }
1221
1222 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1223                                      struct kvm_cpuid2 *cpuid,
1224                                      struct kvm_cpuid_entry2 __user *entries)
1225 {
1226         int r;
1227
1228         r = -E2BIG;
1229         if (cpuid->nent < vcpu->arch.cpuid_nent)
1230                 goto out;
1231         r = -EFAULT;
1232         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1233                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1234                 goto out;
1235         return 0;
1236
1237 out:
1238         cpuid->nent = vcpu->arch.cpuid_nent;
1239         return r;
1240 }
1241
1242 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1243                            u32 index)
1244 {
1245         entry->function = function;
1246         entry->index = index;
1247         cpuid_count(entry->function, entry->index,
1248                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1249         entry->flags = 0;
1250 }
1251
1252 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1253                          u32 index, int *nent, int maxnent)
1254 {
1255         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1256                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1257                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1258                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1259                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1260                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1261                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1262                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1263                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1264                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1265         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1266                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1267                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1268                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1269                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1270                 bit(X86_FEATURE_PGE) |
1271                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1272                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1273                 bit(X86_FEATURE_SYSCALL) |
1274                 (is_efer_nx() ? bit(X86_FEATURE_NX) : 0) |
1275 #ifdef CONFIG_X86_64
1276                 bit(X86_FEATURE_LM) |
1277 #endif
1278                 bit(X86_FEATURE_FXSR_OPT) |
1279                 bit(X86_FEATURE_MMXEXT) |
1280                 bit(X86_FEATURE_3DNOWEXT) |
1281                 bit(X86_FEATURE_3DNOW);
1282         const u32 kvm_supported_word3_x86_features =
1283                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1284         const u32 kvm_supported_word6_x86_features =
1285                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1286                 bit(X86_FEATURE_SVM);
1287
1288         /* all calls to cpuid_count() should be made on the same cpu */
1289         get_cpu();
1290         do_cpuid_1_ent(entry, function, index);
1291         ++*nent;
1292
1293         switch (function) {
1294         case 0:
1295                 entry->eax = min(entry->eax, (u32)0xb);
1296                 break;
1297         case 1:
1298                 entry->edx &= kvm_supported_word0_x86_features;
1299                 entry->ecx &= kvm_supported_word3_x86_features;
1300                 break;
1301         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1302          * may return different values. This forces us to get_cpu() before
1303          * issuing the first command, and also to emulate this annoying behavior
1304          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1305         case 2: {
1306                 int t, times = entry->eax & 0xff;
1307
1308                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1309                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1310                 for (t = 1; t < times && *nent < maxnent; ++t) {
1311                         do_cpuid_1_ent(&entry[t], function, 0);
1312                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1313                         ++*nent;
1314                 }
1315                 break;
1316         }
1317         /* function 4 and 0xb have additional index. */
1318         case 4: {
1319                 int i, cache_type;
1320
1321                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1322                 /* read more entries until cache_type is zero */
1323                 for (i = 1; *nent < maxnent; ++i) {
1324                         cache_type = entry[i - 1].eax & 0x1f;
1325                         if (!cache_type)
1326                                 break;
1327                         do_cpuid_1_ent(&entry[i], function, i);
1328                         entry[i].flags |=
1329                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1330                         ++*nent;
1331                 }
1332                 break;
1333         }
1334         case 0xb: {
1335                 int i, level_type;
1336
1337                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1338                 /* read more entries until level_type is zero */
1339                 for (i = 1; *nent < maxnent; ++i) {
1340                         level_type = entry[i - 1].ecx & 0xff00;
1341                         if (!level_type)
1342                                 break;
1343                         do_cpuid_1_ent(&entry[i], function, i);
1344                         entry[i].flags |=
1345                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1346                         ++*nent;
1347                 }
1348                 break;
1349         }
1350         case 0x80000000:
1351                 entry->eax = min(entry->eax, 0x8000001a);
1352                 break;
1353         case 0x80000001:
1354                 entry->edx &= kvm_supported_word1_x86_features;
1355                 entry->ecx &= kvm_supported_word6_x86_features;
1356                 break;
1357         }
1358         put_cpu();
1359 }
1360
1361 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1362                                      struct kvm_cpuid_entry2 __user *entries)
1363 {
1364         struct kvm_cpuid_entry2 *cpuid_entries;
1365         int limit, nent = 0, r = -E2BIG;
1366         u32 func;
1367
1368         if (cpuid->nent < 1)
1369                 goto out;
1370         r = -ENOMEM;
1371         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1372         if (!cpuid_entries)
1373                 goto out;
1374
1375         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1376         limit = cpuid_entries[0].eax;
1377         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1378                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1379                              &nent, cpuid->nent);
1380         r = -E2BIG;
1381         if (nent >= cpuid->nent)
1382                 goto out_free;
1383
1384         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1385         limit = cpuid_entries[nent - 1].eax;
1386         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1387                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1388                              &nent, cpuid->nent);
1389         r = -EFAULT;
1390         if (copy_to_user(entries, cpuid_entries,
1391                          nent * sizeof(struct kvm_cpuid_entry2)))
1392                 goto out_free;
1393         cpuid->nent = nent;
1394         r = 0;
1395
1396 out_free:
1397         vfree(cpuid_entries);
1398 out:
1399         return r;
1400 }
1401
1402 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1403                                     struct kvm_lapic_state *s)
1404 {
1405         vcpu_load(vcpu);
1406         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1407         vcpu_put(vcpu);
1408
1409         return 0;
1410 }
1411
1412 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1413                                     struct kvm_lapic_state *s)
1414 {
1415         vcpu_load(vcpu);
1416         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1417         kvm_apic_post_state_restore(vcpu);
1418         vcpu_put(vcpu);
1419
1420         return 0;
1421 }
1422
1423 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1424                                     struct kvm_interrupt *irq)
1425 {
1426         if (irq->irq < 0 || irq->irq >= 256)
1427                 return -EINVAL;
1428         if (irqchip_in_kernel(vcpu->kvm))
1429                 return -ENXIO;
1430         vcpu_load(vcpu);
1431
1432         set_bit(irq->irq, vcpu->arch.irq_pending);
1433         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1434
1435         vcpu_put(vcpu);
1436
1437         return 0;
1438 }
1439
1440 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1441 {
1442         vcpu_load(vcpu);
1443         kvm_inject_nmi(vcpu);
1444         vcpu_put(vcpu);
1445
1446         return 0;
1447 }
1448
1449 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1450                                            struct kvm_tpr_access_ctl *tac)
1451 {
1452         if (tac->flags)
1453                 return -EINVAL;
1454         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1455         return 0;
1456 }
1457
1458 long kvm_arch_vcpu_ioctl(struct file *filp,
1459                          unsigned int ioctl, unsigned long arg)
1460 {
1461         struct kvm_vcpu *vcpu = filp->private_data;
1462         void __user *argp = (void __user *)arg;
1463         int r;
1464         struct kvm_lapic_state *lapic = NULL;
1465
1466         switch (ioctl) {
1467         case KVM_GET_LAPIC: {
1468                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1469
1470                 r = -ENOMEM;
1471                 if (!lapic)
1472                         goto out;
1473                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1474                 if (r)
1475                         goto out;
1476                 r = -EFAULT;
1477                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1478                         goto out;
1479                 r = 0;
1480                 break;
1481         }
1482         case KVM_SET_LAPIC: {
1483                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1484                 r = -ENOMEM;
1485                 if (!lapic)
1486                         goto out;
1487                 r = -EFAULT;
1488                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1489                         goto out;
1490                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1491                 if (r)
1492                         goto out;
1493                 r = 0;
1494                 break;
1495         }
1496         case KVM_INTERRUPT: {
1497                 struct kvm_interrupt irq;
1498
1499                 r = -EFAULT;
1500                 if (copy_from_user(&irq, argp, sizeof irq))
1501                         goto out;
1502                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1503                 if (r)
1504                         goto out;
1505                 r = 0;
1506                 break;
1507         }
1508         case KVM_NMI: {
1509                 r = kvm_vcpu_ioctl_nmi(vcpu);
1510                 if (r)
1511                         goto out;
1512                 r = 0;
1513                 break;
1514         }
1515         case KVM_SET_CPUID: {
1516                 struct kvm_cpuid __user *cpuid_arg = argp;
1517                 struct kvm_cpuid cpuid;
1518
1519                 r = -EFAULT;
1520                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1521                         goto out;
1522                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1523                 if (r)
1524                         goto out;
1525                 break;
1526         }
1527         case KVM_SET_CPUID2: {
1528                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1529                 struct kvm_cpuid2 cpuid;
1530
1531                 r = -EFAULT;
1532                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1533                         goto out;
1534                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1535                                               cpuid_arg->entries);
1536                 if (r)
1537                         goto out;
1538                 break;
1539         }
1540         case KVM_GET_CPUID2: {
1541                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1542                 struct kvm_cpuid2 cpuid;
1543
1544                 r = -EFAULT;
1545                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1546                         goto out;
1547                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1548                                               cpuid_arg->entries);
1549                 if (r)
1550                         goto out;
1551                 r = -EFAULT;
1552                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1553                         goto out;
1554                 r = 0;
1555                 break;
1556         }
1557         case KVM_GET_MSRS:
1558                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1559                 break;
1560         case KVM_SET_MSRS:
1561                 r = msr_io(vcpu, argp, do_set_msr, 0);
1562                 break;
1563         case KVM_TPR_ACCESS_REPORTING: {
1564                 struct kvm_tpr_access_ctl tac;
1565
1566                 r = -EFAULT;
1567                 if (copy_from_user(&tac, argp, sizeof tac))
1568                         goto out;
1569                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1570                 if (r)
1571                         goto out;
1572                 r = -EFAULT;
1573                 if (copy_to_user(argp, &tac, sizeof tac))
1574                         goto out;
1575                 r = 0;
1576                 break;
1577         };
1578         case KVM_SET_VAPIC_ADDR: {
1579                 struct kvm_vapic_addr va;
1580
1581                 r = -EINVAL;
1582                 if (!irqchip_in_kernel(vcpu->kvm))
1583                         goto out;
1584                 r = -EFAULT;
1585                 if (copy_from_user(&va, argp, sizeof va))
1586                         goto out;
1587                 r = 0;
1588                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1589                 break;
1590         }
1591         default:
1592                 r = -EINVAL;
1593         }
1594 out:
1595         kfree(lapic);
1596         return r;
1597 }
1598
1599 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1600 {
1601         int ret;
1602
1603         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1604                 return -1;
1605         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1606         return ret;
1607 }
1608
1609 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1610                                           u32 kvm_nr_mmu_pages)
1611 {
1612         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1613                 return -EINVAL;
1614
1615         down_write(&kvm->slots_lock);
1616
1617         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1618         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1619
1620         up_write(&kvm->slots_lock);
1621         return 0;
1622 }
1623
1624 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1625 {
1626         return kvm->arch.n_alloc_mmu_pages;
1627 }
1628
1629 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1630 {
1631         int i;
1632         struct kvm_mem_alias *alias;
1633
1634         for (i = 0; i < kvm->arch.naliases; ++i) {
1635                 alias = &kvm->arch.aliases[i];
1636                 if (gfn >= alias->base_gfn
1637                     && gfn < alias->base_gfn + alias->npages)
1638                         return alias->target_gfn + gfn - alias->base_gfn;
1639         }
1640         return gfn;
1641 }
1642
1643 /*
1644  * Set a new alias region.  Aliases map a portion of physical memory into
1645  * another portion.  This is useful for memory windows, for example the PC
1646  * VGA region.
1647  */
1648 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1649                                          struct kvm_memory_alias *alias)
1650 {
1651         int r, n;
1652         struct kvm_mem_alias *p;
1653
1654         r = -EINVAL;
1655         /* General sanity checks */
1656         if (alias->memory_size & (PAGE_SIZE - 1))
1657                 goto out;
1658         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1659                 goto out;
1660         if (alias->slot >= KVM_ALIAS_SLOTS)
1661                 goto out;
1662         if (alias->guest_phys_addr + alias->memory_size
1663             < alias->guest_phys_addr)
1664                 goto out;
1665         if (alias->target_phys_addr + alias->memory_size
1666             < alias->target_phys_addr)
1667                 goto out;
1668
1669         down_write(&kvm->slots_lock);
1670         spin_lock(&kvm->mmu_lock);
1671
1672         p = &kvm->arch.aliases[alias->slot];
1673         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1674         p->npages = alias->memory_size >> PAGE_SHIFT;
1675         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1676
1677         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1678                 if (kvm->arch.aliases[n - 1].npages)
1679                         break;
1680         kvm->arch.naliases = n;
1681
1682         spin_unlock(&kvm->mmu_lock);
1683         kvm_mmu_zap_all(kvm);
1684
1685         up_write(&kvm->slots_lock);
1686
1687         return 0;
1688
1689 out:
1690         return r;
1691 }
1692
1693 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1694 {
1695         int r;
1696
1697         r = 0;
1698         switch (chip->chip_id) {
1699         case KVM_IRQCHIP_PIC_MASTER:
1700                 memcpy(&chip->chip.pic,
1701                         &pic_irqchip(kvm)->pics[0],
1702                         sizeof(struct kvm_pic_state));
1703                 break;
1704         case KVM_IRQCHIP_PIC_SLAVE:
1705                 memcpy(&chip->chip.pic,
1706                         &pic_irqchip(kvm)->pics[1],
1707                         sizeof(struct kvm_pic_state));
1708                 break;
1709         case KVM_IRQCHIP_IOAPIC:
1710                 memcpy(&chip->chip.ioapic,
1711                         ioapic_irqchip(kvm),
1712                         sizeof(struct kvm_ioapic_state));
1713                 break;
1714         default:
1715                 r = -EINVAL;
1716                 break;
1717         }
1718         return r;
1719 }
1720
1721 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1722 {
1723         int r;
1724
1725         r = 0;
1726         switch (chip->chip_id) {
1727         case KVM_IRQCHIP_PIC_MASTER:
1728                 memcpy(&pic_irqchip(kvm)->pics[0],
1729                         &chip->chip.pic,
1730                         sizeof(struct kvm_pic_state));
1731                 break;
1732         case KVM_IRQCHIP_PIC_SLAVE:
1733                 memcpy(&pic_irqchip(kvm)->pics[1],
1734                         &chip->chip.pic,
1735                         sizeof(struct kvm_pic_state));
1736                 break;
1737         case KVM_IRQCHIP_IOAPIC:
1738                 memcpy(ioapic_irqchip(kvm),
1739                         &chip->chip.ioapic,
1740                         sizeof(struct kvm_ioapic_state));
1741                 break;
1742         default:
1743                 r = -EINVAL;
1744                 break;
1745         }
1746         kvm_pic_update_irq(pic_irqchip(kvm));
1747         return r;
1748 }
1749
1750 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1751 {
1752         int r = 0;
1753
1754         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1755         return r;
1756 }
1757
1758 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1759 {
1760         int r = 0;
1761
1762         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1763         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1764         return r;
1765 }
1766
1767 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1768                                  struct kvm_reinject_control *control)
1769 {
1770         if (!kvm->arch.vpit)
1771                 return -ENXIO;
1772         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1773         return 0;
1774 }
1775
1776 /*
1777  * Get (and clear) the dirty memory log for a memory slot.
1778  */
1779 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1780                                       struct kvm_dirty_log *log)
1781 {
1782         int r;
1783         int n;
1784         struct kvm_memory_slot *memslot;
1785         int is_dirty = 0;
1786
1787         down_write(&kvm->slots_lock);
1788
1789         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1790         if (r)
1791                 goto out;
1792
1793         /* If nothing is dirty, don't bother messing with page tables. */
1794         if (is_dirty) {
1795                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1796                 kvm_flush_remote_tlbs(kvm);
1797                 memslot = &kvm->memslots[log->slot];
1798                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1799                 memset(memslot->dirty_bitmap, 0, n);
1800         }
1801         r = 0;
1802 out:
1803         up_write(&kvm->slots_lock);
1804         return r;
1805 }
1806
1807 long kvm_arch_vm_ioctl(struct file *filp,
1808                        unsigned int ioctl, unsigned long arg)
1809 {
1810         struct kvm *kvm = filp->private_data;
1811         void __user *argp = (void __user *)arg;
1812         int r = -EINVAL;
1813         /*
1814          * This union makes it completely explicit to gcc-3.x
1815          * that these two variables' stack usage should be
1816          * combined, not added together.
1817          */
1818         union {
1819                 struct kvm_pit_state ps;
1820                 struct kvm_memory_alias alias;
1821         } u;
1822
1823         switch (ioctl) {
1824         case KVM_SET_TSS_ADDR:
1825                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1826                 if (r < 0)
1827                         goto out;
1828                 break;
1829         case KVM_SET_MEMORY_REGION: {
1830                 struct kvm_memory_region kvm_mem;
1831                 struct kvm_userspace_memory_region kvm_userspace_mem;
1832
1833                 r = -EFAULT;
1834                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1835                         goto out;
1836                 kvm_userspace_mem.slot = kvm_mem.slot;
1837                 kvm_userspace_mem.flags = kvm_mem.flags;
1838                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1839                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1840                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1841                 if (r)
1842                         goto out;
1843                 break;
1844         }
1845         case KVM_SET_NR_MMU_PAGES:
1846                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1847                 if (r)
1848                         goto out;
1849                 break;
1850         case KVM_GET_NR_MMU_PAGES:
1851                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1852                 break;
1853         case KVM_SET_MEMORY_ALIAS:
1854                 r = -EFAULT;
1855                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1856                         goto out;
1857                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1858                 if (r)
1859                         goto out;
1860                 break;
1861         case KVM_CREATE_IRQCHIP:
1862                 r = -ENOMEM;
1863                 kvm->arch.vpic = kvm_create_pic(kvm);
1864                 if (kvm->arch.vpic) {
1865                         r = kvm_ioapic_init(kvm);
1866                         if (r) {
1867                                 kfree(kvm->arch.vpic);
1868                                 kvm->arch.vpic = NULL;
1869                                 goto out;
1870                         }
1871                 } else
1872                         goto out;
1873                 r = kvm_setup_default_irq_routing(kvm);
1874                 if (r) {
1875                         kfree(kvm->arch.vpic);
1876                         kfree(kvm->arch.vioapic);
1877                         goto out;
1878                 }
1879                 break;
1880         case KVM_CREATE_PIT:
1881                 mutex_lock(&kvm->lock);
1882                 r = -EEXIST;
1883                 if (kvm->arch.vpit)
1884                         goto create_pit_unlock;
1885                 r = -ENOMEM;
1886                 kvm->arch.vpit = kvm_create_pit(kvm);
1887                 if (kvm->arch.vpit)
1888                         r = 0;
1889         create_pit_unlock:
1890                 mutex_unlock(&kvm->lock);
1891                 break;
1892         case KVM_IRQ_LINE_STATUS:
1893         case KVM_IRQ_LINE: {
1894                 struct kvm_irq_level irq_event;
1895
1896                 r = -EFAULT;
1897                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1898                         goto out;
1899                 if (irqchip_in_kernel(kvm)) {
1900                         __s32 status;
1901                         mutex_lock(&kvm->lock);
1902                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1903                                         irq_event.irq, irq_event.level);
1904                         mutex_unlock(&kvm->lock);
1905                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1906                                 irq_event.status = status;
1907                                 if (copy_to_user(argp, &irq_event,
1908                                                         sizeof irq_event))
1909                                         goto out;
1910                         }
1911                         r = 0;
1912                 }
1913                 break;
1914         }
1915         case KVM_GET_IRQCHIP: {
1916                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1917                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1918
1919                 r = -ENOMEM;
1920                 if (!chip)
1921                         goto out;
1922                 r = -EFAULT;
1923                 if (copy_from_user(chip, argp, sizeof *chip))
1924                         goto get_irqchip_out;
1925                 r = -ENXIO;
1926                 if (!irqchip_in_kernel(kvm))
1927                         goto get_irqchip_out;
1928                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1929                 if (r)
1930                         goto get_irqchip_out;
1931                 r = -EFAULT;
1932                 if (copy_to_user(argp, chip, sizeof *chip))
1933                         goto get_irqchip_out;
1934                 r = 0;
1935         get_irqchip_out:
1936                 kfree(chip);
1937                 if (r)
1938                         goto out;
1939                 break;
1940         }
1941         case KVM_SET_IRQCHIP: {
1942                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1943                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1944
1945                 r = -ENOMEM;
1946                 if (!chip)
1947                         goto out;
1948                 r = -EFAULT;
1949                 if (copy_from_user(chip, argp, sizeof *chip))
1950                         goto set_irqchip_out;
1951                 r = -ENXIO;
1952                 if (!irqchip_in_kernel(kvm))
1953                         goto set_irqchip_out;
1954                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1955                 if (r)
1956                         goto set_irqchip_out;
1957                 r = 0;
1958         set_irqchip_out:
1959                 kfree(chip);
1960                 if (r)
1961                         goto out;
1962                 break;
1963         }
1964         case KVM_GET_PIT: {
1965                 r = -EFAULT;
1966                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1967                         goto out;
1968                 r = -ENXIO;
1969                 if (!kvm->arch.vpit)
1970                         goto out;
1971                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1972                 if (r)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         case KVM_SET_PIT: {
1981                 r = -EFAULT;
1982                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1983                         goto out;
1984                 r = -ENXIO;
1985                 if (!kvm->arch.vpit)
1986                         goto out;
1987                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1988                 if (r)
1989                         goto out;
1990                 r = 0;
1991                 break;
1992         }
1993         case KVM_REINJECT_CONTROL: {
1994                 struct kvm_reinject_control control;
1995                 r =  -EFAULT;
1996                 if (copy_from_user(&control, argp, sizeof(control)))
1997                         goto out;
1998                 r = kvm_vm_ioctl_reinject(kvm, &control);
1999                 if (r)
2000                         goto out;
2001                 r = 0;
2002                 break;
2003         }
2004         default:
2005                 ;
2006         }
2007 out:
2008         return r;
2009 }
2010
2011 static void kvm_init_msr_list(void)
2012 {
2013         u32 dummy[2];
2014         unsigned i, j;
2015
2016         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2017                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2018                         continue;
2019                 if (j < i)
2020                         msrs_to_save[j] = msrs_to_save[i];
2021                 j++;
2022         }
2023         num_msrs_to_save = j;
2024 }
2025
2026 /*
2027  * Only apic need an MMIO device hook, so shortcut now..
2028  */
2029 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2030                                                 gpa_t addr, int len,
2031                                                 int is_write)
2032 {
2033         struct kvm_io_device *dev;
2034
2035         if (vcpu->arch.apic) {
2036                 dev = &vcpu->arch.apic->dev;
2037                 if (dev->in_range(dev, addr, len, is_write))
2038                         return dev;
2039         }
2040         return NULL;
2041 }
2042
2043
2044 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2045                                                 gpa_t addr, int len,
2046                                                 int is_write)
2047 {
2048         struct kvm_io_device *dev;
2049
2050         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2051         if (dev == NULL)
2052                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2053                                           is_write);
2054         return dev;
2055 }
2056
2057 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2058                                struct kvm_vcpu *vcpu)
2059 {
2060         void *data = val;
2061         int r = X86EMUL_CONTINUE;
2062
2063         while (bytes) {
2064                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2065                 unsigned offset = addr & (PAGE_SIZE-1);
2066                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2067                 int ret;
2068
2069                 if (gpa == UNMAPPED_GVA) {
2070                         r = X86EMUL_PROPAGATE_FAULT;
2071                         goto out;
2072                 }
2073                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2074                 if (ret < 0) {
2075                         r = X86EMUL_UNHANDLEABLE;
2076                         goto out;
2077                 }
2078
2079                 bytes -= toread;
2080                 data += toread;
2081                 addr += toread;
2082         }
2083 out:
2084         return r;
2085 }
2086
2087 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2088                                 struct kvm_vcpu *vcpu)
2089 {
2090         void *data = val;
2091         int r = X86EMUL_CONTINUE;
2092
2093         while (bytes) {
2094                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2095                 unsigned offset = addr & (PAGE_SIZE-1);
2096                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2097                 int ret;
2098
2099                 if (gpa == UNMAPPED_GVA) {
2100                         r = X86EMUL_PROPAGATE_FAULT;
2101                         goto out;
2102                 }
2103                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2104                 if (ret < 0) {
2105                         r = X86EMUL_UNHANDLEABLE;
2106                         goto out;
2107                 }
2108
2109                 bytes -= towrite;
2110                 data += towrite;
2111                 addr += towrite;
2112         }
2113 out:
2114         return r;
2115 }
2116
2117
2118 static int emulator_read_emulated(unsigned long addr,
2119                                   void *val,
2120                                   unsigned int bytes,
2121                                   struct kvm_vcpu *vcpu)
2122 {
2123         struct kvm_io_device *mmio_dev;
2124         gpa_t                 gpa;
2125
2126         if (vcpu->mmio_read_completed) {
2127                 memcpy(val, vcpu->mmio_data, bytes);
2128                 vcpu->mmio_read_completed = 0;
2129                 return X86EMUL_CONTINUE;
2130         }
2131
2132         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2133
2134         /* For APIC access vmexit */
2135         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2136                 goto mmio;
2137
2138         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2139                                 == X86EMUL_CONTINUE)
2140                 return X86EMUL_CONTINUE;
2141         if (gpa == UNMAPPED_GVA)
2142                 return X86EMUL_PROPAGATE_FAULT;
2143
2144 mmio:
2145         /*
2146          * Is this MMIO handled locally?
2147          */
2148         mutex_lock(&vcpu->kvm->lock);
2149         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2150         if (mmio_dev) {
2151                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2152                 mutex_unlock(&vcpu->kvm->lock);
2153                 return X86EMUL_CONTINUE;
2154         }
2155         mutex_unlock(&vcpu->kvm->lock);
2156
2157         vcpu->mmio_needed = 1;
2158         vcpu->mmio_phys_addr = gpa;
2159         vcpu->mmio_size = bytes;
2160         vcpu->mmio_is_write = 0;
2161
2162         return X86EMUL_UNHANDLEABLE;
2163 }
2164
2165 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2166                           const void *val, int bytes)
2167 {
2168         int ret;
2169
2170         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2171         if (ret < 0)
2172                 return 0;
2173         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2174         return 1;
2175 }
2176
2177 static int emulator_write_emulated_onepage(unsigned long addr,
2178                                            const void *val,
2179                                            unsigned int bytes,
2180                                            struct kvm_vcpu *vcpu)
2181 {
2182         struct kvm_io_device *mmio_dev;
2183         gpa_t                 gpa;
2184
2185         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2186
2187         if (gpa == UNMAPPED_GVA) {
2188                 kvm_inject_page_fault(vcpu, addr, 2);
2189                 return X86EMUL_PROPAGATE_FAULT;
2190         }
2191
2192         /* For APIC access vmexit */
2193         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2194                 goto mmio;
2195
2196         if (emulator_write_phys(vcpu, gpa, val, bytes))
2197                 return X86EMUL_CONTINUE;
2198
2199 mmio:
2200         /*
2201          * Is this MMIO handled locally?
2202          */
2203         mutex_lock(&vcpu->kvm->lock);
2204         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2205         if (mmio_dev) {
2206                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2207                 mutex_unlock(&vcpu->kvm->lock);
2208                 return X86EMUL_CONTINUE;
2209         }
2210         mutex_unlock(&vcpu->kvm->lock);
2211
2212         vcpu->mmio_needed = 1;
2213         vcpu->mmio_phys_addr = gpa;
2214         vcpu->mmio_size = bytes;
2215         vcpu->mmio_is_write = 1;
2216         memcpy(vcpu->mmio_data, val, bytes);
2217
2218         return X86EMUL_CONTINUE;
2219 }
2220
2221 int emulator_write_emulated(unsigned long addr,
2222                                    const void *val,
2223                                    unsigned int bytes,
2224                                    struct kvm_vcpu *vcpu)
2225 {
2226         /* Crossing a page boundary? */
2227         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2228                 int rc, now;
2229
2230                 now = -addr & ~PAGE_MASK;
2231                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2232                 if (rc != X86EMUL_CONTINUE)
2233                         return rc;
2234                 addr += now;
2235                 val += now;
2236                 bytes -= now;
2237         }
2238         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2239 }
2240 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2241
2242 static int emulator_cmpxchg_emulated(unsigned long addr,
2243                                      const void *old,
2244                                      const void *new,
2245                                      unsigned int bytes,
2246                                      struct kvm_vcpu *vcpu)
2247 {
2248         static int reported;
2249
2250         if (!reported) {
2251                 reported = 1;
2252                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2253         }
2254 #ifndef CONFIG_X86_64
2255         /* guests cmpxchg8b have to be emulated atomically */
2256         if (bytes == 8) {
2257                 gpa_t gpa;
2258                 struct page *page;
2259                 char *kaddr;
2260                 u64 val;
2261
2262                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2263
2264                 if (gpa == UNMAPPED_GVA ||
2265                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2266                         goto emul_write;
2267
2268                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2269                         goto emul_write;
2270
2271                 val = *(u64 *)new;
2272
2273                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2274
2275                 kaddr = kmap_atomic(page, KM_USER0);
2276                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2277                 kunmap_atomic(kaddr, KM_USER0);
2278                 kvm_release_page_dirty(page);
2279         }
2280 emul_write:
2281 #endif
2282
2283         return emulator_write_emulated(addr, new, bytes, vcpu);
2284 }
2285
2286 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2287 {
2288         return kvm_x86_ops->get_segment_base(vcpu, seg);
2289 }
2290
2291 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2292 {
2293         kvm_mmu_invlpg(vcpu, address);
2294         return X86EMUL_CONTINUE;
2295 }
2296
2297 int emulate_clts(struct kvm_vcpu *vcpu)
2298 {
2299         KVMTRACE_0D(CLTS, vcpu, handler);
2300         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2301         return X86EMUL_CONTINUE;
2302 }
2303
2304 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2305 {
2306         struct kvm_vcpu *vcpu = ctxt->vcpu;
2307
2308         switch (dr) {
2309         case 0 ... 3:
2310                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2311                 return X86EMUL_CONTINUE;
2312         default:
2313                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2314                 return X86EMUL_UNHANDLEABLE;
2315         }
2316 }
2317
2318 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2319 {
2320         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2321         int exception;
2322
2323         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2324         if (exception) {
2325                 /* FIXME: better handling */
2326                 return X86EMUL_UNHANDLEABLE;
2327         }
2328         return X86EMUL_CONTINUE;
2329 }
2330
2331 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2332 {
2333         u8 opcodes[4];
2334         unsigned long rip = kvm_rip_read(vcpu);
2335         unsigned long rip_linear;
2336
2337         if (!printk_ratelimit())
2338                 return;
2339
2340         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2341
2342         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2343
2344         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2345                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2346 }
2347 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2348
2349 static struct x86_emulate_ops emulate_ops = {
2350         .read_std            = kvm_read_guest_virt,
2351         .read_emulated       = emulator_read_emulated,
2352         .write_emulated      = emulator_write_emulated,
2353         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2354 };
2355
2356 static void cache_all_regs(struct kvm_vcpu *vcpu)
2357 {
2358         kvm_register_read(vcpu, VCPU_REGS_RAX);
2359         kvm_register_read(vcpu, VCPU_REGS_RSP);
2360         kvm_register_read(vcpu, VCPU_REGS_RIP);
2361         vcpu->arch.regs_dirty = ~0;
2362 }
2363
2364 int emulate_instruction(struct kvm_vcpu *vcpu,
2365                         struct kvm_run *run,
2366                         unsigned long cr2,
2367                         u16 error_code,
2368                         int emulation_type)
2369 {
2370         int r;
2371         struct decode_cache *c;
2372
2373         kvm_clear_exception_queue(vcpu);
2374         vcpu->arch.mmio_fault_cr2 = cr2;
2375         /*
2376          * TODO: fix x86_emulate.c to use guest_read/write_register
2377          * instead of direct ->regs accesses, can save hundred cycles
2378          * on Intel for instructions that don't read/change RSP, for
2379          * for example.
2380          */
2381         cache_all_regs(vcpu);
2382
2383         vcpu->mmio_is_write = 0;
2384         vcpu->arch.pio.string = 0;
2385
2386         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2387                 int cs_db, cs_l;
2388                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2389
2390                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2391                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2392                 vcpu->arch.emulate_ctxt.mode =
2393                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2394                         ? X86EMUL_MODE_REAL : cs_l
2395                         ? X86EMUL_MODE_PROT64 : cs_db
2396                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2397
2398                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2399
2400                 /* Reject the instructions other than VMCALL/VMMCALL when
2401                  * try to emulate invalid opcode */
2402                 c = &vcpu->arch.emulate_ctxt.decode;
2403                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2404                     (!(c->twobyte && c->b == 0x01 &&
2405                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2406                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2407                         return EMULATE_FAIL;
2408
2409                 ++vcpu->stat.insn_emulation;
2410                 if (r)  {
2411                         ++vcpu->stat.insn_emulation_fail;
2412                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2413                                 return EMULATE_DONE;
2414                         return EMULATE_FAIL;
2415                 }
2416         }
2417
2418         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2419
2420         if (vcpu->arch.pio.string)
2421                 return EMULATE_DO_MMIO;
2422
2423         if ((r || vcpu->mmio_is_write) && run) {
2424                 run->exit_reason = KVM_EXIT_MMIO;
2425                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2426                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2427                 run->mmio.len = vcpu->mmio_size;
2428                 run->mmio.is_write = vcpu->mmio_is_write;
2429         }
2430
2431         if (r) {
2432                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2433                         return EMULATE_DONE;
2434                 if (!vcpu->mmio_needed) {
2435                         kvm_report_emulation_failure(vcpu, "mmio");
2436                         return EMULATE_FAIL;
2437                 }
2438                 return EMULATE_DO_MMIO;
2439         }
2440
2441         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2442
2443         if (vcpu->mmio_is_write) {
2444                 vcpu->mmio_needed = 0;
2445                 return EMULATE_DO_MMIO;
2446         }
2447
2448         return EMULATE_DONE;
2449 }
2450 EXPORT_SYMBOL_GPL(emulate_instruction);
2451
2452 static int pio_copy_data(struct kvm_vcpu *vcpu)
2453 {
2454         void *p = vcpu->arch.pio_data;
2455         gva_t q = vcpu->arch.pio.guest_gva;
2456         unsigned bytes;
2457         int ret;
2458
2459         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2460         if (vcpu->arch.pio.in)
2461                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2462         else
2463                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2464         return ret;
2465 }
2466
2467 int complete_pio(struct kvm_vcpu *vcpu)
2468 {
2469         struct kvm_pio_request *io = &vcpu->arch.pio;
2470         long delta;
2471         int r;
2472         unsigned long val;
2473
2474         if (!io->string) {
2475                 if (io->in) {
2476                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2477                         memcpy(&val, vcpu->arch.pio_data, io->size);
2478                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2479                 }
2480         } else {
2481                 if (io->in) {
2482                         r = pio_copy_data(vcpu);
2483                         if (r)
2484                                 return r;
2485                 }
2486
2487                 delta = 1;
2488                 if (io->rep) {
2489                         delta *= io->cur_count;
2490                         /*
2491                          * The size of the register should really depend on
2492                          * current address size.
2493                          */
2494                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2495                         val -= delta;
2496                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2497                 }
2498                 if (io->down)
2499                         delta = -delta;
2500                 delta *= io->size;
2501                 if (io->in) {
2502                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2503                         val += delta;
2504                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2505                 } else {
2506                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2507                         val += delta;
2508                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2509                 }
2510         }
2511
2512         io->count -= io->cur_count;
2513         io->cur_count = 0;
2514
2515         return 0;
2516 }
2517
2518 static void kernel_pio(struct kvm_io_device *pio_dev,
2519                        struct kvm_vcpu *vcpu,
2520                        void *pd)
2521 {
2522         /* TODO: String I/O for in kernel device */
2523
2524         mutex_lock(&vcpu->kvm->lock);
2525         if (vcpu->arch.pio.in)
2526                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2527                                   vcpu->arch.pio.size,
2528                                   pd);
2529         else
2530                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2531                                    vcpu->arch.pio.size,
2532                                    pd);
2533         mutex_unlock(&vcpu->kvm->lock);
2534 }
2535
2536 static void pio_string_write(struct kvm_io_device *pio_dev,
2537                              struct kvm_vcpu *vcpu)
2538 {
2539         struct kvm_pio_request *io = &vcpu->arch.pio;
2540         void *pd = vcpu->arch.pio_data;
2541         int i;
2542
2543         mutex_lock(&vcpu->kvm->lock);
2544         for (i = 0; i < io->cur_count; i++) {
2545                 kvm_iodevice_write(pio_dev, io->port,
2546                                    io->size,
2547                                    pd);
2548                 pd += io->size;
2549         }
2550         mutex_unlock(&vcpu->kvm->lock);
2551 }
2552
2553 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2554                                                gpa_t addr, int len,
2555                                                int is_write)
2556 {
2557         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2558 }
2559
2560 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2561                   int size, unsigned port)
2562 {
2563         struct kvm_io_device *pio_dev;
2564         unsigned long val;
2565
2566         vcpu->run->exit_reason = KVM_EXIT_IO;
2567         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2568         vcpu->run->io.size = vcpu->arch.pio.size = size;
2569         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2570         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2571         vcpu->run->io.port = vcpu->arch.pio.port = port;
2572         vcpu->arch.pio.in = in;
2573         vcpu->arch.pio.string = 0;
2574         vcpu->arch.pio.down = 0;
2575         vcpu->arch.pio.rep = 0;
2576
2577         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2578                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2579                             handler);
2580         else
2581                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2582                             handler);
2583
2584         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2585         memcpy(vcpu->arch.pio_data, &val, 4);
2586
2587         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2588         if (pio_dev) {
2589                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2590                 complete_pio(vcpu);
2591                 return 1;
2592         }
2593         return 0;
2594 }
2595 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2596
2597 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2598                   int size, unsigned long count, int down,
2599                   gva_t address, int rep, unsigned port)
2600 {
2601         unsigned now, in_page;
2602         int ret = 0;
2603         struct kvm_io_device *pio_dev;
2604
2605         vcpu->run->exit_reason = KVM_EXIT_IO;
2606         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2607         vcpu->run->io.size = vcpu->arch.pio.size = size;
2608         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2609         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2610         vcpu->run->io.port = vcpu->arch.pio.port = port;
2611         vcpu->arch.pio.in = in;
2612         vcpu->arch.pio.string = 1;
2613         vcpu->arch.pio.down = down;
2614         vcpu->arch.pio.rep = rep;
2615
2616         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2617                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2618                             handler);
2619         else
2620                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2621                             handler);
2622
2623         if (!count) {
2624                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2625                 return 1;
2626         }
2627
2628         if (!down)
2629                 in_page = PAGE_SIZE - offset_in_page(address);
2630         else
2631                 in_page = offset_in_page(address) + size;
2632         now = min(count, (unsigned long)in_page / size);
2633         if (!now)
2634                 now = 1;
2635         if (down) {
2636                 /*
2637                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2638                  */
2639                 pr_unimpl(vcpu, "guest string pio down\n");
2640                 kvm_inject_gp(vcpu, 0);
2641                 return 1;
2642         }
2643         vcpu->run->io.count = now;
2644         vcpu->arch.pio.cur_count = now;
2645
2646         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2647                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2648
2649         vcpu->arch.pio.guest_gva = address;
2650
2651         pio_dev = vcpu_find_pio_dev(vcpu, port,
2652                                     vcpu->arch.pio.cur_count,
2653                                     !vcpu->arch.pio.in);
2654         if (!vcpu->arch.pio.in) {
2655                 /* string PIO write */
2656                 ret = pio_copy_data(vcpu);
2657                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2658                         kvm_inject_gp(vcpu, 0);
2659                         return 1;
2660                 }
2661                 if (ret == 0 && pio_dev) {
2662                         pio_string_write(pio_dev, vcpu);
2663                         complete_pio(vcpu);
2664                         if (vcpu->arch.pio.count == 0)
2665                                 ret = 1;
2666                 }
2667         } else if (pio_dev)
2668                 pr_unimpl(vcpu, "no string pio read support yet, "
2669                        "port %x size %d count %ld\n",
2670                         port, size, count);
2671
2672         return ret;
2673 }
2674 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2675
2676 static void bounce_off(void *info)
2677 {
2678         /* nothing */
2679 }
2680
2681 static unsigned int  ref_freq;
2682 static unsigned long tsc_khz_ref;
2683
2684 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2685                                      void *data)
2686 {
2687         struct cpufreq_freqs *freq = data;
2688         struct kvm *kvm;
2689         struct kvm_vcpu *vcpu;
2690         int i, send_ipi = 0;
2691
2692         if (!ref_freq)
2693                 ref_freq = freq->old;
2694
2695         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2696                 return 0;
2697         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2698                 return 0;
2699         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2700
2701         spin_lock(&kvm_lock);
2702         list_for_each_entry(kvm, &vm_list, vm_list) {
2703                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2704                         vcpu = kvm->vcpus[i];
2705                         if (!vcpu)
2706                                 continue;
2707                         if (vcpu->cpu != freq->cpu)
2708                                 continue;
2709                         if (!kvm_request_guest_time_update(vcpu))
2710                                 continue;
2711                         if (vcpu->cpu != smp_processor_id())
2712                                 send_ipi++;
2713                 }
2714         }
2715         spin_unlock(&kvm_lock);
2716
2717         if (freq->old < freq->new && send_ipi) {
2718                 /*
2719                  * We upscale the frequency.  Must make the guest
2720                  * doesn't see old kvmclock values while running with
2721                  * the new frequency, otherwise we risk the guest sees
2722                  * time go backwards.
2723                  *
2724                  * In case we update the frequency for another cpu
2725                  * (which might be in guest context) send an interrupt
2726                  * to kick the cpu out of guest context.  Next time
2727                  * guest context is entered kvmclock will be updated,
2728                  * so the guest will not see stale values.
2729                  */
2730                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2731         }
2732         return 0;
2733 }
2734
2735 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2736         .notifier_call  = kvmclock_cpufreq_notifier
2737 };
2738
2739 int kvm_arch_init(void *opaque)
2740 {
2741         int r, cpu;
2742         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2743
2744         if (kvm_x86_ops) {
2745                 printk(KERN_ERR "kvm: already loaded the other module\n");
2746                 r = -EEXIST;
2747                 goto out;
2748         }
2749
2750         if (!ops->cpu_has_kvm_support()) {
2751                 printk(KERN_ERR "kvm: no hardware support\n");
2752                 r = -EOPNOTSUPP;
2753                 goto out;
2754         }
2755         if (ops->disabled_by_bios()) {
2756                 printk(KERN_ERR "kvm: disabled by bios\n");
2757                 r = -EOPNOTSUPP;
2758                 goto out;
2759         }
2760
2761         r = kvm_mmu_module_init();
2762         if (r)
2763                 goto out;
2764
2765         kvm_init_msr_list();
2766
2767         kvm_x86_ops = ops;
2768         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2769         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2770         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2771                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2772
2773         for_each_possible_cpu(cpu)
2774                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2775         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2776                 tsc_khz_ref = tsc_khz;
2777                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2778                                           CPUFREQ_TRANSITION_NOTIFIER);
2779         }
2780
2781         return 0;
2782
2783 out:
2784         return r;
2785 }
2786
2787 void kvm_arch_exit(void)
2788 {
2789         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2790                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2791                                             CPUFREQ_TRANSITION_NOTIFIER);
2792         kvm_x86_ops = NULL;
2793         kvm_mmu_module_exit();
2794 }
2795
2796 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2797 {
2798         ++vcpu->stat.halt_exits;
2799         KVMTRACE_0D(HLT, vcpu, handler);
2800         if (irqchip_in_kernel(vcpu->kvm)) {
2801                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2802                 return 1;
2803         } else {
2804                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2805                 return 0;
2806         }
2807 }
2808 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2809
2810 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2811                            unsigned long a1)
2812 {
2813         if (is_long_mode(vcpu))
2814                 return a0;
2815         else
2816                 return a0 | ((gpa_t)a1 << 32);
2817 }
2818
2819 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2820 {
2821         unsigned long nr, a0, a1, a2, a3, ret;
2822         int r = 1;
2823
2824         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2825         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2826         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2827         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2828         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2829
2830         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2831
2832         if (!is_long_mode(vcpu)) {
2833                 nr &= 0xFFFFFFFF;
2834                 a0 &= 0xFFFFFFFF;
2835                 a1 &= 0xFFFFFFFF;
2836                 a2 &= 0xFFFFFFFF;
2837                 a3 &= 0xFFFFFFFF;
2838         }
2839
2840         switch (nr) {
2841         case KVM_HC_VAPIC_POLL_IRQ:
2842                 ret = 0;
2843                 break;
2844         case KVM_HC_MMU_OP:
2845                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2846                 break;
2847         default:
2848                 ret = -KVM_ENOSYS;
2849                 break;
2850         }
2851         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2852         ++vcpu->stat.hypercalls;
2853         return r;
2854 }
2855 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2856
2857 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2858 {
2859         char instruction[3];
2860         int ret = 0;
2861         unsigned long rip = kvm_rip_read(vcpu);
2862
2863
2864         /*
2865          * Blow out the MMU to ensure that no other VCPU has an active mapping
2866          * to ensure that the updated hypercall appears atomically across all
2867          * VCPUs.
2868          */
2869         kvm_mmu_zap_all(vcpu->kvm);
2870
2871         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2872         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2873             != X86EMUL_CONTINUE)
2874                 ret = -EFAULT;
2875
2876         return ret;
2877 }
2878
2879 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2880 {
2881         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2882 }
2883
2884 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2885 {
2886         struct descriptor_table dt = { limit, base };
2887
2888         kvm_x86_ops->set_gdt(vcpu, &dt);
2889 }
2890
2891 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2892 {
2893         struct descriptor_table dt = { limit, base };
2894
2895         kvm_x86_ops->set_idt(vcpu, &dt);
2896 }
2897
2898 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2899                    unsigned long *rflags)
2900 {
2901         kvm_lmsw(vcpu, msw);
2902         *rflags = kvm_x86_ops->get_rflags(vcpu);
2903 }
2904
2905 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2906 {
2907         unsigned long value;
2908
2909         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2910         switch (cr) {
2911         case 0:
2912                 value = vcpu->arch.cr0;
2913                 break;
2914         case 2:
2915                 value = vcpu->arch.cr2;
2916                 break;
2917         case 3:
2918                 value = vcpu->arch.cr3;
2919                 break;
2920         case 4:
2921                 value = vcpu->arch.cr4;
2922                 break;
2923         case 8:
2924                 value = kvm_get_cr8(vcpu);
2925                 break;
2926         default:
2927                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2928                 return 0;
2929         }
2930         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2931                     (u32)((u64)value >> 32), handler);
2932
2933         return value;
2934 }
2935
2936 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2937                      unsigned long *rflags)
2938 {
2939         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2940                     (u32)((u64)val >> 32), handler);
2941
2942         switch (cr) {
2943         case 0:
2944                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2945                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2946                 break;
2947         case 2:
2948                 vcpu->arch.cr2 = val;
2949                 break;
2950         case 3:
2951                 kvm_set_cr3(vcpu, val);
2952                 break;
2953         case 4:
2954                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2955                 break;
2956         case 8:
2957                 kvm_set_cr8(vcpu, val & 0xfUL);
2958                 break;
2959         default:
2960                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2961         }
2962 }
2963
2964 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2965 {
2966         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2967         int j, nent = vcpu->arch.cpuid_nent;
2968
2969         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2970         /* when no next entry is found, the current entry[i] is reselected */
2971         for (j = i + 1; ; j = (j + 1) % nent) {
2972                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2973                 if (ej->function == e->function) {
2974                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2975                         return j;
2976                 }
2977         }
2978         return 0; /* silence gcc, even though control never reaches here */
2979 }
2980
2981 /* find an entry with matching function, matching index (if needed), and that
2982  * should be read next (if it's stateful) */
2983 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2984         u32 function, u32 index)
2985 {
2986         if (e->function != function)
2987                 return 0;
2988         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2989                 return 0;
2990         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2991             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2992                 return 0;
2993         return 1;
2994 }
2995
2996 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2997                                               u32 function, u32 index)
2998 {
2999         int i;
3000         struct kvm_cpuid_entry2 *best = NULL;
3001
3002         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3003                 struct kvm_cpuid_entry2 *e;
3004
3005                 e = &vcpu->arch.cpuid_entries[i];
3006                 if (is_matching_cpuid_entry(e, function, index)) {
3007                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3008                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3009                         best = e;
3010                         break;
3011                 }
3012                 /*
3013                  * Both basic or both extended?
3014                  */
3015                 if (((e->function ^ function) & 0x80000000) == 0)
3016                         if (!best || e->function > best->function)
3017                                 best = e;
3018         }
3019         return best;
3020 }
3021
3022 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3023 {
3024         struct kvm_cpuid_entry2 *best;
3025
3026         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3027         if (best)
3028                 return best->eax & 0xff;
3029         return 36;
3030 }
3031
3032 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3033 {
3034         u32 function, index;
3035         struct kvm_cpuid_entry2 *best;
3036
3037         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3038         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3039         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3040         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3041         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3042         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3043         best = kvm_find_cpuid_entry(vcpu, function, index);
3044         if (best) {
3045                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3046                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3047                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3048                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3049         }
3050         kvm_x86_ops->skip_emulated_instruction(vcpu);
3051         KVMTRACE_5D(CPUID, vcpu, function,
3052                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3053                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3054                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3055                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3056 }
3057 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3058
3059 /*
3060  * Check if userspace requested an interrupt window, and that the
3061  * interrupt window is open.
3062  *
3063  * No need to exit to userspace if we already have an interrupt queued.
3064  */
3065 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3066                                           struct kvm_run *kvm_run)
3067 {
3068         return (!vcpu->arch.irq_summary &&
3069                 kvm_run->request_interrupt_window &&
3070                 vcpu->arch.interrupt_window_open &&
3071                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3072 }
3073
3074 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3075                               struct kvm_run *kvm_run)
3076 {
3077         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3078         kvm_run->cr8 = kvm_get_cr8(vcpu);
3079         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3080         if (irqchip_in_kernel(vcpu->kvm))
3081                 kvm_run->ready_for_interrupt_injection = 1;
3082         else
3083                 kvm_run->ready_for_interrupt_injection =
3084                                         (vcpu->arch.interrupt_window_open &&
3085                                          vcpu->arch.irq_summary == 0);
3086 }
3087
3088 static void vapic_enter(struct kvm_vcpu *vcpu)
3089 {
3090         struct kvm_lapic *apic = vcpu->arch.apic;
3091         struct page *page;
3092
3093         if (!apic || !apic->vapic_addr)
3094                 return;
3095
3096         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3097
3098         vcpu->arch.apic->vapic_page = page;
3099 }
3100
3101 static void vapic_exit(struct kvm_vcpu *vcpu)
3102 {
3103         struct kvm_lapic *apic = vcpu->arch.apic;
3104
3105         if (!apic || !apic->vapic_addr)
3106                 return;
3107
3108         down_read(&vcpu->kvm->slots_lock);
3109         kvm_release_page_dirty(apic->vapic_page);
3110         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3111         up_read(&vcpu->kvm->slots_lock);
3112 }
3113
3114 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3115 {
3116         int r;
3117
3118         if (vcpu->requests)
3119                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3120                         kvm_mmu_unload(vcpu);
3121
3122         r = kvm_mmu_reload(vcpu);
3123         if (unlikely(r))
3124                 goto out;
3125
3126         if (vcpu->requests) {
3127                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3128                         __kvm_migrate_timers(vcpu);
3129                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3130                         kvm_write_guest_time(vcpu);
3131                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3132                         kvm_mmu_sync_roots(vcpu);
3133                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3134                         kvm_x86_ops->tlb_flush(vcpu);
3135                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3136                                        &vcpu->requests)) {
3137                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3138                         r = 0;
3139                         goto out;
3140                 }
3141                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3142                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3143                         r = 0;
3144                         goto out;
3145                 }
3146         }
3147
3148         preempt_disable();
3149
3150         kvm_x86_ops->prepare_guest_switch(vcpu);
3151         kvm_load_guest_fpu(vcpu);
3152
3153         local_irq_disable();
3154
3155         if (vcpu->requests || need_resched() || signal_pending(current)) {
3156                 local_irq_enable();
3157                 preempt_enable();
3158                 r = 1;
3159                 goto out;
3160         }
3161
3162         vcpu->guest_mode = 1;
3163         /*
3164          * Make sure that guest_mode assignment won't happen after
3165          * testing the pending IRQ vector bitmap.
3166          */
3167         smp_wmb();
3168
3169         if (vcpu->arch.exception.pending)
3170                 __queue_exception(vcpu);
3171         else if (irqchip_in_kernel(vcpu->kvm))
3172                 kvm_x86_ops->inject_pending_irq(vcpu);
3173         else
3174                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3175
3176         kvm_lapic_sync_to_vapic(vcpu);
3177
3178         up_read(&vcpu->kvm->slots_lock);
3179
3180         kvm_guest_enter();
3181
3182         get_debugreg(vcpu->arch.host_dr6, 6);
3183         get_debugreg(vcpu->arch.host_dr7, 7);
3184         if (unlikely(vcpu->arch.switch_db_regs)) {
3185                 get_debugreg(vcpu->arch.host_db[0], 0);
3186                 get_debugreg(vcpu->arch.host_db[1], 1);
3187                 get_debugreg(vcpu->arch.host_db[2], 2);
3188                 get_debugreg(vcpu->arch.host_db[3], 3);
3189
3190                 set_debugreg(0, 7);
3191                 set_debugreg(vcpu->arch.eff_db[0], 0);
3192                 set_debugreg(vcpu->arch.eff_db[1], 1);
3193                 set_debugreg(vcpu->arch.eff_db[2], 2);
3194                 set_debugreg(vcpu->arch.eff_db[3], 3);
3195         }
3196
3197         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3198         kvm_x86_ops->run(vcpu, kvm_run);
3199
3200         if (unlikely(vcpu->arch.switch_db_regs)) {
3201                 set_debugreg(0, 7);
3202                 set_debugreg(vcpu->arch.host_db[0], 0);
3203                 set_debugreg(vcpu->arch.host_db[1], 1);
3204                 set_debugreg(vcpu->arch.host_db[2], 2);
3205                 set_debugreg(vcpu->arch.host_db[3], 3);
3206         }
3207         set_debugreg(vcpu->arch.host_dr6, 6);
3208         set_debugreg(vcpu->arch.host_dr7, 7);
3209
3210         vcpu->guest_mode = 0;
3211         local_irq_enable();
3212
3213         ++vcpu->stat.exits;
3214
3215         /*
3216          * We must have an instruction between local_irq_enable() and
3217          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3218          * the interrupt shadow.  The stat.exits increment will do nicely.
3219          * But we need to prevent reordering, hence this barrier():
3220          */
3221         barrier();
3222
3223         kvm_guest_exit();
3224
3225         preempt_enable();
3226
3227         down_read(&vcpu->kvm->slots_lock);
3228
3229         /*
3230          * Profile KVM exit RIPs:
3231          */
3232         if (unlikely(prof_on == KVM_PROFILING)) {
3233                 unsigned long rip = kvm_rip_read(vcpu);
3234                 profile_hit(KVM_PROFILING, (void *)rip);
3235         }
3236
3237         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3238                 vcpu->arch.exception.pending = false;
3239
3240         kvm_lapic_sync_from_vapic(vcpu);
3241
3242         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3243 out:
3244         return r;
3245 }
3246
3247
3248 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3249 {
3250         int r;
3251
3252         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3253                 pr_debug("vcpu %d received sipi with vector # %x\n",
3254                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3255                 kvm_lapic_reset(vcpu);
3256                 r = kvm_arch_vcpu_reset(vcpu);
3257                 if (r)
3258                         return r;
3259                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3260         }
3261
3262         down_read(&vcpu->kvm->slots_lock);
3263         vapic_enter(vcpu);
3264
3265         r = 1;
3266         while (r > 0) {
3267                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3268                         r = vcpu_enter_guest(vcpu, kvm_run);
3269                 else {
3270                         up_read(&vcpu->kvm->slots_lock);
3271                         kvm_vcpu_block(vcpu);
3272                         down_read(&vcpu->kvm->slots_lock);
3273                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3274                         {
3275                                 switch(vcpu->arch.mp_state) {
3276                                 case KVM_MP_STATE_HALTED:
3277                                         vcpu->arch.mp_state =
3278                                                 KVM_MP_STATE_RUNNABLE;
3279                                 case KVM_MP_STATE_RUNNABLE:
3280                                         break;
3281                                 case KVM_MP_STATE_SIPI_RECEIVED:
3282                                 default:
3283                                         r = -EINTR;
3284                                         break;
3285                                 }
3286                         }
3287                 }
3288
3289                 if (r <= 0)
3290                         break;
3291
3292                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3293                 if (kvm_cpu_has_pending_timer(vcpu))
3294                         kvm_inject_pending_timer_irqs(vcpu);
3295
3296                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3297                         r = -EINTR;
3298                         kvm_run->exit_reason = KVM_EXIT_INTR;
3299                         ++vcpu->stat.request_irq_exits;
3300                 }
3301                 if (signal_pending(current)) {
3302                         r = -EINTR;
3303                         kvm_run->exit_reason = KVM_EXIT_INTR;
3304                         ++vcpu->stat.signal_exits;
3305                 }
3306                 if (need_resched()) {
3307                         up_read(&vcpu->kvm->slots_lock);
3308                         kvm_resched(vcpu);
3309                         down_read(&vcpu->kvm->slots_lock);
3310                 }
3311         }
3312
3313         up_read(&vcpu->kvm->slots_lock);
3314         post_kvm_run_save(vcpu, kvm_run);
3315
3316         vapic_exit(vcpu);
3317
3318         return r;
3319 }
3320
3321 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3322 {
3323         int r;
3324         sigset_t sigsaved;
3325
3326         vcpu_load(vcpu);
3327
3328         if (vcpu->sigset_active)
3329                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3330
3331         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3332                 kvm_vcpu_block(vcpu);
3333                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3334                 r = -EAGAIN;
3335                 goto out;
3336         }
3337
3338         /* re-sync apic's tpr */
3339         if (!irqchip_in_kernel(vcpu->kvm))
3340                 kvm_set_cr8(vcpu, kvm_run->cr8);
3341
3342         if (vcpu->arch.pio.cur_count) {
3343                 r = complete_pio(vcpu);
3344                 if (r)
3345                         goto out;
3346         }
3347 #if CONFIG_HAS_IOMEM
3348         if (vcpu->mmio_needed) {
3349                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3350                 vcpu->mmio_read_completed = 1;
3351                 vcpu->mmio_needed = 0;
3352
3353                 down_read(&vcpu->kvm->slots_lock);
3354                 r = emulate_instruction(vcpu, kvm_run,
3355                                         vcpu->arch.mmio_fault_cr2, 0,
3356                                         EMULTYPE_NO_DECODE);
3357                 up_read(&vcpu->kvm->slots_lock);
3358                 if (r == EMULATE_DO_MMIO) {
3359                         /*
3360                          * Read-modify-write.  Back to userspace.
3361                          */
3362                         r = 0;
3363                         goto out;
3364                 }
3365         }
3366 #endif
3367         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3368                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3369                                      kvm_run->hypercall.ret);
3370
3371         r = __vcpu_run(vcpu, kvm_run);
3372
3373 out:
3374         if (vcpu->sigset_active)
3375                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3376
3377         vcpu_put(vcpu);
3378         return r;
3379 }
3380
3381 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3382 {
3383         vcpu_load(vcpu);
3384
3385         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3386         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3387         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3388         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3389         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3390         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3391         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3392         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3393 #ifdef CONFIG_X86_64
3394         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3395         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3396         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3397         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3398         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3399         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3400         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3401         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3402 #endif
3403
3404         regs->rip = kvm_rip_read(vcpu);
3405         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3406
3407         /*
3408          * Don't leak debug flags in case they were set for guest debugging
3409          */
3410         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3411                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3412
3413         vcpu_put(vcpu);
3414
3415         return 0;
3416 }
3417
3418 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3419 {
3420         vcpu_load(vcpu);
3421
3422         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3423         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3424         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3425         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3426         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3427         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3428         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3429         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3430 #ifdef CONFIG_X86_64
3431         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3432         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3433         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3434         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3435         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3436         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3437         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3438         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3439
3440 #endif
3441
3442         kvm_rip_write(vcpu, regs->rip);
3443         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3444
3445
3446         vcpu->arch.exception.pending = false;
3447
3448         vcpu_put(vcpu);
3449
3450         return 0;
3451 }
3452
3453 void kvm_get_segment(struct kvm_vcpu *vcpu,
3454                      struct kvm_segment *var, int seg)
3455 {
3456         kvm_x86_ops->get_segment(vcpu, var, seg);
3457 }
3458
3459 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3460 {
3461         struct kvm_segment cs;
3462
3463         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3464         *db = cs.db;
3465         *l = cs.l;
3466 }
3467 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3468
3469 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3470                                   struct kvm_sregs *sregs)
3471 {
3472         struct descriptor_table dt;
3473         int pending_vec;
3474
3475         vcpu_load(vcpu);
3476
3477         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3478         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3479         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3480         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3481         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3482         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3483
3484         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3485         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3486
3487         kvm_x86_ops->get_idt(vcpu, &dt);
3488         sregs->idt.limit = dt.limit;
3489         sregs->idt.base = dt.base;
3490         kvm_x86_ops->get_gdt(vcpu, &dt);
3491         sregs->gdt.limit = dt.limit;
3492         sregs->gdt.base = dt.base;
3493
3494         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3495         sregs->cr0 = vcpu->arch.cr0;
3496         sregs->cr2 = vcpu->arch.cr2;
3497         sregs->cr3 = vcpu->arch.cr3;
3498         sregs->cr4 = vcpu->arch.cr4;
3499         sregs->cr8 = kvm_get_cr8(vcpu);
3500         sregs->efer = vcpu->arch.shadow_efer;
3501         sregs->apic_base = kvm_get_apic_base(vcpu);
3502
3503         if (irqchip_in_kernel(vcpu->kvm)) {
3504                 memset(sregs->interrupt_bitmap, 0,
3505                        sizeof sregs->interrupt_bitmap);
3506                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3507                 if (pending_vec >= 0)
3508                         set_bit(pending_vec,
3509                                 (unsigned long *)sregs->interrupt_bitmap);
3510         } else
3511                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3512                        sizeof sregs->interrupt_bitmap);
3513
3514         vcpu_put(vcpu);
3515
3516         return 0;
3517 }
3518
3519 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3520                                     struct kvm_mp_state *mp_state)
3521 {
3522         vcpu_load(vcpu);
3523         mp_state->mp_state = vcpu->arch.mp_state;
3524         vcpu_put(vcpu);
3525         return 0;
3526 }
3527
3528 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3529                                     struct kvm_mp_state *mp_state)
3530 {
3531         vcpu_load(vcpu);
3532         vcpu->arch.mp_state = mp_state->mp_state;
3533         vcpu_put(vcpu);
3534         return 0;
3535 }
3536
3537 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3538                         struct kvm_segment *var, int seg)
3539 {
3540         kvm_x86_ops->set_segment(vcpu, var, seg);
3541 }
3542
3543 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3544                                    struct kvm_segment *kvm_desct)
3545 {
3546         kvm_desct->base = seg_desc->base0;
3547         kvm_desct->base |= seg_desc->base1 << 16;
3548         kvm_desct->base |= seg_desc->base2 << 24;
3549         kvm_desct->limit = seg_desc->limit0;
3550         kvm_desct->limit |= seg_desc->limit << 16;
3551         if (seg_desc->g) {
3552                 kvm_desct->limit <<= 12;
3553                 kvm_desct->limit |= 0xfff;
3554         }
3555         kvm_desct->selector = selector;
3556         kvm_desct->type = seg_desc->type;
3557         kvm_desct->present = seg_desc->p;
3558         kvm_desct->dpl = seg_desc->dpl;
3559         kvm_desct->db = seg_desc->d;
3560         kvm_desct->s = seg_desc->s;
3561         kvm_desct->l = seg_desc->l;
3562         kvm_desct->g = seg_desc->g;
3563         kvm_desct->avl = seg_desc->avl;
3564         if (!selector)
3565                 kvm_desct->unusable = 1;
3566         else
3567                 kvm_desct->unusable = 0;
3568         kvm_desct->padding = 0;
3569 }
3570
3571 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3572                                           u16 selector,
3573                                           struct descriptor_table *dtable)
3574 {
3575         if (selector & 1 << 2) {
3576                 struct kvm_segment kvm_seg;
3577
3578                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3579
3580                 if (kvm_seg.unusable)
3581                         dtable->limit = 0;
3582                 else
3583                         dtable->limit = kvm_seg.limit;
3584                 dtable->base = kvm_seg.base;
3585         }
3586         else
3587                 kvm_x86_ops->get_gdt(vcpu, dtable);
3588 }
3589
3590 /* allowed just for 8 bytes segments */
3591 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3592                                          struct desc_struct *seg_desc)
3593 {
3594         gpa_t gpa;
3595         struct descriptor_table dtable;
3596         u16 index = selector >> 3;
3597
3598         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3599
3600         if (dtable.limit < index * 8 + 7) {
3601                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3602                 return 1;
3603         }
3604         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3605         gpa += index * 8;
3606         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3607 }
3608
3609 /* allowed just for 8 bytes segments */
3610 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3611                                          struct desc_struct *seg_desc)
3612 {
3613         gpa_t gpa;
3614         struct descriptor_table dtable;
3615         u16 index = selector >> 3;
3616
3617         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3618
3619         if (dtable.limit < index * 8 + 7)
3620                 return 1;
3621         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3622         gpa += index * 8;
3623         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3624 }
3625
3626 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3627                              struct desc_struct *seg_desc)
3628 {
3629         u32 base_addr;
3630
3631         base_addr = seg_desc->base0;
3632         base_addr |= (seg_desc->base1 << 16);
3633         base_addr |= (seg_desc->base2 << 24);
3634
3635         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3636 }
3637
3638 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3639 {
3640         struct kvm_segment kvm_seg;
3641
3642         kvm_get_segment(vcpu, &kvm_seg, seg);
3643         return kvm_seg.selector;
3644 }
3645
3646 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3647                                                 u16 selector,
3648                                                 struct kvm_segment *kvm_seg)
3649 {
3650         struct desc_struct seg_desc;
3651
3652         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3653                 return 1;
3654         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3655         return 0;
3656 }
3657
3658 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3659 {
3660         struct kvm_segment segvar = {
3661                 .base = selector << 4,
3662                 .limit = 0xffff,
3663                 .selector = selector,
3664                 .type = 3,
3665                 .present = 1,
3666                 .dpl = 3,
3667                 .db = 0,
3668                 .s = 1,
3669                 .l = 0,
3670                 .g = 0,
3671                 .avl = 0,
3672                 .unusable = 0,
3673         };
3674         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3675         return 0;
3676 }
3677
3678 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3679                                 int type_bits, int seg)
3680 {
3681         struct kvm_segment kvm_seg;
3682
3683         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3684                 return kvm_load_realmode_segment(vcpu, selector, seg);
3685         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3686                 return 1;
3687         kvm_seg.type |= type_bits;
3688
3689         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3690             seg != VCPU_SREG_LDTR)
3691                 if (!kvm_seg.s)
3692                         kvm_seg.unusable = 1;
3693
3694         kvm_set_segment(vcpu, &kvm_seg, seg);
3695         return 0;
3696 }
3697
3698 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3699                                 struct tss_segment_32 *tss)
3700 {
3701         tss->cr3 = vcpu->arch.cr3;
3702         tss->eip = kvm_rip_read(vcpu);
3703         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3704         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3705         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3706         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3707         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3708         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3709         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3710         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3711         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3712         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3713         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3714         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3715         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3716         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3717         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3718         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3719         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3720 }
3721
3722 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3723                                   struct tss_segment_32 *tss)
3724 {
3725         kvm_set_cr3(vcpu, tss->cr3);
3726
3727         kvm_rip_write(vcpu, tss->eip);
3728         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3729
3730         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3731         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3732         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3733         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3734         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3735         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3736         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3737         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3738
3739         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3740                 return 1;
3741
3742         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3743                 return 1;
3744
3745         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3746                 return 1;
3747
3748         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3749                 return 1;
3750
3751         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3752                 return 1;
3753
3754         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3755                 return 1;
3756
3757         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3758                 return 1;
3759         return 0;
3760 }
3761
3762 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3763                                 struct tss_segment_16 *tss)
3764 {
3765         tss->ip = kvm_rip_read(vcpu);
3766         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3767         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3768         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3769         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3770         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3771         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3772         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3773         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3774         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3775
3776         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3777         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3778         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3779         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3780         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3781         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3782 }
3783
3784 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3785                                  struct tss_segment_16 *tss)
3786 {
3787         kvm_rip_write(vcpu, tss->ip);
3788         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3789         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3790         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3791         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3792         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3793         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3794         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3795         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3796         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3797
3798         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3799                 return 1;
3800
3801         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3802                 return 1;
3803
3804         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3805                 return 1;
3806
3807         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3808                 return 1;
3809
3810         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3811                 return 1;
3812         return 0;
3813 }
3814
3815 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3816                        u32 old_tss_base,
3817                        struct desc_struct *nseg_desc)
3818 {
3819         struct tss_segment_16 tss_segment_16;
3820         int ret = 0;
3821
3822         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3823                            sizeof tss_segment_16))
3824                 goto out;
3825
3826         save_state_to_tss16(vcpu, &tss_segment_16);
3827
3828         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3829                             sizeof tss_segment_16))
3830                 goto out;
3831
3832         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3833                            &tss_segment_16, sizeof tss_segment_16))
3834                 goto out;
3835
3836         if (load_state_from_tss16(vcpu, &tss_segment_16))
3837                 goto out;
3838
3839         ret = 1;
3840 out:
3841         return ret;
3842 }
3843
3844 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3845                        u32 old_tss_base,
3846                        struct desc_struct *nseg_desc)
3847 {
3848         struct tss_segment_32 tss_segment_32;
3849         int ret = 0;
3850
3851         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3852                            sizeof tss_segment_32))
3853                 goto out;
3854
3855         save_state_to_tss32(vcpu, &tss_segment_32);
3856
3857         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3858                             sizeof tss_segment_32))
3859                 goto out;
3860
3861         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3862                            &tss_segment_32, sizeof tss_segment_32))
3863                 goto out;
3864
3865         if (load_state_from_tss32(vcpu, &tss_segment_32))
3866                 goto out;
3867
3868         ret = 1;
3869 out:
3870         return ret;
3871 }
3872
3873 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3874 {
3875         struct kvm_segment tr_seg;
3876         struct desc_struct cseg_desc;
3877         struct desc_struct nseg_desc;
3878         int ret = 0;
3879         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3880         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3881
3882         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3883
3884         /* FIXME: Handle errors. Failure to read either TSS or their
3885          * descriptors should generate a pagefault.
3886          */
3887         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3888                 goto out;
3889
3890         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3891                 goto out;
3892
3893         if (reason != TASK_SWITCH_IRET) {
3894                 int cpl;
3895
3896                 cpl = kvm_x86_ops->get_cpl(vcpu);
3897                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3898                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3899                         return 1;
3900                 }
3901         }
3902
3903         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3904                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3905                 return 1;
3906         }
3907
3908         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3909                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3910                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3911         }
3912
3913         if (reason == TASK_SWITCH_IRET) {
3914                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3915                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3916         }
3917
3918         kvm_x86_ops->skip_emulated_instruction(vcpu);
3919
3920         if (nseg_desc.type & 8)
3921                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3922                                          &nseg_desc);
3923         else
3924                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3925                                          &nseg_desc);
3926
3927         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3928                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3929                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3930         }
3931
3932         if (reason != TASK_SWITCH_IRET) {
3933                 nseg_desc.type |= (1 << 1);
3934                 save_guest_segment_descriptor(vcpu, tss_selector,
3935                                               &nseg_desc);
3936         }
3937
3938         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3939         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3940         tr_seg.type = 11;
3941         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3942 out:
3943         return ret;
3944 }
3945 EXPORT_SYMBOL_GPL(kvm_task_switch);
3946
3947 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3948                                   struct kvm_sregs *sregs)
3949 {
3950         int mmu_reset_needed = 0;
3951         int i, pending_vec, max_bits;
3952         struct descriptor_table dt;
3953
3954         vcpu_load(vcpu);
3955
3956         dt.limit = sregs->idt.limit;
3957         dt.base = sregs->idt.base;
3958         kvm_x86_ops->set_idt(vcpu, &dt);
3959         dt.limit = sregs->gdt.limit;
3960         dt.base = sregs->gdt.base;
3961         kvm_x86_ops->set_gdt(vcpu, &dt);
3962
3963         vcpu->arch.cr2 = sregs->cr2;
3964         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3965         vcpu->arch.cr3 = sregs->cr3;
3966
3967         kvm_set_cr8(vcpu, sregs->cr8);
3968
3969         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3970         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3971         kvm_set_apic_base(vcpu, sregs->apic_base);
3972
3973         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3974
3975         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3976         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3977         vcpu->arch.cr0 = sregs->cr0;
3978
3979         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3980         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3981         if (!is_long_mode(vcpu) && is_pae(vcpu))
3982                 load_pdptrs(vcpu, vcpu->arch.cr3);
3983
3984         if (mmu_reset_needed)
3985                 kvm_mmu_reset_context(vcpu);
3986
3987         if (!irqchip_in_kernel(vcpu->kvm)) {
3988                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3989                        sizeof vcpu->arch.irq_pending);
3990                 vcpu->arch.irq_summary = 0;
3991                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3992                         if (vcpu->arch.irq_pending[i])
3993                                 __set_bit(i, &vcpu->arch.irq_summary);
3994         } else {
3995                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3996                 pending_vec = find_first_bit(
3997                         (const unsigned long *)sregs->interrupt_bitmap,
3998                         max_bits);
3999                 /* Only pending external irq is handled here */
4000                 if (pending_vec < max_bits) {
4001                         kvm_x86_ops->set_irq(vcpu, pending_vec);
4002                         pr_debug("Set back pending irq %d\n",
4003                                  pending_vec);
4004                 }
4005                 kvm_pic_clear_isr_ack(vcpu->kvm);
4006         }
4007
4008         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4009         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4010         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4011         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4012         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4013         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4014
4015         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4016         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4017
4018         /* Older userspace won't unhalt the vcpu on reset. */
4019         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4020             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4021             !(vcpu->arch.cr0 & X86_CR0_PE))
4022                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4023
4024         vcpu_put(vcpu);
4025
4026         return 0;
4027 }
4028
4029 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4030                                         struct kvm_guest_debug *dbg)
4031 {
4032         int i, r;
4033
4034         vcpu_load(vcpu);
4035
4036         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4037             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4038                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4039                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4040                 vcpu->arch.switch_db_regs =
4041                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4042         } else {
4043                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4044                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4045                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4046         }
4047
4048         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4049
4050         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4051                 kvm_queue_exception(vcpu, DB_VECTOR);
4052         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4053                 kvm_queue_exception(vcpu, BP_VECTOR);
4054
4055         vcpu_put(vcpu);
4056
4057         return r;
4058 }
4059
4060 /*
4061  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4062  * we have asm/x86/processor.h
4063  */
4064 struct fxsave {
4065         u16     cwd;
4066         u16     swd;
4067         u16     twd;
4068         u16     fop;
4069         u64     rip;
4070         u64     rdp;
4071         u32     mxcsr;
4072         u32     mxcsr_mask;
4073         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4074 #ifdef CONFIG_X86_64
4075         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4076 #else
4077         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4078 #endif
4079 };
4080
4081 /*
4082  * Translate a guest virtual address to a guest physical address.
4083  */
4084 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4085                                     struct kvm_translation *tr)
4086 {
4087         unsigned long vaddr = tr->linear_address;
4088         gpa_t gpa;
4089
4090         vcpu_load(vcpu);
4091         down_read(&vcpu->kvm->slots_lock);
4092         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4093         up_read(&vcpu->kvm->slots_lock);
4094         tr->physical_address = gpa;
4095         tr->valid = gpa != UNMAPPED_GVA;
4096         tr->writeable = 1;
4097         tr->usermode = 0;
4098         vcpu_put(vcpu);
4099
4100         return 0;
4101 }
4102
4103 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4104 {
4105         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4106
4107         vcpu_load(vcpu);
4108
4109         memcpy(fpu->fpr, fxsave->st_space, 128);
4110         fpu->fcw = fxsave->cwd;
4111         fpu->fsw = fxsave->swd;
4112         fpu->ftwx = fxsave->twd;
4113         fpu->last_opcode = fxsave->fop;
4114         fpu->last_ip = fxsave->rip;
4115         fpu->last_dp = fxsave->rdp;
4116         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4117
4118         vcpu_put(vcpu);
4119
4120         return 0;
4121 }
4122
4123 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4124 {
4125         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4126
4127         vcpu_load(vcpu);
4128
4129         memcpy(fxsave->st_space, fpu->fpr, 128);
4130         fxsave->cwd = fpu->fcw;
4131         fxsave->swd = fpu->fsw;
4132         fxsave->twd = fpu->ftwx;
4133         fxsave->fop = fpu->last_opcode;
4134         fxsave->rip = fpu->last_ip;
4135         fxsave->rdp = fpu->last_dp;
4136         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4137
4138         vcpu_put(vcpu);
4139
4140         return 0;
4141 }
4142
4143 void fx_init(struct kvm_vcpu *vcpu)
4144 {
4145         unsigned after_mxcsr_mask;
4146
4147         /*
4148          * Touch the fpu the first time in non atomic context as if
4149          * this is the first fpu instruction the exception handler
4150          * will fire before the instruction returns and it'll have to
4151          * allocate ram with GFP_KERNEL.
4152          */
4153         if (!used_math())
4154                 kvm_fx_save(&vcpu->arch.host_fx_image);
4155
4156         /* Initialize guest FPU by resetting ours and saving into guest's */
4157         preempt_disable();
4158         kvm_fx_save(&vcpu->arch.host_fx_image);
4159         kvm_fx_finit();
4160         kvm_fx_save(&vcpu->arch.guest_fx_image);
4161         kvm_fx_restore(&vcpu->arch.host_fx_image);
4162         preempt_enable();
4163
4164         vcpu->arch.cr0 |= X86_CR0_ET;
4165         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4166         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4167         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4168                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4169 }
4170 EXPORT_SYMBOL_GPL(fx_init);
4171
4172 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4173 {
4174         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4175                 return;
4176
4177         vcpu->guest_fpu_loaded = 1;
4178         kvm_fx_save(&vcpu->arch.host_fx_image);
4179         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4180 }
4181 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4182
4183 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4184 {
4185         if (!vcpu->guest_fpu_loaded)
4186                 return;
4187
4188         vcpu->guest_fpu_loaded = 0;
4189         kvm_fx_save(&vcpu->arch.guest_fx_image);
4190         kvm_fx_restore(&vcpu->arch.host_fx_image);
4191         ++vcpu->stat.fpu_reload;
4192 }
4193 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4194
4195 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4196 {
4197         if (vcpu->arch.time_page) {
4198                 kvm_release_page_dirty(vcpu->arch.time_page);
4199                 vcpu->arch.time_page = NULL;
4200         }
4201
4202         kvm_x86_ops->vcpu_free(vcpu);
4203 }
4204
4205 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4206                                                 unsigned int id)
4207 {
4208         return kvm_x86_ops->vcpu_create(kvm, id);
4209 }
4210
4211 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4212 {
4213         int r;
4214
4215         /* We do fxsave: this must be aligned. */
4216         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4217
4218         vcpu->arch.mtrr_state.have_fixed = 1;
4219         vcpu_load(vcpu);
4220         r = kvm_arch_vcpu_reset(vcpu);
4221         if (r == 0)
4222                 r = kvm_mmu_setup(vcpu);
4223         vcpu_put(vcpu);
4224         if (r < 0)
4225                 goto free_vcpu;
4226
4227         return 0;
4228 free_vcpu:
4229         kvm_x86_ops->vcpu_free(vcpu);
4230         return r;
4231 }
4232
4233 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4234 {
4235         vcpu_load(vcpu);
4236         kvm_mmu_unload(vcpu);
4237         vcpu_put(vcpu);
4238
4239         kvm_x86_ops->vcpu_free(vcpu);
4240 }
4241
4242 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4243 {
4244         vcpu->arch.nmi_pending = false;
4245         vcpu->arch.nmi_injected = false;
4246
4247         vcpu->arch.switch_db_regs = 0;
4248         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4249         vcpu->arch.dr6 = DR6_FIXED_1;
4250         vcpu->arch.dr7 = DR7_FIXED_1;
4251
4252         return kvm_x86_ops->vcpu_reset(vcpu);
4253 }
4254
4255 void kvm_arch_hardware_enable(void *garbage)
4256 {
4257         kvm_x86_ops->hardware_enable(garbage);
4258 }
4259
4260 void kvm_arch_hardware_disable(void *garbage)
4261 {
4262         kvm_x86_ops->hardware_disable(garbage);
4263 }
4264
4265 int kvm_arch_hardware_setup(void)
4266 {
4267         return kvm_x86_ops->hardware_setup();
4268 }
4269
4270 void kvm_arch_hardware_unsetup(void)
4271 {
4272         kvm_x86_ops->hardware_unsetup();
4273 }
4274
4275 void kvm_arch_check_processor_compat(void *rtn)
4276 {
4277         kvm_x86_ops->check_processor_compatibility(rtn);
4278 }
4279
4280 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4281 {
4282         struct page *page;
4283         struct kvm *kvm;
4284         int r;
4285
4286         BUG_ON(vcpu->kvm == NULL);
4287         kvm = vcpu->kvm;
4288
4289         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4290         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4291                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4292         else
4293                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4294
4295         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4296         if (!page) {
4297                 r = -ENOMEM;
4298                 goto fail;
4299         }
4300         vcpu->arch.pio_data = page_address(page);
4301
4302         r = kvm_mmu_create(vcpu);
4303         if (r < 0)
4304                 goto fail_free_pio_data;
4305
4306         if (irqchip_in_kernel(kvm)) {
4307                 r = kvm_create_lapic(vcpu);
4308                 if (r < 0)
4309                         goto fail_mmu_destroy;
4310         }
4311
4312         return 0;
4313
4314 fail_mmu_destroy:
4315         kvm_mmu_destroy(vcpu);
4316 fail_free_pio_data:
4317         free_page((unsigned long)vcpu->arch.pio_data);
4318 fail:
4319         return r;
4320 }
4321
4322 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4323 {
4324         kvm_free_lapic(vcpu);
4325         down_read(&vcpu->kvm->slots_lock);
4326         kvm_mmu_destroy(vcpu);
4327         up_read(&vcpu->kvm->slots_lock);
4328         free_page((unsigned long)vcpu->arch.pio_data);
4329 }
4330
4331 struct  kvm *kvm_arch_create_vm(void)
4332 {
4333         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4334
4335         if (!kvm)
4336                 return ERR_PTR(-ENOMEM);
4337
4338         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4339         INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4340         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4341
4342         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4343         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4344
4345         rdtscll(kvm->arch.vm_init_tsc);
4346
4347         return kvm;
4348 }
4349
4350 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4351 {
4352         vcpu_load(vcpu);
4353         kvm_mmu_unload(vcpu);
4354         vcpu_put(vcpu);
4355 }
4356
4357 static void kvm_free_vcpus(struct kvm *kvm)
4358 {
4359         unsigned int i;
4360
4361         /*
4362          * Unpin any mmu pages first.
4363          */
4364         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4365                 if (kvm->vcpus[i])
4366                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4367         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4368                 if (kvm->vcpus[i]) {
4369                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4370                         kvm->vcpus[i] = NULL;
4371                 }
4372         }
4373
4374 }
4375
4376 void kvm_arch_sync_events(struct kvm *kvm)
4377 {
4378         kvm_free_all_assigned_devices(kvm);
4379 }
4380
4381 void kvm_arch_destroy_vm(struct kvm *kvm)
4382 {
4383         kvm_iommu_unmap_guest(kvm);
4384         kvm_free_pit(kvm);
4385         kfree(kvm->arch.vpic);
4386         kfree(kvm->arch.vioapic);
4387         kvm_free_vcpus(kvm);
4388         kvm_free_physmem(kvm);
4389         if (kvm->arch.apic_access_page)
4390                 put_page(kvm->arch.apic_access_page);
4391         if (kvm->arch.ept_identity_pagetable)
4392                 put_page(kvm->arch.ept_identity_pagetable);
4393         kfree(kvm);
4394 }
4395
4396 int kvm_arch_set_memory_region(struct kvm *kvm,
4397                                 struct kvm_userspace_memory_region *mem,
4398                                 struct kvm_memory_slot old,
4399                                 int user_alloc)
4400 {
4401         int npages = mem->memory_size >> PAGE_SHIFT;
4402         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4403
4404         /*To keep backward compatibility with older userspace,
4405          *x86 needs to hanlde !user_alloc case.
4406          */
4407         if (!user_alloc) {
4408                 if (npages && !old.rmap) {
4409                         unsigned long userspace_addr;
4410
4411                         down_write(&current->mm->mmap_sem);
4412                         userspace_addr = do_mmap(NULL, 0,
4413                                                  npages * PAGE_SIZE,
4414                                                  PROT_READ | PROT_WRITE,
4415                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4416                                                  0);
4417                         up_write(&current->mm->mmap_sem);
4418
4419                         if (IS_ERR((void *)userspace_addr))
4420                                 return PTR_ERR((void *)userspace_addr);
4421
4422                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4423                         spin_lock(&kvm->mmu_lock);
4424                         memslot->userspace_addr = userspace_addr;
4425                         spin_unlock(&kvm->mmu_lock);
4426                 } else {
4427                         if (!old.user_alloc && old.rmap) {
4428                                 int ret;
4429
4430                                 down_write(&current->mm->mmap_sem);
4431                                 ret = do_munmap(current->mm, old.userspace_addr,
4432                                                 old.npages * PAGE_SIZE);
4433                                 up_write(&current->mm->mmap_sem);
4434                                 if (ret < 0)
4435                                         printk(KERN_WARNING
4436                                        "kvm_vm_ioctl_set_memory_region: "
4437                                        "failed to munmap memory\n");
4438                         }
4439                 }
4440         }
4441
4442         if (!kvm->arch.n_requested_mmu_pages) {
4443                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4444                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4445         }
4446
4447         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4448         kvm_flush_remote_tlbs(kvm);
4449
4450         return 0;
4451 }
4452
4453 void kvm_arch_flush_shadow(struct kvm *kvm)
4454 {
4455         kvm_mmu_zap_all(kvm);
4456 }
4457
4458 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4459 {
4460         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4461                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4462                || vcpu->arch.nmi_pending;
4463 }
4464
4465 static void vcpu_kick_intr(void *info)
4466 {
4467 #ifdef DEBUG
4468         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4469         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4470 #endif
4471 }
4472
4473 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4474 {
4475         int ipi_pcpu = vcpu->cpu;
4476         int cpu = get_cpu();
4477
4478         if (waitqueue_active(&vcpu->wq)) {
4479                 wake_up_interruptible(&vcpu->wq);
4480                 ++vcpu->stat.halt_wakeup;
4481         }
4482         /*
4483          * We may be called synchronously with irqs disabled in guest mode,
4484          * So need not to call smp_call_function_single() in that case.
4485          */
4486         if (vcpu->guest_mode && vcpu->cpu != cpu)
4487                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4488         put_cpu();
4489 }
4490
4491 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4492 {
4493         return kvm_x86_ops->interrupt_allowed(vcpu);
4494 }