git.oblomov.eu Git - 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         { "irq_exits", VCPU_STAT(irq_exits) },
  95         { "host_state_reload", VCPU_STAT(host_state_reload) },
  96         { "efer_reload", VCPU_STAT(efer_reload) },
  97         { "fpu_reload", VCPU_STAT(fpu_reload) },
  98         { "insn_emulation", VCPU_STAT(insn_emulation) },
  99         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 100         { "irq_injections", VCPU_STAT(irq_injections) },
 101         { "nmi_injections", VCPU_STAT(nmi_injections) },
 102         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 103         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 104         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 105         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 106         { "mmu_flooded", VM_STAT(mmu_flooded) },
 107         { "mmu_recycled", VM_STAT(mmu_recycled) },
 108         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 109         { "mmu_unsync", VM_STAT(mmu_unsync) },
 110         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 111         { "largepages", VM_STAT(lpages) },
 112         { NULL }
 113 };
 114
 115 unsigned long segment_base(u16 selector)
 116 {
 117         struct descriptor_table gdt;
 118         struct desc_struct *d;
 119         unsigned long table_base;
 120         unsigned long v;
 121
 122         if (selector == 0)
 123                 return 0;
 124
 125         asm("sgdt %0" : "=m"(gdt));
 126         table_base = gdt.base;
 127
 128         if (selector & 4) {           /* from ldt */
 129                 u16 ldt_selector;
 130
 131                 asm("sldt %0" : "=g"(ldt_selector));
 132                 table_base = segment_base(ldt_selector);
 133         }
 134         d = (struct desc_struct *)(table_base + (selector & ~7));
 135         v = d->base0 | ((unsigned long)d->base1 << 16) |
 136                 ((unsigned long)d->base2 << 24);
 137 #ifdef CONFIG_X86_64
 138         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 139                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 140 #endif
 141         return v;
 142 }
 143 EXPORT_SYMBOL_GPL(segment_base);
 144
 145 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 146 {
 147         if (irqchip_in_kernel(vcpu->kvm))
 148                 return vcpu->arch.apic_base;
 149         else
 150                 return vcpu->arch.apic_base;
 151 }
 152 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 153
 154 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 155 {
 156         /* TODO: reserve bits check */
 157         if (irqchip_in_kernel(vcpu->kvm))
 158                 kvm_lapic_set_base(vcpu, data);
 159         else
 160                 vcpu->arch.apic_base = data;
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 163
 164 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 165 {
 166         WARN_ON(vcpu->arch.exception.pending);
 167         vcpu->arch.exception.pending = true;
 168         vcpu->arch.exception.has_error_code = false;
 169         vcpu->arch.exception.nr = nr;
 170 }
 171 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 172
 173 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 174                            u32 error_code)
 175 {
 176         ++vcpu->stat.pf_guest;
 177
 178         if (vcpu->arch.exception.pending) {
 179                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 180                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 181                                         " double fault 0x%lx\n", addr);
 182                         vcpu->arch.exception.nr = DF_VECTOR;
 183                         vcpu->arch.exception.error_code = 0;
 184                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 185                         /* triple fault -> shutdown */
 186                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 187                 }
 188                 return;
 189         }
 190         vcpu->arch.cr2 = addr;
 191         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 192 }
 193
 194 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 195 {
 196         vcpu->arch.nmi_pending = 1;
 197 }
 198 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 199
 200 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 201 {
 202         WARN_ON(vcpu->arch.exception.pending);
 203         vcpu->arch.exception.pending = true;
 204         vcpu->arch.exception.has_error_code = true;
 205         vcpu->arch.exception.nr = nr;
 206         vcpu->arch.exception.error_code = error_code;
 207 }
 208 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 209
 210 static void __queue_exception(struct kvm_vcpu *vcpu)
 211 {
 212         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 213                                      vcpu->arch.exception.has_error_code,
 214                                      vcpu->arch.exception.error_code);
 215 }
 216
 217 /*
 218  * Load the pae pdptrs.  Return true is they are all valid.
 219  */
 220 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 221 {
 222         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 223         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 224         int i;
 225         int ret;
 226         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 227
 228         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 229                                   offset * sizeof(u64), sizeof(pdpte));
 230         if (ret < 0) {
 231                 ret = 0;
 232                 goto out;
 233         }
 234         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 235                 if (is_present_pte(pdpte[i]) &&
 236                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 237                         ret = 0;
 238                         goto out;
 239                 }
 240         }
 241         ret = 1;
 242
 243         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 244 out:
 245
 246         return ret;
 247 }
 248 EXPORT_SYMBOL_GPL(load_pdptrs);
 249
 250 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 251 {
 252         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 253         bool changed = true;
 254         int r;
 255
 256         if (is_long_mode(vcpu) || !is_pae(vcpu))
 257                 return false;
 258
 259         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 260         if (r < 0)
 261                 goto out;
 262         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 263 out:
 264
 265         return changed;
 266 }
 267
 268 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 269 {
 270         if (cr0 & CR0_RESERVED_BITS) {
 271                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 272                        cr0, vcpu->arch.cr0);
 273                 kvm_inject_gp(vcpu, 0);
 274                 return;
 275         }
 276
 277         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 278                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 279                 kvm_inject_gp(vcpu, 0);
 280                 return;
 281         }
 282
 283         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 284                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 285                        "and a clear PE flag\n");
 286                 kvm_inject_gp(vcpu, 0);
 287                 return;
 288         }
 289
 290         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 291 #ifdef CONFIG_X86_64
 292                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 293                         int cs_db, cs_l;
 294
 295                         if (!is_pae(vcpu)) {
 296                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 297                                        "in long mode while PAE is disabled\n");
 298                                 kvm_inject_gp(vcpu, 0);
 299                                 return;
 300                         }
 301                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 302                         if (cs_l) {
 303                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 304                                        "in long mode while CS.L == 1\n");
 305                                 kvm_inject_gp(vcpu, 0);
 306                                 return;
 307
 308                         }
 309                 } else
 310 #endif
 311                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 312                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 313                                "reserved bits\n");
 314                         kvm_inject_gp(vcpu, 0);
 315                         return;
 316                 }
 317
 318         }
 319
 320         kvm_x86_ops->set_cr0(vcpu, cr0);
 321         vcpu->arch.cr0 = cr0;
 322
 323         kvm_mmu_reset_context(vcpu);
 324         return;
 325 }
 326 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 327
 328 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 329 {
 330         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 331         KVMTRACE_1D(LMSW, vcpu,
 332                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 333                     handler);
 334 }
 335 EXPORT_SYMBOL_GPL(kvm_lmsw);
 336
 337 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 338 {
 339         unsigned long old_cr4 = vcpu->arch.cr4;
 340         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 341
 342         if (cr4 & CR4_RESERVED_BITS) {
 343                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 344                 kvm_inject_gp(vcpu, 0);
 345                 return;
 346         }
 347
 348         if (is_long_mode(vcpu)) {
 349                 if (!(cr4 & X86_CR4_PAE)) {
 350                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 351                                "in long mode\n");
 352                         kvm_inject_gp(vcpu, 0);
 353                         return;
 354                 }
 355         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 356                    && ((cr4 ^ old_cr4) & pdptr_bits)
 357                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 358                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 359                 kvm_inject_gp(vcpu, 0);
 360                 return;
 361         }
 362
 363         if (cr4 & X86_CR4_VMXE) {
 364                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 365                 kvm_inject_gp(vcpu, 0);
 366                 return;
 367         }
 368         kvm_x86_ops->set_cr4(vcpu, cr4);
 369         vcpu->arch.cr4 = cr4;
 370         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 371         kvm_mmu_reset_context(vcpu);
 372 }
 373 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 374
 375 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 376 {
 377         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 378                 kvm_mmu_sync_roots(vcpu);
 379                 kvm_mmu_flush_tlb(vcpu);
 380                 return;
 381         }
 382
 383         if (is_long_mode(vcpu)) {
 384                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 385                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 386                         kvm_inject_gp(vcpu, 0);
 387                         return;
 388                 }
 389         } else {
 390                 if (is_pae(vcpu)) {
 391                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 392                                 printk(KERN_DEBUG
 393                                        "set_cr3: #GP, reserved bits\n");
 394                                 kvm_inject_gp(vcpu, 0);
 395                                 return;
 396                         }
 397                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 398                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 399                                        "reserved bits\n");
 400                                 kvm_inject_gp(vcpu, 0);
 401                                 return;
 402                         }
 403                 }
 404                 /*
 405                  * We don't check reserved bits in nonpae mode, because
 406                  * this isn't enforced, and VMware depends on this.
 407                  */
 408         }
 409
 410         /*
 411          * Does the new cr3 value map to physical memory? (Note, we
 412          * catch an invalid cr3 even in real-mode, because it would
 413          * cause trouble later on when we turn on paging anyway.)
 414          *
 415          * A real CPU would silently accept an invalid cr3 and would
 416          * attempt to use it - with largely undefined (and often hard
 417          * to debug) behavior on the guest side.
 418          */
 419         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 420                 kvm_inject_gp(vcpu, 0);
 421         else {
 422                 vcpu->arch.cr3 = cr3;
 423                 vcpu->arch.mmu.new_cr3(vcpu);
 424         }
 425 }
 426 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 427
 428 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 429 {
 430         if (cr8 & CR8_RESERVED_BITS) {
 431                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 432                 kvm_inject_gp(vcpu, 0);
 433                 return;
 434         }
 435         if (irqchip_in_kernel(vcpu->kvm))
 436                 kvm_lapic_set_tpr(vcpu, cr8);
 437         else
 438                 vcpu->arch.cr8 = cr8;
 439 }
 440 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 441
 442 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 443 {
 444         if (irqchip_in_kernel(vcpu->kvm))
 445                 return kvm_lapic_get_cr8(vcpu);
 446         else
 447                 return vcpu->arch.cr8;
 448 }
 449 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 450
 451 static inline u32 bit(int bitno)
 452 {
 453         return 1 << (bitno & 31);
 454 }
 455
 456 /*
 457  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 458  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 459  *
 460  * This list is modified at module load time to reflect the
 461  * capabilities of the host cpu.
 462  */
 463 static u32 msrs_to_save[] = {
 464         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 465         MSR_K6_STAR,
 466 #ifdef CONFIG_X86_64
 467         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 468 #endif
 469         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 470         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 471 };
 472
 473 static unsigned num_msrs_to_save;
 474
 475 static u32 emulated_msrs[] = {
 476         MSR_IA32_MISC_ENABLE,
 477 };
 478
 479 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 480 {
 481         if (efer & efer_reserved_bits) {
 482                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 483                        efer);
 484                 kvm_inject_gp(vcpu, 0);
 485                 return;
 486         }
 487
 488         if (is_paging(vcpu)
 489             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 490                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 491                 kvm_inject_gp(vcpu, 0);
 492                 return;
 493         }
 494
 495         if (efer & EFER_FFXSR) {
 496                 struct kvm_cpuid_entry2 *feat;
 497
 498                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 499                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 500                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 501                         kvm_inject_gp(vcpu, 0);
 502                         return;
 503                 }
 504         }
 505
 506         if (efer & EFER_SVME) {
 507                 struct kvm_cpuid_entry2 *feat;
 508
 509                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 510                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 511                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 512                         kvm_inject_gp(vcpu, 0);
 513                         return;
 514                 }
 515         }
 516
 517         kvm_x86_ops->set_efer(vcpu, efer);
 518
 519         efer &= ~EFER_LMA;
 520         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 521
 522         vcpu->arch.shadow_efer = efer;
 523
 524         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 525         kvm_mmu_reset_context(vcpu);
 526 }
 527
 528 void kvm_enable_efer_bits(u64 mask)
 529 {
 530        efer_reserved_bits &= ~mask;
 531 }
 532 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 533
 534
 535 /*
 536  * Writes msr value into into the appropriate "register".
 537  * Returns 0 on success, non-0 otherwise.
 538  * Assumes vcpu_load() was already called.
 539  */
 540 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 541 {
 542         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 543 }
 544
 545 /*
 546  * Adapt set_msr() to msr_io()'s calling convention
 547  */
 548 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 549 {
 550         return kvm_set_msr(vcpu, index, *data);
 551 }
 552
 553 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 554 {
 555         static int version;
 556         struct pvclock_wall_clock wc;
 557         struct timespec now, sys, boot;
 558
 559         if (!wall_clock)
 560                 return;
 561
 562         version++;
 563
 564         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 565
 566         /*
 567          * The guest calculates current wall clock time by adding
 568          * system time (updated by kvm_write_guest_time below) to the
 569          * wall clock specified here.  guest system time equals host
 570          * system time for us, thus we must fill in host boot time here.
 571          */
 572         now = current_kernel_time();
 573         ktime_get_ts(&sys);
 574         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 575
 576         wc.sec = boot.tv_sec;
 577         wc.nsec = boot.tv_nsec;
 578         wc.version = version;
 579
 580         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 581
 582         version++;
 583         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 584 }
 585
 586 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 587 {
 588         uint32_t quotient, remainder;
 589
 590         /* Don't try to replace with do_div(), this one calculates
 591          * "(dividend << 32) / divisor" */
 592         __asm__ ( "divl %4"
 593                   : "=a" (quotient), "=d" (remainder)
 594                   : "0" (0), "1" (dividend), "r" (divisor) );
 595         return quotient;
 596 }
 597
 598 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 599 {
 600         uint64_t nsecs = 1000000000LL;
 601         int32_t  shift = 0;
 602         uint64_t tps64;
 603         uint32_t tps32;
 604
 605         tps64 = tsc_khz * 1000LL;
 606         while (tps64 > nsecs*2) {
 607                 tps64 >>= 1;
 608                 shift--;
 609         }
 610
 611         tps32 = (uint32_t)tps64;
 612         while (tps32 <= (uint32_t)nsecs) {
 613                 tps32 <<= 1;
 614                 shift++;
 615         }
 616
 617         hv_clock->tsc_shift = shift;
 618         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 619
 620         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 621                  __func__, tsc_khz, hv_clock->tsc_shift,
 622                  hv_clock->tsc_to_system_mul);
 623 }
 624
 625 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 626
 627 static void kvm_write_guest_time(struct kvm_vcpu *v)
 628 {
 629         struct timespec ts;
 630         unsigned long flags;
 631         struct kvm_vcpu_arch *vcpu = &v->arch;
 632         void *shared_kaddr;
 633         unsigned long this_tsc_khz;
 634
 635         if ((!vcpu->time_page))
 636                 return;
 637
 638         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 639         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 640                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 641                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 642         }
 643         put_cpu_var(cpu_tsc_khz);
 644
 645         /* Keep irq disabled to prevent changes to the clock */
 646         local_irq_save(flags);
 647         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 648                           &vcpu->hv_clock.tsc_timestamp);
 649         ktime_get_ts(&ts);
 650         local_irq_restore(flags);
 651
 652         /* With all the info we got, fill in the values */
 653
 654         vcpu->hv_clock.system_time = ts.tv_nsec +
 655                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 656         /*
 657          * The interface expects us to write an even number signaling that the
 658          * update is finished. Since the guest won't see the intermediate
 659          * state, we just increase by 2 at the end.
 660          */
 661         vcpu->hv_clock.version += 2;
 662
 663         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 664
 665         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 666                sizeof(vcpu->hv_clock));
 667
 668         kunmap_atomic(shared_kaddr, KM_USER0);
 669
 670         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 671 }
 672
 673 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 674 {
 675         struct kvm_vcpu_arch *vcpu = &v->arch;
 676
 677         if (!vcpu->time_page)
 678                 return 0;
 679         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 680         return 1;
 681 }
 682
 683 static bool msr_mtrr_valid(unsigned msr)
 684 {
 685         switch (msr) {
 686         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 687         case MSR_MTRRfix64K_00000:
 688         case MSR_MTRRfix16K_80000:
 689         case MSR_MTRRfix16K_A0000:
 690         case MSR_MTRRfix4K_C0000:
 691         case MSR_MTRRfix4K_C8000:
 692         case MSR_MTRRfix4K_D0000:
 693         case MSR_MTRRfix4K_D8000:
 694         case MSR_MTRRfix4K_E0000:
 695         case MSR_MTRRfix4K_E8000:
 696         case MSR_MTRRfix4K_F0000:
 697         case MSR_MTRRfix4K_F8000:
 698         case MSR_MTRRdefType:
 699         case MSR_IA32_CR_PAT:
 700                 return true;
 701         case 0x2f8:
 702                 return true;
 703         }
 704         return false;
 705 }
 706
 707 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 708 {
 709         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 710
 711         if (!msr_mtrr_valid(msr))
 712                 return 1;
 713
 714         if (msr == MSR_MTRRdefType) {
 715                 vcpu->arch.mtrr_state.def_type = data;
 716                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 717         } else if (msr == MSR_MTRRfix64K_00000)
 718                 p[0] = data;
 719         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 720                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 721         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 722                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 723         else if (msr == MSR_IA32_CR_PAT)
 724                 vcpu->arch.pat = data;
 725         else {  /* Variable MTRRs */
 726                 int idx, is_mtrr_mask;
 727                 u64 *pt;
 728
 729                 idx = (msr - 0x200) / 2;
 730                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 731                 if (!is_mtrr_mask)
 732                         pt =
 733                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 734                 else
 735                         pt =
 736                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 737                 *pt = data;
 738         }
 739
 740         kvm_mmu_reset_context(vcpu);
 741         return 0;
 742 }
 743
 744 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 745 {
 746         switch (msr) {
 747         case MSR_EFER:
 748                 set_efer(vcpu, data);
 749                 break;
 750         case MSR_IA32_MC0_STATUS:
 751                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 752                        __func__, data);
 753                 break;
 754         case MSR_IA32_MCG_STATUS:
 755                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 756                         __func__, data);
 757                 break;
 758         case MSR_IA32_MCG_CTL:
 759                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 760                         __func__, data);
 761                 break;
 762         case MSR_IA32_DEBUGCTLMSR:
 763                 if (!data) {
 764                         /* We support the non-activated case already */
 765                         break;
 766                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 767                         /* Values other than LBR and BTF are vendor-specific,
 768                            thus reserved and should throw a #GP */
 769                         return 1;
 770                 }
 771                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 772                         __func__, data);
 773                 break;
 774         case MSR_IA32_UCODE_REV:
 775         case MSR_IA32_UCODE_WRITE:
 776         case MSR_VM_HSAVE_PA:
 777                 break;
 778         case 0x200 ... 0x2ff:
 779                 return set_msr_mtrr(vcpu, msr, data);
 780         case MSR_IA32_APICBASE:
 781                 kvm_set_apic_base(vcpu, data);
 782                 break;
 783         case MSR_IA32_MISC_ENABLE:
 784                 vcpu->arch.ia32_misc_enable_msr = data;
 785                 break;
 786         case MSR_KVM_WALL_CLOCK:
 787                 vcpu->kvm->arch.wall_clock = data;
 788                 kvm_write_wall_clock(vcpu->kvm, data);
 789                 break;
 790         case MSR_KVM_SYSTEM_TIME: {
 791                 if (vcpu->arch.time_page) {
 792                         kvm_release_page_dirty(vcpu->arch.time_page);
 793                         vcpu->arch.time_page = NULL;
 794                 }
 795
 796                 vcpu->arch.time = data;
 797
 798                 /* we verify if the enable bit is set... */
 799                 if (!(data & 1))
 800                         break;
 801
 802                 /* ...but clean it before doing the actual write */
 803                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 804
 805                 vcpu->arch.time_page =
 806                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 807
 808                 if (is_error_page(vcpu->arch.time_page)) {
 809                         kvm_release_page_clean(vcpu->arch.time_page);
 810                         vcpu->arch.time_page = NULL;
 811                 }
 812
 813                 kvm_request_guest_time_update(vcpu);
 814                 break;
 815         }
 816         default:
 817                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 818                 return 1;
 819         }
 820         return 0;
 821 }
 822 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 823
 824
 825 /*
 826  * Reads an msr value (of 'msr_index') into 'pdata'.
 827  * Returns 0 on success, non-0 otherwise.
 828  * Assumes vcpu_load() was already called.
 829  */
 830 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 831 {
 832         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 833 }
 834
 835 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 836 {
 837         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 838
 839         if (!msr_mtrr_valid(msr))
 840                 return 1;
 841
 842         if (msr == MSR_MTRRdefType)
 843                 *pdata = vcpu->arch.mtrr_state.def_type +
 844                          (vcpu->arch.mtrr_state.enabled << 10);
 845         else if (msr == MSR_MTRRfix64K_00000)
 846                 *pdata = p[0];
 847         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 848                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 849         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 850                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 851         else if (msr == MSR_IA32_CR_PAT)
 852                 *pdata = vcpu->arch.pat;
 853         else {  /* Variable MTRRs */
 854                 int idx, is_mtrr_mask;
 855                 u64 *pt;
 856
 857                 idx = (msr - 0x200) / 2;
 858                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 859                 if (!is_mtrr_mask)
 860                         pt =
 861                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 862                 else
 863                         pt =
 864                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 865                 *pdata = *pt;
 866         }
 867
 868         return 0;
 869 }
 870
 871 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 872 {
 873         u64 data;
 874
 875         switch (msr) {
 876         case 0xc0010010: /* SYSCFG */
 877         case 0xc0010015: /* HWCR */
 878         case MSR_IA32_PLATFORM_ID:
 879         case MSR_IA32_P5_MC_ADDR:
 880         case MSR_IA32_P5_MC_TYPE:
 881         case MSR_IA32_MC0_CTL:
 882         case MSR_IA32_MCG_STATUS:
 883         case MSR_IA32_MCG_CAP:
 884         case MSR_IA32_MCG_CTL:
 885         case MSR_IA32_MC0_MISC:
 886         case MSR_IA32_MC0_MISC+4:
 887         case MSR_IA32_MC0_MISC+8:
 888         case MSR_IA32_MC0_MISC+12:
 889         case MSR_IA32_MC0_MISC+16:
 890         case MSR_IA32_MC0_MISC+20:
 891         case MSR_IA32_UCODE_REV:
 892         case MSR_IA32_EBL_CR_POWERON:
 893         case MSR_IA32_DEBUGCTLMSR:
 894         case MSR_IA32_LASTBRANCHFROMIP:
 895         case MSR_IA32_LASTBRANCHTOIP:
 896         case MSR_IA32_LASTINTFROMIP:
 897         case MSR_IA32_LASTINTTOIP:
 898         case MSR_VM_HSAVE_PA:
 899         case MSR_P6_EVNTSEL0:
 900         case MSR_P6_EVNTSEL1:
 901                 data = 0;
 902                 break;
 903         case MSR_MTRRcap:
 904                 data = 0x500 | KVM_NR_VAR_MTRR;
 905                 break;
 906         case 0x200 ... 0x2ff:
 907                 return get_msr_mtrr(vcpu, msr, pdata);
 908         case 0xcd: /* fsb frequency */
 909                 data = 3;
 910                 break;
 911         case MSR_IA32_APICBASE:
 912                 data = kvm_get_apic_base(vcpu);
 913                 break;
 914         case MSR_IA32_MISC_ENABLE:
 915                 data = vcpu->arch.ia32_misc_enable_msr;
 916                 break;
 917         case MSR_IA32_PERF_STATUS:
 918                 /* TSC increment by tick */
 919                 data = 1000ULL;
 920                 /* CPU multiplier */
 921                 data |= (((uint64_t)4ULL) << 40);
 922                 break;
 923         case MSR_EFER:
 924                 data = vcpu->arch.shadow_efer;
 925                 break;
 926         case MSR_KVM_WALL_CLOCK:
 927                 data = vcpu->kvm->arch.wall_clock;
 928                 break;
 929         case MSR_KVM_SYSTEM_TIME:
 930                 data = vcpu->arch.time;
 931                 break;
 932         default:
 933                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 934                 return 1;
 935         }
 936         *pdata = data;
 937         return 0;
 938 }
 939 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 940
 941 /*
 942  * Read or write a bunch of msrs. All parameters are kernel addresses.
 943  *
 944  * @return number of msrs set successfully.
 945  */
 946 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 947                     struct kvm_msr_entry *entries,
 948                     int (*do_msr)(struct kvm_vcpu *vcpu,
 949                                   unsigned index, u64 *data))
 950 {
 951         int i;
 952
 953         vcpu_load(vcpu);
 954
 955         down_read(&vcpu->kvm->slots_lock);
 956         for (i = 0; i < msrs->nmsrs; ++i)
 957                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 958                         break;
 959         up_read(&vcpu->kvm->slots_lock);
 960
 961         vcpu_put(vcpu);
 962
 963         return i;
 964 }
 965
 966 /*
 967  * Read or write a bunch of msrs. Parameters are user addresses.
 968  *
 969  * @return number of msrs set successfully.
 970  */
 971 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 972                   int (*do_msr)(struct kvm_vcpu *vcpu,
 973                                 unsigned index, u64 *data),
 974                   int writeback)
 975 {
 976         struct kvm_msrs msrs;
 977         struct kvm_msr_entry *entries;
 978         int r, n;
 979         unsigned size;
 980
 981         r = -EFAULT;
 982         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 983                 goto out;
 984
 985         r = -E2BIG;
 986         if (msrs.nmsrs >= MAX_IO_MSRS)
 987                 goto out;
 988
 989         r = -ENOMEM;
 990         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 991         entries = vmalloc(size);
 992         if (!entries)
 993                 goto out;
 994
 995         r = -EFAULT;
 996         if (copy_from_user(entries, user_msrs->entries, size))
 997                 goto out_free;
 998
 999         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1000         if (r < 0)
1001                 goto out_free;
1002
1003         r = -EFAULT;
1004         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1005                 goto out_free;
1006
1007         r = n;
1008
1009 out_free:
1010         vfree(entries);
1011 out:
1012         return r;
1013 }
1014
1015 int kvm_dev_ioctl_check_extension(long ext)
1016 {
1017         int r;
1018
1019         switch (ext) {
1020         case KVM_CAP_IRQCHIP:
1021         case KVM_CAP_HLT:
1022         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1023         case KVM_CAP_SET_TSS_ADDR:
1024         case KVM_CAP_EXT_CPUID:
1025         case KVM_CAP_CLOCKSOURCE:
1026         case KVM_CAP_PIT:
1027         case KVM_CAP_NOP_IO_DELAY:
1028         case KVM_CAP_MP_STATE:
1029         case KVM_CAP_SYNC_MMU:
1030         case KVM_CAP_REINJECT_CONTROL:
1031         case KVM_CAP_IRQ_INJECT_STATUS:
1032         case KVM_CAP_ASSIGN_DEV_IRQ:
1033                 r = 1;
1034                 break;
1035         case KVM_CAP_COALESCED_MMIO:
1036                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1037                 break;
1038         case KVM_CAP_VAPIC:
1039                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1040                 break;
1041         case KVM_CAP_NR_VCPUS:
1042                 r = KVM_MAX_VCPUS;
1043                 break;
1044         case KVM_CAP_NR_MEMSLOTS:
1045                 r = KVM_MEMORY_SLOTS;
1046                 break;
1047         case KVM_CAP_PV_MMU:
1048                 r = !tdp_enabled;
1049                 break;
1050         case KVM_CAP_IOMMU:
1051                 r = iommu_found();
1052                 break;
1053         default:
1054                 r = 0;
1055                 break;
1056         }
1057         return r;
1058
1059 }
1060
1061 long kvm_arch_dev_ioctl(struct file *filp,
1062                         unsigned int ioctl, unsigned long arg)
1063 {
1064         void __user *argp = (void __user *)arg;
1065         long r;
1066
1067         switch (ioctl) {
1068         case KVM_GET_MSR_INDEX_LIST: {
1069                 struct kvm_msr_list __user *user_msr_list = argp;
1070                 struct kvm_msr_list msr_list;
1071                 unsigned n;
1072
1073                 r = -EFAULT;
1074                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1075                         goto out;
1076                 n = msr_list.nmsrs;
1077                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1078                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1079                         goto out;
1080                 r = -E2BIG;
1081                 if (n < num_msrs_to_save)
1082                         goto out;
1083                 r = -EFAULT;
1084                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1085                                  num_msrs_to_save * sizeof(u32)))
1086                         goto out;
1087                 if (copy_to_user(user_msr_list->indices
1088                                  + num_msrs_to_save * sizeof(u32),
1089                                  &emulated_msrs,
1090                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1091                         goto out;
1092                 r = 0;
1093                 break;
1094         }
1095         case KVM_GET_SUPPORTED_CPUID: {
1096                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1097                 struct kvm_cpuid2 cpuid;
1098
1099                 r = -EFAULT;
1100                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1101                         goto out;
1102                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1103                                                       cpuid_arg->entries);
1104                 if (r)
1105                         goto out;
1106
1107                 r = -EFAULT;
1108                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1109                         goto out;
1110                 r = 0;
1111                 break;
1112         }
1113         default:
1114                 r = -EINVAL;
1115         }
1116 out:
1117         return r;
1118 }
1119
1120 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1121 {
1122         kvm_x86_ops->vcpu_load(vcpu, cpu);
1123         kvm_request_guest_time_update(vcpu);
1124 }
1125
1126 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1127 {
1128         kvm_x86_ops->vcpu_put(vcpu);
1129         kvm_put_guest_fpu(vcpu);
1130 }
1131
1132 static int is_efer_nx(void)
1133 {
1134         unsigned long long efer = 0;
1135
1136         rdmsrl_safe(MSR_EFER, &efer);
1137         return efer & EFER_NX;
1138 }
1139
1140 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1141 {
1142         int i;
1143         struct kvm_cpuid_entry2 *e, *entry;
1144
1145         entry = NULL;
1146         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1147                 e = &vcpu->arch.cpuid_entries[i];
1148                 if (e->function == 0x80000001) {
1149                         entry = e;
1150                         break;
1151                 }
1152         }
1153         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1154                 entry->edx &= ~(1 << 20);
1155                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1156         }
1157 }
1158
1159 /* when an old userspace process fills a new kernel module */
1160 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1161                                     struct kvm_cpuid *cpuid,
1162                                     struct kvm_cpuid_entry __user *entries)
1163 {
1164         int r, i;
1165         struct kvm_cpuid_entry *cpuid_entries;
1166
1167         r = -E2BIG;
1168         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1169                 goto out;
1170         r = -ENOMEM;
1171         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1172         if (!cpuid_entries)
1173                 goto out;
1174         r = -EFAULT;
1175         if (copy_from_user(cpuid_entries, entries,
1176                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1177                 goto out_free;
1178         for (i = 0; i < cpuid->nent; i++) {
1179                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1180                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1181                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1182                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1183                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1184                 vcpu->arch.cpuid_entries[i].index = 0;
1185                 vcpu->arch.cpuid_entries[i].flags = 0;
1186                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1187                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1188                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1189         }
1190         vcpu->arch.cpuid_nent = cpuid->nent;
1191         cpuid_fix_nx_cap(vcpu);
1192         r = 0;
1193
1194 out_free:
1195         vfree(cpuid_entries);
1196 out:
1197         return r;
1198 }
1199
1200 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1201                                      struct kvm_cpuid2 *cpuid,
1202                                      struct kvm_cpuid_entry2 __user *entries)
1203 {
1204         int r;
1205
1206         r = -E2BIG;
1207         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1208                 goto out;
1209         r = -EFAULT;
1210         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1211                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1212                 goto out;
1213         vcpu->arch.cpuid_nent = cpuid->nent;
1214         return 0;
1215
1216 out:
1217         return r;
1218 }
1219
1220 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1221                                      struct kvm_cpuid2 *cpuid,
1222                                      struct kvm_cpuid_entry2 __user *entries)
1223 {
1224         int r;
1225
1226         r = -E2BIG;
1227         if (cpuid->nent < vcpu->arch.cpuid_nent)
1228                 goto out;
1229         r = -EFAULT;
1230         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1231                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1232                 goto out;
1233         return 0;
1234
1235 out:
1236         cpuid->nent = vcpu->arch.cpuid_nent;
1237         return r;
1238 }
1239
1240 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1241                            u32 index)
1242 {
1243         entry->function = function;
1244         entry->index = index;
1245         cpuid_count(entry->function, entry->index,
1246                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1247         entry->flags = 0;
1248 }
1249
1250 #define F(x) bit(X86_FEATURE_##x)
1251
1252 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1253                          u32 index, int *nent, int maxnent)
1254 {
1255         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1256 #ifdef CONFIG_X86_64
1257         unsigned f_lm = F(LM);
1258 #else
1259         unsigned f_lm = 0;
1260 #endif
1261
1262         /* cpuid 1.edx */
1263         const u32 kvm_supported_word0_x86_features =
1264                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1265                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1266                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1267                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1268                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1269                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1270                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1271                 0 /* HTT, TM, Reserved, PBE */;
1272         /* cpuid 0x80000001.edx */
1273         const u32 kvm_supported_word1_x86_features =
1274                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1275                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1276                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1277                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1278                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1279                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1280                 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1281                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1282         /* cpuid 1.ecx */
1283         const u32 kvm_supported_word4_x86_features =
1284                 F(XMM3) | F(CX16);
1285         /* cpuid 0x80000001.ecx */
1286         const u32 kvm_supported_word6_x86_features =
1287                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1288                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1289                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1290                 0 /* SKINIT */ | 0 /* WDT */;
1291
1292         /* all calls to cpuid_count() should be made on the same cpu */
1293         get_cpu();
1294         do_cpuid_1_ent(entry, function, index);
1295         ++*nent;
1296
1297         switch (function) {
1298         case 0:
1299                 entry->eax = min(entry->eax, (u32)0xb);
1300                 break;
1301         case 1:
1302                 entry->edx &= kvm_supported_word0_x86_features;
1303                 entry->ecx &= kvm_supported_word4_x86_features;
1304                 break;
1305         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1306          * may return different values. This forces us to get_cpu() before
1307          * issuing the first command, and also to emulate this annoying behavior
1308          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1309         case 2: {
1310                 int t, times = entry->eax & 0xff;
1311
1312                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1313                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1314                 for (t = 1; t < times && *nent < maxnent; ++t) {
1315                         do_cpuid_1_ent(&entry[t], function, 0);
1316                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1317                         ++*nent;
1318                 }
1319                 break;
1320         }
1321         /* function 4 and 0xb have additional index. */
1322         case 4: {
1323                 int i, cache_type;
1324
1325                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1326                 /* read more entries until cache_type is zero */
1327                 for (i = 1; *nent < maxnent; ++i) {
1328                         cache_type = entry[i - 1].eax & 0x1f;
1329                         if (!cache_type)
1330                                 break;
1331                         do_cpuid_1_ent(&entry[i], function, i);
1332                         entry[i].flags |=
1333                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1334                         ++*nent;
1335                 }
1336                 break;
1337         }
1338         case 0xb: {
1339                 int i, level_type;
1340
1341                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1342                 /* read more entries until level_type is zero */
1343                 for (i = 1; *nent < maxnent; ++i) {
1344                         level_type = entry[i - 1].ecx & 0xff00;
1345                         if (!level_type)
1346                                 break;
1347                         do_cpuid_1_ent(&entry[i], function, i);
1348                         entry[i].flags |=
1349                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1350                         ++*nent;
1351                 }
1352                 break;
1353         }
1354         case 0x80000000:
1355                 entry->eax = min(entry->eax, 0x8000001a);
1356                 break;
1357         case 0x80000001:
1358                 entry->edx &= kvm_supported_word1_x86_features;
1359                 entry->ecx &= kvm_supported_word6_x86_features;
1360                 break;
1361         }
1362         put_cpu();
1363 }
1364
1365 #undef F
1366
1367 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1368                                      struct kvm_cpuid_entry2 __user *entries)
1369 {
1370         struct kvm_cpuid_entry2 *cpuid_entries;
1371         int limit, nent = 0, r = -E2BIG;
1372         u32 func;
1373
1374         if (cpuid->nent < 1)
1375                 goto out;
1376         r = -ENOMEM;
1377         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1378         if (!cpuid_entries)
1379                 goto out;
1380
1381         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1382         limit = cpuid_entries[0].eax;
1383         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1384                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1385                              &nent, cpuid->nent);
1386         r = -E2BIG;
1387         if (nent >= cpuid->nent)
1388                 goto out_free;
1389
1390         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1391         limit = cpuid_entries[nent - 1].eax;
1392         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1393                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1394                              &nent, cpuid->nent);
1395         r = -EFAULT;
1396         if (copy_to_user(entries, cpuid_entries,
1397                          nent * sizeof(struct kvm_cpuid_entry2)))
1398                 goto out_free;
1399         cpuid->nent = nent;
1400         r = 0;
1401
1402 out_free:
1403         vfree(cpuid_entries);
1404 out:
1405         return r;
1406 }
1407
1408 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1409                                     struct kvm_lapic_state *s)
1410 {
1411         vcpu_load(vcpu);
1412         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1413         vcpu_put(vcpu);
1414
1415         return 0;
1416 }
1417
1418 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1419                                     struct kvm_lapic_state *s)
1420 {
1421         vcpu_load(vcpu);
1422         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1423         kvm_apic_post_state_restore(vcpu);
1424         vcpu_put(vcpu);
1425
1426         return 0;
1427 }
1428
1429 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1430                                     struct kvm_interrupt *irq)
1431 {
1432         if (irq->irq < 0 || irq->irq >= 256)
1433                 return -EINVAL;
1434         if (irqchip_in_kernel(vcpu->kvm))
1435                 return -ENXIO;
1436         vcpu_load(vcpu);
1437
1438         set_bit(irq->irq, vcpu->arch.irq_pending);
1439         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1440
1441         vcpu_put(vcpu);
1442
1443         return 0;
1444 }
1445
1446 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1447 {
1448         vcpu_load(vcpu);
1449         kvm_inject_nmi(vcpu);
1450         vcpu_put(vcpu);
1451
1452         return 0;
1453 }
1454
1455 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1456                                            struct kvm_tpr_access_ctl *tac)
1457 {
1458         if (tac->flags)
1459                 return -EINVAL;
1460         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1461         return 0;
1462 }
1463
1464 long kvm_arch_vcpu_ioctl(struct file *filp,
1465                          unsigned int ioctl, unsigned long arg)
1466 {
1467         struct kvm_vcpu *vcpu = filp->private_data;
1468         void __user *argp = (void __user *)arg;
1469         int r;
1470         struct kvm_lapic_state *lapic = NULL;
1471
1472         switch (ioctl) {
1473         case KVM_GET_LAPIC: {
1474                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1475
1476                 r = -ENOMEM;
1477                 if (!lapic)
1478                         goto out;
1479                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1480                 if (r)
1481                         goto out;
1482                 r = -EFAULT;
1483                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1484                         goto out;
1485                 r = 0;
1486                 break;
1487         }
1488         case KVM_SET_LAPIC: {
1489                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1490                 r = -ENOMEM;
1491                 if (!lapic)
1492                         goto out;
1493                 r = -EFAULT;
1494                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1495                         goto out;
1496                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1497                 if (r)
1498                         goto out;
1499                 r = 0;
1500                 break;
1501         }
1502         case KVM_INTERRUPT: {
1503                 struct kvm_interrupt irq;
1504
1505                 r = -EFAULT;
1506                 if (copy_from_user(&irq, argp, sizeof irq))
1507                         goto out;
1508                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1509                 if (r)
1510                         goto out;
1511                 r = 0;
1512                 break;
1513         }
1514         case KVM_NMI: {
1515                 r = kvm_vcpu_ioctl_nmi(vcpu);
1516                 if (r)
1517                         goto out;
1518                 r = 0;
1519                 break;
1520         }
1521         case KVM_SET_CPUID: {
1522                 struct kvm_cpuid __user *cpuid_arg = argp;
1523                 struct kvm_cpuid cpuid;
1524
1525                 r = -EFAULT;
1526                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1527                         goto out;
1528                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1529                 if (r)
1530                         goto out;
1531                 break;
1532         }
1533         case KVM_SET_CPUID2: {
1534                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1535                 struct kvm_cpuid2 cpuid;
1536
1537                 r = -EFAULT;
1538                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1539                         goto out;
1540                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1541                                               cpuid_arg->entries);
1542                 if (r)
1543                         goto out;
1544                 break;
1545         }
1546         case KVM_GET_CPUID2: {
1547                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1548                 struct kvm_cpuid2 cpuid;
1549
1550                 r = -EFAULT;
1551                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1552                         goto out;
1553                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1554                                               cpuid_arg->entries);
1555                 if (r)
1556                         goto out;
1557                 r = -EFAULT;
1558                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1559                         goto out;
1560                 r = 0;
1561                 break;
1562         }
1563         case KVM_GET_MSRS:
1564                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1565                 break;
1566         case KVM_SET_MSRS:
1567                 r = msr_io(vcpu, argp, do_set_msr, 0);
1568                 break;
1569         case KVM_TPR_ACCESS_REPORTING: {
1570                 struct kvm_tpr_access_ctl tac;
1571
1572                 r = -EFAULT;
1573                 if (copy_from_user(&tac, argp, sizeof tac))
1574                         goto out;
1575                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1576                 if (r)
1577                         goto out;
1578                 r = -EFAULT;
1579                 if (copy_to_user(argp, &tac, sizeof tac))
1580                         goto out;
1581                 r = 0;
1582                 break;
1583         };
1584         case KVM_SET_VAPIC_ADDR: {
1585                 struct kvm_vapic_addr va;
1586
1587                 r = -EINVAL;
1588                 if (!irqchip_in_kernel(vcpu->kvm))
1589                         goto out;
1590                 r = -EFAULT;
1591                 if (copy_from_user(&va, argp, sizeof va))
1592                         goto out;
1593                 r = 0;
1594                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1595                 break;
1596         }
1597         default:
1598                 r = -EINVAL;
1599         }
1600 out:
1601         kfree(lapic);
1602         return r;
1603 }
1604
1605 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1606 {
1607         int ret;
1608
1609         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1610                 return -1;
1611         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1612         return ret;
1613 }
1614
1615 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1616                                           u32 kvm_nr_mmu_pages)
1617 {
1618         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1619                 return -EINVAL;
1620
1621         down_write(&kvm->slots_lock);
1622
1623         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1624         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1625
1626         up_write(&kvm->slots_lock);
1627         return 0;
1628 }
1629
1630 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1631 {
1632         return kvm->arch.n_alloc_mmu_pages;
1633 }
1634
1635 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1636 {
1637         int i;
1638         struct kvm_mem_alias *alias;
1639
1640         for (i = 0; i < kvm->arch.naliases; ++i) {
1641                 alias = &kvm->arch.aliases[i];
1642                 if (gfn >= alias->base_gfn
1643                     && gfn < alias->base_gfn + alias->npages)
1644                         return alias->target_gfn + gfn - alias->base_gfn;
1645         }
1646         return gfn;
1647 }
1648
1649 /*
1650  * Set a new alias region.  Aliases map a portion of physical memory into
1651  * another portion.  This is useful for memory windows, for example the PC
1652  * VGA region.
1653  */
1654 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1655                                          struct kvm_memory_alias *alias)
1656 {
1657         int r, n;
1658         struct kvm_mem_alias *p;
1659
1660         r = -EINVAL;
1661         /* General sanity checks */
1662         if (alias->memory_size & (PAGE_SIZE - 1))
1663                 goto out;
1664         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1665                 goto out;
1666         if (alias->slot >= KVM_ALIAS_SLOTS)
1667                 goto out;
1668         if (alias->guest_phys_addr + alias->memory_size
1669             < alias->guest_phys_addr)
1670                 goto out;
1671         if (alias->target_phys_addr + alias->memory_size
1672             < alias->target_phys_addr)
1673                 goto out;
1674
1675         down_write(&kvm->slots_lock);
1676         spin_lock(&kvm->mmu_lock);
1677
1678         p = &kvm->arch.aliases[alias->slot];
1679         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1680         p->npages = alias->memory_size >> PAGE_SHIFT;
1681         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1682
1683         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1684                 if (kvm->arch.aliases[n - 1].npages)
1685                         break;
1686         kvm->arch.naliases = n;
1687
1688         spin_unlock(&kvm->mmu_lock);
1689         kvm_mmu_zap_all(kvm);
1690
1691         up_write(&kvm->slots_lock);
1692
1693         return 0;
1694
1695 out:
1696         return r;
1697 }
1698
1699 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1700 {
1701         int r;
1702
1703         r = 0;
1704         switch (chip->chip_id) {
1705         case KVM_IRQCHIP_PIC_MASTER:
1706                 memcpy(&chip->chip.pic,
1707                         &pic_irqchip(kvm)->pics[0],
1708                         sizeof(struct kvm_pic_state));
1709                 break;
1710         case KVM_IRQCHIP_PIC_SLAVE:
1711                 memcpy(&chip->chip.pic,
1712                         &pic_irqchip(kvm)->pics[1],
1713                         sizeof(struct kvm_pic_state));
1714                 break;
1715         case KVM_IRQCHIP_IOAPIC:
1716                 memcpy(&chip->chip.ioapic,
1717                         ioapic_irqchip(kvm),
1718                         sizeof(struct kvm_ioapic_state));
1719                 break;
1720         default:
1721                 r = -EINVAL;
1722                 break;
1723         }
1724         return r;
1725 }
1726
1727 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1728 {
1729         int r;
1730
1731         r = 0;
1732         switch (chip->chip_id) {
1733         case KVM_IRQCHIP_PIC_MASTER:
1734                 memcpy(&pic_irqchip(kvm)->pics[0],
1735                         &chip->chip.pic,
1736                         sizeof(struct kvm_pic_state));
1737                 break;
1738         case KVM_IRQCHIP_PIC_SLAVE:
1739                 memcpy(&pic_irqchip(kvm)->pics[1],
1740                         &chip->chip.pic,
1741                         sizeof(struct kvm_pic_state));
1742                 break;
1743         case KVM_IRQCHIP_IOAPIC:
1744                 memcpy(ioapic_irqchip(kvm),
1745                         &chip->chip.ioapic,
1746                         sizeof(struct kvm_ioapic_state));
1747                 break;
1748         default:
1749                 r = -EINVAL;
1750                 break;
1751         }
1752         kvm_pic_update_irq(pic_irqchip(kvm));
1753         return r;
1754 }
1755
1756 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1757 {
1758         int r = 0;
1759
1760         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1761         return r;
1762 }
1763
1764 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1765 {
1766         int r = 0;
1767
1768         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1769         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1770         return r;
1771 }
1772
1773 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1774                                  struct kvm_reinject_control *control)
1775 {
1776         if (!kvm->arch.vpit)
1777                 return -ENXIO;
1778         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1779         return 0;
1780 }
1781
1782 /*
1783  * Get (and clear) the dirty memory log for a memory slot.
1784  */
1785 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1786                                       struct kvm_dirty_log *log)
1787 {
1788         int r;
1789         int n;
1790         struct kvm_memory_slot *memslot;
1791         int is_dirty = 0;
1792
1793         down_write(&kvm->slots_lock);
1794
1795         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1796         if (r)
1797                 goto out;
1798
1799         /* If nothing is dirty, don't bother messing with page tables. */
1800         if (is_dirty) {
1801                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1802                 kvm_flush_remote_tlbs(kvm);
1803                 memslot = &kvm->memslots[log->slot];
1804                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1805                 memset(memslot->dirty_bitmap, 0, n);
1806         }
1807         r = 0;
1808 out:
1809         up_write(&kvm->slots_lock);
1810         return r;
1811 }
1812
1813 long kvm_arch_vm_ioctl(struct file *filp,
1814                        unsigned int ioctl, unsigned long arg)
1815 {
1816         struct kvm *kvm = filp->private_data;
1817         void __user *argp = (void __user *)arg;
1818         int r = -EINVAL;
1819         /*
1820          * This union makes it completely explicit to gcc-3.x
1821          * that these two variables' stack usage should be
1822          * combined, not added together.
1823          */
1824         union {
1825                 struct kvm_pit_state ps;
1826                 struct kvm_memory_alias alias;
1827         } u;
1828
1829         switch (ioctl) {
1830         case KVM_SET_TSS_ADDR:
1831                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1832                 if (r < 0)
1833                         goto out;
1834                 break;
1835         case KVM_SET_MEMORY_REGION: {
1836                 struct kvm_memory_region kvm_mem;
1837                 struct kvm_userspace_memory_region kvm_userspace_mem;
1838
1839                 r = -EFAULT;
1840                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1841                         goto out;
1842                 kvm_userspace_mem.slot = kvm_mem.slot;
1843                 kvm_userspace_mem.flags = kvm_mem.flags;
1844                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1845                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1846                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1847                 if (r)
1848                         goto out;
1849                 break;
1850         }
1851         case KVM_SET_NR_MMU_PAGES:
1852                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1853                 if (r)
1854                         goto out;
1855                 break;
1856         case KVM_GET_NR_MMU_PAGES:
1857                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1858                 break;
1859         case KVM_SET_MEMORY_ALIAS:
1860                 r = -EFAULT;
1861                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1862                         goto out;
1863                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1864                 if (r)
1865                         goto out;
1866                 break;
1867         case KVM_CREATE_IRQCHIP:
1868                 r = -ENOMEM;
1869                 kvm->arch.vpic = kvm_create_pic(kvm);
1870                 if (kvm->arch.vpic) {
1871                         r = kvm_ioapic_init(kvm);
1872                         if (r) {
1873                                 kfree(kvm->arch.vpic);
1874                                 kvm->arch.vpic = NULL;
1875                                 goto out;
1876                         }
1877                 } else
1878                         goto out;
1879                 r = kvm_setup_default_irq_routing(kvm);
1880                 if (r) {
1881                         kfree(kvm->arch.vpic);
1882                         kfree(kvm->arch.vioapic);
1883                         goto out;
1884                 }
1885                 break;
1886         case KVM_CREATE_PIT:
1887                 mutex_lock(&kvm->lock);
1888                 r = -EEXIST;
1889                 if (kvm->arch.vpit)
1890                         goto create_pit_unlock;
1891                 r = -ENOMEM;
1892                 kvm->arch.vpit = kvm_create_pit(kvm);
1893                 if (kvm->arch.vpit)
1894                         r = 0;
1895         create_pit_unlock:
1896                 mutex_unlock(&kvm->lock);
1897                 break;
1898         case KVM_IRQ_LINE_STATUS:
1899         case KVM_IRQ_LINE: {
1900                 struct kvm_irq_level irq_event;
1901
1902                 r = -EFAULT;
1903                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1904                         goto out;
1905                 if (irqchip_in_kernel(kvm)) {
1906                         __s32 status;
1907                         mutex_lock(&kvm->lock);
1908                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1909                                         irq_event.irq, irq_event.level);
1910                         mutex_unlock(&kvm->lock);
1911                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1912                                 irq_event.status = status;
1913                                 if (copy_to_user(argp, &irq_event,
1914                                                         sizeof irq_event))
1915                                         goto out;
1916                         }
1917                         r = 0;
1918                 }
1919                 break;
1920         }
1921         case KVM_GET_IRQCHIP: {
1922                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1923                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1924
1925                 r = -ENOMEM;
1926                 if (!chip)
1927                         goto out;
1928                 r = -EFAULT;
1929                 if (copy_from_user(chip, argp, sizeof *chip))
1930                         goto get_irqchip_out;
1931                 r = -ENXIO;
1932                 if (!irqchip_in_kernel(kvm))
1933                         goto get_irqchip_out;
1934                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1935                 if (r)
1936                         goto get_irqchip_out;
1937                 r = -EFAULT;
1938                 if (copy_to_user(argp, chip, sizeof *chip))
1939                         goto get_irqchip_out;
1940                 r = 0;
1941         get_irqchip_out:
1942                 kfree(chip);
1943                 if (r)
1944                         goto out;
1945                 break;
1946         }
1947         case KVM_SET_IRQCHIP: {
1948                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1949                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1950
1951                 r = -ENOMEM;
1952                 if (!chip)
1953                         goto out;
1954                 r = -EFAULT;
1955                 if (copy_from_user(chip, argp, sizeof *chip))
1956                         goto set_irqchip_out;
1957                 r = -ENXIO;
1958                 if (!irqchip_in_kernel(kvm))
1959                         goto set_irqchip_out;
1960                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1961                 if (r)
1962                         goto set_irqchip_out;
1963                 r = 0;
1964         set_irqchip_out:
1965                 kfree(chip);
1966                 if (r)
1967                         goto out;
1968                 break;
1969         }
1970         case KVM_GET_PIT: {
1971                 r = -EFAULT;
1972                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1973                         goto out;
1974                 r = -ENXIO;
1975                 if (!kvm->arch.vpit)
1976                         goto out;
1977                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1978                 if (r)
1979                         goto out;
1980                 r = -EFAULT;
1981                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1982                         goto out;
1983                 r = 0;
1984                 break;
1985         }
1986         case KVM_SET_PIT: {
1987                 r = -EFAULT;
1988                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1989                         goto out;
1990                 r = -ENXIO;
1991                 if (!kvm->arch.vpit)
1992                         goto out;
1993                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1994                 if (r)
1995                         goto out;
1996                 r = 0;
1997                 break;
1998         }
1999         case KVM_REINJECT_CONTROL: {
2000                 struct kvm_reinject_control control;
2001                 r =  -EFAULT;
2002                 if (copy_from_user(&control, argp, sizeof(control)))
2003                         goto out;
2004                 r = kvm_vm_ioctl_reinject(kvm, &control);
2005                 if (r)
2006                         goto out;
2007                 r = 0;
2008                 break;
2009         }
2010         default:
2011                 ;
2012         }
2013 out:
2014         return r;
2015 }
2016
2017 static void kvm_init_msr_list(void)
2018 {
2019         u32 dummy[2];
2020         unsigned i, j;
2021
2022         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2023                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2024                         continue;
2025                 if (j < i)
2026                         msrs_to_save[j] = msrs_to_save[i];
2027                 j++;
2028         }
2029         num_msrs_to_save = j;
2030 }
2031
2032 /*
2033  * Only apic need an MMIO device hook, so shortcut now..
2034  */
2035 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2036                                                 gpa_t addr, int len,
2037                                                 int is_write)
2038 {
2039         struct kvm_io_device *dev;
2040
2041         if (vcpu->arch.apic) {
2042                 dev = &vcpu->arch.apic->dev;
2043                 if (dev->in_range(dev, addr, len, is_write))
2044                         return dev;
2045         }
2046         return NULL;
2047 }
2048
2049
2050 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2051                                                 gpa_t addr, int len,
2052                                                 int is_write)
2053 {
2054         struct kvm_io_device *dev;
2055
2056         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2057         if (dev == NULL)
2058                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2059                                           is_write);
2060         return dev;
2061 }
2062
2063 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2064                                struct kvm_vcpu *vcpu)
2065 {
2066         void *data = val;
2067         int r = X86EMUL_CONTINUE;
2068
2069         while (bytes) {
2070                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2071                 unsigned offset = addr & (PAGE_SIZE-1);
2072                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2073                 int ret;
2074
2075                 if (gpa == UNMAPPED_GVA) {
2076                         r = X86EMUL_PROPAGATE_FAULT;
2077                         goto out;
2078                 }
2079                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2080                 if (ret < 0) {
2081                         r = X86EMUL_UNHANDLEABLE;
2082                         goto out;
2083                 }
2084
2085                 bytes -= toread;
2086                 data += toread;
2087                 addr += toread;
2088         }
2089 out:
2090         return r;
2091 }
2092
2093 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2094                                 struct kvm_vcpu *vcpu)
2095 {
2096         void *data = val;
2097         int r = X86EMUL_CONTINUE;
2098
2099         while (bytes) {
2100                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2101                 unsigned offset = addr & (PAGE_SIZE-1);
2102                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2103                 int ret;
2104
2105                 if (gpa == UNMAPPED_GVA) {
2106                         r = X86EMUL_PROPAGATE_FAULT;
2107                         goto out;
2108                 }
2109                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2110                 if (ret < 0) {
2111                         r = X86EMUL_UNHANDLEABLE;
2112                         goto out;
2113                 }
2114
2115                 bytes -= towrite;
2116                 data += towrite;
2117                 addr += towrite;
2118         }
2119 out:
2120         return r;
2121 }
2122
2123
2124 static int emulator_read_emulated(unsigned long addr,
2125                                   void *val,
2126                                   unsigned int bytes,
2127                                   struct kvm_vcpu *vcpu)
2128 {
2129         struct kvm_io_device *mmio_dev;
2130         gpa_t                 gpa;
2131
2132         if (vcpu->mmio_read_completed) {
2133                 memcpy(val, vcpu->mmio_data, bytes);
2134                 vcpu->mmio_read_completed = 0;
2135                 return X86EMUL_CONTINUE;
2136         }
2137
2138         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2139
2140         /* For APIC access vmexit */
2141         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2142                 goto mmio;
2143
2144         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2145                                 == X86EMUL_CONTINUE)
2146                 return X86EMUL_CONTINUE;
2147         if (gpa == UNMAPPED_GVA)
2148                 return X86EMUL_PROPAGATE_FAULT;
2149
2150 mmio:
2151         /*
2152          * Is this MMIO handled locally?
2153          */
2154         mutex_lock(&vcpu->kvm->lock);
2155         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2156         if (mmio_dev) {
2157                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2158                 mutex_unlock(&vcpu->kvm->lock);
2159                 return X86EMUL_CONTINUE;
2160         }
2161         mutex_unlock(&vcpu->kvm->lock);
2162
2163         vcpu->mmio_needed = 1;
2164         vcpu->mmio_phys_addr = gpa;
2165         vcpu->mmio_size = bytes;
2166         vcpu->mmio_is_write = 0;
2167
2168         return X86EMUL_UNHANDLEABLE;
2169 }
2170
2171 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2172                           const void *val, int bytes)
2173 {
2174         int ret;
2175
2176         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2177         if (ret < 0)
2178                 return 0;
2179         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2180         return 1;
2181 }
2182
2183 static int emulator_write_emulated_onepage(unsigned long addr,
2184                                            const void *val,
2185                                            unsigned int bytes,
2186                                            struct kvm_vcpu *vcpu)
2187 {
2188         struct kvm_io_device *mmio_dev;
2189         gpa_t                 gpa;
2190
2191         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2192
2193         if (gpa == UNMAPPED_GVA) {
2194                 kvm_inject_page_fault(vcpu, addr, 2);
2195                 return X86EMUL_PROPAGATE_FAULT;
2196         }
2197
2198         /* For APIC access vmexit */
2199         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2200                 goto mmio;
2201
2202         if (emulator_write_phys(vcpu, gpa, val, bytes))
2203                 return X86EMUL_CONTINUE;
2204
2205 mmio:
2206         /*
2207          * Is this MMIO handled locally?
2208          */
2209         mutex_lock(&vcpu->kvm->lock);
2210         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2211         if (mmio_dev) {
2212                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2213                 mutex_unlock(&vcpu->kvm->lock);
2214                 return X86EMUL_CONTINUE;
2215         }
2216         mutex_unlock(&vcpu->kvm->lock);
2217
2218         vcpu->mmio_needed = 1;
2219         vcpu->mmio_phys_addr = gpa;
2220         vcpu->mmio_size = bytes;
2221         vcpu->mmio_is_write = 1;
2222         memcpy(vcpu->mmio_data, val, bytes);
2223
2224         return X86EMUL_CONTINUE;
2225 }
2226
2227 int emulator_write_emulated(unsigned long addr,
2228                                    const void *val,
2229                                    unsigned int bytes,
2230                                    struct kvm_vcpu *vcpu)
2231 {
2232         /* Crossing a page boundary? */
2233         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2234                 int rc, now;
2235
2236                 now = -addr & ~PAGE_MASK;
2237                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2238                 if (rc != X86EMUL_CONTINUE)
2239                         return rc;
2240                 addr += now;
2241                 val += now;
2242                 bytes -= now;
2243         }
2244         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2245 }
2246 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2247
2248 static int emulator_cmpxchg_emulated(unsigned long addr,
2249                                      const void *old,
2250                                      const void *new,
2251                                      unsigned int bytes,
2252                                      struct kvm_vcpu *vcpu)
2253 {
2254         static int reported;
2255
2256         if (!reported) {
2257                 reported = 1;
2258                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2259         }
2260 #ifndef CONFIG_X86_64
2261         /* guests cmpxchg8b have to be emulated atomically */
2262         if (bytes == 8) {
2263                 gpa_t gpa;
2264                 struct page *page;
2265                 char *kaddr;
2266                 u64 val;
2267
2268                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2269
2270                 if (gpa == UNMAPPED_GVA ||
2271                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2272                         goto emul_write;
2273
2274                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2275                         goto emul_write;
2276
2277                 val = *(u64 *)new;
2278
2279                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2280
2281                 kaddr = kmap_atomic(page, KM_USER0);
2282                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2283                 kunmap_atomic(kaddr, KM_USER0);
2284                 kvm_release_page_dirty(page);
2285         }
2286 emul_write:
2287 #endif
2288
2289         return emulator_write_emulated(addr, new, bytes, vcpu);
2290 }
2291
2292 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2293 {
2294         return kvm_x86_ops->get_segment_base(vcpu, seg);
2295 }
2296
2297 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2298 {
2299         kvm_mmu_invlpg(vcpu, address);
2300         return X86EMUL_CONTINUE;
2301 }
2302
2303 int emulate_clts(struct kvm_vcpu *vcpu)
2304 {
2305         KVMTRACE_0D(CLTS, vcpu, handler);
2306         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2307         return X86EMUL_CONTINUE;
2308 }
2309
2310 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2311 {
2312         struct kvm_vcpu *vcpu = ctxt->vcpu;
2313
2314         switch (dr) {
2315         case 0 ... 3:
2316                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2317                 return X86EMUL_CONTINUE;
2318         default:
2319                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2320                 return X86EMUL_UNHANDLEABLE;
2321         }
2322 }
2323
2324 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2325 {
2326         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2327         int exception;
2328
2329         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2330         if (exception) {
2331                 /* FIXME: better handling */
2332                 return X86EMUL_UNHANDLEABLE;
2333         }
2334         return X86EMUL_CONTINUE;
2335 }
2336
2337 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2338 {
2339         u8 opcodes[4];
2340         unsigned long rip = kvm_rip_read(vcpu);
2341         unsigned long rip_linear;
2342
2343         if (!printk_ratelimit())
2344                 return;
2345
2346         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2347
2348         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2349
2350         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2351                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2352 }
2353 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2354
2355 static struct x86_emulate_ops emulate_ops = {
2356         .read_std            = kvm_read_guest_virt,
2357         .read_emulated       = emulator_read_emulated,
2358         .write_emulated      = emulator_write_emulated,
2359         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2360 };
2361
2362 static void cache_all_regs(struct kvm_vcpu *vcpu)
2363 {
2364         kvm_register_read(vcpu, VCPU_REGS_RAX);
2365         kvm_register_read(vcpu, VCPU_REGS_RSP);
2366         kvm_register_read(vcpu, VCPU_REGS_RIP);
2367         vcpu->arch.regs_dirty = ~0;
2368 }
2369
2370 int emulate_instruction(struct kvm_vcpu *vcpu,
2371                         struct kvm_run *run,
2372                         unsigned long cr2,
2373                         u16 error_code,
2374                         int emulation_type)
2375 {
2376         int r;
2377         struct decode_cache *c;
2378
2379         kvm_clear_exception_queue(vcpu);
2380         vcpu->arch.mmio_fault_cr2 = cr2;
2381         /*
2382          * TODO: fix x86_emulate.c to use guest_read/write_register
2383          * instead of direct ->regs accesses, can save hundred cycles
2384          * on Intel for instructions that don't read/change RSP, for
2385          * for example.
2386          */
2387         cache_all_regs(vcpu);
2388
2389         vcpu->mmio_is_write = 0;
2390         vcpu->arch.pio.string = 0;
2391
2392         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2393                 int cs_db, cs_l;
2394                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2395
2396                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2397                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2398                 vcpu->arch.emulate_ctxt.mode =
2399                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2400                         ? X86EMUL_MODE_REAL : cs_l
2401                         ? X86EMUL_MODE_PROT64 : cs_db
2402                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2403
2404                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2405
2406                 /* Reject the instructions other than VMCALL/VMMCALL when
2407                  * try to emulate invalid opcode */
2408                 c = &vcpu->arch.emulate_ctxt.decode;
2409                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2410                     (!(c->twobyte && c->b == 0x01 &&
2411                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2412                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2413                         return EMULATE_FAIL;
2414
2415                 ++vcpu->stat.insn_emulation;
2416                 if (r)  {
2417                         ++vcpu->stat.insn_emulation_fail;
2418                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2419                                 return EMULATE_DONE;
2420                         return EMULATE_FAIL;
2421                 }
2422         }
2423
2424         if (emulation_type & EMULTYPE_SKIP) {
2425                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2426                 return EMULATE_DONE;
2427         }
2428
2429         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2430
2431         if (vcpu->arch.pio.string)
2432                 return EMULATE_DO_MMIO;
2433
2434         if ((r || vcpu->mmio_is_write) && run) {
2435                 run->exit_reason = KVM_EXIT_MMIO;
2436                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2437                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2438                 run->mmio.len = vcpu->mmio_size;
2439                 run->mmio.is_write = vcpu->mmio_is_write;
2440         }
2441
2442         if (r) {
2443                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2444                         return EMULATE_DONE;
2445                 if (!vcpu->mmio_needed) {
2446                         kvm_report_emulation_failure(vcpu, "mmio");
2447                         return EMULATE_FAIL;
2448                 }
2449                 return EMULATE_DO_MMIO;
2450         }
2451
2452         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2453
2454         if (vcpu->mmio_is_write) {
2455                 vcpu->mmio_needed = 0;
2456                 return EMULATE_DO_MMIO;
2457         }
2458
2459         return EMULATE_DONE;
2460 }
2461 EXPORT_SYMBOL_GPL(emulate_instruction);
2462
2463 static int pio_copy_data(struct kvm_vcpu *vcpu)
2464 {
2465         void *p = vcpu->arch.pio_data;
2466         gva_t q = vcpu->arch.pio.guest_gva;
2467         unsigned bytes;
2468         int ret;
2469
2470         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2471         if (vcpu->arch.pio.in)
2472                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2473         else
2474                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2475         return ret;
2476 }
2477
2478 int complete_pio(struct kvm_vcpu *vcpu)
2479 {
2480         struct kvm_pio_request *io = &vcpu->arch.pio;
2481         long delta;
2482         int r;
2483         unsigned long val;
2484
2485         if (!io->string) {
2486                 if (io->in) {
2487                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2488                         memcpy(&val, vcpu->arch.pio_data, io->size);
2489                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2490                 }
2491         } else {
2492                 if (io->in) {
2493                         r = pio_copy_data(vcpu);
2494                         if (r)
2495                                 return r;
2496                 }
2497
2498                 delta = 1;
2499                 if (io->rep) {
2500                         delta *= io->cur_count;
2501                         /*
2502                          * The size of the register should really depend on
2503                          * current address size.
2504                          */
2505                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2506                         val -= delta;
2507                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2508                 }
2509                 if (io->down)
2510                         delta = -delta;
2511                 delta *= io->size;
2512                 if (io->in) {
2513                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2514                         val += delta;
2515                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2516                 } else {
2517                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2518                         val += delta;
2519                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2520                 }
2521         }
2522
2523         io->count -= io->cur_count;
2524         io->cur_count = 0;
2525
2526         return 0;
2527 }
2528
2529 static void kernel_pio(struct kvm_io_device *pio_dev,
2530                        struct kvm_vcpu *vcpu,
2531                        void *pd)
2532 {
2533         /* TODO: String I/O for in kernel device */
2534
2535         mutex_lock(&vcpu->kvm->lock);
2536         if (vcpu->arch.pio.in)
2537                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2538                                   vcpu->arch.pio.size,
2539                                   pd);
2540         else
2541                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2542                                    vcpu->arch.pio.size,
2543                                    pd);
2544         mutex_unlock(&vcpu->kvm->lock);
2545 }
2546
2547 static void pio_string_write(struct kvm_io_device *pio_dev,
2548                              struct kvm_vcpu *vcpu)
2549 {
2550         struct kvm_pio_request *io = &vcpu->arch.pio;
2551         void *pd = vcpu->arch.pio_data;
2552         int i;
2553
2554         mutex_lock(&vcpu->kvm->lock);
2555         for (i = 0; i < io->cur_count; i++) {
2556                 kvm_iodevice_write(pio_dev, io->port,
2557                                    io->size,
2558                                    pd);
2559                 pd += io->size;
2560         }
2561         mutex_unlock(&vcpu->kvm->lock);
2562 }
2563
2564 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2565                                                gpa_t addr, int len,
2566                                                int is_write)
2567 {
2568         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2569 }
2570
2571 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2572                   int size, unsigned port)
2573 {
2574         struct kvm_io_device *pio_dev;
2575         unsigned long val;
2576
2577         vcpu->run->exit_reason = KVM_EXIT_IO;
2578         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2579         vcpu->run->io.size = vcpu->arch.pio.size = size;
2580         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2581         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2582         vcpu->run->io.port = vcpu->arch.pio.port = port;
2583         vcpu->arch.pio.in = in;
2584         vcpu->arch.pio.string = 0;
2585         vcpu->arch.pio.down = 0;
2586         vcpu->arch.pio.rep = 0;
2587
2588         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2589                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2590                             handler);
2591         else
2592                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2593                             handler);
2594
2595         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2596         memcpy(vcpu->arch.pio_data, &val, 4);
2597
2598         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2599         if (pio_dev) {
2600                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2601                 complete_pio(vcpu);
2602                 return 1;
2603         }
2604         return 0;
2605 }
2606 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2607
2608 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2609                   int size, unsigned long count, int down,
2610                   gva_t address, int rep, unsigned port)
2611 {
2612         unsigned now, in_page;
2613         int ret = 0;
2614         struct kvm_io_device *pio_dev;
2615
2616         vcpu->run->exit_reason = KVM_EXIT_IO;
2617         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2618         vcpu->run->io.size = vcpu->arch.pio.size = size;
2619         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2620         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2621         vcpu->run->io.port = vcpu->arch.pio.port = port;
2622         vcpu->arch.pio.in = in;
2623         vcpu->arch.pio.string = 1;
2624         vcpu->arch.pio.down = down;
2625         vcpu->arch.pio.rep = rep;
2626
2627         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2628                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2629                             handler);
2630         else
2631                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2632                             handler);
2633
2634         if (!count) {
2635                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2636                 return 1;
2637         }
2638
2639         if (!down)
2640                 in_page = PAGE_SIZE - offset_in_page(address);
2641         else
2642                 in_page = offset_in_page(address) + size;
2643         now = min(count, (unsigned long)in_page / size);
2644         if (!now)
2645                 now = 1;
2646         if (down) {
2647                 /*
2648                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2649                  */
2650                 pr_unimpl(vcpu, "guest string pio down\n");
2651                 kvm_inject_gp(vcpu, 0);
2652                 return 1;
2653         }
2654         vcpu->run->io.count = now;
2655         vcpu->arch.pio.cur_count = now;
2656
2657         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2658                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2659
2660         vcpu->arch.pio.guest_gva = address;
2661
2662         pio_dev = vcpu_find_pio_dev(vcpu, port,
2663                                     vcpu->arch.pio.cur_count,
2664                                     !vcpu->arch.pio.in);
2665         if (!vcpu->arch.pio.in) {
2666                 /* string PIO write */
2667                 ret = pio_copy_data(vcpu);
2668                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2669                         kvm_inject_gp(vcpu, 0);
2670                         return 1;
2671                 }
2672                 if (ret == 0 && pio_dev) {
2673                         pio_string_write(pio_dev, vcpu);
2674                         complete_pio(vcpu);
2675                         if (vcpu->arch.pio.count == 0)
2676                                 ret = 1;
2677                 }
2678         } else if (pio_dev)
2679                 pr_unimpl(vcpu, "no string pio read support yet, "
2680                        "port %x size %d count %ld\n",
2681                         port, size, count);
2682
2683         return ret;
2684 }
2685 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2686
2687 static void bounce_off(void *info)
2688 {
2689         /* nothing */
2690 }
2691
2692 static unsigned int  ref_freq;
2693 static unsigned long tsc_khz_ref;
2694
2695 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2696                                      void *data)
2697 {
2698         struct cpufreq_freqs *freq = data;
2699         struct kvm *kvm;
2700         struct kvm_vcpu *vcpu;
2701         int i, send_ipi = 0;
2702
2703         if (!ref_freq)
2704                 ref_freq = freq->old;
2705
2706         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2707                 return 0;
2708         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2709                 return 0;
2710         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2711
2712         spin_lock(&kvm_lock);
2713         list_for_each_entry(kvm, &vm_list, vm_list) {
2714                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2715                         vcpu = kvm->vcpus[i];
2716                         if (!vcpu)
2717                                 continue;
2718                         if (vcpu->cpu != freq->cpu)
2719                                 continue;
2720                         if (!kvm_request_guest_time_update(vcpu))
2721                                 continue;
2722                         if (vcpu->cpu != smp_processor_id())
2723                                 send_ipi++;
2724                 }
2725         }
2726         spin_unlock(&kvm_lock);
2727
2728         if (freq->old < freq->new && send_ipi) {
2729                 /*
2730                  * We upscale the frequency.  Must make the guest
2731                  * doesn't see old kvmclock values while running with
2732                  * the new frequency, otherwise we risk the guest sees
2733                  * time go backwards.
2734                  *
2735                  * In case we update the frequency for another cpu
2736                  * (which might be in guest context) send an interrupt
2737                  * to kick the cpu out of guest context.  Next time
2738                  * guest context is entered kvmclock will be updated,
2739                  * so the guest will not see stale values.
2740                  */
2741                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2742         }
2743         return 0;
2744 }
2745
2746 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2747         .notifier_call  = kvmclock_cpufreq_notifier
2748 };
2749
2750 int kvm_arch_init(void *opaque)
2751 {
2752         int r, cpu;
2753         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2754
2755         if (kvm_x86_ops) {
2756                 printk(KERN_ERR "kvm: already loaded the other module\n");
2757                 r = -EEXIST;
2758                 goto out;
2759         }
2760
2761         if (!ops->cpu_has_kvm_support()) {
2762                 printk(KERN_ERR "kvm: no hardware support\n");
2763                 r = -EOPNOTSUPP;
2764                 goto out;
2765         }
2766         if (ops->disabled_by_bios()) {
2767                 printk(KERN_ERR "kvm: disabled by bios\n");
2768                 r = -EOPNOTSUPP;
2769                 goto out;
2770         }
2771
2772         r = kvm_mmu_module_init();
2773         if (r)
2774                 goto out;
2775
2776         kvm_init_msr_list();
2777
2778         kvm_x86_ops = ops;
2779         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2780         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2781         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2782                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2783
2784         for_each_possible_cpu(cpu)
2785                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2786         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2787                 tsc_khz_ref = tsc_khz;
2788                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2789                                           CPUFREQ_TRANSITION_NOTIFIER);
2790         }
2791
2792         return 0;
2793
2794 out:
2795         return r;
2796 }
2797
2798 void kvm_arch_exit(void)
2799 {
2800         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2801                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2802                                             CPUFREQ_TRANSITION_NOTIFIER);
2803         kvm_x86_ops = NULL;
2804         kvm_mmu_module_exit();
2805 }
2806
2807 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2808 {
2809         ++vcpu->stat.halt_exits;
2810         KVMTRACE_0D(HLT, vcpu, handler);
2811         if (irqchip_in_kernel(vcpu->kvm)) {
2812                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2813                 return 1;
2814         } else {
2815                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2816                 return 0;
2817         }
2818 }
2819 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2820
2821 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2822                            unsigned long a1)
2823 {
2824         if (is_long_mode(vcpu))
2825                 return a0;
2826         else
2827                 return a0 | ((gpa_t)a1 << 32);
2828 }
2829
2830 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2831 {
2832         unsigned long nr, a0, a1, a2, a3, ret;
2833         int r = 1;
2834
2835         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2836         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2837         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2838         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2839         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2840
2841         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2842
2843         if (!is_long_mode(vcpu)) {
2844                 nr &= 0xFFFFFFFF;
2845                 a0 &= 0xFFFFFFFF;
2846                 a1 &= 0xFFFFFFFF;
2847                 a2 &= 0xFFFFFFFF;
2848                 a3 &= 0xFFFFFFFF;
2849         }
2850
2851         switch (nr) {
2852         case KVM_HC_VAPIC_POLL_IRQ:
2853                 ret = 0;
2854                 break;
2855         case KVM_HC_MMU_OP:
2856                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2857                 break;
2858         default:
2859                 ret = -KVM_ENOSYS;
2860                 break;
2861         }
2862         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2863         ++vcpu->stat.hypercalls;
2864         return r;
2865 }
2866 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2867
2868 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2869 {
2870         char instruction[3];
2871         int ret = 0;
2872         unsigned long rip = kvm_rip_read(vcpu);
2873
2874
2875         /*
2876          * Blow out the MMU to ensure that no other VCPU has an active mapping
2877          * to ensure that the updated hypercall appears atomically across all
2878          * VCPUs.
2879          */
2880         kvm_mmu_zap_all(vcpu->kvm);
2881
2882         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2883         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2884             != X86EMUL_CONTINUE)
2885                 ret = -EFAULT;
2886
2887         return ret;
2888 }
2889
2890 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2891 {
2892         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2893 }
2894
2895 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2896 {
2897         struct descriptor_table dt = { limit, base };
2898
2899         kvm_x86_ops->set_gdt(vcpu, &dt);
2900 }
2901
2902 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2903 {
2904         struct descriptor_table dt = { limit, base };
2905
2906         kvm_x86_ops->set_idt(vcpu, &dt);
2907 }
2908
2909 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2910                    unsigned long *rflags)
2911 {
2912         kvm_lmsw(vcpu, msw);
2913         *rflags = kvm_x86_ops->get_rflags(vcpu);
2914 }
2915
2916 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2917 {
2918         unsigned long value;
2919
2920         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2921         switch (cr) {
2922         case 0:
2923                 value = vcpu->arch.cr0;
2924                 break;
2925         case 2:
2926                 value = vcpu->arch.cr2;
2927                 break;
2928         case 3:
2929                 value = vcpu->arch.cr3;
2930                 break;
2931         case 4:
2932                 value = vcpu->arch.cr4;
2933                 break;
2934         case 8:
2935                 value = kvm_get_cr8(vcpu);
2936                 break;
2937         default:
2938                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2939                 return 0;
2940         }
2941         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2942                     (u32)((u64)value >> 32), handler);
2943
2944         return value;
2945 }
2946
2947 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2948                      unsigned long *rflags)
2949 {
2950         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2951                     (u32)((u64)val >> 32), handler);
2952
2953         switch (cr) {
2954         case 0:
2955                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2956                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2957                 break;
2958         case 2:
2959                 vcpu->arch.cr2 = val;
2960                 break;
2961         case 3:
2962                 kvm_set_cr3(vcpu, val);
2963                 break;
2964         case 4:
2965                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2966                 break;
2967         case 8:
2968                 kvm_set_cr8(vcpu, val & 0xfUL);
2969                 break;
2970         default:
2971                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2972         }
2973 }
2974
2975 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2976 {
2977         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2978         int j, nent = vcpu->arch.cpuid_nent;
2979
2980         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2981         /* when no next entry is found, the current entry[i] is reselected */
2982         for (j = i + 1; ; j = (j + 1) % nent) {
2983                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2984                 if (ej->function == e->function) {
2985                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2986                         return j;
2987                 }
2988         }
2989         return 0; /* silence gcc, even though control never reaches here */
2990 }
2991
2992 /* find an entry with matching function, matching index (if needed), and that
2993  * should be read next (if it's stateful) */
2994 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2995         u32 function, u32 index)
2996 {
2997         if (e->function != function)
2998                 return 0;
2999         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3000                 return 0;
3001         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3002             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3003                 return 0;
3004         return 1;
3005 }
3006
3007 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3008                                               u32 function, u32 index)
3009 {
3010         int i;
3011         struct kvm_cpuid_entry2 *best = NULL;
3012
3013         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3014                 struct kvm_cpuid_entry2 *e;
3015
3016                 e = &vcpu->arch.cpuid_entries[i];
3017                 if (is_matching_cpuid_entry(e, function, index)) {
3018                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3019                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3020                         best = e;
3021                         break;
3022                 }
3023                 /*
3024                  * Both basic or both extended?
3025                  */
3026                 if (((e->function ^ function) & 0x80000000) == 0)
3027                         if (!best || e->function > best->function)
3028                                 best = e;
3029         }
3030         return best;
3031 }
3032
3033 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3034 {
3035         struct kvm_cpuid_entry2 *best;
3036
3037         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3038         if (best)
3039                 return best->eax & 0xff;
3040         return 36;
3041 }
3042
3043 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3044 {
3045         u32 function, index;
3046         struct kvm_cpuid_entry2 *best;
3047
3048         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3049         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3050         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3051         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3052         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3053         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3054         best = kvm_find_cpuid_entry(vcpu, function, index);
3055         if (best) {
3056                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3057                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3058                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3059                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3060         }
3061         kvm_x86_ops->skip_emulated_instruction(vcpu);
3062         KVMTRACE_5D(CPUID, vcpu, function,
3063                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3064                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3065                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3066                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3067 }
3068 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3069
3070 /*
3071  * Check if userspace requested an interrupt window, and that the
3072  * interrupt window is open.
3073  *
3074  * No need to exit to userspace if we already have an interrupt queued.
3075  */
3076 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3077                                           struct kvm_run *kvm_run)
3078 {
3079         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3080                 kvm_run->request_interrupt_window &&
3081                 kvm_arch_interrupt_allowed(vcpu));
3082 }
3083
3084 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3085                               struct kvm_run *kvm_run)
3086 {
3087         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3088         kvm_run->cr8 = kvm_get_cr8(vcpu);
3089         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3090         if (irqchip_in_kernel(vcpu->kvm))
3091                 kvm_run->ready_for_interrupt_injection = 1;
3092         else
3093                 kvm_run->ready_for_interrupt_injection =
3094                                         (kvm_arch_interrupt_allowed(vcpu) &&
3095                                          !kvm_cpu_has_interrupt(vcpu));
3096 }
3097
3098 static void vapic_enter(struct kvm_vcpu *vcpu)
3099 {
3100         struct kvm_lapic *apic = vcpu->arch.apic;
3101         struct page *page;
3102
3103         if (!apic || !apic->vapic_addr)
3104                 return;
3105
3106         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3107
3108         vcpu->arch.apic->vapic_page = page;
3109 }
3110
3111 static void vapic_exit(struct kvm_vcpu *vcpu)
3112 {
3113         struct kvm_lapic *apic = vcpu->arch.apic;
3114
3115         if (!apic || !apic->vapic_addr)
3116                 return;
3117
3118         down_read(&vcpu->kvm->slots_lock);
3119         kvm_release_page_dirty(apic->vapic_page);
3120         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3121         up_read(&vcpu->kvm->slots_lock);
3122 }
3123
3124 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3125 {
3126         int max_irr, tpr;
3127
3128         if (!kvm_x86_ops->update_cr8_intercept)
3129                 return;
3130
3131         max_irr = kvm_lapic_find_highest_irr(vcpu);
3132
3133         if (max_irr != -1)
3134                 max_irr >>= 4;
3135
3136         tpr = kvm_lapic_get_cr8(vcpu);
3137
3138         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3139 }
3140
3141 static void inject_irq(struct kvm_vcpu *vcpu)
3142 {
3143         /* try to reinject previous events if any */
3144         if (vcpu->arch.nmi_injected) {
3145                 kvm_x86_ops->set_nmi(vcpu);
3146                 return;
3147         }
3148
3149         if (vcpu->arch.interrupt.pending) {
3150                 kvm_x86_ops->set_irq(vcpu, vcpu->arch.interrupt.nr);
3151                 return;
3152         }
3153
3154         /* try to inject new event if pending */
3155         if (vcpu->arch.nmi_pending) {
3156                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3157                         vcpu->arch.nmi_pending = false;
3158                         vcpu->arch.nmi_injected = true;
3159                         kvm_x86_ops->set_nmi(vcpu);
3160                 }
3161         } else if (kvm_cpu_has_interrupt(vcpu)) {
3162                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3163                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu));
3164                         kvm_x86_ops->set_irq(vcpu, vcpu->arch.interrupt.nr);
3165                 }
3166         }
3167 }
3168
3169 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3170 {
3171         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3172                 kvm_run->request_interrupt_window;
3173
3174         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3175                 kvm_x86_ops->drop_interrupt_shadow(vcpu);
3176
3177         inject_irq(vcpu);
3178
3179         /* enable NMI/IRQ window open exits if needed */
3180         if (vcpu->arch.nmi_pending)
3181                 kvm_x86_ops->enable_nmi_window(vcpu);
3182         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3183                 kvm_x86_ops->enable_irq_window(vcpu);
3184 }
3185
3186 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3187 {
3188         int r;
3189
3190         if (vcpu->requests)
3191                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3192                         kvm_mmu_unload(vcpu);
3193
3194         r = kvm_mmu_reload(vcpu);
3195         if (unlikely(r))
3196                 goto out;
3197
3198         if (vcpu->requests) {
3199                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3200                         __kvm_migrate_timers(vcpu);
3201                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3202                         kvm_write_guest_time(vcpu);
3203                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3204                         kvm_mmu_sync_roots(vcpu);
3205                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3206                         kvm_x86_ops->tlb_flush(vcpu);
3207                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3208                                        &vcpu->requests)) {
3209                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3210                         r = 0;
3211                         goto out;
3212                 }
3213                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3214                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3215                         r = 0;
3216                         goto out;
3217                 }
3218         }
3219
3220         preempt_disable();
3221
3222         kvm_x86_ops->prepare_guest_switch(vcpu);
3223         kvm_load_guest_fpu(vcpu);
3224
3225         local_irq_disable();
3226
3227         if (vcpu->requests || need_resched() || signal_pending(current)) {
3228                 local_irq_enable();
3229                 preempt_enable();
3230                 r = 1;
3231                 goto out;
3232         }
3233
3234         vcpu->guest_mode = 1;
3235         /*
3236          * Make sure that guest_mode assignment won't happen after
3237          * testing the pending IRQ vector bitmap.
3238          */
3239         smp_wmb();
3240
3241         if (vcpu->arch.exception.pending)
3242                 __queue_exception(vcpu);
3243         else
3244                 inject_pending_irq(vcpu, kvm_run);
3245
3246         if (kvm_lapic_enabled(vcpu)) {
3247                 if (!vcpu->arch.apic->vapic_addr)
3248                         update_cr8_intercept(vcpu);
3249                 else
3250                         kvm_lapic_sync_to_vapic(vcpu);
3251         }
3252
3253         up_read(&vcpu->kvm->slots_lock);
3254
3255         kvm_guest_enter();
3256
3257         get_debugreg(vcpu->arch.host_dr6, 6);
3258         get_debugreg(vcpu->arch.host_dr7, 7);
3259         if (unlikely(vcpu->arch.switch_db_regs)) {
3260                 get_debugreg(vcpu->arch.host_db[0], 0);
3261                 get_debugreg(vcpu->arch.host_db[1], 1);
3262                 get_debugreg(vcpu->arch.host_db[2], 2);
3263                 get_debugreg(vcpu->arch.host_db[3], 3);
3264
3265                 set_debugreg(0, 7);
3266                 set_debugreg(vcpu->arch.eff_db[0], 0);
3267                 set_debugreg(vcpu->arch.eff_db[1], 1);
3268                 set_debugreg(vcpu->arch.eff_db[2], 2);
3269                 set_debugreg(vcpu->arch.eff_db[3], 3);
3270         }
3271
3272         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3273         kvm_x86_ops->run(vcpu, kvm_run);
3274
3275         if (unlikely(vcpu->arch.switch_db_regs)) {
3276                 set_debugreg(0, 7);
3277                 set_debugreg(vcpu->arch.host_db[0], 0);
3278                 set_debugreg(vcpu->arch.host_db[1], 1);
3279                 set_debugreg(vcpu->arch.host_db[2], 2);
3280                 set_debugreg(vcpu->arch.host_db[3], 3);
3281         }
3282         set_debugreg(vcpu->arch.host_dr6, 6);
3283         set_debugreg(vcpu->arch.host_dr7, 7);
3284
3285         vcpu->guest_mode = 0;
3286         local_irq_enable();
3287
3288         ++vcpu->stat.exits;
3289
3290         /*
3291          * We must have an instruction between local_irq_enable() and
3292          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3293          * the interrupt shadow.  The stat.exits increment will do nicely.
3294          * But we need to prevent reordering, hence this barrier():
3295          */
3296         barrier();
3297
3298         kvm_guest_exit();
3299
3300         preempt_enable();
3301
3302         down_read(&vcpu->kvm->slots_lock);
3303
3304         /*
3305          * Profile KVM exit RIPs:
3306          */
3307         if (unlikely(prof_on == KVM_PROFILING)) {
3308                 unsigned long rip = kvm_rip_read(vcpu);
3309                 profile_hit(KVM_PROFILING, (void *)rip);
3310         }
3311
3312
3313         kvm_lapic_sync_from_vapic(vcpu);
3314
3315         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3316 out:
3317         return r;
3318 }
3319
3320
3321 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3322 {
3323         int r;
3324
3325         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3326                 pr_debug("vcpu %d received sipi with vector # %x\n",
3327                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3328                 kvm_lapic_reset(vcpu);
3329                 r = kvm_arch_vcpu_reset(vcpu);
3330                 if (r)
3331                         return r;
3332                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3333         }
3334
3335         down_read(&vcpu->kvm->slots_lock);
3336         vapic_enter(vcpu);
3337
3338         r = 1;
3339         while (r > 0) {
3340                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3341                         r = vcpu_enter_guest(vcpu, kvm_run);
3342                 else {
3343                         up_read(&vcpu->kvm->slots_lock);
3344                         kvm_vcpu_block(vcpu);
3345                         down_read(&vcpu->kvm->slots_lock);
3346                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3347                         {
3348                                 switch(vcpu->arch.mp_state) {
3349                                 case KVM_MP_STATE_HALTED:
3350                                         vcpu->arch.mp_state =
3351                                                 KVM_MP_STATE_RUNNABLE;
3352                                 case KVM_MP_STATE_RUNNABLE:
3353                                         break;
3354                                 case KVM_MP_STATE_SIPI_RECEIVED:
3355                                 default:
3356                                         r = -EINTR;
3357                                         break;
3358                                 }
3359                         }
3360                 }
3361
3362                 if (r <= 0)
3363                         break;
3364
3365                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3366                 if (kvm_cpu_has_pending_timer(vcpu))
3367                         kvm_inject_pending_timer_irqs(vcpu);
3368
3369                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3370                         r = -EINTR;
3371                         kvm_run->exit_reason = KVM_EXIT_INTR;
3372                         ++vcpu->stat.request_irq_exits;
3373                 }
3374                 if (signal_pending(current)) {
3375                         r = -EINTR;
3376                         kvm_run->exit_reason = KVM_EXIT_INTR;
3377                         ++vcpu->stat.signal_exits;
3378                 }
3379                 if (need_resched()) {
3380                         up_read(&vcpu->kvm->slots_lock);
3381                         kvm_resched(vcpu);
3382                         down_read(&vcpu->kvm->slots_lock);
3383                 }
3384         }
3385
3386         up_read(&vcpu->kvm->slots_lock);
3387         post_kvm_run_save(vcpu, kvm_run);
3388
3389         vapic_exit(vcpu);
3390
3391         return r;
3392 }
3393
3394 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3395 {
3396         int r;
3397         sigset_t sigsaved;
3398
3399         vcpu_load(vcpu);
3400
3401         if (vcpu->sigset_active)
3402                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3403
3404         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3405                 kvm_vcpu_block(vcpu);
3406                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3407                 r = -EAGAIN;
3408                 goto out;
3409         }
3410
3411         /* re-sync apic's tpr */
3412         if (!irqchip_in_kernel(vcpu->kvm))
3413                 kvm_set_cr8(vcpu, kvm_run->cr8);
3414
3415         if (vcpu->arch.pio.cur_count) {
3416                 r = complete_pio(vcpu);
3417                 if (r)
3418                         goto out;
3419         }
3420 #if CONFIG_HAS_IOMEM
3421         if (vcpu->mmio_needed) {
3422                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3423                 vcpu->mmio_read_completed = 1;
3424                 vcpu->mmio_needed = 0;
3425
3426                 down_read(&vcpu->kvm->slots_lock);
3427                 r = emulate_instruction(vcpu, kvm_run,
3428                                         vcpu->arch.mmio_fault_cr2, 0,
3429                                         EMULTYPE_NO_DECODE);
3430                 up_read(&vcpu->kvm->slots_lock);
3431                 if (r == EMULATE_DO_MMIO) {
3432                         /*
3433                          * Read-modify-write.  Back to userspace.
3434                          */
3435                         r = 0;
3436                         goto out;
3437                 }
3438         }
3439 #endif
3440         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3441                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3442                                      kvm_run->hypercall.ret);
3443
3444         r = __vcpu_run(vcpu, kvm_run);
3445
3446 out:
3447         if (vcpu->sigset_active)
3448                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3449
3450         vcpu_put(vcpu);
3451         return r;
3452 }
3453
3454 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3455 {
3456         vcpu_load(vcpu);
3457
3458         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3459         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3460         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3461         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3462         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3463         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3464         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3465         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3466 #ifdef CONFIG_X86_64
3467         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3468         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3469         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3470         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3471         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3472         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3473         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3474         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3475 #endif
3476
3477         regs->rip = kvm_rip_read(vcpu);
3478         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3479
3480         /*
3481          * Don't leak debug flags in case they were set for guest debugging
3482          */
3483         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3484                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3485
3486         vcpu_put(vcpu);
3487
3488         return 0;
3489 }
3490
3491 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3492 {
3493         vcpu_load(vcpu);
3494
3495         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3496         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3497         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3498         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3499         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3500         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3501         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3502         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3503 #ifdef CONFIG_X86_64
3504         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3505         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3506         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3507         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3508         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3509         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3510         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3511         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3512
3513 #endif
3514
3515         kvm_rip_write(vcpu, regs->rip);
3516         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3517
3518
3519         vcpu->arch.exception.pending = false;
3520
3521         vcpu_put(vcpu);
3522
3523         return 0;
3524 }
3525
3526 void kvm_get_segment(struct kvm_vcpu *vcpu,
3527                      struct kvm_segment *var, int seg)
3528 {
3529         kvm_x86_ops->get_segment(vcpu, var, seg);
3530 }
3531
3532 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3533 {
3534         struct kvm_segment cs;
3535
3536         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3537         *db = cs.db;
3538         *l = cs.l;
3539 }
3540 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3541
3542 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3543                                   struct kvm_sregs *sregs)
3544 {
3545         struct descriptor_table dt;
3546
3547         vcpu_load(vcpu);
3548
3549         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3550         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3551         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3552         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3553         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3554         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3555
3556         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3557         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3558
3559         kvm_x86_ops->get_idt(vcpu, &dt);
3560         sregs->idt.limit = dt.limit;
3561         sregs->idt.base = dt.base;
3562         kvm_x86_ops->get_gdt(vcpu, &dt);
3563         sregs->gdt.limit = dt.limit;
3564         sregs->gdt.base = dt.base;
3565
3566         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3567         sregs->cr0 = vcpu->arch.cr0;
3568         sregs->cr2 = vcpu->arch.cr2;
3569         sregs->cr3 = vcpu->arch.cr3;
3570         sregs->cr4 = vcpu->arch.cr4;
3571         sregs->cr8 = kvm_get_cr8(vcpu);
3572         sregs->efer = vcpu->arch.shadow_efer;
3573         sregs->apic_base = kvm_get_apic_base(vcpu);
3574
3575         if (irqchip_in_kernel(vcpu->kvm))
3576                 memset(sregs->interrupt_bitmap, 0,
3577                        sizeof sregs->interrupt_bitmap);
3578         else
3579                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3580                        sizeof sregs->interrupt_bitmap);
3581
3582         if (vcpu->arch.interrupt.pending)
3583                 set_bit(vcpu->arch.interrupt.nr,
3584                         (unsigned long *)sregs->interrupt_bitmap);
3585
3586         vcpu_put(vcpu);
3587
3588         return 0;
3589 }
3590
3591 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3592                                     struct kvm_mp_state *mp_state)
3593 {
3594         vcpu_load(vcpu);
3595         mp_state->mp_state = vcpu->arch.mp_state;
3596         vcpu_put(vcpu);
3597         return 0;
3598 }
3599
3600 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3601                                     struct kvm_mp_state *mp_state)
3602 {
3603         vcpu_load(vcpu);
3604         vcpu->arch.mp_state = mp_state->mp_state;
3605         vcpu_put(vcpu);
3606         return 0;
3607 }
3608
3609 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3610                         struct kvm_segment *var, int seg)
3611 {
3612         kvm_x86_ops->set_segment(vcpu, var, seg);
3613 }
3614
3615 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3616                                    struct kvm_segment *kvm_desct)
3617 {
3618         kvm_desct->base = seg_desc->base0;
3619         kvm_desct->base |= seg_desc->base1 << 16;
3620         kvm_desct->base |= seg_desc->base2 << 24;
3621         kvm_desct->limit = seg_desc->limit0;
3622         kvm_desct->limit |= seg_desc->limit << 16;
3623         if (seg_desc->g) {
3624                 kvm_desct->limit <<= 12;
3625                 kvm_desct->limit |= 0xfff;
3626         }
3627         kvm_desct->selector = selector;
3628         kvm_desct->type = seg_desc->type;
3629         kvm_desct->present = seg_desc->p;
3630         kvm_desct->dpl = seg_desc->dpl;
3631         kvm_desct->db = seg_desc->d;
3632         kvm_desct->s = seg_desc->s;
3633         kvm_desct->l = seg_desc->l;
3634         kvm_desct->g = seg_desc->g;
3635         kvm_desct->avl = seg_desc->avl;
3636         if (!selector)
3637                 kvm_desct->unusable = 1;
3638         else
3639                 kvm_desct->unusable = 0;
3640         kvm_desct->padding = 0;
3641 }
3642
3643 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3644                                           u16 selector,
3645                                           struct descriptor_table *dtable)
3646 {
3647         if (selector & 1 << 2) {
3648                 struct kvm_segment kvm_seg;
3649
3650                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3651
3652                 if (kvm_seg.unusable)
3653                         dtable->limit = 0;
3654                 else
3655                         dtable->limit = kvm_seg.limit;
3656                 dtable->base = kvm_seg.base;
3657         }
3658         else
3659                 kvm_x86_ops->get_gdt(vcpu, dtable);
3660 }
3661
3662 /* allowed just for 8 bytes segments */
3663 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3664                                          struct desc_struct *seg_desc)
3665 {
3666         gpa_t gpa;
3667         struct descriptor_table dtable;
3668         u16 index = selector >> 3;
3669
3670         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3671
3672         if (dtable.limit < index * 8 + 7) {
3673                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3674                 return 1;
3675         }
3676         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3677         gpa += index * 8;
3678         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3679 }
3680
3681 /* allowed just for 8 bytes segments */
3682 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3683                                          struct desc_struct *seg_desc)
3684 {
3685         gpa_t gpa;
3686         struct descriptor_table dtable;
3687         u16 index = selector >> 3;
3688
3689         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3690
3691         if (dtable.limit < index * 8 + 7)
3692                 return 1;
3693         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3694         gpa += index * 8;
3695         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3696 }
3697
3698 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3699                              struct desc_struct *seg_desc)
3700 {
3701         u32 base_addr;
3702
3703         base_addr = seg_desc->base0;
3704         base_addr |= (seg_desc->base1 << 16);
3705         base_addr |= (seg_desc->base2 << 24);
3706
3707         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3708 }
3709
3710 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3711 {
3712         struct kvm_segment kvm_seg;
3713
3714         kvm_get_segment(vcpu, &kvm_seg, seg);
3715         return kvm_seg.selector;
3716 }
3717
3718 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3719                                                 u16 selector,
3720                                                 struct kvm_segment *kvm_seg)
3721 {
3722         struct desc_struct seg_desc;
3723
3724         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3725                 return 1;
3726         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3727         return 0;
3728 }
3729
3730 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3731 {
3732         struct kvm_segment segvar = {
3733                 .base = selector << 4,
3734                 .limit = 0xffff,
3735                 .selector = selector,
3736                 .type = 3,
3737                 .present = 1,
3738                 .dpl = 3,
3739                 .db = 0,
3740                 .s = 1,
3741                 .l = 0,
3742                 .g = 0,
3743                 .avl = 0,
3744                 .unusable = 0,
3745         };
3746         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3747         return 0;
3748 }
3749
3750 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3751                                 int type_bits, int seg)
3752 {
3753         struct kvm_segment kvm_seg;
3754
3755         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3756                 return kvm_load_realmode_segment(vcpu, selector, seg);
3757         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3758                 return 1;
3759         kvm_seg.type |= type_bits;
3760
3761         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3762             seg != VCPU_SREG_LDTR)
3763                 if (!kvm_seg.s)
3764                         kvm_seg.unusable = 1;
3765
3766         kvm_set_segment(vcpu, &kvm_seg, seg);
3767         return 0;
3768 }
3769
3770 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3771                                 struct tss_segment_32 *tss)
3772 {
3773         tss->cr3 = vcpu->arch.cr3;
3774         tss->eip = kvm_rip_read(vcpu);
3775         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3776         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3777         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3778         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3779         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3780         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3781         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3782         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3783         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3784         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3785         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3786         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3787         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3788         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3789         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3790         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3791 }
3792
3793 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3794                                   struct tss_segment_32 *tss)
3795 {
3796         kvm_set_cr3(vcpu, tss->cr3);
3797
3798         kvm_rip_write(vcpu, tss->eip);
3799         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3800
3801         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3802         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3803         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3804         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3805         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3806         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3807         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3808         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3809
3810         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3811                 return 1;
3812
3813         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3814                 return 1;
3815
3816         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3817                 return 1;
3818
3819         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3820                 return 1;
3821
3822         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3823                 return 1;
3824
3825         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3826                 return 1;
3827
3828         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3829                 return 1;
3830         return 0;
3831 }
3832
3833 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3834                                 struct tss_segment_16 *tss)
3835 {
3836         tss->ip = kvm_rip_read(vcpu);
3837         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3838         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3839         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3840         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3841         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3842         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3843         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3844         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3845         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3846
3847         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3848         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3849         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3850         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3851         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3852         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3853 }
3854
3855 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3856                                  struct tss_segment_16 *tss)
3857 {
3858         kvm_rip_write(vcpu, tss->ip);
3859         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3860         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3861         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3862         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3863         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3864         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3865         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3866         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3867         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3868
3869         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3870                 return 1;
3871
3872         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3873                 return 1;
3874
3875         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3876                 return 1;
3877
3878         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3879                 return 1;
3880
3881         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3882                 return 1;
3883         return 0;
3884 }
3885
3886 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3887                               u16 old_tss_sel, u32 old_tss_base,
3888                               struct desc_struct *nseg_desc)
3889 {
3890         struct tss_segment_16 tss_segment_16;
3891         int ret = 0;
3892
3893         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3894                            sizeof tss_segment_16))
3895                 goto out;
3896
3897         save_state_to_tss16(vcpu, &tss_segment_16);
3898
3899         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3900                             sizeof tss_segment_16))
3901                 goto out;
3902
3903         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3904                            &tss_segment_16, sizeof tss_segment_16))
3905                 goto out;
3906
3907         if (old_tss_sel != 0xffff) {
3908                 tss_segment_16.prev_task_link = old_tss_sel;
3909
3910                 if (kvm_write_guest(vcpu->kvm,
3911                                     get_tss_base_addr(vcpu, nseg_desc),
3912                                     &tss_segment_16.prev_task_link,
3913                                     sizeof tss_segment_16.prev_task_link))
3914                         goto out;
3915         }
3916
3917         if (load_state_from_tss16(vcpu, &tss_segment_16))
3918                 goto out;
3919
3920         ret = 1;
3921 out:
3922         return ret;
3923 }
3924
3925 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3926                        u16 old_tss_sel, u32 old_tss_base,
3927                        struct desc_struct *nseg_desc)
3928 {
3929         struct tss_segment_32 tss_segment_32;
3930         int ret = 0;
3931
3932         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3933                            sizeof tss_segment_32))
3934                 goto out;
3935
3936         save_state_to_tss32(vcpu, &tss_segment_32);
3937
3938         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3939                             sizeof tss_segment_32))
3940                 goto out;
3941
3942         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3943                            &tss_segment_32, sizeof tss_segment_32))
3944                 goto out;
3945
3946         if (old_tss_sel != 0xffff) {
3947                 tss_segment_32.prev_task_link = old_tss_sel;
3948
3949                 if (kvm_write_guest(vcpu->kvm,
3950                                     get_tss_base_addr(vcpu, nseg_desc),
3951                                     &tss_segment_32.prev_task_link,
3952                                     sizeof tss_segment_32.prev_task_link))
3953                         goto out;
3954         }
3955
3956         if (load_state_from_tss32(vcpu, &tss_segment_32))
3957                 goto out;
3958
3959         ret = 1;
3960 out:
3961         return ret;
3962 }
3963
3964 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3965 {
3966         struct kvm_segment tr_seg;
3967         struct desc_struct cseg_desc;
3968         struct desc_struct nseg_desc;
3969         int ret = 0;
3970         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3971         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3972
3973         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3974
3975         /* FIXME: Handle errors. Failure to read either TSS or their
3976          * descriptors should generate a pagefault.
3977          */
3978         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3979                 goto out;
3980
3981         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3982                 goto out;
3983
3984         if (reason != TASK_SWITCH_IRET) {
3985                 int cpl;
3986
3987                 cpl = kvm_x86_ops->get_cpl(vcpu);
3988                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3989                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3990                         return 1;
3991                 }
3992         }
3993
3994         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3995                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3996                 return 1;
3997         }
3998
3999         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4000                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4001                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4002         }
4003
4004         if (reason == TASK_SWITCH_IRET) {
4005                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4006                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4007         }
4008
4009         /* set back link to prev task only if NT bit is set in eflags
4010            note that old_tss_sel is not used afetr this point */
4011         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4012                 old_tss_sel = 0xffff;
4013
4014         /* set back link to prev task only if NT bit is set in eflags
4015            note that old_tss_sel is not used afetr this point */
4016         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4017                 old_tss_sel = 0xffff;
4018
4019         if (nseg_desc.type & 8)
4020                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4021                                          old_tss_base, &nseg_desc);
4022         else
4023                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4024                                          old_tss_base, &nseg_desc);
4025
4026         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4027                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4028                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4029         }
4030
4031         if (reason != TASK_SWITCH_IRET) {
4032                 nseg_desc.type |= (1 << 1);
4033                 save_guest_segment_descriptor(vcpu, tss_selector,
4034                                               &nseg_desc);
4035         }
4036
4037         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4038         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4039         tr_seg.type = 11;
4040         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4041 out:
4042         return ret;
4043 }
4044 EXPORT_SYMBOL_GPL(kvm_task_switch);
4045
4046 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4047                                   struct kvm_sregs *sregs)
4048 {
4049         int mmu_reset_needed = 0;
4050         int i, pending_vec, max_bits;
4051         struct descriptor_table dt;
4052
4053         vcpu_load(vcpu);
4054
4055         dt.limit = sregs->idt.limit;
4056         dt.base = sregs->idt.base;
4057         kvm_x86_ops->set_idt(vcpu, &dt);
4058         dt.limit = sregs->gdt.limit;
4059         dt.base = sregs->gdt.base;
4060         kvm_x86_ops->set_gdt(vcpu, &dt);
4061
4062         vcpu->arch.cr2 = sregs->cr2;
4063         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4064
4065         down_read(&vcpu->kvm->slots_lock);
4066         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4067                 vcpu->arch.cr3 = sregs->cr3;
4068         else
4069                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4070         up_read(&vcpu->kvm->slots_lock);
4071
4072         kvm_set_cr8(vcpu, sregs->cr8);
4073
4074         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4075         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4076         kvm_set_apic_base(vcpu, sregs->apic_base);
4077
4078         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4079
4080         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4081         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4082         vcpu->arch.cr0 = sregs->cr0;
4083
4084         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4085         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4086         if (!is_long_mode(vcpu) && is_pae(vcpu))
4087                 load_pdptrs(vcpu, vcpu->arch.cr3);
4088
4089         if (mmu_reset_needed)
4090                 kvm_mmu_reset_context(vcpu);
4091
4092         if (!irqchip_in_kernel(vcpu->kvm)) {
4093                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
4094                        sizeof vcpu->arch.irq_pending);
4095                 vcpu->arch.irq_summary = 0;
4096                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
4097                         if (vcpu->arch.irq_pending[i])
4098                                 __set_bit(i, &vcpu->arch.irq_summary);
4099         } else {
4100                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4101                 pending_vec = find_first_bit(
4102                         (const unsigned long *)sregs->interrupt_bitmap,
4103                         max_bits);
4104                 /* Only pending external irq is handled here */
4105                 if (pending_vec < max_bits) {
4106                         kvm_queue_interrupt(vcpu, pending_vec);
4107                         pr_debug("Set back pending irq %d\n", pending_vec);
4108                 }
4109                 kvm_pic_clear_isr_ack(vcpu->kvm);
4110         }
4111
4112         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4113         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4114         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4115         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4116         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4117         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4118
4119         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4120         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4121
4122         /* Older userspace won't unhalt the vcpu on reset. */
4123         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4124             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4125             !(vcpu->arch.cr0 & X86_CR0_PE))
4126                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4127
4128         vcpu_put(vcpu);
4129
4130         return 0;
4131 }
4132
4133 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4134                                         struct kvm_guest_debug *dbg)
4135 {
4136         int i, r;
4137
4138         vcpu_load(vcpu);
4139
4140         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4141             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4142                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4143                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4144                 vcpu->arch.switch_db_regs =
4145                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4146         } else {
4147                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4148                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4149                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4150         }
4151
4152         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4153
4154         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4155                 kvm_queue_exception(vcpu, DB_VECTOR);
4156         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4157                 kvm_queue_exception(vcpu, BP_VECTOR);
4158
4159         vcpu_put(vcpu);
4160
4161         return r;
4162 }
4163
4164 /*
4165  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4166  * we have asm/x86/processor.h
4167  */
4168 struct fxsave {
4169         u16     cwd;
4170         u16     swd;
4171         u16     twd;
4172         u16     fop;
4173         u64     rip;
4174         u64     rdp;
4175         u32     mxcsr;
4176         u32     mxcsr_mask;
4177         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4178 #ifdef CONFIG_X86_64
4179         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4180 #else
4181         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4182 #endif
4183 };
4184
4185 /*
4186  * Translate a guest virtual address to a guest physical address.
4187  */
4188 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4189                                     struct kvm_translation *tr)
4190 {
4191         unsigned long vaddr = tr->linear_address;
4192         gpa_t gpa;
4193
4194         vcpu_load(vcpu);
4195         down_read(&vcpu->kvm->slots_lock);
4196         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4197         up_read(&vcpu->kvm->slots_lock);
4198         tr->physical_address = gpa;
4199         tr->valid = gpa != UNMAPPED_GVA;
4200         tr->writeable = 1;
4201         tr->usermode = 0;
4202         vcpu_put(vcpu);
4203
4204         return 0;
4205 }
4206
4207 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4208 {
4209         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4210
4211         vcpu_load(vcpu);
4212
4213         memcpy(fpu->fpr, fxsave->st_space, 128);
4214         fpu->fcw = fxsave->cwd;
4215         fpu->fsw = fxsave->swd;
4216         fpu->ftwx = fxsave->twd;
4217         fpu->last_opcode = fxsave->fop;
4218         fpu->last_ip = fxsave->rip;
4219         fpu->last_dp = fxsave->rdp;
4220         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4221
4222         vcpu_put(vcpu);
4223
4224         return 0;
4225 }
4226
4227 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4228 {
4229         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4230
4231         vcpu_load(vcpu);
4232
4233         memcpy(fxsave->st_space, fpu->fpr, 128);
4234         fxsave->cwd = fpu->fcw;
4235         fxsave->swd = fpu->fsw;
4236         fxsave->twd = fpu->ftwx;
4237         fxsave->fop = fpu->last_opcode;
4238         fxsave->rip = fpu->last_ip;
4239         fxsave->rdp = fpu->last_dp;
4240         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4241
4242         vcpu_put(vcpu);
4243
4244         return 0;
4245 }
4246
4247 void fx_init(struct kvm_vcpu *vcpu)
4248 {
4249         unsigned after_mxcsr_mask;
4250
4251         /*
4252          * Touch the fpu the first time in non atomic context as if
4253          * this is the first fpu instruction the exception handler
4254          * will fire before the instruction returns and it'll have to
4255          * allocate ram with GFP_KERNEL.
4256          */
4257         if (!used_math())
4258                 kvm_fx_save(&vcpu->arch.host_fx_image);
4259
4260         /* Initialize guest FPU by resetting ours and saving into guest's */
4261         preempt_disable();
4262         kvm_fx_save(&vcpu->arch.host_fx_image);
4263         kvm_fx_finit();
4264         kvm_fx_save(&vcpu->arch.guest_fx_image);
4265         kvm_fx_restore(&vcpu->arch.host_fx_image);
4266         preempt_enable();
4267
4268         vcpu->arch.cr0 |= X86_CR0_ET;
4269         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4270         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4271         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4272                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4273 }
4274 EXPORT_SYMBOL_GPL(fx_init);
4275
4276 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4277 {
4278         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4279                 return;
4280
4281         vcpu->guest_fpu_loaded = 1;
4282         kvm_fx_save(&vcpu->arch.host_fx_image);
4283         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4284 }
4285 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4286
4287 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4288 {
4289         if (!vcpu->guest_fpu_loaded)
4290                 return;
4291
4292         vcpu->guest_fpu_loaded = 0;
4293         kvm_fx_save(&vcpu->arch.guest_fx_image);
4294         kvm_fx_restore(&vcpu->arch.host_fx_image);
4295         ++vcpu->stat.fpu_reload;
4296 }
4297 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4298
4299 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4300 {
4301         if (vcpu->arch.time_page) {
4302                 kvm_release_page_dirty(vcpu->arch.time_page);
4303                 vcpu->arch.time_page = NULL;
4304         }
4305
4306         kvm_x86_ops->vcpu_free(vcpu);
4307 }
4308
4309 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4310                                                 unsigned int id)
4311 {
4312         return kvm_x86_ops->vcpu_create(kvm, id);
4313 }
4314
4315 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4316 {
4317         int r;
4318
4319         /* We do fxsave: this must be aligned. */
4320         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4321
4322         vcpu->arch.mtrr_state.have_fixed = 1;
4323         vcpu_load(vcpu);
4324         r = kvm_arch_vcpu_reset(vcpu);
4325         if (r == 0)
4326                 r = kvm_mmu_setup(vcpu);
4327         vcpu_put(vcpu);
4328         if (r < 0)
4329                 goto free_vcpu;
4330
4331         return 0;
4332 free_vcpu:
4333         kvm_x86_ops->vcpu_free(vcpu);
4334         return r;
4335 }
4336
4337 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4338 {
4339         vcpu_load(vcpu);
4340         kvm_mmu_unload(vcpu);
4341         vcpu_put(vcpu);
4342
4343         kvm_x86_ops->vcpu_free(vcpu);
4344 }
4345
4346 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4347 {
4348         vcpu->arch.nmi_pending = false;
4349         vcpu->arch.nmi_injected = false;
4350
4351         vcpu->arch.switch_db_regs = 0;
4352         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4353         vcpu->arch.dr6 = DR6_FIXED_1;
4354         vcpu->arch.dr7 = DR7_FIXED_1;
4355
4356         return kvm_x86_ops->vcpu_reset(vcpu);
4357 }
4358
4359 void kvm_arch_hardware_enable(void *garbage)
4360 {
4361         kvm_x86_ops->hardware_enable(garbage);
4362 }
4363
4364 void kvm_arch_hardware_disable(void *garbage)
4365 {
4366         kvm_x86_ops->hardware_disable(garbage);
4367 }
4368
4369 int kvm_arch_hardware_setup(void)
4370 {
4371         return kvm_x86_ops->hardware_setup();
4372 }
4373
4374 void kvm_arch_hardware_unsetup(void)
4375 {
4376         kvm_x86_ops->hardware_unsetup();
4377 }
4378
4379 void kvm_arch_check_processor_compat(void *rtn)
4380 {
4381         kvm_x86_ops->check_processor_compatibility(rtn);
4382 }
4383
4384 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4385 {
4386         struct page *page;
4387         struct kvm *kvm;
4388         int r;
4389
4390         BUG_ON(vcpu->kvm == NULL);
4391         kvm = vcpu->kvm;
4392
4393         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4394         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4395                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4396         else
4397                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4398
4399         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4400         if (!page) {
4401                 r = -ENOMEM;
4402                 goto fail;
4403         }
4404         vcpu->arch.pio_data = page_address(page);
4405
4406         r = kvm_mmu_create(vcpu);
4407         if (r < 0)
4408                 goto fail_free_pio_data;
4409
4410         if (irqchip_in_kernel(kvm)) {
4411                 r = kvm_create_lapic(vcpu);
4412                 if (r < 0)
4413                         goto fail_mmu_destroy;
4414         }
4415
4416         return 0;
4417
4418 fail_mmu_destroy:
4419         kvm_mmu_destroy(vcpu);
4420 fail_free_pio_data:
4421         free_page((unsigned long)vcpu->arch.pio_data);
4422 fail:
4423         return r;
4424 }
4425
4426 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4427 {
4428         kvm_free_lapic(vcpu);
4429         down_read(&vcpu->kvm->slots_lock);
4430         kvm_mmu_destroy(vcpu);
4431         up_read(&vcpu->kvm->slots_lock);
4432         free_page((unsigned long)vcpu->arch.pio_data);
4433 }
4434
4435 struct  kvm *kvm_arch_create_vm(void)
4436 {
4437         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4438
4439         if (!kvm)
4440                 return ERR_PTR(-ENOMEM);
4441
4442         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4443         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4444
4445         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4446         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4447
4448         rdtscll(kvm->arch.vm_init_tsc);
4449
4450         return kvm;
4451 }
4452
4453 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4454 {
4455         vcpu_load(vcpu);
4456         kvm_mmu_unload(vcpu);
4457         vcpu_put(vcpu);
4458 }
4459
4460 static void kvm_free_vcpus(struct kvm *kvm)
4461 {
4462         unsigned int i;
4463
4464         /*
4465          * Unpin any mmu pages first.
4466          */
4467         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4468                 if (kvm->vcpus[i])
4469                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4470         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4471                 if (kvm->vcpus[i]) {
4472                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4473                         kvm->vcpus[i] = NULL;
4474                 }
4475         }
4476
4477 }
4478
4479 void kvm_arch_sync_events(struct kvm *kvm)
4480 {
4481         kvm_free_all_assigned_devices(kvm);
4482 }
4483
4484 void kvm_arch_destroy_vm(struct kvm *kvm)
4485 {
4486         kvm_iommu_unmap_guest(kvm);
4487         kvm_free_pit(kvm);
4488         kfree(kvm->arch.vpic);
4489         kfree(kvm->arch.vioapic);
4490         kvm_free_vcpus(kvm);
4491         kvm_free_physmem(kvm);
4492         if (kvm->arch.apic_access_page)
4493                 put_page(kvm->arch.apic_access_page);
4494         if (kvm->arch.ept_identity_pagetable)
4495                 put_page(kvm->arch.ept_identity_pagetable);
4496         kfree(kvm);
4497 }
4498
4499 int kvm_arch_set_memory_region(struct kvm *kvm,
4500                                 struct kvm_userspace_memory_region *mem,
4501                                 struct kvm_memory_slot old,
4502                                 int user_alloc)
4503 {
4504         int npages = mem->memory_size >> PAGE_SHIFT;
4505         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4506
4507         /*To keep backward compatibility with older userspace,
4508          *x86 needs to hanlde !user_alloc case.
4509          */
4510         if (!user_alloc) {
4511                 if (npages && !old.rmap) {
4512                         unsigned long userspace_addr;
4513
4514                         down_write(&current->mm->mmap_sem);
4515                         userspace_addr = do_mmap(NULL, 0,
4516                                                  npages * PAGE_SIZE,
4517                                                  PROT_READ | PROT_WRITE,
4518                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4519                                                  0);
4520                         up_write(&current->mm->mmap_sem);
4521
4522                         if (IS_ERR((void *)userspace_addr))
4523                                 return PTR_ERR((void *)userspace_addr);
4524
4525                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4526                         spin_lock(&kvm->mmu_lock);
4527                         memslot->userspace_addr = userspace_addr;
4528                         spin_unlock(&kvm->mmu_lock);
4529                 } else {
4530                         if (!old.user_alloc && old.rmap) {
4531                                 int ret;
4532
4533                                 down_write(&current->mm->mmap_sem);
4534                                 ret = do_munmap(current->mm, old.userspace_addr,
4535                                                 old.npages * PAGE_SIZE);
4536                                 up_write(&current->mm->mmap_sem);
4537                                 if (ret < 0)
4538                                         printk(KERN_WARNING
4539                                        "kvm_vm_ioctl_set_memory_region: "
4540                                        "failed to munmap memory\n");
4541                         }
4542                 }
4543         }
4544
4545         if (!kvm->arch.n_requested_mmu_pages) {
4546                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4547                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4548         }
4549
4550         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4551         kvm_flush_remote_tlbs(kvm);
4552
4553         return 0;
4554 }
4555
4556 void kvm_arch_flush_shadow(struct kvm *kvm)
4557 {
4558         kvm_mmu_zap_all(kvm);
4559 }
4560
4561 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4562 {
4563         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4564                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4565                || vcpu->arch.nmi_pending;
4566 }
4567
4568 static void vcpu_kick_intr(void *info)
4569 {
4570 #ifdef DEBUG
4571         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4572         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4573 #endif
4574 }
4575
4576 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4577 {
4578         int ipi_pcpu = vcpu->cpu;
4579         int cpu;
4580
4581         if (waitqueue_active(&vcpu->wq)) {
4582                 wake_up_interruptible(&vcpu->wq);
4583                 ++vcpu->stat.halt_wakeup;
4584         }
4585         /*
4586          * We may be called synchronously with irqs disabled in guest mode,
4587          * So need not to call smp_call_function_single() in that case.
4588          */
4589         cpu = get_cpu();
4590         if (vcpu->guest_mode && vcpu->cpu != cpu)
4591                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4592         put_cpu();
4593 }
4594
4595 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4596 {
4597         return kvm_x86_ops->interrupt_allowed(vcpu);
4598 }