SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/msr.h>
  43 #include <asm/desc.h>
  44 #include <asm/mtrr.h>
  45
  46 #define MAX_IO_MSRS 256
  47 #define CR0_RESERVED_BITS                                               \
  48         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  49                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  50                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  51 #define CR4_RESERVED_BITS                                               \
  52         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  53                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  54                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  55                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  56
  57 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  58 /* EFER defaults:
  59  * - enable syscall per default because its emulated by KVM
  60  * - enable LME and LMA per default on 64 bit KVM
  61  */
  62 #ifdef CONFIG_X86_64
  63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  64 #else
  65 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  66 #endif
  67
  68 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  69 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  70
  71 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  72                                     struct kvm_cpuid_entry2 __user *entries);
  73 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
  74                                               u32 function, u32 index);
  75
  76 struct kvm_x86_ops *kvm_x86_ops;
  77 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  78
  79 struct kvm_stats_debugfs_item debugfs_entries[] = {
  80         { "pf_fixed", VCPU_STAT(pf_fixed) },
  81         { "pf_guest", VCPU_STAT(pf_guest) },
  82         { "tlb_flush", VCPU_STAT(tlb_flush) },
  83         { "invlpg", VCPU_STAT(invlpg) },
  84         { "exits", VCPU_STAT(exits) },
  85         { "io_exits", VCPU_STAT(io_exits) },
  86         { "mmio_exits", VCPU_STAT(mmio_exits) },
  87         { "signal_exits", VCPU_STAT(signal_exits) },
  88         { "irq_window", VCPU_STAT(irq_window_exits) },
  89         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  90         { "halt_exits", VCPU_STAT(halt_exits) },
  91         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  92         { "hypercalls", VCPU_STAT(hypercalls) },
  93         { "request_irq", VCPU_STAT(request_irq_exits) },
  94         { "request_nmi", VCPU_STAT(request_nmi_exits) },
  95         { "irq_exits", VCPU_STAT(irq_exits) },
  96         { "host_state_reload", VCPU_STAT(host_state_reload) },
  97         { "efer_reload", VCPU_STAT(efer_reload) },
  98         { "fpu_reload", VCPU_STAT(fpu_reload) },
  99         { "insn_emulation", VCPU_STAT(insn_emulation) },
 100         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 101         { "irq_injections", VCPU_STAT(irq_injections) },
 102         { "nmi_injections", VCPU_STAT(nmi_injections) },
 103         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 104         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 105         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 106         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 107         { "mmu_flooded", VM_STAT(mmu_flooded) },
 108         { "mmu_recycled", VM_STAT(mmu_recycled) },
 109         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 110         { "mmu_unsync", VM_STAT(mmu_unsync) },
 111         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 112         { "largepages", VM_STAT(lpages) },
 113         { NULL }
 114 };
 115
 116 unsigned long segment_base(u16 selector)
 117 {
 118         struct descriptor_table gdt;
 119         struct desc_struct *d;
 120         unsigned long table_base;
 121         unsigned long v;
 122
 123         if (selector == 0)
 124                 return 0;
 125
 126         asm("sgdt %0" : "=m"(gdt));
 127         table_base = gdt.base;
 128
 129         if (selector & 4) {           /* from ldt */
 130                 u16 ldt_selector;
 131
 132                 asm("sldt %0" : "=g"(ldt_selector));
 133                 table_base = segment_base(ldt_selector);
 134         }
 135         d = (struct desc_struct *)(table_base + (selector & ~7));
 136         v = d->base0 | ((unsigned long)d->base1 << 16) |
 137                 ((unsigned long)d->base2 << 24);
 138 #ifdef CONFIG_X86_64
 139         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 140                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 141 #endif
 142         return v;
 143 }
 144 EXPORT_SYMBOL_GPL(segment_base);
 145
 146 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 147 {
 148         if (irqchip_in_kernel(vcpu->kvm))
 149                 return vcpu->arch.apic_base;
 150         else
 151                 return vcpu->arch.apic_base;
 152 }
 153 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 154
 155 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 156 {
 157         /* TODO: reserve bits check */
 158         if (irqchip_in_kernel(vcpu->kvm))
 159                 kvm_lapic_set_base(vcpu, data);
 160         else
 161                 vcpu->arch.apic_base = data;
 162 }
 163 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 164
 165 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 166 {
 167         WARN_ON(vcpu->arch.exception.pending);
 168         vcpu->arch.exception.pending = true;
 169         vcpu->arch.exception.has_error_code = false;
 170         vcpu->arch.exception.nr = nr;
 171 }
 172 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 173
 174 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 175                            u32 error_code)
 176 {
 177         ++vcpu->stat.pf_guest;
 178
 179         if (vcpu->arch.exception.pending) {
 180                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 181                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 182                                         " double fault 0x%lx\n", addr);
 183                         vcpu->arch.exception.nr = DF_VECTOR;
 184                         vcpu->arch.exception.error_code = 0;
 185                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 186                         /* triple fault -> shutdown */
 187                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 188                 }
 189                 return;
 190         }
 191         vcpu->arch.cr2 = addr;
 192         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 193 }
 194
 195 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 196 {
 197         vcpu->arch.nmi_pending = 1;
 198 }
 199 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 200
 201 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 202 {
 203         WARN_ON(vcpu->arch.exception.pending);
 204         vcpu->arch.exception.pending = true;
 205         vcpu->arch.exception.has_error_code = true;
 206         vcpu->arch.exception.nr = nr;
 207         vcpu->arch.exception.error_code = error_code;
 208 }
 209 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 210
 211 static void __queue_exception(struct kvm_vcpu *vcpu)
 212 {
 213         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 214                                      vcpu->arch.exception.has_error_code,
 215                                      vcpu->arch.exception.error_code);
 216 }
 217
 218 /*
 219  * Load the pae pdptrs.  Return true is they are all valid.
 220  */
 221 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 222 {
 223         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 224         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 225         int i;
 226         int ret;
 227         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 228
 229         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 230                                   offset * sizeof(u64), sizeof(pdpte));
 231         if (ret < 0) {
 232                 ret = 0;
 233                 goto out;
 234         }
 235         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 236                 if (is_present_pte(pdpte[i]) &&
 237                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 238                         ret = 0;
 239                         goto out;
 240                 }
 241         }
 242         ret = 1;
 243
 244         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 245 out:
 246
 247         return ret;
 248 }
 249 EXPORT_SYMBOL_GPL(load_pdptrs);
 250
 251 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 252 {
 253         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 254         bool changed = true;
 255         int r;
 256
 257         if (is_long_mode(vcpu) || !is_pae(vcpu))
 258                 return false;
 259
 260         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 261         if (r < 0)
 262                 goto out;
 263         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 264 out:
 265
 266         return changed;
 267 }
 268
 269 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 270 {
 271         if (cr0 & CR0_RESERVED_BITS) {
 272                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 273                        cr0, vcpu->arch.cr0);
 274                 kvm_inject_gp(vcpu, 0);
 275                 return;
 276         }
 277
 278         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 279                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 280                 kvm_inject_gp(vcpu, 0);
 281                 return;
 282         }
 283
 284         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 285                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 286                        "and a clear PE flag\n");
 287                 kvm_inject_gp(vcpu, 0);
 288                 return;
 289         }
 290
 291         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 292 #ifdef CONFIG_X86_64
 293                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 294                         int cs_db, cs_l;
 295
 296                         if (!is_pae(vcpu)) {
 297                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 298                                        "in long mode while PAE is disabled\n");
 299                                 kvm_inject_gp(vcpu, 0);
 300                                 return;
 301                         }
 302                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 303                         if (cs_l) {
 304                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 305                                        "in long mode while CS.L == 1\n");
 306                                 kvm_inject_gp(vcpu, 0);
 307                                 return;
 308
 309                         }
 310                 } else
 311 #endif
 312                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 313                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 314                                "reserved bits\n");
 315                         kvm_inject_gp(vcpu, 0);
 316                         return;
 317                 }
 318
 319         }
 320
 321         kvm_x86_ops->set_cr0(vcpu, cr0);
 322         vcpu->arch.cr0 = cr0;
 323
 324         kvm_mmu_reset_context(vcpu);
 325         return;
 326 }
 327 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 328
 329 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 330 {
 331         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 332         KVMTRACE_1D(LMSW, vcpu,
 333                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 334                     handler);
 335 }
 336 EXPORT_SYMBOL_GPL(kvm_lmsw);
 337
 338 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 339 {
 340         unsigned long old_cr4 = vcpu->arch.cr4;
 341         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 342
 343         if (cr4 & CR4_RESERVED_BITS) {
 344                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 345                 kvm_inject_gp(vcpu, 0);
 346                 return;
 347         }
 348
 349         if (is_long_mode(vcpu)) {
 350                 if (!(cr4 & X86_CR4_PAE)) {
 351                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 352                                "in long mode\n");
 353                         kvm_inject_gp(vcpu, 0);
 354                         return;
 355                 }
 356         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 357                    && ((cr4 ^ old_cr4) & pdptr_bits)
 358                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 359                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 360                 kvm_inject_gp(vcpu, 0);
 361                 return;
 362         }
 363
 364         if (cr4 & X86_CR4_VMXE) {
 365                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 366                 kvm_inject_gp(vcpu, 0);
 367                 return;
 368         }
 369         kvm_x86_ops->set_cr4(vcpu, cr4);
 370         vcpu->arch.cr4 = cr4;
 371         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 372         kvm_mmu_reset_context(vcpu);
 373 }
 374 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 375
 376 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 377 {
 378         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 379                 kvm_mmu_sync_roots(vcpu);
 380                 kvm_mmu_flush_tlb(vcpu);
 381                 return;
 382         }
 383
 384         if (is_long_mode(vcpu)) {
 385                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 386                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 387                         kvm_inject_gp(vcpu, 0);
 388                         return;
 389                 }
 390         } else {
 391                 if (is_pae(vcpu)) {
 392                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 393                                 printk(KERN_DEBUG
 394                                        "set_cr3: #GP, reserved bits\n");
 395                                 kvm_inject_gp(vcpu, 0);
 396                                 return;
 397                         }
 398                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 399                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 400                                        "reserved bits\n");
 401                                 kvm_inject_gp(vcpu, 0);
 402                                 return;
 403                         }
 404                 }
 405                 /*
 406                  * We don't check reserved bits in nonpae mode, because
 407                  * this isn't enforced, and VMware depends on this.
 408                  */
 409         }
 410
 411         /*
 412          * Does the new cr3 value map to physical memory? (Note, we
 413          * catch an invalid cr3 even in real-mode, because it would
 414          * cause trouble later on when we turn on paging anyway.)
 415          *
 416          * A real CPU would silently accept an invalid cr3 and would
 417          * attempt to use it - with largely undefined (and often hard
 418          * to debug) behavior on the guest side.
 419          */
 420         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 421                 kvm_inject_gp(vcpu, 0);
 422         else {
 423                 vcpu->arch.cr3 = cr3;
 424                 vcpu->arch.mmu.new_cr3(vcpu);
 425         }
 426 }
 427 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 428
 429 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 430 {
 431         if (cr8 & CR8_RESERVED_BITS) {
 432                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 433                 kvm_inject_gp(vcpu, 0);
 434                 return;
 435         }
 436         if (irqchip_in_kernel(vcpu->kvm))
 437                 kvm_lapic_set_tpr(vcpu, cr8);
 438         else
 439                 vcpu->arch.cr8 = cr8;
 440 }
 441 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 442
 443 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 444 {
 445         if (irqchip_in_kernel(vcpu->kvm))
 446                 return kvm_lapic_get_cr8(vcpu);
 447         else
 448                 return vcpu->arch.cr8;
 449 }
 450 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 451
 452 static inline u32 bit(int bitno)
 453 {
 454         return 1 << (bitno & 31);
 455 }
 456
 457 /*
 458  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 459  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 460  *
 461  * This list is modified at module load time to reflect the
 462  * capabilities of the host cpu.
 463  */
 464 static u32 msrs_to_save[] = {
 465         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 466         MSR_K6_STAR,
 467 #ifdef CONFIG_X86_64
 468         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 469 #endif
 470         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 471         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 472 };
 473
 474 static unsigned num_msrs_to_save;
 475
 476 static u32 emulated_msrs[] = {
 477         MSR_IA32_MISC_ENABLE,
 478 };
 479
 480 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 481 {
 482         if (efer & efer_reserved_bits) {
 483                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 484                        efer);
 485                 kvm_inject_gp(vcpu, 0);
 486                 return;
 487         }
 488
 489         if (is_paging(vcpu)
 490             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 491                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 492                 kvm_inject_gp(vcpu, 0);
 493                 return;
 494         }
 495
 496         if (efer & EFER_FFXSR) {
 497                 struct kvm_cpuid_entry2 *feat;
 498
 499                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 500                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 501                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 502                         kvm_inject_gp(vcpu, 0);
 503                         return;
 504                 }
 505         }
 506
 507         if (efer & EFER_SVME) {
 508                 struct kvm_cpuid_entry2 *feat;
 509
 510                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 511                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 512                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 513                         kvm_inject_gp(vcpu, 0);
 514                         return;
 515                 }
 516         }
 517
 518         kvm_x86_ops->set_efer(vcpu, efer);
 519
 520         efer &= ~EFER_LMA;
 521         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 522
 523         vcpu->arch.shadow_efer = efer;
 524
 525         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 526         kvm_mmu_reset_context(vcpu);
 527 }
 528
 529 void kvm_enable_efer_bits(u64 mask)
 530 {
 531        efer_reserved_bits &= ~mask;
 532 }
 533 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 534
 535
 536 /*
 537  * Writes msr value into into the appropriate "register".
 538  * Returns 0 on success, non-0 otherwise.
 539  * Assumes vcpu_load() was already called.
 540  */
 541 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 542 {
 543         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 544 }
 545
 546 /*
 547  * Adapt set_msr() to msr_io()'s calling convention
 548  */
 549 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 550 {
 551         return kvm_set_msr(vcpu, index, *data);
 552 }
 553
 554 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 555 {
 556         static int version;
 557         struct pvclock_wall_clock wc;
 558         struct timespec now, sys, boot;
 559
 560         if (!wall_clock)
 561                 return;
 562
 563         version++;
 564
 565         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 566
 567         /*
 568          * The guest calculates current wall clock time by adding
 569          * system time (updated by kvm_write_guest_time below) to the
 570          * wall clock specified here.  guest system time equals host
 571          * system time for us, thus we must fill in host boot time here.
 572          */
 573         now = current_kernel_time();
 574         ktime_get_ts(&sys);
 575         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 576
 577         wc.sec = boot.tv_sec;
 578         wc.nsec = boot.tv_nsec;
 579         wc.version = version;
 580
 581         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 582
 583         version++;
 584         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 585 }
 586
 587 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 588 {
 589         uint32_t quotient, remainder;
 590
 591         /* Don't try to replace with do_div(), this one calculates
 592          * "(dividend << 32) / divisor" */
 593         __asm__ ( "divl %4"
 594                   : "=a" (quotient), "=d" (remainder)
 595                   : "0" (0), "1" (dividend), "r" (divisor) );
 596         return quotient;
 597 }
 598
 599 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 600 {
 601         uint64_t nsecs = 1000000000LL;
 602         int32_t  shift = 0;
 603         uint64_t tps64;
 604         uint32_t tps32;
 605
 606         tps64 = tsc_khz * 1000LL;
 607         while (tps64 > nsecs*2) {
 608                 tps64 >>= 1;
 609                 shift--;
 610         }
 611
 612         tps32 = (uint32_t)tps64;
 613         while (tps32 <= (uint32_t)nsecs) {
 614                 tps32 <<= 1;
 615                 shift++;
 616         }
 617
 618         hv_clock->tsc_shift = shift;
 619         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 620
 621         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 622                  __func__, tsc_khz, hv_clock->tsc_shift,
 623                  hv_clock->tsc_to_system_mul);
 624 }
 625
 626 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 627
 628 static void kvm_write_guest_time(struct kvm_vcpu *v)
 629 {
 630         struct timespec ts;
 631         unsigned long flags;
 632         struct kvm_vcpu_arch *vcpu = &v->arch;
 633         void *shared_kaddr;
 634         unsigned long this_tsc_khz;
 635
 636         if ((!vcpu->time_page))
 637                 return;
 638
 639         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 640         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 641                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 642                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 643         }
 644         put_cpu_var(cpu_tsc_khz);
 645
 646         /* Keep irq disabled to prevent changes to the clock */
 647         local_irq_save(flags);
 648         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 649                           &vcpu->hv_clock.tsc_timestamp);
 650         ktime_get_ts(&ts);
 651         local_irq_restore(flags);
 652
 653         /* With all the info we got, fill in the values */
 654
 655         vcpu->hv_clock.system_time = ts.tv_nsec +
 656                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 657         /*
 658          * The interface expects us to write an even number signaling that the
 659          * update is finished. Since the guest won't see the intermediate
 660          * state, we just increase by 2 at the end.
 661          */
 662         vcpu->hv_clock.version += 2;
 663
 664         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 665
 666         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 667                sizeof(vcpu->hv_clock));
 668
 669         kunmap_atomic(shared_kaddr, KM_USER0);
 670
 671         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 672 }
 673
 674 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 675 {
 676         struct kvm_vcpu_arch *vcpu = &v->arch;
 677
 678         if (!vcpu->time_page)
 679                 return 0;
 680         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 681         return 1;
 682 }
 683
 684 static bool msr_mtrr_valid(unsigned msr)
 685 {
 686         switch (msr) {
 687         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 688         case MSR_MTRRfix64K_00000:
 689         case MSR_MTRRfix16K_80000:
 690         case MSR_MTRRfix16K_A0000:
 691         case MSR_MTRRfix4K_C0000:
 692         case MSR_MTRRfix4K_C8000:
 693         case MSR_MTRRfix4K_D0000:
 694         case MSR_MTRRfix4K_D8000:
 695         case MSR_MTRRfix4K_E0000:
 696         case MSR_MTRRfix4K_E8000:
 697         case MSR_MTRRfix4K_F0000:
 698         case MSR_MTRRfix4K_F8000:
 699         case MSR_MTRRdefType:
 700         case MSR_IA32_CR_PAT:
 701                 return true;
 702         case 0x2f8:
 703                 return true;
 704         }
 705         return false;
 706 }
 707
 708 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 709 {
 710         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 711
 712         if (!msr_mtrr_valid(msr))
 713                 return 1;
 714
 715         if (msr == MSR_MTRRdefType) {
 716                 vcpu->arch.mtrr_state.def_type = data;
 717                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 718         } else if (msr == MSR_MTRRfix64K_00000)
 719                 p[0] = data;
 720         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 721                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 722         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 723                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 724         else if (msr == MSR_IA32_CR_PAT)
 725                 vcpu->arch.pat = data;
 726         else {  /* Variable MTRRs */
 727                 int idx, is_mtrr_mask;
 728                 u64 *pt;
 729
 730                 idx = (msr - 0x200) / 2;
 731                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 732                 if (!is_mtrr_mask)
 733                         pt =
 734                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 735                 else
 736                         pt =
 737                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 738                 *pt = data;
 739         }
 740
 741         kvm_mmu_reset_context(vcpu);
 742         return 0;
 743 }
 744
 745 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 746 {
 747         switch (msr) {
 748         case MSR_EFER:
 749                 set_efer(vcpu, data);
 750                 break;
 751         case MSR_IA32_MC0_STATUS:
 752                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 753                        __func__, data);
 754                 break;
 755         case MSR_IA32_MCG_STATUS:
 756                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 757                         __func__, data);
 758                 break;
 759         case MSR_IA32_MCG_CTL:
 760                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 761                         __func__, data);
 762                 break;
 763         case MSR_IA32_DEBUGCTLMSR:
 764                 if (!data) {
 765                         /* We support the non-activated case already */
 766                         break;
 767                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 768                         /* Values other than LBR and BTF are vendor-specific,
 769                            thus reserved and should throw a #GP */
 770                         return 1;
 771                 }
 772                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 773                         __func__, data);
 774                 break;
 775         case MSR_IA32_UCODE_REV:
 776         case MSR_IA32_UCODE_WRITE:
 777         case MSR_VM_HSAVE_PA:
 778                 break;
 779         case 0x200 ... 0x2ff:
 780                 return set_msr_mtrr(vcpu, msr, data);
 781         case MSR_IA32_APICBASE:
 782                 kvm_set_apic_base(vcpu, data);
 783                 break;
 784         case MSR_IA32_MISC_ENABLE:
 785                 vcpu->arch.ia32_misc_enable_msr = data;
 786                 break;
 787         case MSR_KVM_WALL_CLOCK:
 788                 vcpu->kvm->arch.wall_clock = data;
 789                 kvm_write_wall_clock(vcpu->kvm, data);
 790                 break;
 791         case MSR_KVM_SYSTEM_TIME: {
 792                 if (vcpu->arch.time_page) {
 793                         kvm_release_page_dirty(vcpu->arch.time_page);
 794                         vcpu->arch.time_page = NULL;
 795                 }
 796
 797                 vcpu->arch.time = data;
 798
 799                 /* we verify if the enable bit is set... */
 800                 if (!(data & 1))
 801                         break;
 802
 803                 /* ...but clean it before doing the actual write */
 804                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 805
 806                 vcpu->arch.time_page =
 807                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 808
 809                 if (is_error_page(vcpu->arch.time_page)) {
 810                         kvm_release_page_clean(vcpu->arch.time_page);
 811                         vcpu->arch.time_page = NULL;
 812                 }
 813
 814                 kvm_request_guest_time_update(vcpu);
 815                 break;
 816         }
 817         default:
 818                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 819                 return 1;
 820         }
 821         return 0;
 822 }
 823 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 824
 825
 826 /*
 827  * Reads an msr value (of 'msr_index') into 'pdata'.
 828  * Returns 0 on success, non-0 otherwise.
 829  * Assumes vcpu_load() was already called.
 830  */
 831 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 832 {
 833         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 834 }
 835
 836 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 837 {
 838         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 839
 840         if (!msr_mtrr_valid(msr))
 841                 return 1;
 842
 843         if (msr == MSR_MTRRdefType)
 844                 *pdata = vcpu->arch.mtrr_state.def_type +
 845                          (vcpu->arch.mtrr_state.enabled << 10);
 846         else if (msr == MSR_MTRRfix64K_00000)
 847                 *pdata = p[0];
 848         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 849                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 850         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 851                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 852         else if (msr == MSR_IA32_CR_PAT)
 853                 *pdata = vcpu->arch.pat;
 854         else {  /* Variable MTRRs */
 855                 int idx, is_mtrr_mask;
 856                 u64 *pt;
 857
 858                 idx = (msr - 0x200) / 2;
 859                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 860                 if (!is_mtrr_mask)
 861                         pt =
 862                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 863                 else
 864                         pt =
 865                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 866                 *pdata = *pt;
 867         }
 868
 869         return 0;
 870 }
 871
 872 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 873 {
 874         u64 data;
 875
 876         switch (msr) {
 877         case 0xc0010010: /* SYSCFG */
 878         case 0xc0010015: /* HWCR */
 879         case MSR_IA32_PLATFORM_ID:
 880         case MSR_IA32_P5_MC_ADDR:
 881         case MSR_IA32_P5_MC_TYPE:
 882         case MSR_IA32_MC0_CTL:
 883         case MSR_IA32_MCG_STATUS:
 884         case MSR_IA32_MCG_CAP:
 885         case MSR_IA32_MCG_CTL:
 886         case MSR_IA32_MC0_MISC:
 887         case MSR_IA32_MC0_MISC+4:
 888         case MSR_IA32_MC0_MISC+8:
 889         case MSR_IA32_MC0_MISC+12:
 890         case MSR_IA32_MC0_MISC+16:
 891         case MSR_IA32_MC0_MISC+20:
 892         case MSR_IA32_UCODE_REV:
 893         case MSR_IA32_EBL_CR_POWERON:
 894         case MSR_IA32_DEBUGCTLMSR:
 895         case MSR_IA32_LASTBRANCHFROMIP:
 896         case MSR_IA32_LASTBRANCHTOIP:
 897         case MSR_IA32_LASTINTFROMIP:
 898         case MSR_IA32_LASTINTTOIP:
 899         case MSR_VM_HSAVE_PA:
 900         case MSR_P6_EVNTSEL0:
 901         case MSR_P6_EVNTSEL1:
 902                 data = 0;
 903                 break;
 904         case MSR_MTRRcap:
 905                 data = 0x500 | KVM_NR_VAR_MTRR;
 906                 break;
 907         case 0x200 ... 0x2ff:
 908                 return get_msr_mtrr(vcpu, msr, pdata);
 909         case 0xcd: /* fsb frequency */
 910                 data = 3;
 911                 break;
 912         case MSR_IA32_APICBASE:
 913                 data = kvm_get_apic_base(vcpu);
 914                 break;
 915         case MSR_IA32_MISC_ENABLE:
 916                 data = vcpu->arch.ia32_misc_enable_msr;
 917                 break;
 918         case MSR_IA32_PERF_STATUS:
 919                 /* TSC increment by tick */
 920                 data = 1000ULL;
 921                 /* CPU multiplier */
 922                 data |= (((uint64_t)4ULL) << 40);
 923                 break;
 924         case MSR_EFER:
 925                 data = vcpu->arch.shadow_efer;
 926                 break;
 927         case MSR_KVM_WALL_CLOCK:
 928                 data = vcpu->kvm->arch.wall_clock;
 929                 break;
 930         case MSR_KVM_SYSTEM_TIME:
 931                 data = vcpu->arch.time;
 932                 break;
 933         default:
 934                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 935                 return 1;
 936         }
 937         *pdata = data;
 938         return 0;
 939 }
 940 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 941
 942 /*
 943  * Read or write a bunch of msrs. All parameters are kernel addresses.
 944  *
 945  * @return number of msrs set successfully.
 946  */
 947 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 948                     struct kvm_msr_entry *entries,
 949                     int (*do_msr)(struct kvm_vcpu *vcpu,
 950                                   unsigned index, u64 *data))
 951 {
 952         int i;
 953
 954         vcpu_load(vcpu);
 955
 956         down_read(&vcpu->kvm->slots_lock);
 957         for (i = 0; i < msrs->nmsrs; ++i)
 958                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 959                         break;
 960         up_read(&vcpu->kvm->slots_lock);
 961
 962         vcpu_put(vcpu);
 963
 964         return i;
 965 }
 966
 967 /*
 968  * Read or write a bunch of msrs. Parameters are user addresses.
 969  *
 970  * @return number of msrs set successfully.
 971  */
 972 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 973                   int (*do_msr)(struct kvm_vcpu *vcpu,
 974                                 unsigned index, u64 *data),
 975                   int writeback)
 976 {
 977         struct kvm_msrs msrs;
 978         struct kvm_msr_entry *entries;
 979         int r, n;
 980         unsigned size;
 981
 982         r = -EFAULT;
 983         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 984                 goto out;
 985
 986         r = -E2BIG;
 987         if (msrs.nmsrs >= MAX_IO_MSRS)
 988                 goto out;
 989
 990         r = -ENOMEM;
 991         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 992         entries = vmalloc(size);
 993         if (!entries)
 994                 goto out;
 995
 996         r = -EFAULT;
 997         if (copy_from_user(entries, user_msrs->entries, size))
 998                 goto out_free;
 999
1000         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1001         if (r < 0)
1002                 goto out_free;
1003
1004         r = -EFAULT;
1005         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1006                 goto out_free;
1007
1008         r = n;
1009
1010 out_free:
1011         vfree(entries);
1012 out:
1013         return r;
1014 }
1015
1016 int kvm_dev_ioctl_check_extension(long ext)
1017 {
1018         int r;
1019
1020         switch (ext) {
1021         case KVM_CAP_IRQCHIP:
1022         case KVM_CAP_HLT:
1023         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1024         case KVM_CAP_SET_TSS_ADDR:
1025         case KVM_CAP_EXT_CPUID:
1026         case KVM_CAP_CLOCKSOURCE:
1027         case KVM_CAP_PIT:
1028         case KVM_CAP_NOP_IO_DELAY:
1029         case KVM_CAP_MP_STATE:
1030         case KVM_CAP_SYNC_MMU:
1031         case KVM_CAP_REINJECT_CONTROL:
1032         case KVM_CAP_IRQ_INJECT_STATUS:
1033         case KVM_CAP_ASSIGN_DEV_IRQ:
1034                 r = 1;
1035                 break;
1036         case KVM_CAP_COALESCED_MMIO:
1037                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1038                 break;
1039         case KVM_CAP_VAPIC:
1040                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1041                 break;
1042         case KVM_CAP_NR_VCPUS:
1043                 r = KVM_MAX_VCPUS;
1044                 break;
1045         case KVM_CAP_NR_MEMSLOTS:
1046                 r = KVM_MEMORY_SLOTS;
1047                 break;
1048         case KVM_CAP_PV_MMU:
1049                 r = !tdp_enabled;
1050                 break;
1051         case KVM_CAP_IOMMU:
1052                 r = iommu_found();
1053                 break;
1054         default:
1055                 r = 0;
1056                 break;
1057         }
1058         return r;
1059
1060 }
1061
1062 long kvm_arch_dev_ioctl(struct file *filp,
1063                         unsigned int ioctl, unsigned long arg)
1064 {
1065         void __user *argp = (void __user *)arg;
1066         long r;
1067
1068         switch (ioctl) {
1069         case KVM_GET_MSR_INDEX_LIST: {
1070                 struct kvm_msr_list __user *user_msr_list = argp;
1071                 struct kvm_msr_list msr_list;
1072                 unsigned n;
1073
1074                 r = -EFAULT;
1075                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1076                         goto out;
1077                 n = msr_list.nmsrs;
1078                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1079                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1080                         goto out;
1081                 r = -E2BIG;
1082                 if (n < num_msrs_to_save)
1083                         goto out;
1084                 r = -EFAULT;
1085                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1086                                  num_msrs_to_save * sizeof(u32)))
1087                         goto out;
1088                 if (copy_to_user(user_msr_list->indices
1089                                  + num_msrs_to_save * sizeof(u32),
1090                                  &emulated_msrs,
1091                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1092                         goto out;
1093                 r = 0;
1094                 break;
1095         }
1096         case KVM_GET_SUPPORTED_CPUID: {
1097                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1098                 struct kvm_cpuid2 cpuid;
1099
1100                 r = -EFAULT;
1101                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1102                         goto out;
1103                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1104                                                       cpuid_arg->entries);
1105                 if (r)
1106                         goto out;
1107
1108                 r = -EFAULT;
1109                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1110                         goto out;
1111                 r = 0;
1112                 break;
1113         }
1114         default:
1115                 r = -EINVAL;
1116         }
1117 out:
1118         return r;
1119 }
1120
1121 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1122 {
1123         kvm_x86_ops->vcpu_load(vcpu, cpu);
1124         kvm_request_guest_time_update(vcpu);
1125 }
1126
1127 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1128 {
1129         kvm_x86_ops->vcpu_put(vcpu);
1130         kvm_put_guest_fpu(vcpu);
1131 }
1132
1133 static int is_efer_nx(void)
1134 {
1135         unsigned long long efer = 0;
1136
1137         rdmsrl_safe(MSR_EFER, &efer);
1138         return efer & EFER_NX;
1139 }
1140
1141 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1142 {
1143         int i;
1144         struct kvm_cpuid_entry2 *e, *entry;
1145
1146         entry = NULL;
1147         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1148                 e = &vcpu->arch.cpuid_entries[i];
1149                 if (e->function == 0x80000001) {
1150                         entry = e;
1151                         break;
1152                 }
1153         }
1154         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1155                 entry->edx &= ~(1 << 20);
1156                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1157         }
1158 }
1159
1160 /* when an old userspace process fills a new kernel module */
1161 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1162                                     struct kvm_cpuid *cpuid,
1163                                     struct kvm_cpuid_entry __user *entries)
1164 {
1165         int r, i;
1166         struct kvm_cpuid_entry *cpuid_entries;
1167
1168         r = -E2BIG;
1169         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1170                 goto out;
1171         r = -ENOMEM;
1172         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1173         if (!cpuid_entries)
1174                 goto out;
1175         r = -EFAULT;
1176         if (copy_from_user(cpuid_entries, entries,
1177                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1178                 goto out_free;
1179         for (i = 0; i < cpuid->nent; i++) {
1180                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1181                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1182                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1183                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1184                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1185                 vcpu->arch.cpuid_entries[i].index = 0;
1186                 vcpu->arch.cpuid_entries[i].flags = 0;
1187                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1188                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1189                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1190         }
1191         vcpu->arch.cpuid_nent = cpuid->nent;
1192         cpuid_fix_nx_cap(vcpu);
1193         r = 0;
1194
1195 out_free:
1196         vfree(cpuid_entries);
1197 out:
1198         return r;
1199 }
1200
1201 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1202                                      struct kvm_cpuid2 *cpuid,
1203                                      struct kvm_cpuid_entry2 __user *entries)
1204 {
1205         int r;
1206
1207         r = -E2BIG;
1208         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1209                 goto out;
1210         r = -EFAULT;
1211         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1212                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1213                 goto out;
1214         vcpu->arch.cpuid_nent = cpuid->nent;
1215         return 0;
1216
1217 out:
1218         return r;
1219 }
1220
1221 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1222                                      struct kvm_cpuid2 *cpuid,
1223                                      struct kvm_cpuid_entry2 __user *entries)
1224 {
1225         int r;
1226
1227         r = -E2BIG;
1228         if (cpuid->nent < vcpu->arch.cpuid_nent)
1229                 goto out;
1230         r = -EFAULT;
1231         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1232                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1233                 goto out;
1234         return 0;
1235
1236 out:
1237         cpuid->nent = vcpu->arch.cpuid_nent;
1238         return r;
1239 }
1240
1241 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1242                            u32 index)
1243 {
1244         entry->function = function;
1245         entry->index = index;
1246         cpuid_count(entry->function, entry->index,
1247                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1248         entry->flags = 0;
1249 }
1250
1251 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1252                          u32 index, int *nent, int maxnent)
1253 {
1254         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1255                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1256                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1257                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1258                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1259                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1260                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1261                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1262                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1263                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1264         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1265                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1266                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1267                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1268                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1269                 bit(X86_FEATURE_PGE) |
1270                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1271                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1272                 bit(X86_FEATURE_SYSCALL) |
1273                 (is_efer_nx() ? bit(X86_FEATURE_NX) : 0) |
1274 #ifdef CONFIG_X86_64
1275                 bit(X86_FEATURE_LM) |
1276 #endif
1277                 bit(X86_FEATURE_FXSR_OPT) |
1278                 bit(X86_FEATURE_MMXEXT) |
1279                 bit(X86_FEATURE_3DNOWEXT) |
1280                 bit(X86_FEATURE_3DNOW);
1281         const u32 kvm_supported_word3_x86_features =
1282                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1283         const u32 kvm_supported_word6_x86_features =
1284                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1285                 bit(X86_FEATURE_SVM);
1286
1287         /* all calls to cpuid_count() should be made on the same cpu */
1288         get_cpu();
1289         do_cpuid_1_ent(entry, function, index);
1290         ++*nent;
1291
1292         switch (function) {
1293         case 0:
1294                 entry->eax = min(entry->eax, (u32)0xb);
1295                 break;
1296         case 1:
1297                 entry->edx &= kvm_supported_word0_x86_features;
1298                 entry->ecx &= kvm_supported_word3_x86_features;
1299                 break;
1300         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1301          * may return different values. This forces us to get_cpu() before
1302          * issuing the first command, and also to emulate this annoying behavior
1303          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1304         case 2: {
1305                 int t, times = entry->eax & 0xff;
1306
1307                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1308                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1309                 for (t = 1; t < times && *nent < maxnent; ++t) {
1310                         do_cpuid_1_ent(&entry[t], function, 0);
1311                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1312                         ++*nent;
1313                 }
1314                 break;
1315         }
1316         /* function 4 and 0xb have additional index. */
1317         case 4: {
1318                 int i, cache_type;
1319
1320                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1321                 /* read more entries until cache_type is zero */
1322                 for (i = 1; *nent < maxnent; ++i) {
1323                         cache_type = entry[i - 1].eax & 0x1f;
1324                         if (!cache_type)
1325                                 break;
1326                         do_cpuid_1_ent(&entry[i], function, i);
1327                         entry[i].flags |=
1328                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1329                         ++*nent;
1330                 }
1331                 break;
1332         }
1333         case 0xb: {
1334                 int i, level_type;
1335
1336                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1337                 /* read more entries until level_type is zero */
1338                 for (i = 1; *nent < maxnent; ++i) {
1339                         level_type = entry[i - 1].ecx & 0xff00;
1340                         if (!level_type)
1341                                 break;
1342                         do_cpuid_1_ent(&entry[i], function, i);
1343                         entry[i].flags |=
1344                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1345                         ++*nent;
1346                 }
1347                 break;
1348         }
1349         case 0x80000000:
1350                 entry->eax = min(entry->eax, 0x8000001a);
1351                 break;
1352         case 0x80000001:
1353                 entry->edx &= kvm_supported_word1_x86_features;
1354                 entry->ecx &= kvm_supported_word6_x86_features;
1355                 break;
1356         }
1357         put_cpu();
1358 }
1359
1360 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1361                                      struct kvm_cpuid_entry2 __user *entries)
1362 {
1363         struct kvm_cpuid_entry2 *cpuid_entries;
1364         int limit, nent = 0, r = -E2BIG;
1365         u32 func;
1366
1367         if (cpuid->nent < 1)
1368                 goto out;
1369         r = -ENOMEM;
1370         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1371         if (!cpuid_entries)
1372                 goto out;
1373
1374         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1375         limit = cpuid_entries[0].eax;
1376         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1377                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1378                              &nent, cpuid->nent);
1379         r = -E2BIG;
1380         if (nent >= cpuid->nent)
1381                 goto out_free;
1382
1383         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1384         limit = cpuid_entries[nent - 1].eax;
1385         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1386                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1387                              &nent, cpuid->nent);
1388         r = -EFAULT;
1389         if (copy_to_user(entries, cpuid_entries,
1390                          nent * sizeof(struct kvm_cpuid_entry2)))
1391                 goto out_free;
1392         cpuid->nent = nent;
1393         r = 0;
1394
1395 out_free:
1396         vfree(cpuid_entries);
1397 out:
1398         return r;
1399 }
1400
1401 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1402                                     struct kvm_lapic_state *s)
1403 {
1404         vcpu_load(vcpu);
1405         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1406         vcpu_put(vcpu);
1407
1408         return 0;
1409 }
1410
1411 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1412                                     struct kvm_lapic_state *s)
1413 {
1414         vcpu_load(vcpu);
1415         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1416         kvm_apic_post_state_restore(vcpu);
1417         vcpu_put(vcpu);
1418
1419         return 0;
1420 }
1421
1422 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1423                                     struct kvm_interrupt *irq)
1424 {
1425         if (irq->irq < 0 || irq->irq >= 256)
1426                 return -EINVAL;
1427         if (irqchip_in_kernel(vcpu->kvm))
1428                 return -ENXIO;
1429         vcpu_load(vcpu);
1430
1431         set_bit(irq->irq, vcpu->arch.irq_pending);
1432         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1433
1434         vcpu_put(vcpu);
1435
1436         return 0;
1437 }
1438
1439 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1440 {
1441         vcpu_load(vcpu);
1442         kvm_inject_nmi(vcpu);
1443         vcpu_put(vcpu);
1444
1445         return 0;
1446 }
1447
1448 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1449                                            struct kvm_tpr_access_ctl *tac)
1450 {
1451         if (tac->flags)
1452                 return -EINVAL;
1453         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1454         return 0;
1455 }
1456
1457 long kvm_arch_vcpu_ioctl(struct file *filp,
1458                          unsigned int ioctl, unsigned long arg)
1459 {
1460         struct kvm_vcpu *vcpu = filp->private_data;
1461         void __user *argp = (void __user *)arg;
1462         int r;
1463         struct kvm_lapic_state *lapic = NULL;
1464
1465         switch (ioctl) {
1466         case KVM_GET_LAPIC: {
1467                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1468
1469                 r = -ENOMEM;
1470                 if (!lapic)
1471                         goto out;
1472                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1473                 if (r)
1474                         goto out;
1475                 r = -EFAULT;
1476                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1477                         goto out;
1478                 r = 0;
1479                 break;
1480         }
1481         case KVM_SET_LAPIC: {
1482                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1483                 r = -ENOMEM;
1484                 if (!lapic)
1485                         goto out;
1486                 r = -EFAULT;
1487                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1488                         goto out;
1489                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1490                 if (r)
1491                         goto out;
1492                 r = 0;
1493                 break;
1494         }
1495         case KVM_INTERRUPT: {
1496                 struct kvm_interrupt irq;
1497
1498                 r = -EFAULT;
1499                 if (copy_from_user(&irq, argp, sizeof irq))
1500                         goto out;
1501                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1502                 if (r)
1503                         goto out;
1504                 r = 0;
1505                 break;
1506         }
1507         case KVM_NMI: {
1508                 r = kvm_vcpu_ioctl_nmi(vcpu);
1509                 if (r)
1510                         goto out;
1511                 r = 0;
1512                 break;
1513         }
1514         case KVM_SET_CPUID: {
1515                 struct kvm_cpuid __user *cpuid_arg = argp;
1516                 struct kvm_cpuid cpuid;
1517
1518                 r = -EFAULT;
1519                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1520                         goto out;
1521                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1522                 if (r)
1523                         goto out;
1524                 break;
1525         }
1526         case KVM_SET_CPUID2: {
1527                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1528                 struct kvm_cpuid2 cpuid;
1529
1530                 r = -EFAULT;
1531                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1532                         goto out;
1533                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1534                                               cpuid_arg->entries);
1535                 if (r)
1536                         goto out;
1537                 break;
1538         }
1539         case KVM_GET_CPUID2: {
1540                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1541                 struct kvm_cpuid2 cpuid;
1542
1543                 r = -EFAULT;
1544                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1545                         goto out;
1546                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1547                                               cpuid_arg->entries);
1548                 if (r)
1549                         goto out;
1550                 r = -EFAULT;
1551                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1552                         goto out;
1553                 r = 0;
1554                 break;
1555         }
1556         case KVM_GET_MSRS:
1557                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1558                 break;
1559         case KVM_SET_MSRS:
1560                 r = msr_io(vcpu, argp, do_set_msr, 0);
1561                 break;
1562         case KVM_TPR_ACCESS_REPORTING: {
1563                 struct kvm_tpr_access_ctl tac;
1564
1565                 r = -EFAULT;
1566                 if (copy_from_user(&tac, argp, sizeof tac))
1567                         goto out;
1568                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1569                 if (r)
1570                         goto out;
1571                 r = -EFAULT;
1572                 if (copy_to_user(argp, &tac, sizeof tac))
1573                         goto out;
1574                 r = 0;
1575                 break;
1576         };
1577         case KVM_SET_VAPIC_ADDR: {
1578                 struct kvm_vapic_addr va;
1579
1580                 r = -EINVAL;
1581                 if (!irqchip_in_kernel(vcpu->kvm))
1582                         goto out;
1583                 r = -EFAULT;
1584                 if (copy_from_user(&va, argp, sizeof va))
1585                         goto out;
1586                 r = 0;
1587                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1588                 break;
1589         }
1590         default:
1591                 r = -EINVAL;
1592         }
1593 out:
1594         kfree(lapic);
1595         return r;
1596 }
1597
1598 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1599 {
1600         int ret;
1601
1602         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1603                 return -1;
1604         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1605         return ret;
1606 }
1607
1608 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1609                                           u32 kvm_nr_mmu_pages)
1610 {
1611         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1612                 return -EINVAL;
1613
1614         down_write(&kvm->slots_lock);
1615
1616         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1617         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1618
1619         up_write(&kvm->slots_lock);
1620         return 0;
1621 }
1622
1623 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1624 {
1625         return kvm->arch.n_alloc_mmu_pages;
1626 }
1627
1628 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1629 {
1630         int i;
1631         struct kvm_mem_alias *alias;
1632
1633         for (i = 0; i < kvm->arch.naliases; ++i) {
1634                 alias = &kvm->arch.aliases[i];
1635                 if (gfn >= alias->base_gfn
1636                     && gfn < alias->base_gfn + alias->npages)
1637                         return alias->target_gfn + gfn - alias->base_gfn;
1638         }
1639         return gfn;
1640 }
1641
1642 /*
1643  * Set a new alias region.  Aliases map a portion of physical memory into
1644  * another portion.  This is useful for memory windows, for example the PC
1645  * VGA region.
1646  */
1647 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1648                                          struct kvm_memory_alias *alias)
1649 {
1650         int r, n;
1651         struct kvm_mem_alias *p;
1652
1653         r = -EINVAL;
1654         /* General sanity checks */
1655         if (alias->memory_size & (PAGE_SIZE - 1))
1656                 goto out;
1657         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1658                 goto out;
1659         if (alias->slot >= KVM_ALIAS_SLOTS)
1660                 goto out;
1661         if (alias->guest_phys_addr + alias->memory_size
1662             < alias->guest_phys_addr)
1663                 goto out;
1664         if (alias->target_phys_addr + alias->memory_size
1665             < alias->target_phys_addr)
1666                 goto out;
1667
1668         down_write(&kvm->slots_lock);
1669         spin_lock(&kvm->mmu_lock);
1670
1671         p = &kvm->arch.aliases[alias->slot];
1672         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1673         p->npages = alias->memory_size >> PAGE_SHIFT;
1674         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1675
1676         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1677                 if (kvm->arch.aliases[n - 1].npages)
1678                         break;
1679         kvm->arch.naliases = n;
1680
1681         spin_unlock(&kvm->mmu_lock);
1682         kvm_mmu_zap_all(kvm);
1683
1684         up_write(&kvm->slots_lock);
1685
1686         return 0;
1687
1688 out:
1689         return r;
1690 }
1691
1692 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1693 {
1694         int r;
1695
1696         r = 0;
1697         switch (chip->chip_id) {
1698         case KVM_IRQCHIP_PIC_MASTER:
1699                 memcpy(&chip->chip.pic,
1700                         &pic_irqchip(kvm)->pics[0],
1701                         sizeof(struct kvm_pic_state));
1702                 break;
1703         case KVM_IRQCHIP_PIC_SLAVE:
1704                 memcpy(&chip->chip.pic,
1705                         &pic_irqchip(kvm)->pics[1],
1706                         sizeof(struct kvm_pic_state));
1707                 break;
1708         case KVM_IRQCHIP_IOAPIC:
1709                 memcpy(&chip->chip.ioapic,
1710                         ioapic_irqchip(kvm),
1711                         sizeof(struct kvm_ioapic_state));
1712                 break;
1713         default:
1714                 r = -EINVAL;
1715                 break;
1716         }
1717         return r;
1718 }
1719
1720 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1721 {
1722         int r;
1723
1724         r = 0;
1725         switch (chip->chip_id) {
1726         case KVM_IRQCHIP_PIC_MASTER:
1727                 memcpy(&pic_irqchip(kvm)->pics[0],
1728                         &chip->chip.pic,
1729                         sizeof(struct kvm_pic_state));
1730                 break;
1731         case KVM_IRQCHIP_PIC_SLAVE:
1732                 memcpy(&pic_irqchip(kvm)->pics[1],
1733                         &chip->chip.pic,
1734                         sizeof(struct kvm_pic_state));
1735                 break;
1736         case KVM_IRQCHIP_IOAPIC:
1737                 memcpy(ioapic_irqchip(kvm),
1738                         &chip->chip.ioapic,
1739                         sizeof(struct kvm_ioapic_state));
1740                 break;
1741         default:
1742                 r = -EINVAL;
1743                 break;
1744         }
1745         kvm_pic_update_irq(pic_irqchip(kvm));
1746         return r;
1747 }
1748
1749 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1750 {
1751         int r = 0;
1752
1753         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1754         return r;
1755 }
1756
1757 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1758 {
1759         int r = 0;
1760
1761         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1762         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1763         return r;
1764 }
1765
1766 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1767                                  struct kvm_reinject_control *control)
1768 {
1769         if (!kvm->arch.vpit)
1770                 return -ENXIO;
1771         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1772         return 0;
1773 }
1774
1775 /*
1776  * Get (and clear) the dirty memory log for a memory slot.
1777  */
1778 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1779                                       struct kvm_dirty_log *log)
1780 {
1781         int r;
1782         int n;
1783         struct kvm_memory_slot *memslot;
1784         int is_dirty = 0;
1785
1786         down_write(&kvm->slots_lock);
1787
1788         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1789         if (r)
1790                 goto out;
1791
1792         /* If nothing is dirty, don't bother messing with page tables. */
1793         if (is_dirty) {
1794                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1795                 kvm_flush_remote_tlbs(kvm);
1796                 memslot = &kvm->memslots[log->slot];
1797                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1798                 memset(memslot->dirty_bitmap, 0, n);
1799         }
1800         r = 0;
1801 out:
1802         up_write(&kvm->slots_lock);
1803         return r;
1804 }
1805
1806 long kvm_arch_vm_ioctl(struct file *filp,
1807                        unsigned int ioctl, unsigned long arg)
1808 {
1809         struct kvm *kvm = filp->private_data;
1810         void __user *argp = (void __user *)arg;
1811         int r = -EINVAL;
1812         /*
1813          * This union makes it completely explicit to gcc-3.x
1814          * that these two variables' stack usage should be
1815          * combined, not added together.
1816          */
1817         union {
1818                 struct kvm_pit_state ps;
1819                 struct kvm_memory_alias alias;
1820         } u;
1821
1822         switch (ioctl) {
1823         case KVM_SET_TSS_ADDR:
1824                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1825                 if (r < 0)
1826                         goto out;
1827                 break;
1828         case KVM_SET_MEMORY_REGION: {
1829                 struct kvm_memory_region kvm_mem;
1830                 struct kvm_userspace_memory_region kvm_userspace_mem;
1831
1832                 r = -EFAULT;
1833                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1834                         goto out;
1835                 kvm_userspace_mem.slot = kvm_mem.slot;
1836                 kvm_userspace_mem.flags = kvm_mem.flags;
1837                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1838                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1839                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1840                 if (r)
1841                         goto out;
1842                 break;
1843         }
1844         case KVM_SET_NR_MMU_PAGES:
1845                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1846                 if (r)
1847                         goto out;
1848                 break;
1849         case KVM_GET_NR_MMU_PAGES:
1850                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1851                 break;
1852         case KVM_SET_MEMORY_ALIAS:
1853                 r = -EFAULT;
1854                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1855                         goto out;
1856                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1857                 if (r)
1858                         goto out;
1859                 break;
1860         case KVM_CREATE_IRQCHIP:
1861                 r = -ENOMEM;
1862                 kvm->arch.vpic = kvm_create_pic(kvm);
1863                 if (kvm->arch.vpic) {
1864                         r = kvm_ioapic_init(kvm);
1865                         if (r) {
1866                                 kfree(kvm->arch.vpic);
1867                                 kvm->arch.vpic = NULL;
1868                                 goto out;
1869                         }
1870                 } else
1871                         goto out;
1872                 r = kvm_setup_default_irq_routing(kvm);
1873                 if (r) {
1874                         kfree(kvm->arch.vpic);
1875                         kfree(kvm->arch.vioapic);
1876                         goto out;
1877                 }
1878                 break;
1879         case KVM_CREATE_PIT:
1880                 mutex_lock(&kvm->lock);
1881                 r = -EEXIST;
1882                 if (kvm->arch.vpit)
1883                         goto create_pit_unlock;
1884                 r = -ENOMEM;
1885                 kvm->arch.vpit = kvm_create_pit(kvm);
1886                 if (kvm->arch.vpit)
1887                         r = 0;
1888         create_pit_unlock:
1889                 mutex_unlock(&kvm->lock);
1890                 break;
1891         case KVM_IRQ_LINE_STATUS:
1892         case KVM_IRQ_LINE: {
1893                 struct kvm_irq_level irq_event;
1894
1895                 r = -EFAULT;
1896                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1897                         goto out;
1898                 if (irqchip_in_kernel(kvm)) {
1899                         __s32 status;
1900                         mutex_lock(&kvm->lock);
1901                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1902                                         irq_event.irq, irq_event.level);
1903                         mutex_unlock(&kvm->lock);
1904                         if (ioctl == KVM_IRQ_LINE_STATUS) {
1905                                 irq_event.status = status;
1906                                 if (copy_to_user(argp, &irq_event,
1907                                                         sizeof irq_event))
1908                                         goto out;
1909                         }
1910                         r = 0;
1911                 }
1912                 break;
1913         }
1914         case KVM_GET_IRQCHIP: {
1915                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1916                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1917
1918                 r = -ENOMEM;
1919                 if (!chip)
1920                         goto out;
1921                 r = -EFAULT;
1922                 if (copy_from_user(chip, argp, sizeof *chip))
1923                         goto get_irqchip_out;
1924                 r = -ENXIO;
1925                 if (!irqchip_in_kernel(kvm))
1926                         goto get_irqchip_out;
1927                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1928                 if (r)
1929                         goto get_irqchip_out;
1930                 r = -EFAULT;
1931                 if (copy_to_user(argp, chip, sizeof *chip))
1932                         goto get_irqchip_out;
1933                 r = 0;
1934         get_irqchip_out:
1935                 kfree(chip);
1936                 if (r)
1937                         goto out;
1938                 break;
1939         }
1940         case KVM_SET_IRQCHIP: {
1941                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1942                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1943
1944                 r = -ENOMEM;
1945                 if (!chip)
1946                         goto out;
1947                 r = -EFAULT;
1948                 if (copy_from_user(chip, argp, sizeof *chip))
1949                         goto set_irqchip_out;
1950                 r = -ENXIO;
1951                 if (!irqchip_in_kernel(kvm))
1952                         goto set_irqchip_out;
1953                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1954                 if (r)
1955                         goto set_irqchip_out;
1956                 r = 0;
1957         set_irqchip_out:
1958                 kfree(chip);
1959                 if (r)
1960                         goto out;
1961                 break;
1962         }
1963         case KVM_GET_PIT: {
1964                 r = -EFAULT;
1965                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1966                         goto out;
1967                 r = -ENXIO;
1968                 if (!kvm->arch.vpit)
1969                         goto out;
1970                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1971                 if (r)
1972                         goto out;
1973                 r = -EFAULT;
1974                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1975                         goto out;
1976                 r = 0;
1977                 break;
1978         }
1979         case KVM_SET_PIT: {
1980                 r = -EFAULT;
1981                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1982                         goto out;
1983                 r = -ENXIO;
1984                 if (!kvm->arch.vpit)
1985                         goto out;
1986                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1987                 if (r)
1988                         goto out;
1989                 r = 0;
1990                 break;
1991         }
1992         case KVM_REINJECT_CONTROL: {
1993                 struct kvm_reinject_control control;
1994                 r =  -EFAULT;
1995                 if (copy_from_user(&control, argp, sizeof(control)))
1996                         goto out;
1997                 r = kvm_vm_ioctl_reinject(kvm, &control);
1998                 if (r)
1999                         goto out;
2000                 r = 0;
2001                 break;
2002         }
2003         default:
2004                 ;
2005         }
2006 out:
2007         return r;
2008 }
2009
2010 static void kvm_init_msr_list(void)
2011 {
2012         u32 dummy[2];
2013         unsigned i, j;
2014
2015         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2016                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2017                         continue;
2018                 if (j < i)
2019                         msrs_to_save[j] = msrs_to_save[i];
2020                 j++;
2021         }
2022         num_msrs_to_save = j;
2023 }
2024
2025 /*
2026  * Only apic need an MMIO device hook, so shortcut now..
2027  */
2028 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2029                                                 gpa_t addr, int len,
2030                                                 int is_write)
2031 {
2032         struct kvm_io_device *dev;
2033
2034         if (vcpu->arch.apic) {
2035                 dev = &vcpu->arch.apic->dev;
2036                 if (dev->in_range(dev, addr, len, is_write))
2037                         return dev;
2038         }
2039         return NULL;
2040 }
2041
2042
2043 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2044                                                 gpa_t addr, int len,
2045                                                 int is_write)
2046 {
2047         struct kvm_io_device *dev;
2048
2049         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2050         if (dev == NULL)
2051                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2052                                           is_write);
2053         return dev;
2054 }
2055
2056 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2057                                struct kvm_vcpu *vcpu)
2058 {
2059         void *data = val;
2060         int r = X86EMUL_CONTINUE;
2061
2062         while (bytes) {
2063                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2064                 unsigned offset = addr & (PAGE_SIZE-1);
2065                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2066                 int ret;
2067
2068                 if (gpa == UNMAPPED_GVA) {
2069                         r = X86EMUL_PROPAGATE_FAULT;
2070                         goto out;
2071                 }
2072                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2073                 if (ret < 0) {
2074                         r = X86EMUL_UNHANDLEABLE;
2075                         goto out;
2076                 }
2077
2078                 bytes -= toread;
2079                 data += toread;
2080                 addr += toread;
2081         }
2082 out:
2083         return r;
2084 }
2085
2086 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2087                                 struct kvm_vcpu *vcpu)
2088 {
2089         void *data = val;
2090         int r = X86EMUL_CONTINUE;
2091
2092         while (bytes) {
2093                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2094                 unsigned offset = addr & (PAGE_SIZE-1);
2095                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2096                 int ret;
2097
2098                 if (gpa == UNMAPPED_GVA) {
2099                         r = X86EMUL_PROPAGATE_FAULT;
2100                         goto out;
2101                 }
2102                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2103                 if (ret < 0) {
2104                         r = X86EMUL_UNHANDLEABLE;
2105                         goto out;
2106                 }
2107
2108                 bytes -= towrite;
2109                 data += towrite;
2110                 addr += towrite;
2111         }
2112 out:
2113         return r;
2114 }
2115
2116
2117 static int emulator_read_emulated(unsigned long addr,
2118                                   void *val,
2119                                   unsigned int bytes,
2120                                   struct kvm_vcpu *vcpu)
2121 {
2122         struct kvm_io_device *mmio_dev;
2123         gpa_t                 gpa;
2124
2125         if (vcpu->mmio_read_completed) {
2126                 memcpy(val, vcpu->mmio_data, bytes);
2127                 vcpu->mmio_read_completed = 0;
2128                 return X86EMUL_CONTINUE;
2129         }
2130
2131         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2132
2133         /* For APIC access vmexit */
2134         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2135                 goto mmio;
2136
2137         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2138                                 == X86EMUL_CONTINUE)
2139                 return X86EMUL_CONTINUE;
2140         if (gpa == UNMAPPED_GVA)
2141                 return X86EMUL_PROPAGATE_FAULT;
2142
2143 mmio:
2144         /*
2145          * Is this MMIO handled locally?
2146          */
2147         mutex_lock(&vcpu->kvm->lock);
2148         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2149         if (mmio_dev) {
2150                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2151                 mutex_unlock(&vcpu->kvm->lock);
2152                 return X86EMUL_CONTINUE;
2153         }
2154         mutex_unlock(&vcpu->kvm->lock);
2155
2156         vcpu->mmio_needed = 1;
2157         vcpu->mmio_phys_addr = gpa;
2158         vcpu->mmio_size = bytes;
2159         vcpu->mmio_is_write = 0;
2160
2161         return X86EMUL_UNHANDLEABLE;
2162 }
2163
2164 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2165                           const void *val, int bytes)
2166 {
2167         int ret;
2168
2169         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2170         if (ret < 0)
2171                 return 0;
2172         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2173         return 1;
2174 }
2175
2176 static int emulator_write_emulated_onepage(unsigned long addr,
2177                                            const void *val,
2178                                            unsigned int bytes,
2179                                            struct kvm_vcpu *vcpu)
2180 {
2181         struct kvm_io_device *mmio_dev;
2182         gpa_t                 gpa;
2183
2184         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2185
2186         if (gpa == UNMAPPED_GVA) {
2187                 kvm_inject_page_fault(vcpu, addr, 2);
2188                 return X86EMUL_PROPAGATE_FAULT;
2189         }
2190
2191         /* For APIC access vmexit */
2192         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2193                 goto mmio;
2194
2195         if (emulator_write_phys(vcpu, gpa, val, bytes))
2196                 return X86EMUL_CONTINUE;
2197
2198 mmio:
2199         /*
2200          * Is this MMIO handled locally?
2201          */
2202         mutex_lock(&vcpu->kvm->lock);
2203         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2204         if (mmio_dev) {
2205                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2206                 mutex_unlock(&vcpu->kvm->lock);
2207                 return X86EMUL_CONTINUE;
2208         }
2209         mutex_unlock(&vcpu->kvm->lock);
2210
2211         vcpu->mmio_needed = 1;
2212         vcpu->mmio_phys_addr = gpa;
2213         vcpu->mmio_size = bytes;
2214         vcpu->mmio_is_write = 1;
2215         memcpy(vcpu->mmio_data, val, bytes);
2216
2217         return X86EMUL_CONTINUE;
2218 }
2219
2220 int emulator_write_emulated(unsigned long addr,
2221                                    const void *val,
2222                                    unsigned int bytes,
2223                                    struct kvm_vcpu *vcpu)
2224 {
2225         /* Crossing a page boundary? */
2226         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2227                 int rc, now;
2228
2229                 now = -addr & ~PAGE_MASK;
2230                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2231                 if (rc != X86EMUL_CONTINUE)
2232                         return rc;
2233                 addr += now;
2234                 val += now;
2235                 bytes -= now;
2236         }
2237         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2238 }
2239 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2240
2241 static int emulator_cmpxchg_emulated(unsigned long addr,
2242                                      const void *old,
2243                                      const void *new,
2244                                      unsigned int bytes,
2245                                      struct kvm_vcpu *vcpu)
2246 {
2247         static int reported;
2248
2249         if (!reported) {
2250                 reported = 1;
2251                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2252         }
2253 #ifndef CONFIG_X86_64
2254         /* guests cmpxchg8b have to be emulated atomically */
2255         if (bytes == 8) {
2256                 gpa_t gpa;
2257                 struct page *page;
2258                 char *kaddr;
2259                 u64 val;
2260
2261                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2262
2263                 if (gpa == UNMAPPED_GVA ||
2264                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2265                         goto emul_write;
2266
2267                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2268                         goto emul_write;
2269
2270                 val = *(u64 *)new;
2271
2272                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2273
2274                 kaddr = kmap_atomic(page, KM_USER0);
2275                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2276                 kunmap_atomic(kaddr, KM_USER0);
2277                 kvm_release_page_dirty(page);
2278         }
2279 emul_write:
2280 #endif
2281
2282         return emulator_write_emulated(addr, new, bytes, vcpu);
2283 }
2284
2285 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2286 {
2287         return kvm_x86_ops->get_segment_base(vcpu, seg);
2288 }
2289
2290 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2291 {
2292         kvm_mmu_invlpg(vcpu, address);
2293         return X86EMUL_CONTINUE;
2294 }
2295
2296 int emulate_clts(struct kvm_vcpu *vcpu)
2297 {
2298         KVMTRACE_0D(CLTS, vcpu, handler);
2299         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2300         return X86EMUL_CONTINUE;
2301 }
2302
2303 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2304 {
2305         struct kvm_vcpu *vcpu = ctxt->vcpu;
2306
2307         switch (dr) {
2308         case 0 ... 3:
2309                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2310                 return X86EMUL_CONTINUE;
2311         default:
2312                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2313                 return X86EMUL_UNHANDLEABLE;
2314         }
2315 }
2316
2317 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2318 {
2319         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2320         int exception;
2321
2322         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2323         if (exception) {
2324                 /* FIXME: better handling */
2325                 return X86EMUL_UNHANDLEABLE;
2326         }
2327         return X86EMUL_CONTINUE;
2328 }
2329
2330 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2331 {
2332         u8 opcodes[4];
2333         unsigned long rip = kvm_rip_read(vcpu);
2334         unsigned long rip_linear;
2335
2336         if (!printk_ratelimit())
2337                 return;
2338
2339         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2340
2341         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2342
2343         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2344                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2345 }
2346 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2347
2348 static struct x86_emulate_ops emulate_ops = {
2349         .read_std            = kvm_read_guest_virt,
2350         .read_emulated       = emulator_read_emulated,
2351         .write_emulated      = emulator_write_emulated,
2352         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2353 };
2354
2355 static void cache_all_regs(struct kvm_vcpu *vcpu)
2356 {
2357         kvm_register_read(vcpu, VCPU_REGS_RAX);
2358         kvm_register_read(vcpu, VCPU_REGS_RSP);
2359         kvm_register_read(vcpu, VCPU_REGS_RIP);
2360         vcpu->arch.regs_dirty = ~0;
2361 }
2362
2363 int emulate_instruction(struct kvm_vcpu *vcpu,
2364                         struct kvm_run *run,
2365                         unsigned long cr2,
2366                         u16 error_code,
2367                         int emulation_type)
2368 {
2369         int r;
2370         struct decode_cache *c;
2371
2372         kvm_clear_exception_queue(vcpu);
2373         vcpu->arch.mmio_fault_cr2 = cr2;
2374         /*
2375          * TODO: fix x86_emulate.c to use guest_read/write_register
2376          * instead of direct ->regs accesses, can save hundred cycles
2377          * on Intel for instructions that don't read/change RSP, for
2378          * for example.
2379          */
2380         cache_all_regs(vcpu);
2381
2382         vcpu->mmio_is_write = 0;
2383         vcpu->arch.pio.string = 0;
2384
2385         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2386                 int cs_db, cs_l;
2387                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2388
2389                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2390                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2391                 vcpu->arch.emulate_ctxt.mode =
2392                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2393                         ? X86EMUL_MODE_REAL : cs_l
2394                         ? X86EMUL_MODE_PROT64 : cs_db
2395                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2396
2397                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2398
2399                 /* Reject the instructions other than VMCALL/VMMCALL when
2400                  * try to emulate invalid opcode */
2401                 c = &vcpu->arch.emulate_ctxt.decode;
2402                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2403                     (!(c->twobyte && c->b == 0x01 &&
2404                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2405                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2406                         return EMULATE_FAIL;
2407
2408                 ++vcpu->stat.insn_emulation;
2409                 if (r)  {
2410                         ++vcpu->stat.insn_emulation_fail;
2411                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2412                                 return EMULATE_DONE;
2413                         return EMULATE_FAIL;
2414                 }
2415         }
2416
2417         if (emulation_type & EMULTYPE_SKIP) {
2418                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2419                 return EMULATE_DONE;
2420         }
2421
2422         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2423
2424         if (vcpu->arch.pio.string)
2425                 return EMULATE_DO_MMIO;
2426
2427         if ((r || vcpu->mmio_is_write) && run) {
2428                 run->exit_reason = KVM_EXIT_MMIO;
2429                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2430                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2431                 run->mmio.len = vcpu->mmio_size;
2432                 run->mmio.is_write = vcpu->mmio_is_write;
2433         }
2434
2435         if (r) {
2436                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2437                         return EMULATE_DONE;
2438                 if (!vcpu->mmio_needed) {
2439                         kvm_report_emulation_failure(vcpu, "mmio");
2440                         return EMULATE_FAIL;
2441                 }
2442                 return EMULATE_DO_MMIO;
2443         }
2444
2445         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2446
2447         if (vcpu->mmio_is_write) {
2448                 vcpu->mmio_needed = 0;
2449                 return EMULATE_DO_MMIO;
2450         }
2451
2452         return EMULATE_DONE;
2453 }
2454 EXPORT_SYMBOL_GPL(emulate_instruction);
2455
2456 static int pio_copy_data(struct kvm_vcpu *vcpu)
2457 {
2458         void *p = vcpu->arch.pio_data;
2459         gva_t q = vcpu->arch.pio.guest_gva;
2460         unsigned bytes;
2461         int ret;
2462
2463         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2464         if (vcpu->arch.pio.in)
2465                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2466         else
2467                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2468         return ret;
2469 }
2470
2471 int complete_pio(struct kvm_vcpu *vcpu)
2472 {
2473         struct kvm_pio_request *io = &vcpu->arch.pio;
2474         long delta;
2475         int r;
2476         unsigned long val;
2477
2478         if (!io->string) {
2479                 if (io->in) {
2480                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2481                         memcpy(&val, vcpu->arch.pio_data, io->size);
2482                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2483                 }
2484         } else {
2485                 if (io->in) {
2486                         r = pio_copy_data(vcpu);
2487                         if (r)
2488                                 return r;
2489                 }
2490
2491                 delta = 1;
2492                 if (io->rep) {
2493                         delta *= io->cur_count;
2494                         /*
2495                          * The size of the register should really depend on
2496                          * current address size.
2497                          */
2498                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2499                         val -= delta;
2500                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2501                 }
2502                 if (io->down)
2503                         delta = -delta;
2504                 delta *= io->size;
2505                 if (io->in) {
2506                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2507                         val += delta;
2508                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2509                 } else {
2510                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2511                         val += delta;
2512                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2513                 }
2514         }
2515
2516         io->count -= io->cur_count;
2517         io->cur_count = 0;
2518
2519         return 0;
2520 }
2521
2522 static void kernel_pio(struct kvm_io_device *pio_dev,
2523                        struct kvm_vcpu *vcpu,
2524                        void *pd)
2525 {
2526         /* TODO: String I/O for in kernel device */
2527
2528         mutex_lock(&vcpu->kvm->lock);
2529         if (vcpu->arch.pio.in)
2530                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2531                                   vcpu->arch.pio.size,
2532                                   pd);
2533         else
2534                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2535                                    vcpu->arch.pio.size,
2536                                    pd);
2537         mutex_unlock(&vcpu->kvm->lock);
2538 }
2539
2540 static void pio_string_write(struct kvm_io_device *pio_dev,
2541                              struct kvm_vcpu *vcpu)
2542 {
2543         struct kvm_pio_request *io = &vcpu->arch.pio;
2544         void *pd = vcpu->arch.pio_data;
2545         int i;
2546
2547         mutex_lock(&vcpu->kvm->lock);
2548         for (i = 0; i < io->cur_count; i++) {
2549                 kvm_iodevice_write(pio_dev, io->port,
2550                                    io->size,
2551                                    pd);
2552                 pd += io->size;
2553         }
2554         mutex_unlock(&vcpu->kvm->lock);
2555 }
2556
2557 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2558                                                gpa_t addr, int len,
2559                                                int is_write)
2560 {
2561         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2562 }
2563
2564 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2565                   int size, unsigned port)
2566 {
2567         struct kvm_io_device *pio_dev;
2568         unsigned long val;
2569
2570         vcpu->run->exit_reason = KVM_EXIT_IO;
2571         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2572         vcpu->run->io.size = vcpu->arch.pio.size = size;
2573         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2574         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2575         vcpu->run->io.port = vcpu->arch.pio.port = port;
2576         vcpu->arch.pio.in = in;
2577         vcpu->arch.pio.string = 0;
2578         vcpu->arch.pio.down = 0;
2579         vcpu->arch.pio.rep = 0;
2580
2581         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2582                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2583                             handler);
2584         else
2585                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2586                             handler);
2587
2588         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2589         memcpy(vcpu->arch.pio_data, &val, 4);
2590
2591         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2592         if (pio_dev) {
2593                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2594                 complete_pio(vcpu);
2595                 return 1;
2596         }
2597         return 0;
2598 }
2599 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2600
2601 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2602                   int size, unsigned long count, int down,
2603                   gva_t address, int rep, unsigned port)
2604 {
2605         unsigned now, in_page;
2606         int ret = 0;
2607         struct kvm_io_device *pio_dev;
2608
2609         vcpu->run->exit_reason = KVM_EXIT_IO;
2610         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2611         vcpu->run->io.size = vcpu->arch.pio.size = size;
2612         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2613         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2614         vcpu->run->io.port = vcpu->arch.pio.port = port;
2615         vcpu->arch.pio.in = in;
2616         vcpu->arch.pio.string = 1;
2617         vcpu->arch.pio.down = down;
2618         vcpu->arch.pio.rep = rep;
2619
2620         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2621                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2622                             handler);
2623         else
2624                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2625                             handler);
2626
2627         if (!count) {
2628                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2629                 return 1;
2630         }
2631
2632         if (!down)
2633                 in_page = PAGE_SIZE - offset_in_page(address);
2634         else
2635                 in_page = offset_in_page(address) + size;
2636         now = min(count, (unsigned long)in_page / size);
2637         if (!now)
2638                 now = 1;
2639         if (down) {
2640                 /*
2641                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2642                  */
2643                 pr_unimpl(vcpu, "guest string pio down\n");
2644                 kvm_inject_gp(vcpu, 0);
2645                 return 1;
2646         }
2647         vcpu->run->io.count = now;
2648         vcpu->arch.pio.cur_count = now;
2649
2650         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2651                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2652
2653         vcpu->arch.pio.guest_gva = address;
2654
2655         pio_dev = vcpu_find_pio_dev(vcpu, port,
2656                                     vcpu->arch.pio.cur_count,
2657                                     !vcpu->arch.pio.in);
2658         if (!vcpu->arch.pio.in) {
2659                 /* string PIO write */
2660                 ret = pio_copy_data(vcpu);
2661                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2662                         kvm_inject_gp(vcpu, 0);
2663                         return 1;
2664                 }
2665                 if (ret == 0 && pio_dev) {
2666                         pio_string_write(pio_dev, vcpu);
2667                         complete_pio(vcpu);
2668                         if (vcpu->arch.pio.count == 0)
2669                                 ret = 1;
2670                 }
2671         } else if (pio_dev)
2672                 pr_unimpl(vcpu, "no string pio read support yet, "
2673                        "port %x size %d count %ld\n",
2674                         port, size, count);
2675
2676         return ret;
2677 }
2678 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2679
2680 static void bounce_off(void *info)
2681 {
2682         /* nothing */
2683 }
2684
2685 static unsigned int  ref_freq;
2686 static unsigned long tsc_khz_ref;
2687
2688 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2689                                      void *data)
2690 {
2691         struct cpufreq_freqs *freq = data;
2692         struct kvm *kvm;
2693         struct kvm_vcpu *vcpu;
2694         int i, send_ipi = 0;
2695
2696         if (!ref_freq)
2697                 ref_freq = freq->old;
2698
2699         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2700                 return 0;
2701         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2702                 return 0;
2703         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2704
2705         spin_lock(&kvm_lock);
2706         list_for_each_entry(kvm, &vm_list, vm_list) {
2707                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2708                         vcpu = kvm->vcpus[i];
2709                         if (!vcpu)
2710                                 continue;
2711                         if (vcpu->cpu != freq->cpu)
2712                                 continue;
2713                         if (!kvm_request_guest_time_update(vcpu))
2714                                 continue;
2715                         if (vcpu->cpu != smp_processor_id())
2716                                 send_ipi++;
2717                 }
2718         }
2719         spin_unlock(&kvm_lock);
2720
2721         if (freq->old < freq->new && send_ipi) {
2722                 /*
2723                  * We upscale the frequency.  Must make the guest
2724                  * doesn't see old kvmclock values while running with
2725                  * the new frequency, otherwise we risk the guest sees
2726                  * time go backwards.
2727                  *
2728                  * In case we update the frequency for another cpu
2729                  * (which might be in guest context) send an interrupt
2730                  * to kick the cpu out of guest context.  Next time
2731                  * guest context is entered kvmclock will be updated,
2732                  * so the guest will not see stale values.
2733                  */
2734                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2735         }
2736         return 0;
2737 }
2738
2739 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2740         .notifier_call  = kvmclock_cpufreq_notifier
2741 };
2742
2743 int kvm_arch_init(void *opaque)
2744 {
2745         int r, cpu;
2746         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2747
2748         if (kvm_x86_ops) {
2749                 printk(KERN_ERR "kvm: already loaded the other module\n");
2750                 r = -EEXIST;
2751                 goto out;
2752         }
2753
2754         if (!ops->cpu_has_kvm_support()) {
2755                 printk(KERN_ERR "kvm: no hardware support\n");
2756                 r = -EOPNOTSUPP;
2757                 goto out;
2758         }
2759         if (ops->disabled_by_bios()) {
2760                 printk(KERN_ERR "kvm: disabled by bios\n");
2761                 r = -EOPNOTSUPP;
2762                 goto out;
2763         }
2764
2765         r = kvm_mmu_module_init();
2766         if (r)
2767                 goto out;
2768
2769         kvm_init_msr_list();
2770
2771         kvm_x86_ops = ops;
2772         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2773         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2774         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2775                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2776
2777         for_each_possible_cpu(cpu)
2778                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2779         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2780                 tsc_khz_ref = tsc_khz;
2781                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2782                                           CPUFREQ_TRANSITION_NOTIFIER);
2783         }
2784
2785         return 0;
2786
2787 out:
2788         return r;
2789 }
2790
2791 void kvm_arch_exit(void)
2792 {
2793         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2794                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2795                                             CPUFREQ_TRANSITION_NOTIFIER);
2796         kvm_x86_ops = NULL;
2797         kvm_mmu_module_exit();
2798 }
2799
2800 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2801 {
2802         ++vcpu->stat.halt_exits;
2803         KVMTRACE_0D(HLT, vcpu, handler);
2804         if (irqchip_in_kernel(vcpu->kvm)) {
2805                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2806                 return 1;
2807         } else {
2808                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2809                 return 0;
2810         }
2811 }
2812 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2813
2814 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2815                            unsigned long a1)
2816 {
2817         if (is_long_mode(vcpu))
2818                 return a0;
2819         else
2820                 return a0 | ((gpa_t)a1 << 32);
2821 }
2822
2823 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2824 {
2825         unsigned long nr, a0, a1, a2, a3, ret;
2826         int r = 1;
2827
2828         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2829         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2830         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2831         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2832         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2833
2834         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2835
2836         if (!is_long_mode(vcpu)) {
2837                 nr &= 0xFFFFFFFF;
2838                 a0 &= 0xFFFFFFFF;
2839                 a1 &= 0xFFFFFFFF;
2840                 a2 &= 0xFFFFFFFF;
2841                 a3 &= 0xFFFFFFFF;
2842         }
2843
2844         switch (nr) {
2845         case KVM_HC_VAPIC_POLL_IRQ:
2846                 ret = 0;
2847                 break;
2848         case KVM_HC_MMU_OP:
2849                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2850                 break;
2851         default:
2852                 ret = -KVM_ENOSYS;
2853                 break;
2854         }
2855         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2856         ++vcpu->stat.hypercalls;
2857         return r;
2858 }
2859 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2860
2861 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2862 {
2863         char instruction[3];
2864         int ret = 0;
2865         unsigned long rip = kvm_rip_read(vcpu);
2866
2867
2868         /*
2869          * Blow out the MMU to ensure that no other VCPU has an active mapping
2870          * to ensure that the updated hypercall appears atomically across all
2871          * VCPUs.
2872          */
2873         kvm_mmu_zap_all(vcpu->kvm);
2874
2875         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2876         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2877             != X86EMUL_CONTINUE)
2878                 ret = -EFAULT;
2879
2880         return ret;
2881 }
2882
2883 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2884 {
2885         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2886 }
2887
2888 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2889 {
2890         struct descriptor_table dt = { limit, base };
2891
2892         kvm_x86_ops->set_gdt(vcpu, &dt);
2893 }
2894
2895 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2896 {
2897         struct descriptor_table dt = { limit, base };
2898
2899         kvm_x86_ops->set_idt(vcpu, &dt);
2900 }
2901
2902 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2903                    unsigned long *rflags)
2904 {
2905         kvm_lmsw(vcpu, msw);
2906         *rflags = kvm_x86_ops->get_rflags(vcpu);
2907 }
2908
2909 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2910 {
2911         unsigned long value;
2912
2913         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2914         switch (cr) {
2915         case 0:
2916                 value = vcpu->arch.cr0;
2917                 break;
2918         case 2:
2919                 value = vcpu->arch.cr2;
2920                 break;
2921         case 3:
2922                 value = vcpu->arch.cr3;
2923                 break;
2924         case 4:
2925                 value = vcpu->arch.cr4;
2926                 break;
2927         case 8:
2928                 value = kvm_get_cr8(vcpu);
2929                 break;
2930         default:
2931                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2932                 return 0;
2933         }
2934         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2935                     (u32)((u64)value >> 32), handler);
2936
2937         return value;
2938 }
2939
2940 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2941                      unsigned long *rflags)
2942 {
2943         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2944                     (u32)((u64)val >> 32), handler);
2945
2946         switch (cr) {
2947         case 0:
2948                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2949                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2950                 break;
2951         case 2:
2952                 vcpu->arch.cr2 = val;
2953                 break;
2954         case 3:
2955                 kvm_set_cr3(vcpu, val);
2956                 break;
2957         case 4:
2958                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2959                 break;
2960         case 8:
2961                 kvm_set_cr8(vcpu, val & 0xfUL);
2962                 break;
2963         default:
2964                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2965         }
2966 }
2967
2968 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2969 {
2970         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2971         int j, nent = vcpu->arch.cpuid_nent;
2972
2973         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2974         /* when no next entry is found, the current entry[i] is reselected */
2975         for (j = i + 1; ; j = (j + 1) % nent) {
2976                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2977                 if (ej->function == e->function) {
2978                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2979                         return j;
2980                 }
2981         }
2982         return 0; /* silence gcc, even though control never reaches here */
2983 }
2984
2985 /* find an entry with matching function, matching index (if needed), and that
2986  * should be read next (if it's stateful) */
2987 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2988         u32 function, u32 index)
2989 {
2990         if (e->function != function)
2991                 return 0;
2992         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2993                 return 0;
2994         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2995             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2996                 return 0;
2997         return 1;
2998 }
2999
3000 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3001                                               u32 function, u32 index)
3002 {
3003         int i;
3004         struct kvm_cpuid_entry2 *best = NULL;
3005
3006         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3007                 struct kvm_cpuid_entry2 *e;
3008
3009                 e = &vcpu->arch.cpuid_entries[i];
3010                 if (is_matching_cpuid_entry(e, function, index)) {
3011                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3012                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3013                         best = e;
3014                         break;
3015                 }
3016                 /*
3017                  * Both basic or both extended?
3018                  */
3019                 if (((e->function ^ function) & 0x80000000) == 0)
3020                         if (!best || e->function > best->function)
3021                                 best = e;
3022         }
3023         return best;
3024 }
3025
3026 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3027 {
3028         struct kvm_cpuid_entry2 *best;
3029
3030         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3031         if (best)
3032                 return best->eax & 0xff;
3033         return 36;
3034 }
3035
3036 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3037 {
3038         u32 function, index;
3039         struct kvm_cpuid_entry2 *best;
3040
3041         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3042         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3043         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3044         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3045         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3046         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3047         best = kvm_find_cpuid_entry(vcpu, function, index);
3048         if (best) {
3049                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3050                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3051                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3052                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3053         }
3054         kvm_x86_ops->skip_emulated_instruction(vcpu);
3055         KVMTRACE_5D(CPUID, vcpu, function,
3056                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3057                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3058                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3059                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3060 }
3061 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3062
3063 /*
3064  * Check if userspace requested an interrupt window, and that the
3065  * interrupt window is open.
3066  *
3067  * No need to exit to userspace if we already have an interrupt queued.
3068  */
3069 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3070                                           struct kvm_run *kvm_run)
3071 {
3072         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3073                 kvm_run->request_interrupt_window &&
3074                 vcpu->arch.interrupt_window_open &&
3075                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3076 }
3077
3078 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3079                               struct kvm_run *kvm_run)
3080 {
3081         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3082         kvm_run->cr8 = kvm_get_cr8(vcpu);
3083         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3084         if (irqchip_in_kernel(vcpu->kvm))
3085                 kvm_run->ready_for_interrupt_injection = 1;
3086         else
3087                 kvm_run->ready_for_interrupt_injection =
3088                                         (vcpu->arch.interrupt_window_open &&
3089                                          !kvm_cpu_has_interrupt(vcpu));
3090 }
3091
3092 static void vapic_enter(struct kvm_vcpu *vcpu)
3093 {
3094         struct kvm_lapic *apic = vcpu->arch.apic;
3095         struct page *page;
3096
3097         if (!apic || !apic->vapic_addr)
3098                 return;
3099
3100         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3101
3102         vcpu->arch.apic->vapic_page = page;
3103 }
3104
3105 static void vapic_exit(struct kvm_vcpu *vcpu)
3106 {
3107         struct kvm_lapic *apic = vcpu->arch.apic;
3108
3109         if (!apic || !apic->vapic_addr)
3110                 return;
3111
3112         down_read(&vcpu->kvm->slots_lock);
3113         kvm_release_page_dirty(apic->vapic_page);
3114         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3115         up_read(&vcpu->kvm->slots_lock);
3116 }
3117
3118 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3119 {
3120         int r;
3121
3122         if (vcpu->requests)
3123                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3124                         kvm_mmu_unload(vcpu);
3125
3126         r = kvm_mmu_reload(vcpu);
3127         if (unlikely(r))
3128                 goto out;
3129
3130         if (vcpu->requests) {
3131                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3132                         __kvm_migrate_timers(vcpu);
3133                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3134                         kvm_write_guest_time(vcpu);
3135                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3136                         kvm_mmu_sync_roots(vcpu);
3137                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3138                         kvm_x86_ops->tlb_flush(vcpu);
3139                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3140                                        &vcpu->requests)) {
3141                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3142                         r = 0;
3143                         goto out;
3144                 }
3145                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3146                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3147                         r = 0;
3148                         goto out;
3149                 }
3150         }
3151
3152         preempt_disable();
3153
3154         kvm_x86_ops->prepare_guest_switch(vcpu);
3155         kvm_load_guest_fpu(vcpu);
3156
3157         local_irq_disable();
3158
3159         if (vcpu->requests || need_resched() || signal_pending(current)) {
3160                 local_irq_enable();
3161                 preempt_enable();
3162                 r = 1;
3163                 goto out;
3164         }
3165
3166         vcpu->guest_mode = 1;
3167         /*
3168          * Make sure that guest_mode assignment won't happen after
3169          * testing the pending IRQ vector bitmap.
3170          */
3171         smp_wmb();
3172
3173         if (vcpu->arch.exception.pending)
3174                 __queue_exception(vcpu);
3175         else if (irqchip_in_kernel(vcpu->kvm))
3176                 kvm_x86_ops->inject_pending_irq(vcpu);
3177         else
3178                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3179
3180         kvm_lapic_sync_to_vapic(vcpu);
3181
3182         up_read(&vcpu->kvm->slots_lock);
3183
3184         kvm_guest_enter();
3185
3186         get_debugreg(vcpu->arch.host_dr6, 6);
3187         get_debugreg(vcpu->arch.host_dr7, 7);
3188         if (unlikely(vcpu->arch.switch_db_regs)) {
3189                 get_debugreg(vcpu->arch.host_db[0], 0);
3190                 get_debugreg(vcpu->arch.host_db[1], 1);
3191                 get_debugreg(vcpu->arch.host_db[2], 2);
3192                 get_debugreg(vcpu->arch.host_db[3], 3);
3193
3194                 set_debugreg(0, 7);
3195                 set_debugreg(vcpu->arch.eff_db[0], 0);
3196                 set_debugreg(vcpu->arch.eff_db[1], 1);
3197                 set_debugreg(vcpu->arch.eff_db[2], 2);
3198                 set_debugreg(vcpu->arch.eff_db[3], 3);
3199         }
3200
3201         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3202         kvm_x86_ops->run(vcpu, kvm_run);
3203
3204         if (unlikely(vcpu->arch.switch_db_regs)) {
3205                 set_debugreg(0, 7);
3206                 set_debugreg(vcpu->arch.host_db[0], 0);
3207                 set_debugreg(vcpu->arch.host_db[1], 1);
3208                 set_debugreg(vcpu->arch.host_db[2], 2);
3209                 set_debugreg(vcpu->arch.host_db[3], 3);
3210         }
3211         set_debugreg(vcpu->arch.host_dr6, 6);
3212         set_debugreg(vcpu->arch.host_dr7, 7);
3213
3214         vcpu->guest_mode = 0;
3215         local_irq_enable();
3216
3217         ++vcpu->stat.exits;
3218
3219         /*
3220          * We must have an instruction between local_irq_enable() and
3221          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3222          * the interrupt shadow.  The stat.exits increment will do nicely.
3223          * But we need to prevent reordering, hence this barrier():
3224          */
3225         barrier();
3226
3227         kvm_guest_exit();
3228
3229         preempt_enable();
3230
3231         down_read(&vcpu->kvm->slots_lock);
3232
3233         /*
3234          * Profile KVM exit RIPs:
3235          */
3236         if (unlikely(prof_on == KVM_PROFILING)) {
3237                 unsigned long rip = kvm_rip_read(vcpu);
3238                 profile_hit(KVM_PROFILING, (void *)rip);
3239         }
3240
3241         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3242                 vcpu->arch.exception.pending = false;
3243
3244         kvm_lapic_sync_from_vapic(vcpu);
3245
3246         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3247 out:
3248         return r;
3249 }
3250
3251
3252 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3253 {
3254         int r;
3255
3256         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3257                 pr_debug("vcpu %d received sipi with vector # %x\n",
3258                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3259                 kvm_lapic_reset(vcpu);
3260                 r = kvm_arch_vcpu_reset(vcpu);
3261                 if (r)
3262                         return r;
3263                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3264         }
3265
3266         down_read(&vcpu->kvm->slots_lock);
3267         vapic_enter(vcpu);
3268
3269         r = 1;
3270         while (r > 0) {
3271                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3272                         r = vcpu_enter_guest(vcpu, kvm_run);
3273                 else {
3274                         up_read(&vcpu->kvm->slots_lock);
3275                         kvm_vcpu_block(vcpu);
3276                         down_read(&vcpu->kvm->slots_lock);
3277                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3278                         {
3279                                 switch(vcpu->arch.mp_state) {
3280                                 case KVM_MP_STATE_HALTED:
3281                                         vcpu->arch.mp_state =
3282                                                 KVM_MP_STATE_RUNNABLE;
3283                                 case KVM_MP_STATE_RUNNABLE:
3284                                         break;
3285                                 case KVM_MP_STATE_SIPI_RECEIVED:
3286                                 default:
3287                                         r = -EINTR;
3288                                         break;
3289                                 }
3290                         }
3291                 }
3292
3293                 if (r <= 0)
3294                         break;
3295
3296                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3297                 if (kvm_cpu_has_pending_timer(vcpu))
3298                         kvm_inject_pending_timer_irqs(vcpu);
3299
3300                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3301                         r = -EINTR;
3302                         kvm_run->exit_reason = KVM_EXIT_INTR;
3303                         ++vcpu->stat.request_irq_exits;
3304                 }
3305                 if (signal_pending(current)) {
3306                         r = -EINTR;
3307                         kvm_run->exit_reason = KVM_EXIT_INTR;
3308                         ++vcpu->stat.signal_exits;
3309                 }
3310                 if (need_resched()) {
3311                         up_read(&vcpu->kvm->slots_lock);
3312                         kvm_resched(vcpu);
3313                         down_read(&vcpu->kvm->slots_lock);
3314                 }
3315         }
3316
3317         up_read(&vcpu->kvm->slots_lock);
3318         post_kvm_run_save(vcpu, kvm_run);
3319
3320         vapic_exit(vcpu);
3321
3322         return r;
3323 }
3324
3325 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3326 {
3327         int r;
3328         sigset_t sigsaved;
3329
3330         vcpu_load(vcpu);
3331
3332         if (vcpu->sigset_active)
3333                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3334
3335         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3336                 kvm_vcpu_block(vcpu);
3337                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3338                 r = -EAGAIN;
3339                 goto out;
3340         }
3341
3342         /* re-sync apic's tpr */
3343         if (!irqchip_in_kernel(vcpu->kvm))
3344                 kvm_set_cr8(vcpu, kvm_run->cr8);
3345
3346         if (vcpu->arch.pio.cur_count) {
3347                 r = complete_pio(vcpu);
3348                 if (r)
3349                         goto out;
3350         }
3351 #if CONFIG_HAS_IOMEM
3352         if (vcpu->mmio_needed) {
3353                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3354                 vcpu->mmio_read_completed = 1;
3355                 vcpu->mmio_needed = 0;
3356
3357                 down_read(&vcpu->kvm->slots_lock);
3358                 r = emulate_instruction(vcpu, kvm_run,
3359                                         vcpu->arch.mmio_fault_cr2, 0,
3360                                         EMULTYPE_NO_DECODE);
3361                 up_read(&vcpu->kvm->slots_lock);
3362                 if (r == EMULATE_DO_MMIO) {
3363                         /*
3364                          * Read-modify-write.  Back to userspace.
3365                          */
3366                         r = 0;
3367                         goto out;
3368                 }
3369         }
3370 #endif
3371         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3372                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3373                                      kvm_run->hypercall.ret);
3374
3375         r = __vcpu_run(vcpu, kvm_run);
3376
3377 out:
3378         if (vcpu->sigset_active)
3379                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3380
3381         vcpu_put(vcpu);
3382         return r;
3383 }
3384
3385 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3386 {
3387         vcpu_load(vcpu);
3388
3389         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3390         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3391         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3392         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3393         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3394         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3395         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3396         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3397 #ifdef CONFIG_X86_64
3398         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3399         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3400         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3401         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3402         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3403         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3404         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3405         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3406 #endif
3407
3408         regs->rip = kvm_rip_read(vcpu);
3409         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3410
3411         /*
3412          * Don't leak debug flags in case they were set for guest debugging
3413          */
3414         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3415                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3416
3417         vcpu_put(vcpu);
3418
3419         return 0;
3420 }
3421
3422 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3423 {
3424         vcpu_load(vcpu);
3425
3426         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3427         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3428         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3429         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3430         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3431         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3432         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3433         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3434 #ifdef CONFIG_X86_64
3435         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3436         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3437         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3438         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3439         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3440         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3441         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3442         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3443
3444 #endif
3445
3446         kvm_rip_write(vcpu, regs->rip);
3447         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3448
3449
3450         vcpu->arch.exception.pending = false;
3451
3452         vcpu_put(vcpu);
3453
3454         return 0;
3455 }
3456
3457 void kvm_get_segment(struct kvm_vcpu *vcpu,
3458                      struct kvm_segment *var, int seg)
3459 {
3460         kvm_x86_ops->get_segment(vcpu, var, seg);
3461 }
3462
3463 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3464 {
3465         struct kvm_segment cs;
3466
3467         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3468         *db = cs.db;
3469         *l = cs.l;
3470 }
3471 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3472
3473 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3474                                   struct kvm_sregs *sregs)
3475 {
3476         struct descriptor_table dt;
3477         int pending_vec;
3478
3479         vcpu_load(vcpu);
3480
3481         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3482         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3483         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3484         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3485         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3486         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3487
3488         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3489         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3490
3491         kvm_x86_ops->get_idt(vcpu, &dt);
3492         sregs->idt.limit = dt.limit;
3493         sregs->idt.base = dt.base;
3494         kvm_x86_ops->get_gdt(vcpu, &dt);
3495         sregs->gdt.limit = dt.limit;
3496         sregs->gdt.base = dt.base;
3497
3498         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3499         sregs->cr0 = vcpu->arch.cr0;
3500         sregs->cr2 = vcpu->arch.cr2;
3501         sregs->cr3 = vcpu->arch.cr3;
3502         sregs->cr4 = vcpu->arch.cr4;
3503         sregs->cr8 = kvm_get_cr8(vcpu);
3504         sregs->efer = vcpu->arch.shadow_efer;
3505         sregs->apic_base = kvm_get_apic_base(vcpu);
3506
3507         if (irqchip_in_kernel(vcpu->kvm)) {
3508                 memset(sregs->interrupt_bitmap, 0,
3509                        sizeof sregs->interrupt_bitmap);
3510                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3511                 if (pending_vec >= 0)
3512                         set_bit(pending_vec,
3513                                 (unsigned long *)sregs->interrupt_bitmap);
3514         } else
3515                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3516                        sizeof sregs->interrupt_bitmap);
3517
3518         vcpu_put(vcpu);
3519
3520         return 0;
3521 }
3522
3523 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3524                                     struct kvm_mp_state *mp_state)
3525 {
3526         vcpu_load(vcpu);
3527         mp_state->mp_state = vcpu->arch.mp_state;
3528         vcpu_put(vcpu);
3529         return 0;
3530 }
3531
3532 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3533                                     struct kvm_mp_state *mp_state)
3534 {
3535         vcpu_load(vcpu);
3536         vcpu->arch.mp_state = mp_state->mp_state;
3537         vcpu_put(vcpu);
3538         return 0;
3539 }
3540
3541 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3542                         struct kvm_segment *var, int seg)
3543 {
3544         kvm_x86_ops->set_segment(vcpu, var, seg);
3545 }
3546
3547 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3548                                    struct kvm_segment *kvm_desct)
3549 {
3550         kvm_desct->base = seg_desc->base0;
3551         kvm_desct->base |= seg_desc->base1 << 16;
3552         kvm_desct->base |= seg_desc->base2 << 24;
3553         kvm_desct->limit = seg_desc->limit0;
3554         kvm_desct->limit |= seg_desc->limit << 16;
3555         if (seg_desc->g) {
3556                 kvm_desct->limit <<= 12;
3557                 kvm_desct->limit |= 0xfff;
3558         }
3559         kvm_desct->selector = selector;
3560         kvm_desct->type = seg_desc->type;
3561         kvm_desct->present = seg_desc->p;
3562         kvm_desct->dpl = seg_desc->dpl;
3563         kvm_desct->db = seg_desc->d;
3564         kvm_desct->s = seg_desc->s;
3565         kvm_desct->l = seg_desc->l;
3566         kvm_desct->g = seg_desc->g;
3567         kvm_desct->avl = seg_desc->avl;
3568         if (!selector)
3569                 kvm_desct->unusable = 1;
3570         else
3571                 kvm_desct->unusable = 0;
3572         kvm_desct->padding = 0;
3573 }
3574
3575 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3576                                           u16 selector,
3577                                           struct descriptor_table *dtable)
3578 {
3579         if (selector & 1 << 2) {
3580                 struct kvm_segment kvm_seg;
3581
3582                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3583
3584                 if (kvm_seg.unusable)
3585                         dtable->limit = 0;
3586                 else
3587                         dtable->limit = kvm_seg.limit;
3588                 dtable->base = kvm_seg.base;
3589         }
3590         else
3591                 kvm_x86_ops->get_gdt(vcpu, dtable);
3592 }
3593
3594 /* allowed just for 8 bytes segments */
3595 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3596                                          struct desc_struct *seg_desc)
3597 {
3598         gpa_t gpa;
3599         struct descriptor_table dtable;
3600         u16 index = selector >> 3;
3601
3602         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3603
3604         if (dtable.limit < index * 8 + 7) {
3605                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3606                 return 1;
3607         }
3608         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3609         gpa += index * 8;
3610         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3611 }
3612
3613 /* allowed just for 8 bytes segments */
3614 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3615                                          struct desc_struct *seg_desc)
3616 {
3617         gpa_t gpa;
3618         struct descriptor_table dtable;
3619         u16 index = selector >> 3;
3620
3621         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3622
3623         if (dtable.limit < index * 8 + 7)
3624                 return 1;
3625         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3626         gpa += index * 8;
3627         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3628 }
3629
3630 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3631                              struct desc_struct *seg_desc)
3632 {
3633         u32 base_addr;
3634
3635         base_addr = seg_desc->base0;
3636         base_addr |= (seg_desc->base1 << 16);
3637         base_addr |= (seg_desc->base2 << 24);
3638
3639         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3640 }
3641
3642 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3643 {
3644         struct kvm_segment kvm_seg;
3645
3646         kvm_get_segment(vcpu, &kvm_seg, seg);
3647         return kvm_seg.selector;
3648 }
3649
3650 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3651                                                 u16 selector,
3652                                                 struct kvm_segment *kvm_seg)
3653 {
3654         struct desc_struct seg_desc;
3655
3656         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3657                 return 1;
3658         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3659         return 0;
3660 }
3661
3662 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3663 {
3664         struct kvm_segment segvar = {
3665                 .base = selector << 4,
3666                 .limit = 0xffff,
3667                 .selector = selector,
3668                 .type = 3,
3669                 .present = 1,
3670                 .dpl = 3,
3671                 .db = 0,
3672                 .s = 1,
3673                 .l = 0,
3674                 .g = 0,
3675                 .avl = 0,
3676                 .unusable = 0,
3677         };
3678         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3679         return 0;
3680 }
3681
3682 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3683                                 int type_bits, int seg)
3684 {
3685         struct kvm_segment kvm_seg;
3686
3687         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3688                 return kvm_load_realmode_segment(vcpu, selector, seg);
3689         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3690                 return 1;
3691         kvm_seg.type |= type_bits;
3692
3693         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3694             seg != VCPU_SREG_LDTR)
3695                 if (!kvm_seg.s)
3696                         kvm_seg.unusable = 1;
3697
3698         kvm_set_segment(vcpu, &kvm_seg, seg);
3699         return 0;
3700 }
3701
3702 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3703                                 struct tss_segment_32 *tss)
3704 {
3705         tss->cr3 = vcpu->arch.cr3;
3706         tss->eip = kvm_rip_read(vcpu);
3707         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3708         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3709         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3710         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3711         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3712         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3713         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3714         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3715         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3716         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3717         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3718         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3719         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3720         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3721         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3722         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3723 }
3724
3725 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3726                                   struct tss_segment_32 *tss)
3727 {
3728         kvm_set_cr3(vcpu, tss->cr3);
3729
3730         kvm_rip_write(vcpu, tss->eip);
3731         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3732
3733         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3734         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3735         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3736         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3737         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3738         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3739         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3740         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3741
3742         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3743                 return 1;
3744
3745         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3746                 return 1;
3747
3748         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3749                 return 1;
3750
3751         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3752                 return 1;
3753
3754         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3755                 return 1;
3756
3757         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3758                 return 1;
3759
3760         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3761                 return 1;
3762         return 0;
3763 }
3764
3765 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3766                                 struct tss_segment_16 *tss)
3767 {
3768         tss->ip = kvm_rip_read(vcpu);
3769         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3770         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3771         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3772         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3773         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3774         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3775         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3776         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3777         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3778
3779         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3780         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3781         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3782         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3783         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3784         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3785 }
3786
3787 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3788                                  struct tss_segment_16 *tss)
3789 {
3790         kvm_rip_write(vcpu, tss->ip);
3791         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3792         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3793         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3794         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3795         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3796         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3797         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3798         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3799         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3800
3801         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3802                 return 1;
3803
3804         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3805                 return 1;
3806
3807         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3808                 return 1;
3809
3810         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3811                 return 1;
3812
3813         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3814                 return 1;
3815         return 0;
3816 }
3817
3818 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3819                               u16 old_tss_sel, u32 old_tss_base,
3820                               struct desc_struct *nseg_desc)
3821 {
3822         struct tss_segment_16 tss_segment_16;
3823         int ret = 0;
3824
3825         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3826                            sizeof tss_segment_16))
3827                 goto out;
3828
3829         save_state_to_tss16(vcpu, &tss_segment_16);
3830
3831         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3832                             sizeof tss_segment_16))
3833                 goto out;
3834
3835         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3836                            &tss_segment_16, sizeof tss_segment_16))
3837                 goto out;
3838
3839         if (old_tss_sel != 0xffff) {
3840                 tss_segment_16.prev_task_link = old_tss_sel;
3841
3842                 if (kvm_write_guest(vcpu->kvm,
3843                                     get_tss_base_addr(vcpu, nseg_desc),
3844                                     &tss_segment_16.prev_task_link,
3845                                     sizeof tss_segment_16.prev_task_link))
3846                         goto out;
3847         }
3848
3849         if (load_state_from_tss16(vcpu, &tss_segment_16))
3850                 goto out;
3851
3852         ret = 1;
3853 out:
3854         return ret;
3855 }
3856
3857 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3858                        u16 old_tss_sel, u32 old_tss_base,
3859                        struct desc_struct *nseg_desc)
3860 {
3861         struct tss_segment_32 tss_segment_32;
3862         int ret = 0;
3863
3864         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3865                            sizeof tss_segment_32))
3866                 goto out;
3867
3868         save_state_to_tss32(vcpu, &tss_segment_32);
3869
3870         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3871                             sizeof tss_segment_32))
3872                 goto out;
3873
3874         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3875                            &tss_segment_32, sizeof tss_segment_32))
3876                 goto out;
3877
3878         if (old_tss_sel != 0xffff) {
3879                 tss_segment_32.prev_task_link = old_tss_sel;
3880
3881                 if (kvm_write_guest(vcpu->kvm,
3882                                     get_tss_base_addr(vcpu, nseg_desc),
3883                                     &tss_segment_32.prev_task_link,
3884                                     sizeof tss_segment_32.prev_task_link))
3885                         goto out;
3886         }
3887
3888         if (load_state_from_tss32(vcpu, &tss_segment_32))
3889                 goto out;
3890
3891         ret = 1;
3892 out:
3893         return ret;
3894 }
3895
3896 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3897 {
3898         struct kvm_segment tr_seg;
3899         struct desc_struct cseg_desc;
3900         struct desc_struct nseg_desc;
3901         int ret = 0;
3902         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3903         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3904
3905         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3906
3907         /* FIXME: Handle errors. Failure to read either TSS or their
3908          * descriptors should generate a pagefault.
3909          */
3910         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3911                 goto out;
3912
3913         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3914                 goto out;
3915
3916         if (reason != TASK_SWITCH_IRET) {
3917                 int cpl;
3918
3919                 cpl = kvm_x86_ops->get_cpl(vcpu);
3920                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3921                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3922                         return 1;
3923                 }
3924         }
3925
3926         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3927                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3928                 return 1;
3929         }
3930
3931         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3932                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3933                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3934         }
3935
3936         if (reason == TASK_SWITCH_IRET) {
3937                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3938                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3939         }
3940
3941         /* set back link to prev task only if NT bit is set in eflags
3942            note that old_tss_sel is not used afetr this point */
3943         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3944                 old_tss_sel = 0xffff;
3945
3946         /* set back link to prev task only if NT bit is set in eflags
3947            note that old_tss_sel is not used afetr this point */
3948         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3949                 old_tss_sel = 0xffff;
3950
3951         if (nseg_desc.type & 8)
3952                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
3953                                          old_tss_base, &nseg_desc);
3954         else
3955                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
3956                                          old_tss_base, &nseg_desc);
3957
3958         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3959                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3960                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3961         }
3962
3963         if (reason != TASK_SWITCH_IRET) {
3964                 nseg_desc.type |= (1 << 1);
3965                 save_guest_segment_descriptor(vcpu, tss_selector,
3966                                               &nseg_desc);
3967         }
3968
3969         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3970         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3971         tr_seg.type = 11;
3972         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3973 out:
3974         return ret;
3975 }
3976 EXPORT_SYMBOL_GPL(kvm_task_switch);
3977
3978 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3979                                   struct kvm_sregs *sregs)
3980 {
3981         int mmu_reset_needed = 0;
3982         int i, pending_vec, max_bits;
3983         struct descriptor_table dt;
3984
3985         vcpu_load(vcpu);
3986
3987         dt.limit = sregs->idt.limit;
3988         dt.base = sregs->idt.base;
3989         kvm_x86_ops->set_idt(vcpu, &dt);
3990         dt.limit = sregs->gdt.limit;
3991         dt.base = sregs->gdt.base;
3992         kvm_x86_ops->set_gdt(vcpu, &dt);
3993
3994         vcpu->arch.cr2 = sregs->cr2;
3995         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3996
3997         down_read(&vcpu->kvm->slots_lock);
3998         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
3999                 vcpu->arch.cr3 = sregs->cr3;
4000         else
4001                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4002         up_read(&vcpu->kvm->slots_lock);
4003
4004         kvm_set_cr8(vcpu, sregs->cr8);
4005
4006         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4007         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4008         kvm_set_apic_base(vcpu, sregs->apic_base);
4009
4010         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4011
4012         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4013         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4014         vcpu->arch.cr0 = sregs->cr0;
4015
4016         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4017         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4018         if (!is_long_mode(vcpu) && is_pae(vcpu))
4019                 load_pdptrs(vcpu, vcpu->arch.cr3);
4020
4021         if (mmu_reset_needed)
4022                 kvm_mmu_reset_context(vcpu);
4023
4024         if (!irqchip_in_kernel(vcpu->kvm)) {
4025                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
4026                        sizeof vcpu->arch.irq_pending);
4027                 vcpu->arch.irq_summary = 0;
4028                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
4029                         if (vcpu->arch.irq_pending[i])
4030                                 __set_bit(i, &vcpu->arch.irq_summary);
4031         } else {
4032                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4033                 pending_vec = find_first_bit(
4034                         (const unsigned long *)sregs->interrupt_bitmap,
4035                         max_bits);
4036                 /* Only pending external irq is handled here */
4037                 if (pending_vec < max_bits) {
4038                         kvm_x86_ops->set_irq(vcpu, pending_vec);
4039                         pr_debug("Set back pending irq %d\n",
4040                                  pending_vec);
4041                 }
4042                 kvm_pic_clear_isr_ack(vcpu->kvm);
4043         }
4044
4045         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4046         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4047         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4048         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4049         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4050         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4051
4052         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4053         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4054
4055         /* Older userspace won't unhalt the vcpu on reset. */
4056         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4057             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4058             !(vcpu->arch.cr0 & X86_CR0_PE))
4059                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4060
4061         vcpu_put(vcpu);
4062
4063         return 0;
4064 }
4065
4066 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4067                                         struct kvm_guest_debug *dbg)
4068 {
4069         int i, r;
4070
4071         vcpu_load(vcpu);
4072
4073         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4074             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4075                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4076                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4077                 vcpu->arch.switch_db_regs =
4078                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4079         } else {
4080                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4081                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4082                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4083         }
4084
4085         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4086
4087         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4088                 kvm_queue_exception(vcpu, DB_VECTOR);
4089         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4090                 kvm_queue_exception(vcpu, BP_VECTOR);
4091
4092         vcpu_put(vcpu);
4093
4094         return r;
4095 }
4096
4097 /*
4098  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4099  * we have asm/x86/processor.h
4100  */
4101 struct fxsave {
4102         u16     cwd;
4103         u16     swd;
4104         u16     twd;
4105         u16     fop;
4106         u64     rip;
4107         u64     rdp;
4108         u32     mxcsr;
4109         u32     mxcsr_mask;
4110         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4111 #ifdef CONFIG_X86_64
4112         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4113 #else
4114         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4115 #endif
4116 };
4117
4118 /*
4119  * Translate a guest virtual address to a guest physical address.
4120  */
4121 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4122                                     struct kvm_translation *tr)
4123 {
4124         unsigned long vaddr = tr->linear_address;
4125         gpa_t gpa;
4126
4127         vcpu_load(vcpu);
4128         down_read(&vcpu->kvm->slots_lock);
4129         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4130         up_read(&vcpu->kvm->slots_lock);
4131         tr->physical_address = gpa;
4132         tr->valid = gpa != UNMAPPED_GVA;
4133         tr->writeable = 1;
4134         tr->usermode = 0;
4135         vcpu_put(vcpu);
4136
4137         return 0;
4138 }
4139
4140 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4141 {
4142         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4143
4144         vcpu_load(vcpu);
4145
4146         memcpy(fpu->fpr, fxsave->st_space, 128);
4147         fpu->fcw = fxsave->cwd;
4148         fpu->fsw = fxsave->swd;
4149         fpu->ftwx = fxsave->twd;
4150         fpu->last_opcode = fxsave->fop;
4151         fpu->last_ip = fxsave->rip;
4152         fpu->last_dp = fxsave->rdp;
4153         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4154
4155         vcpu_put(vcpu);
4156
4157         return 0;
4158 }
4159
4160 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4161 {
4162         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4163
4164         vcpu_load(vcpu);
4165
4166         memcpy(fxsave->st_space, fpu->fpr, 128);
4167         fxsave->cwd = fpu->fcw;
4168         fxsave->swd = fpu->fsw;
4169         fxsave->twd = fpu->ftwx;
4170         fxsave->fop = fpu->last_opcode;
4171         fxsave->rip = fpu->last_ip;
4172         fxsave->rdp = fpu->last_dp;
4173         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4174
4175         vcpu_put(vcpu);
4176
4177         return 0;
4178 }
4179
4180 void fx_init(struct kvm_vcpu *vcpu)
4181 {
4182         unsigned after_mxcsr_mask;
4183
4184         /*
4185          * Touch the fpu the first time in non atomic context as if
4186          * this is the first fpu instruction the exception handler
4187          * will fire before the instruction returns and it'll have to
4188          * allocate ram with GFP_KERNEL.
4189          */
4190         if (!used_math())
4191                 kvm_fx_save(&vcpu->arch.host_fx_image);
4192
4193         /* Initialize guest FPU by resetting ours and saving into guest's */
4194         preempt_disable();
4195         kvm_fx_save(&vcpu->arch.host_fx_image);
4196         kvm_fx_finit();
4197         kvm_fx_save(&vcpu->arch.guest_fx_image);
4198         kvm_fx_restore(&vcpu->arch.host_fx_image);
4199         preempt_enable();
4200
4201         vcpu->arch.cr0 |= X86_CR0_ET;
4202         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4203         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4204         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4205                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4206 }
4207 EXPORT_SYMBOL_GPL(fx_init);
4208
4209 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4210 {
4211         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4212                 return;
4213
4214         vcpu->guest_fpu_loaded = 1;
4215         kvm_fx_save(&vcpu->arch.host_fx_image);
4216         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4217 }
4218 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4219
4220 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4221 {
4222         if (!vcpu->guest_fpu_loaded)
4223                 return;
4224
4225         vcpu->guest_fpu_loaded = 0;
4226         kvm_fx_save(&vcpu->arch.guest_fx_image);
4227         kvm_fx_restore(&vcpu->arch.host_fx_image);
4228         ++vcpu->stat.fpu_reload;
4229 }
4230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4231
4232 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4233 {
4234         if (vcpu->arch.time_page) {
4235                 kvm_release_page_dirty(vcpu->arch.time_page);
4236                 vcpu->arch.time_page = NULL;
4237         }
4238
4239         kvm_x86_ops->vcpu_free(vcpu);
4240 }
4241
4242 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4243                                                 unsigned int id)
4244 {
4245         return kvm_x86_ops->vcpu_create(kvm, id);
4246 }
4247
4248 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4249 {
4250         int r;
4251
4252         /* We do fxsave: this must be aligned. */
4253         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4254
4255         vcpu->arch.mtrr_state.have_fixed = 1;
4256         vcpu_load(vcpu);
4257         r = kvm_arch_vcpu_reset(vcpu);
4258         if (r == 0)
4259                 r = kvm_mmu_setup(vcpu);
4260         vcpu_put(vcpu);
4261         if (r < 0)
4262                 goto free_vcpu;
4263
4264         return 0;
4265 free_vcpu:
4266         kvm_x86_ops->vcpu_free(vcpu);
4267         return r;
4268 }
4269
4270 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4271 {
4272         vcpu_load(vcpu);
4273         kvm_mmu_unload(vcpu);
4274         vcpu_put(vcpu);
4275
4276         kvm_x86_ops->vcpu_free(vcpu);
4277 }
4278
4279 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4280 {
4281         vcpu->arch.nmi_pending = false;
4282         vcpu->arch.nmi_injected = false;
4283
4284         vcpu->arch.switch_db_regs = 0;
4285         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4286         vcpu->arch.dr6 = DR6_FIXED_1;
4287         vcpu->arch.dr7 = DR7_FIXED_1;
4288
4289         return kvm_x86_ops->vcpu_reset(vcpu);
4290 }
4291
4292 void kvm_arch_hardware_enable(void *garbage)
4293 {
4294         kvm_x86_ops->hardware_enable(garbage);
4295 }
4296
4297 void kvm_arch_hardware_disable(void *garbage)
4298 {
4299         kvm_x86_ops->hardware_disable(garbage);
4300 }
4301
4302 int kvm_arch_hardware_setup(void)
4303 {
4304         return kvm_x86_ops->hardware_setup();
4305 }
4306
4307 void kvm_arch_hardware_unsetup(void)
4308 {
4309         kvm_x86_ops->hardware_unsetup();
4310 }
4311
4312 void kvm_arch_check_processor_compat(void *rtn)
4313 {
4314         kvm_x86_ops->check_processor_compatibility(rtn);
4315 }
4316
4317 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4318 {
4319         struct page *page;
4320         struct kvm *kvm;
4321         int r;
4322
4323         BUG_ON(vcpu->kvm == NULL);
4324         kvm = vcpu->kvm;
4325
4326         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4327         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4328                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4329         else
4330                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4331
4332         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4333         if (!page) {
4334                 r = -ENOMEM;
4335                 goto fail;
4336         }
4337         vcpu->arch.pio_data = page_address(page);
4338
4339         r = kvm_mmu_create(vcpu);
4340         if (r < 0)
4341                 goto fail_free_pio_data;
4342
4343         if (irqchip_in_kernel(kvm)) {
4344                 r = kvm_create_lapic(vcpu);
4345                 if (r < 0)
4346                         goto fail_mmu_destroy;
4347         }
4348
4349         return 0;
4350
4351 fail_mmu_destroy:
4352         kvm_mmu_destroy(vcpu);
4353 fail_free_pio_data:
4354         free_page((unsigned long)vcpu->arch.pio_data);
4355 fail:
4356         return r;
4357 }
4358
4359 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4360 {
4361         kvm_free_lapic(vcpu);
4362         down_read(&vcpu->kvm->slots_lock);
4363         kvm_mmu_destroy(vcpu);
4364         up_read(&vcpu->kvm->slots_lock);
4365         free_page((unsigned long)vcpu->arch.pio_data);
4366 }
4367
4368 struct  kvm *kvm_arch_create_vm(void)
4369 {
4370         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4371
4372         if (!kvm)
4373                 return ERR_PTR(-ENOMEM);
4374
4375         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4376         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4377
4378         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4379         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4380
4381         rdtscll(kvm->arch.vm_init_tsc);
4382
4383         return kvm;
4384 }
4385
4386 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4387 {
4388         vcpu_load(vcpu);
4389         kvm_mmu_unload(vcpu);
4390         vcpu_put(vcpu);
4391 }
4392
4393 static void kvm_free_vcpus(struct kvm *kvm)
4394 {
4395         unsigned int i;
4396
4397         /*
4398          * Unpin any mmu pages first.
4399          */
4400         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4401                 if (kvm->vcpus[i])
4402                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4403         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4404                 if (kvm->vcpus[i]) {
4405                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4406                         kvm->vcpus[i] = NULL;
4407                 }
4408         }
4409
4410 }
4411
4412 void kvm_arch_sync_events(struct kvm *kvm)
4413 {
4414         kvm_free_all_assigned_devices(kvm);
4415 }
4416
4417 void kvm_arch_destroy_vm(struct kvm *kvm)
4418 {
4419         kvm_iommu_unmap_guest(kvm);
4420         kvm_free_pit(kvm);
4421         kfree(kvm->arch.vpic);
4422         kfree(kvm->arch.vioapic);
4423         kvm_free_vcpus(kvm);
4424         kvm_free_physmem(kvm);
4425         if (kvm->arch.apic_access_page)
4426                 put_page(kvm->arch.apic_access_page);
4427         if (kvm->arch.ept_identity_pagetable)
4428                 put_page(kvm->arch.ept_identity_pagetable);
4429         kfree(kvm);
4430 }
4431
4432 int kvm_arch_set_memory_region(struct kvm *kvm,
4433                                 struct kvm_userspace_memory_region *mem,
4434                                 struct kvm_memory_slot old,
4435                                 int user_alloc)
4436 {
4437         int npages = mem->memory_size >> PAGE_SHIFT;
4438         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4439
4440         /*To keep backward compatibility with older userspace,
4441          *x86 needs to hanlde !user_alloc case.
4442          */
4443         if (!user_alloc) {
4444                 if (npages && !old.rmap) {
4445                         unsigned long userspace_addr;
4446
4447                         down_write(&current->mm->mmap_sem);
4448                         userspace_addr = do_mmap(NULL, 0,
4449                                                  npages * PAGE_SIZE,
4450                                                  PROT_READ | PROT_WRITE,
4451                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4452                                                  0);
4453                         up_write(&current->mm->mmap_sem);
4454
4455                         if (IS_ERR((void *)userspace_addr))
4456                                 return PTR_ERR((void *)userspace_addr);
4457
4458                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4459                         spin_lock(&kvm->mmu_lock);
4460                         memslot->userspace_addr = userspace_addr;
4461                         spin_unlock(&kvm->mmu_lock);
4462                 } else {
4463                         if (!old.user_alloc && old.rmap) {
4464                                 int ret;
4465
4466                                 down_write(&current->mm->mmap_sem);
4467                                 ret = do_munmap(current->mm, old.userspace_addr,
4468                                                 old.npages * PAGE_SIZE);
4469                                 up_write(&current->mm->mmap_sem);
4470                                 if (ret < 0)
4471                                         printk(KERN_WARNING
4472                                        "kvm_vm_ioctl_set_memory_region: "
4473                                        "failed to munmap memory\n");
4474                         }
4475                 }
4476         }
4477
4478         if (!kvm->arch.n_requested_mmu_pages) {
4479                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4480                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4481         }
4482
4483         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4484         kvm_flush_remote_tlbs(kvm);
4485
4486         return 0;
4487 }
4488
4489 void kvm_arch_flush_shadow(struct kvm *kvm)
4490 {
4491         kvm_mmu_zap_all(kvm);
4492 }
4493
4494 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4495 {
4496         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4497                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4498                || vcpu->arch.nmi_pending;
4499 }
4500
4501 static void vcpu_kick_intr(void *info)
4502 {
4503 #ifdef DEBUG
4504         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4505         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4506 #endif
4507 }
4508
4509 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4510 {
4511         int ipi_pcpu = vcpu->cpu;
4512         int cpu = get_cpu();
4513
4514         if (waitqueue_active(&vcpu->wq)) {
4515                 wake_up_interruptible(&vcpu->wq);
4516                 ++vcpu->stat.halt_wakeup;
4517         }
4518         /*
4519          * We may be called synchronously with irqs disabled in guest mode,
4520          * So need not to call smp_call_function_single() in that case.
4521          */
4522         if (vcpu->guest_mode && vcpu->cpu != cpu)
4523                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4524         put_cpu();
4525 }
4526
4527 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4528 {
4529         return kvm_x86_ops->interrupt_allowed(vcpu);
4530 }