err.no Git - linux-2.6/blob - arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   8  * Authors:
   9  *   Avi Kivity   <avi@qumranet.com>
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include <linux/kvm_host.h>
  18 #include "irq.h"
  19 #include "mmu.h"
  20 #include "i8254.h"
  21 #include "tss.h"
  22
  23 #include <linux/clocksource.h>
  24 #include <linux/kvm.h>
  25 #include <linux/fs.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/module.h>
  28 #include <linux/mman.h>
  29 #include <linux/highmem.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/msr.h>
  33 #include <asm/desc.h>
  34
  35 #define MAX_IO_MSRS 256
  36 #define CR0_RESERVED_BITS                                               \
  37         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  38                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  39                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  40 #define CR4_RESERVED_BITS                                               \
  41         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  42                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  43                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  44                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  45
  46 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  47 /* EFER defaults:
  48  * - enable syscall per default because its emulated by KVM
  49  * - enable LME and LMA per default on 64 bit KVM
  50  */
  51 #ifdef CONFIG_X86_64
  52 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  53 #else
  54 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  55 #endif
  56
  57 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  59
  60 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  61                                     struct kvm_cpuid_entry2 __user *entries);
  62
  63 struct kvm_x86_ops *kvm_x86_ops;
  64
  65 struct kvm_stats_debugfs_item debugfs_entries[] = {
  66         { "pf_fixed", VCPU_STAT(pf_fixed) },
  67         { "pf_guest", VCPU_STAT(pf_guest) },
  68         { "tlb_flush", VCPU_STAT(tlb_flush) },
  69         { "invlpg", VCPU_STAT(invlpg) },
  70         { "exits", VCPU_STAT(exits) },
  71         { "io_exits", VCPU_STAT(io_exits) },
  72         { "mmio_exits", VCPU_STAT(mmio_exits) },
  73         { "signal_exits", VCPU_STAT(signal_exits) },
  74         { "irq_window", VCPU_STAT(irq_window_exits) },
  75         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  76         { "halt_exits", VCPU_STAT(halt_exits) },
  77         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  78         { "hypercalls", VCPU_STAT(hypercalls) },
  79         { "request_irq", VCPU_STAT(request_irq_exits) },
  80         { "irq_exits", VCPU_STAT(irq_exits) },
  81         { "host_state_reload", VCPU_STAT(host_state_reload) },
  82         { "efer_reload", VCPU_STAT(efer_reload) },
  83         { "fpu_reload", VCPU_STAT(fpu_reload) },
  84         { "insn_emulation", VCPU_STAT(insn_emulation) },
  85         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  86         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  87         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  88         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  89         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  90         { "mmu_flooded", VM_STAT(mmu_flooded) },
  91         { "mmu_recycled", VM_STAT(mmu_recycled) },
  92         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
  93         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  94         { "largepages", VM_STAT(lpages) },
  95         { NULL }
  96 };
  97
  98
  99 unsigned long segment_base(u16 selector)
 100 {
 101         struct descriptor_table gdt;
 102         struct desc_struct *d;
 103         unsigned long table_base;
 104         unsigned long v;
 105
 106         if (selector == 0)
 107                 return 0;
 108
 109         asm("sgdt %0" : "=m"(gdt));
 110         table_base = gdt.base;
 111
 112         if (selector & 4) {           /* from ldt */
 113                 u16 ldt_selector;
 114
 115                 asm("sldt %0" : "=g"(ldt_selector));
 116                 table_base = segment_base(ldt_selector);
 117         }
 118         d = (struct desc_struct *)(table_base + (selector & ~7));
 119         v = d->base0 | ((unsigned long)d->base1 << 16) |
 120                 ((unsigned long)d->base2 << 24);
 121 #ifdef CONFIG_X86_64
 122         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 123                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 124 #endif
 125         return v;
 126 }
 127 EXPORT_SYMBOL_GPL(segment_base);
 128
 129 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 130 {
 131         if (irqchip_in_kernel(vcpu->kvm))
 132                 return vcpu->arch.apic_base;
 133         else
 134                 return vcpu->arch.apic_base;
 135 }
 136 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 137
 138 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 139 {
 140         /* TODO: reserve bits check */
 141         if (irqchip_in_kernel(vcpu->kvm))
 142                 kvm_lapic_set_base(vcpu, data);
 143         else
 144                 vcpu->arch.apic_base = data;
 145 }
 146 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 147
 148 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 149 {
 150         WARN_ON(vcpu->arch.exception.pending);
 151         vcpu->arch.exception.pending = true;
 152         vcpu->arch.exception.has_error_code = false;
 153         vcpu->arch.exception.nr = nr;
 154 }
 155 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 156
 157 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 158                            u32 error_code)
 159 {
 160         ++vcpu->stat.pf_guest;
 161         if (vcpu->arch.exception.pending) {
 162                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 163                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 164                                         " double fault 0x%lx\n", addr);
 165                         vcpu->arch.exception.nr = DF_VECTOR;
 166                         vcpu->arch.exception.error_code = 0;
 167                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 168                         /* triple fault -> shutdown */
 169                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 170                 }
 171                 return;
 172         }
 173         vcpu->arch.cr2 = addr;
 174         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 175 }
 176
 177 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 178 {
 179         vcpu->arch.nmi_pending = 1;
 180 }
 181 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 182
 183 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 184 {
 185         WARN_ON(vcpu->arch.exception.pending);
 186         vcpu->arch.exception.pending = true;
 187         vcpu->arch.exception.has_error_code = true;
 188         vcpu->arch.exception.nr = nr;
 189         vcpu->arch.exception.error_code = error_code;
 190 }
 191 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 192
 193 static void __queue_exception(struct kvm_vcpu *vcpu)
 194 {
 195         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 196                                      vcpu->arch.exception.has_error_code,
 197                                      vcpu->arch.exception.error_code);
 198 }
 199
 200 /*
 201  * Load the pae pdptrs.  Return true is they are all valid.
 202  */
 203 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 204 {
 205         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 206         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 207         int i;
 208         int ret;
 209         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 210
 211         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 212                                   offset * sizeof(u64), sizeof(pdpte));
 213         if (ret < 0) {
 214                 ret = 0;
 215                 goto out;
 216         }
 217         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 218                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 219                         ret = 0;
 220                         goto out;
 221                 }
 222         }
 223         ret = 1;
 224
 225         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 226 out:
 227
 228         return ret;
 229 }
 230 EXPORT_SYMBOL_GPL(load_pdptrs);
 231
 232 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 233 {
 234         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 235         bool changed = true;
 236         int r;
 237
 238         if (is_long_mode(vcpu) || !is_pae(vcpu))
 239                 return false;
 240
 241         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 242         if (r < 0)
 243                 goto out;
 244         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 245 out:
 246
 247         return changed;
 248 }
 249
 250 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 251 {
 252         if (cr0 & CR0_RESERVED_BITS) {
 253                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 254                        cr0, vcpu->arch.cr0);
 255                 kvm_inject_gp(vcpu, 0);
 256                 return;
 257         }
 258
 259         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 260                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 261                 kvm_inject_gp(vcpu, 0);
 262                 return;
 263         }
 264
 265         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 266                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 267                        "and a clear PE flag\n");
 268                 kvm_inject_gp(vcpu, 0);
 269                 return;
 270         }
 271
 272         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 273 #ifdef CONFIG_X86_64
 274                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 275                         int cs_db, cs_l;
 276
 277                         if (!is_pae(vcpu)) {
 278                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 279                                        "in long mode while PAE is disabled\n");
 280                                 kvm_inject_gp(vcpu, 0);
 281                                 return;
 282                         }
 283                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 284                         if (cs_l) {
 285                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 286                                        "in long mode while CS.L == 1\n");
 287                                 kvm_inject_gp(vcpu, 0);
 288                                 return;
 289
 290                         }
 291                 } else
 292 #endif
 293                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 294                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 295                                "reserved bits\n");
 296                         kvm_inject_gp(vcpu, 0);
 297                         return;
 298                 }
 299
 300         }
 301
 302         kvm_x86_ops->set_cr0(vcpu, cr0);
 303         vcpu->arch.cr0 = cr0;
 304
 305         kvm_mmu_reset_context(vcpu);
 306         return;
 307 }
 308 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 309
 310 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 311 {
 312         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 313         KVMTRACE_1D(LMSW, vcpu,
 314                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 315                     handler);
 316 }
 317 EXPORT_SYMBOL_GPL(kvm_lmsw);
 318
 319 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 320 {
 321         if (cr4 & CR4_RESERVED_BITS) {
 322                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 323                 kvm_inject_gp(vcpu, 0);
 324                 return;
 325         }
 326
 327         if (is_long_mode(vcpu)) {
 328                 if (!(cr4 & X86_CR4_PAE)) {
 329                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 330                                "in long mode\n");
 331                         kvm_inject_gp(vcpu, 0);
 332                         return;
 333                 }
 334         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 335                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 336                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 337                 kvm_inject_gp(vcpu, 0);
 338                 return;
 339         }
 340
 341         if (cr4 & X86_CR4_VMXE) {
 342                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 343                 kvm_inject_gp(vcpu, 0);
 344                 return;
 345         }
 346         kvm_x86_ops->set_cr4(vcpu, cr4);
 347         vcpu->arch.cr4 = cr4;
 348         kvm_mmu_reset_context(vcpu);
 349 }
 350 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 351
 352 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 353 {
 354         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 355                 kvm_mmu_flush_tlb(vcpu);
 356                 return;
 357         }
 358
 359         if (is_long_mode(vcpu)) {
 360                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 361                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 362                         kvm_inject_gp(vcpu, 0);
 363                         return;
 364                 }
 365         } else {
 366                 if (is_pae(vcpu)) {
 367                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 368                                 printk(KERN_DEBUG
 369                                        "set_cr3: #GP, reserved bits\n");
 370                                 kvm_inject_gp(vcpu, 0);
 371                                 return;
 372                         }
 373                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 374                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 375                                        "reserved bits\n");
 376                                 kvm_inject_gp(vcpu, 0);
 377                                 return;
 378                         }
 379                 }
 380                 /*
 381                  * We don't check reserved bits in nonpae mode, because
 382                  * this isn't enforced, and VMware depends on this.
 383                  */
 384         }
 385
 386         /*
 387          * Does the new cr3 value map to physical memory? (Note, we
 388          * catch an invalid cr3 even in real-mode, because it would
 389          * cause trouble later on when we turn on paging anyway.)
 390          *
 391          * A real CPU would silently accept an invalid cr3 and would
 392          * attempt to use it - with largely undefined (and often hard
 393          * to debug) behavior on the guest side.
 394          */
 395         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 396                 kvm_inject_gp(vcpu, 0);
 397         else {
 398                 vcpu->arch.cr3 = cr3;
 399                 vcpu->arch.mmu.new_cr3(vcpu);
 400         }
 401 }
 402 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 403
 404 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 405 {
 406         if (cr8 & CR8_RESERVED_BITS) {
 407                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 408                 kvm_inject_gp(vcpu, 0);
 409                 return;
 410         }
 411         if (irqchip_in_kernel(vcpu->kvm))
 412                 kvm_lapic_set_tpr(vcpu, cr8);
 413         else
 414                 vcpu->arch.cr8 = cr8;
 415 }
 416 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 417
 418 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 419 {
 420         if (irqchip_in_kernel(vcpu->kvm))
 421                 return kvm_lapic_get_cr8(vcpu);
 422         else
 423                 return vcpu->arch.cr8;
 424 }
 425 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 426
 427 /*
 428  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 429  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 430  *
 431  * This list is modified at module load time to reflect the
 432  * capabilities of the host cpu.
 433  */
 434 static u32 msrs_to_save[] = {
 435         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 436         MSR_K6_STAR,
 437 #ifdef CONFIG_X86_64
 438         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 439 #endif
 440         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 441         MSR_IA32_PERF_STATUS,
 442 };
 443
 444 static unsigned num_msrs_to_save;
 445
 446 static u32 emulated_msrs[] = {
 447         MSR_IA32_MISC_ENABLE,
 448 };
 449
 450 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 451 {
 452         if (efer & efer_reserved_bits) {
 453                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 454                        efer);
 455                 kvm_inject_gp(vcpu, 0);
 456                 return;
 457         }
 458
 459         if (is_paging(vcpu)
 460             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 461                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 462                 kvm_inject_gp(vcpu, 0);
 463                 return;
 464         }
 465
 466         kvm_x86_ops->set_efer(vcpu, efer);
 467
 468         efer &= ~EFER_LMA;
 469         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 470
 471         vcpu->arch.shadow_efer = efer;
 472 }
 473
 474 void kvm_enable_efer_bits(u64 mask)
 475 {
 476        efer_reserved_bits &= ~mask;
 477 }
 478 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 479
 480
 481 /*
 482  * Writes msr value into into the appropriate "register".
 483  * Returns 0 on success, non-0 otherwise.
 484  * Assumes vcpu_load() was already called.
 485  */
 486 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 487 {
 488         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 489 }
 490
 491 /*
 492  * Adapt set_msr() to msr_io()'s calling convention
 493  */
 494 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 495 {
 496         return kvm_set_msr(vcpu, index, *data);
 497 }
 498
 499 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 500 {
 501         static int version;
 502         struct pvclock_wall_clock wc;
 503         struct timespec now, sys, boot;
 504
 505         if (!wall_clock)
 506                 return;
 507
 508         version++;
 509
 510         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 511
 512         /*
 513          * The guest calculates current wall clock time by adding
 514          * system time (updated by kvm_write_guest_time below) to the
 515          * wall clock specified here.  guest system time equals host
 516          * system time for us, thus we must fill in host boot time here.
 517          */
 518         now = current_kernel_time();
 519         ktime_get_ts(&sys);
 520         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 521
 522         wc.sec = boot.tv_sec;
 523         wc.nsec = boot.tv_nsec;
 524         wc.version = version;
 525
 526         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 527
 528         version++;
 529         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 530 }
 531
 532 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 533 {
 534         uint32_t quotient, remainder;
 535
 536         /* Don't try to replace with do_div(), this one calculates
 537          * "(dividend << 32) / divisor" */
 538         __asm__ ( "divl %4"
 539                   : "=a" (quotient), "=d" (remainder)
 540                   : "0" (0), "1" (dividend), "r" (divisor) );
 541         return quotient;
 542 }
 543
 544 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 545 {
 546         uint64_t nsecs = 1000000000LL;
 547         int32_t  shift = 0;
 548         uint64_t tps64;
 549         uint32_t tps32;
 550
 551         tps64 = tsc_khz * 1000LL;
 552         while (tps64 > nsecs*2) {
 553                 tps64 >>= 1;
 554                 shift--;
 555         }
 556
 557         tps32 = (uint32_t)tps64;
 558         while (tps32 <= (uint32_t)nsecs) {
 559                 tps32 <<= 1;
 560                 shift++;
 561         }
 562
 563         hv_clock->tsc_shift = shift;
 564         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 565
 566         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 567                  __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
 568                  hv_clock->tsc_to_system_mul);
 569 }
 570
 571 static void kvm_write_guest_time(struct kvm_vcpu *v)
 572 {
 573         struct timespec ts;
 574         unsigned long flags;
 575         struct kvm_vcpu_arch *vcpu = &v->arch;
 576         void *shared_kaddr;
 577
 578         if ((!vcpu->time_page))
 579                 return;
 580
 581         if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
 582                 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
 583                 vcpu->hv_clock_tsc_khz = tsc_khz;
 584         }
 585
 586         /* Keep irq disabled to prevent changes to the clock */
 587         local_irq_save(flags);
 588         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 589                           &vcpu->hv_clock.tsc_timestamp);
 590         ktime_get_ts(&ts);
 591         local_irq_restore(flags);
 592
 593         /* With all the info we got, fill in the values */
 594
 595         vcpu->hv_clock.system_time = ts.tv_nsec +
 596                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 597         /*
 598          * The interface expects us to write an even number signaling that the
 599          * update is finished. Since the guest won't see the intermediate
 600          * state, we just increase by 2 at the end.
 601          */
 602         vcpu->hv_clock.version += 2;
 603
 604         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 605
 606         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 607                sizeof(vcpu->hv_clock));
 608
 609         kunmap_atomic(shared_kaddr, KM_USER0);
 610
 611         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 612 }
 613
 614 static bool msr_mtrr_valid(unsigned msr)
 615 {
 616         switch (msr) {
 617         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 618         case MSR_MTRRfix64K_00000:
 619         case MSR_MTRRfix16K_80000:
 620         case MSR_MTRRfix16K_A0000:
 621         case MSR_MTRRfix4K_C0000:
 622         case MSR_MTRRfix4K_C8000:
 623         case MSR_MTRRfix4K_D0000:
 624         case MSR_MTRRfix4K_D8000:
 625         case MSR_MTRRfix4K_E0000:
 626         case MSR_MTRRfix4K_E8000:
 627         case MSR_MTRRfix4K_F0000:
 628         case MSR_MTRRfix4K_F8000:
 629         case MSR_MTRRdefType:
 630         case MSR_IA32_CR_PAT:
 631                 return true;
 632         case 0x2f8:
 633                 return true;
 634         }
 635         return false;
 636 }
 637
 638 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 639 {
 640         if (!msr_mtrr_valid(msr))
 641                 return 1;
 642
 643         vcpu->arch.mtrr[msr - 0x200] = data;
 644         return 0;
 645 }
 646
 647 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 648 {
 649         switch (msr) {
 650         case MSR_EFER:
 651                 set_efer(vcpu, data);
 652                 break;
 653         case MSR_IA32_MC0_STATUS:
 654                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 655                        __func__, data);
 656                 break;
 657         case MSR_IA32_MCG_STATUS:
 658                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 659                         __func__, data);
 660                 break;
 661         case MSR_IA32_MCG_CTL:
 662                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 663                         __func__, data);
 664                 break;
 665         case MSR_IA32_UCODE_REV:
 666         case MSR_IA32_UCODE_WRITE:
 667                 break;
 668         case 0x200 ... 0x2ff:
 669                 return set_msr_mtrr(vcpu, msr, data);
 670         case MSR_IA32_APICBASE:
 671                 kvm_set_apic_base(vcpu, data);
 672                 break;
 673         case MSR_IA32_MISC_ENABLE:
 674                 vcpu->arch.ia32_misc_enable_msr = data;
 675                 break;
 676         case MSR_KVM_WALL_CLOCK:
 677                 vcpu->kvm->arch.wall_clock = data;
 678                 kvm_write_wall_clock(vcpu->kvm, data);
 679                 break;
 680         case MSR_KVM_SYSTEM_TIME: {
 681                 if (vcpu->arch.time_page) {
 682                         kvm_release_page_dirty(vcpu->arch.time_page);
 683                         vcpu->arch.time_page = NULL;
 684                 }
 685
 686                 vcpu->arch.time = data;
 687
 688                 /* we verify if the enable bit is set... */
 689                 if (!(data & 1))
 690                         break;
 691
 692                 /* ...but clean it before doing the actual write */
 693                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 694
 695                 down_read(&current->mm->mmap_sem);
 696                 vcpu->arch.time_page =
 697                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 698                 up_read(&current->mm->mmap_sem);
 699
 700                 if (is_error_page(vcpu->arch.time_page)) {
 701                         kvm_release_page_clean(vcpu->arch.time_page);
 702                         vcpu->arch.time_page = NULL;
 703                 }
 704
 705                 kvm_write_guest_time(vcpu);
 706                 break;
 707         }
 708         default:
 709                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 710                 return 1;
 711         }
 712         return 0;
 713 }
 714 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 715
 716
 717 /*
 718  * Reads an msr value (of 'msr_index') into 'pdata'.
 719  * Returns 0 on success, non-0 otherwise.
 720  * Assumes vcpu_load() was already called.
 721  */
 722 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 723 {
 724         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 725 }
 726
 727 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 728 {
 729         if (!msr_mtrr_valid(msr))
 730                 return 1;
 731
 732         *pdata = vcpu->arch.mtrr[msr - 0x200];
 733         return 0;
 734 }
 735
 736 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 737 {
 738         u64 data;
 739
 740         switch (msr) {
 741         case 0xc0010010: /* SYSCFG */
 742         case 0xc0010015: /* HWCR */
 743         case MSR_IA32_PLATFORM_ID:
 744         case MSR_IA32_P5_MC_ADDR:
 745         case MSR_IA32_P5_MC_TYPE:
 746         case MSR_IA32_MC0_CTL:
 747         case MSR_IA32_MCG_STATUS:
 748         case MSR_IA32_MCG_CAP:
 749         case MSR_IA32_MCG_CTL:
 750         case MSR_IA32_MC0_MISC:
 751         case MSR_IA32_MC0_MISC+4:
 752         case MSR_IA32_MC0_MISC+8:
 753         case MSR_IA32_MC0_MISC+12:
 754         case MSR_IA32_MC0_MISC+16:
 755         case MSR_IA32_UCODE_REV:
 756         case MSR_IA32_EBL_CR_POWERON:
 757                 data = 0;
 758                 break;
 759         case MSR_MTRRcap:
 760                 data = 0x500 | KVM_NR_VAR_MTRR;
 761                 break;
 762         case 0x200 ... 0x2ff:
 763                 return get_msr_mtrr(vcpu, msr, pdata);
 764         case 0xcd: /* fsb frequency */
 765                 data = 3;
 766                 break;
 767         case MSR_IA32_APICBASE:
 768                 data = kvm_get_apic_base(vcpu);
 769                 break;
 770         case MSR_IA32_MISC_ENABLE:
 771                 data = vcpu->arch.ia32_misc_enable_msr;
 772                 break;
 773         case MSR_IA32_PERF_STATUS:
 774                 /* TSC increment by tick */
 775                 data = 1000ULL;
 776                 /* CPU multiplier */
 777                 data |= (((uint64_t)4ULL) << 40);
 778                 break;
 779         case MSR_EFER:
 780                 data = vcpu->arch.shadow_efer;
 781                 break;
 782         case MSR_KVM_WALL_CLOCK:
 783                 data = vcpu->kvm->arch.wall_clock;
 784                 break;
 785         case MSR_KVM_SYSTEM_TIME:
 786                 data = vcpu->arch.time;
 787                 break;
 788         default:
 789                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 790                 return 1;
 791         }
 792         *pdata = data;
 793         return 0;
 794 }
 795 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 796
 797 /*
 798  * Read or write a bunch of msrs. All parameters are kernel addresses.
 799  *
 800  * @return number of msrs set successfully.
 801  */
 802 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 803                     struct kvm_msr_entry *entries,
 804                     int (*do_msr)(struct kvm_vcpu *vcpu,
 805                                   unsigned index, u64 *data))
 806 {
 807         int i;
 808
 809         vcpu_load(vcpu);
 810
 811         down_read(&vcpu->kvm->slots_lock);
 812         for (i = 0; i < msrs->nmsrs; ++i)
 813                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 814                         break;
 815         up_read(&vcpu->kvm->slots_lock);
 816
 817         vcpu_put(vcpu);
 818
 819         return i;
 820 }
 821
 822 /*
 823  * Read or write a bunch of msrs. Parameters are user addresses.
 824  *
 825  * @return number of msrs set successfully.
 826  */
 827 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 828                   int (*do_msr)(struct kvm_vcpu *vcpu,
 829                                 unsigned index, u64 *data),
 830                   int writeback)
 831 {
 832         struct kvm_msrs msrs;
 833         struct kvm_msr_entry *entries;
 834         int r, n;
 835         unsigned size;
 836
 837         r = -EFAULT;
 838         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 839                 goto out;
 840
 841         r = -E2BIG;
 842         if (msrs.nmsrs >= MAX_IO_MSRS)
 843                 goto out;
 844
 845         r = -ENOMEM;
 846         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 847         entries = vmalloc(size);
 848         if (!entries)
 849                 goto out;
 850
 851         r = -EFAULT;
 852         if (copy_from_user(entries, user_msrs->entries, size))
 853                 goto out_free;
 854
 855         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 856         if (r < 0)
 857                 goto out_free;
 858
 859         r = -EFAULT;
 860         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 861                 goto out_free;
 862
 863         r = n;
 864
 865 out_free:
 866         vfree(entries);
 867 out:
 868         return r;
 869 }
 870
 871 int kvm_dev_ioctl_check_extension(long ext)
 872 {
 873         int r;
 874
 875         switch (ext) {
 876         case KVM_CAP_IRQCHIP:
 877         case KVM_CAP_HLT:
 878         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 879         case KVM_CAP_USER_MEMORY:
 880         case KVM_CAP_SET_TSS_ADDR:
 881         case KVM_CAP_EXT_CPUID:
 882         case KVM_CAP_CLOCKSOURCE:
 883         case KVM_CAP_PIT:
 884         case KVM_CAP_NOP_IO_DELAY:
 885         case KVM_CAP_MP_STATE:
 886                 r = 1;
 887                 break;
 888         case KVM_CAP_COALESCED_MMIO:
 889                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 890                 break;
 891         case KVM_CAP_VAPIC:
 892                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 893                 break;
 894         case KVM_CAP_NR_VCPUS:
 895                 r = KVM_MAX_VCPUS;
 896                 break;
 897         case KVM_CAP_NR_MEMSLOTS:
 898                 r = KVM_MEMORY_SLOTS;
 899                 break;
 900         case KVM_CAP_PV_MMU:
 901                 r = !tdp_enabled;
 902                 break;
 903         default:
 904                 r = 0;
 905                 break;
 906         }
 907         return r;
 908
 909 }
 910
 911 long kvm_arch_dev_ioctl(struct file *filp,
 912                         unsigned int ioctl, unsigned long arg)
 913 {
 914         void __user *argp = (void __user *)arg;
 915         long r;
 916
 917         switch (ioctl) {
 918         case KVM_GET_MSR_INDEX_LIST: {
 919                 struct kvm_msr_list __user *user_msr_list = argp;
 920                 struct kvm_msr_list msr_list;
 921                 unsigned n;
 922
 923                 r = -EFAULT;
 924                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 925                         goto out;
 926                 n = msr_list.nmsrs;
 927                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 928                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 929                         goto out;
 930                 r = -E2BIG;
 931                 if (n < num_msrs_to_save)
 932                         goto out;
 933                 r = -EFAULT;
 934                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 935                                  num_msrs_to_save * sizeof(u32)))
 936                         goto out;
 937                 if (copy_to_user(user_msr_list->indices
 938                                  + num_msrs_to_save * sizeof(u32),
 939                                  &emulated_msrs,
 940                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 941                         goto out;
 942                 r = 0;
 943                 break;
 944         }
 945         case KVM_GET_SUPPORTED_CPUID: {
 946                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 947                 struct kvm_cpuid2 cpuid;
 948
 949                 r = -EFAULT;
 950                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 951                         goto out;
 952                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 953                         cpuid_arg->entries);
 954                 if (r)
 955                         goto out;
 956
 957                 r = -EFAULT;
 958                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 959                         goto out;
 960                 r = 0;
 961                 break;
 962         }
 963         default:
 964                 r = -EINVAL;
 965         }
 966 out:
 967         return r;
 968 }
 969
 970 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 971 {
 972         kvm_x86_ops->vcpu_load(vcpu, cpu);
 973         kvm_write_guest_time(vcpu);
 974 }
 975
 976 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 977 {
 978         kvm_x86_ops->vcpu_put(vcpu);
 979         kvm_put_guest_fpu(vcpu);
 980 }
 981
 982 static int is_efer_nx(void)
 983 {
 984         u64 efer;
 985
 986         rdmsrl(MSR_EFER, efer);
 987         return efer & EFER_NX;
 988 }
 989
 990 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 991 {
 992         int i;
 993         struct kvm_cpuid_entry2 *e, *entry;
 994
 995         entry = NULL;
 996         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 997                 e = &vcpu->arch.cpuid_entries[i];
 998                 if (e->function == 0x80000001) {
 999                         entry = e;
1000                         break;
1001                 }
1002         }
1003         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1004                 entry->edx &= ~(1 << 20);
1005                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1006         }
1007 }
1008
1009 /* when an old userspace process fills a new kernel module */
1010 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1011                                     struct kvm_cpuid *cpuid,
1012                                     struct kvm_cpuid_entry __user *entries)
1013 {
1014         int r, i;
1015         struct kvm_cpuid_entry *cpuid_entries;
1016
1017         r = -E2BIG;
1018         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1019                 goto out;
1020         r = -ENOMEM;
1021         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1022         if (!cpuid_entries)
1023                 goto out;
1024         r = -EFAULT;
1025         if (copy_from_user(cpuid_entries, entries,
1026                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1027                 goto out_free;
1028         for (i = 0; i < cpuid->nent; i++) {
1029                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1030                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1031                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1032                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1033                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1034                 vcpu->arch.cpuid_entries[i].index = 0;
1035                 vcpu->arch.cpuid_entries[i].flags = 0;
1036                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1037                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1038                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1039         }
1040         vcpu->arch.cpuid_nent = cpuid->nent;
1041         cpuid_fix_nx_cap(vcpu);
1042         r = 0;
1043
1044 out_free:
1045         vfree(cpuid_entries);
1046 out:
1047         return r;
1048 }
1049
1050 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1051                                     struct kvm_cpuid2 *cpuid,
1052                                     struct kvm_cpuid_entry2 __user *entries)
1053 {
1054         int r;
1055
1056         r = -E2BIG;
1057         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1058                 goto out;
1059         r = -EFAULT;
1060         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1061                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1062                 goto out;
1063         vcpu->arch.cpuid_nent = cpuid->nent;
1064         return 0;
1065
1066 out:
1067         return r;
1068 }
1069
1070 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1071                                     struct kvm_cpuid2 *cpuid,
1072                                     struct kvm_cpuid_entry2 __user *entries)
1073 {
1074         int r;
1075
1076         r = -E2BIG;
1077         if (cpuid->nent < vcpu->arch.cpuid_nent)
1078                 goto out;
1079         r = -EFAULT;
1080         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1081                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1082                 goto out;
1083         return 0;
1084
1085 out:
1086         cpuid->nent = vcpu->arch.cpuid_nent;
1087         return r;
1088 }
1089
1090 static inline u32 bit(int bitno)
1091 {
1092         return 1 << (bitno & 31);
1093 }
1094
1095 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1096                           u32 index)
1097 {
1098         entry->function = function;
1099         entry->index = index;
1100         cpuid_count(entry->function, entry->index,
1101                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1102         entry->flags = 0;
1103 }
1104
1105 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1106                          u32 index, int *nent, int maxnent)
1107 {
1108         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1109                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1110                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1111                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1112                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1113                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1114                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1115                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1116                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1117                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1118         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1119                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1120                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1121                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1122                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1123                 bit(X86_FEATURE_PGE) |
1124                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1125                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1126                 bit(X86_FEATURE_SYSCALL) |
1127                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1128 #ifdef CONFIG_X86_64
1129                 bit(X86_FEATURE_LM) |
1130 #endif
1131                 bit(X86_FEATURE_MMXEXT) |
1132                 bit(X86_FEATURE_3DNOWEXT) |
1133                 bit(X86_FEATURE_3DNOW);
1134         const u32 kvm_supported_word3_x86_features =
1135                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1136         const u32 kvm_supported_word6_x86_features =
1137                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1138
1139         /* all func 2 cpuid_count() should be called on the same cpu */
1140         get_cpu();
1141         do_cpuid_1_ent(entry, function, index);
1142         ++*nent;
1143
1144         switch (function) {
1145         case 0:
1146                 entry->eax = min(entry->eax, (u32)0xb);
1147                 break;
1148         case 1:
1149                 entry->edx &= kvm_supported_word0_x86_features;
1150                 entry->ecx &= kvm_supported_word3_x86_features;
1151                 break;
1152         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1153          * may return different values. This forces us to get_cpu() before
1154          * issuing the first command, and also to emulate this annoying behavior
1155          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1156         case 2: {
1157                 int t, times = entry->eax & 0xff;
1158
1159                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1160                 for (t = 1; t < times && *nent < maxnent; ++t) {
1161                         do_cpuid_1_ent(&entry[t], function, 0);
1162                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1163                         ++*nent;
1164                 }
1165                 break;
1166         }
1167         /* function 4 and 0xb have additional index. */
1168         case 4: {
1169                 int i, cache_type;
1170
1171                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1172                 /* read more entries until cache_type is zero */
1173                 for (i = 1; *nent < maxnent; ++i) {
1174                         cache_type = entry[i - 1].eax & 0x1f;
1175                         if (!cache_type)
1176                                 break;
1177                         do_cpuid_1_ent(&entry[i], function, i);
1178                         entry[i].flags |=
1179                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1180                         ++*nent;
1181                 }
1182                 break;
1183         }
1184         case 0xb: {
1185                 int i, level_type;
1186
1187                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1188                 /* read more entries until level_type is zero */
1189                 for (i = 1; *nent < maxnent; ++i) {
1190                         level_type = entry[i - 1].ecx & 0xff;
1191                         if (!level_type)
1192                                 break;
1193                         do_cpuid_1_ent(&entry[i], function, i);
1194                         entry[i].flags |=
1195                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1196                         ++*nent;
1197                 }
1198                 break;
1199         }
1200         case 0x80000000:
1201                 entry->eax = min(entry->eax, 0x8000001a);
1202                 break;
1203         case 0x80000001:
1204                 entry->edx &= kvm_supported_word1_x86_features;
1205                 entry->ecx &= kvm_supported_word6_x86_features;
1206                 break;
1207         }
1208         put_cpu();
1209 }
1210
1211 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1212                                     struct kvm_cpuid_entry2 __user *entries)
1213 {
1214         struct kvm_cpuid_entry2 *cpuid_entries;
1215         int limit, nent = 0, r = -E2BIG;
1216         u32 func;
1217
1218         if (cpuid->nent < 1)
1219                 goto out;
1220         r = -ENOMEM;
1221         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1222         if (!cpuid_entries)
1223                 goto out;
1224
1225         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1226         limit = cpuid_entries[0].eax;
1227         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1228                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1229                                 &nent, cpuid->nent);
1230         r = -E2BIG;
1231         if (nent >= cpuid->nent)
1232                 goto out_free;
1233
1234         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1235         limit = cpuid_entries[nent - 1].eax;
1236         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1237                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1238                                &nent, cpuid->nent);
1239         r = -EFAULT;
1240         if (copy_to_user(entries, cpuid_entries,
1241                         nent * sizeof(struct kvm_cpuid_entry2)))
1242                 goto out_free;
1243         cpuid->nent = nent;
1244         r = 0;
1245
1246 out_free:
1247         vfree(cpuid_entries);
1248 out:
1249         return r;
1250 }
1251
1252 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1253                                     struct kvm_lapic_state *s)
1254 {
1255         vcpu_load(vcpu);
1256         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1257         vcpu_put(vcpu);
1258
1259         return 0;
1260 }
1261
1262 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1263                                     struct kvm_lapic_state *s)
1264 {
1265         vcpu_load(vcpu);
1266         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1267         kvm_apic_post_state_restore(vcpu);
1268         vcpu_put(vcpu);
1269
1270         return 0;
1271 }
1272
1273 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1274                                     struct kvm_interrupt *irq)
1275 {
1276         if (irq->irq < 0 || irq->irq >= 256)
1277                 return -EINVAL;
1278         if (irqchip_in_kernel(vcpu->kvm))
1279                 return -ENXIO;
1280         vcpu_load(vcpu);
1281
1282         set_bit(irq->irq, vcpu->arch.irq_pending);
1283         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1284
1285         vcpu_put(vcpu);
1286
1287         return 0;
1288 }
1289
1290 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1291                                            struct kvm_tpr_access_ctl *tac)
1292 {
1293         if (tac->flags)
1294                 return -EINVAL;
1295         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1296         return 0;
1297 }
1298
1299 long kvm_arch_vcpu_ioctl(struct file *filp,
1300                          unsigned int ioctl, unsigned long arg)
1301 {
1302         struct kvm_vcpu *vcpu = filp->private_data;
1303         void __user *argp = (void __user *)arg;
1304         int r;
1305
1306         switch (ioctl) {
1307         case KVM_GET_LAPIC: {
1308                 struct kvm_lapic_state lapic;
1309
1310                 memset(&lapic, 0, sizeof lapic);
1311                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1312                 if (r)
1313                         goto out;
1314                 r = -EFAULT;
1315                 if (copy_to_user(argp, &lapic, sizeof lapic))
1316                         goto out;
1317                 r = 0;
1318                 break;
1319         }
1320         case KVM_SET_LAPIC: {
1321                 struct kvm_lapic_state lapic;
1322
1323                 r = -EFAULT;
1324                 if (copy_from_user(&lapic, argp, sizeof lapic))
1325                         goto out;
1326                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1327                 if (r)
1328                         goto out;
1329                 r = 0;
1330                 break;
1331         }
1332         case KVM_INTERRUPT: {
1333                 struct kvm_interrupt irq;
1334
1335                 r = -EFAULT;
1336                 if (copy_from_user(&irq, argp, sizeof irq))
1337                         goto out;
1338                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1339                 if (r)
1340                         goto out;
1341                 r = 0;
1342                 break;
1343         }
1344         case KVM_SET_CPUID: {
1345                 struct kvm_cpuid __user *cpuid_arg = argp;
1346                 struct kvm_cpuid cpuid;
1347
1348                 r = -EFAULT;
1349                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1350                         goto out;
1351                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1352                 if (r)
1353                         goto out;
1354                 break;
1355         }
1356         case KVM_SET_CPUID2: {
1357                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1358                 struct kvm_cpuid2 cpuid;
1359
1360                 r = -EFAULT;
1361                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1362                         goto out;
1363                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1364                                 cpuid_arg->entries);
1365                 if (r)
1366                         goto out;
1367                 break;
1368         }
1369         case KVM_GET_CPUID2: {
1370                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1371                 struct kvm_cpuid2 cpuid;
1372
1373                 r = -EFAULT;
1374                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1375                         goto out;
1376                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1377                                 cpuid_arg->entries);
1378                 if (r)
1379                         goto out;
1380                 r = -EFAULT;
1381                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1382                         goto out;
1383                 r = 0;
1384                 break;
1385         }
1386         case KVM_GET_MSRS:
1387                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1388                 break;
1389         case KVM_SET_MSRS:
1390                 r = msr_io(vcpu, argp, do_set_msr, 0);
1391                 break;
1392         case KVM_TPR_ACCESS_REPORTING: {
1393                 struct kvm_tpr_access_ctl tac;
1394
1395                 r = -EFAULT;
1396                 if (copy_from_user(&tac, argp, sizeof tac))
1397                         goto out;
1398                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1399                 if (r)
1400                         goto out;
1401                 r = -EFAULT;
1402                 if (copy_to_user(argp, &tac, sizeof tac))
1403                         goto out;
1404                 r = 0;
1405                 break;
1406         };
1407         case KVM_SET_VAPIC_ADDR: {
1408                 struct kvm_vapic_addr va;
1409
1410                 r = -EINVAL;
1411                 if (!irqchip_in_kernel(vcpu->kvm))
1412                         goto out;
1413                 r = -EFAULT;
1414                 if (copy_from_user(&va, argp, sizeof va))
1415                         goto out;
1416                 r = 0;
1417                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1418                 break;
1419         }
1420         default:
1421                 r = -EINVAL;
1422         }
1423 out:
1424         return r;
1425 }
1426
1427 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1428 {
1429         int ret;
1430
1431         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1432                 return -1;
1433         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1434         return ret;
1435 }
1436
1437 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1438                                           u32 kvm_nr_mmu_pages)
1439 {
1440         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1441                 return -EINVAL;
1442
1443         down_write(&kvm->slots_lock);
1444
1445         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1446         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1447
1448         up_write(&kvm->slots_lock);
1449         return 0;
1450 }
1451
1452 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1453 {
1454         return kvm->arch.n_alloc_mmu_pages;
1455 }
1456
1457 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1458 {
1459         int i;
1460         struct kvm_mem_alias *alias;
1461
1462         for (i = 0; i < kvm->arch.naliases; ++i) {
1463                 alias = &kvm->arch.aliases[i];
1464                 if (gfn >= alias->base_gfn
1465                     && gfn < alias->base_gfn + alias->npages)
1466                         return alias->target_gfn + gfn - alias->base_gfn;
1467         }
1468         return gfn;
1469 }
1470
1471 /*
1472  * Set a new alias region.  Aliases map a portion of physical memory into
1473  * another portion.  This is useful for memory windows, for example the PC
1474  * VGA region.
1475  */
1476 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1477                                          struct kvm_memory_alias *alias)
1478 {
1479         int r, n;
1480         struct kvm_mem_alias *p;
1481
1482         r = -EINVAL;
1483         /* General sanity checks */
1484         if (alias->memory_size & (PAGE_SIZE - 1))
1485                 goto out;
1486         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1487                 goto out;
1488         if (alias->slot >= KVM_ALIAS_SLOTS)
1489                 goto out;
1490         if (alias->guest_phys_addr + alias->memory_size
1491             < alias->guest_phys_addr)
1492                 goto out;
1493         if (alias->target_phys_addr + alias->memory_size
1494             < alias->target_phys_addr)
1495                 goto out;
1496
1497         down_write(&kvm->slots_lock);
1498
1499         p = &kvm->arch.aliases[alias->slot];
1500         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1501         p->npages = alias->memory_size >> PAGE_SHIFT;
1502         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1503
1504         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1505                 if (kvm->arch.aliases[n - 1].npages)
1506                         break;
1507         kvm->arch.naliases = n;
1508
1509         kvm_mmu_zap_all(kvm);
1510
1511         up_write(&kvm->slots_lock);
1512
1513         return 0;
1514
1515 out:
1516         return r;
1517 }
1518
1519 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1520 {
1521         int r;
1522
1523         r = 0;
1524         switch (chip->chip_id) {
1525         case KVM_IRQCHIP_PIC_MASTER:
1526                 memcpy(&chip->chip.pic,
1527                         &pic_irqchip(kvm)->pics[0],
1528                         sizeof(struct kvm_pic_state));
1529                 break;
1530         case KVM_IRQCHIP_PIC_SLAVE:
1531                 memcpy(&chip->chip.pic,
1532                         &pic_irqchip(kvm)->pics[1],
1533                         sizeof(struct kvm_pic_state));
1534                 break;
1535         case KVM_IRQCHIP_IOAPIC:
1536                 memcpy(&chip->chip.ioapic,
1537                         ioapic_irqchip(kvm),
1538                         sizeof(struct kvm_ioapic_state));
1539                 break;
1540         default:
1541                 r = -EINVAL;
1542                 break;
1543         }
1544         return r;
1545 }
1546
1547 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1548 {
1549         int r;
1550
1551         r = 0;
1552         switch (chip->chip_id) {
1553         case KVM_IRQCHIP_PIC_MASTER:
1554                 memcpy(&pic_irqchip(kvm)->pics[0],
1555                         &chip->chip.pic,
1556                         sizeof(struct kvm_pic_state));
1557                 break;
1558         case KVM_IRQCHIP_PIC_SLAVE:
1559                 memcpy(&pic_irqchip(kvm)->pics[1],
1560                         &chip->chip.pic,
1561                         sizeof(struct kvm_pic_state));
1562                 break;
1563         case KVM_IRQCHIP_IOAPIC:
1564                 memcpy(ioapic_irqchip(kvm),
1565                         &chip->chip.ioapic,
1566                         sizeof(struct kvm_ioapic_state));
1567                 break;
1568         default:
1569                 r = -EINVAL;
1570                 break;
1571         }
1572         kvm_pic_update_irq(pic_irqchip(kvm));
1573         return r;
1574 }
1575
1576 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1577 {
1578         int r = 0;
1579
1580         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1581         return r;
1582 }
1583
1584 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1585 {
1586         int r = 0;
1587
1588         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1589         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1590         return r;
1591 }
1592
1593 /*
1594  * Get (and clear) the dirty memory log for a memory slot.
1595  */
1596 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1597                                       struct kvm_dirty_log *log)
1598 {
1599         int r;
1600         int n;
1601         struct kvm_memory_slot *memslot;
1602         int is_dirty = 0;
1603
1604         down_write(&kvm->slots_lock);
1605
1606         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1607         if (r)
1608                 goto out;
1609
1610         /* If nothing is dirty, don't bother messing with page tables. */
1611         if (is_dirty) {
1612                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1613                 kvm_flush_remote_tlbs(kvm);
1614                 memslot = &kvm->memslots[log->slot];
1615                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1616                 memset(memslot->dirty_bitmap, 0, n);
1617         }
1618         r = 0;
1619 out:
1620         up_write(&kvm->slots_lock);
1621         return r;
1622 }
1623
1624 long kvm_arch_vm_ioctl(struct file *filp,
1625                        unsigned int ioctl, unsigned long arg)
1626 {
1627         struct kvm *kvm = filp->private_data;
1628         void __user *argp = (void __user *)arg;
1629         int r = -EINVAL;
1630
1631         switch (ioctl) {
1632         case KVM_SET_TSS_ADDR:
1633                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1634                 if (r < 0)
1635                         goto out;
1636                 break;
1637         case KVM_SET_MEMORY_REGION: {
1638                 struct kvm_memory_region kvm_mem;
1639                 struct kvm_userspace_memory_region kvm_userspace_mem;
1640
1641                 r = -EFAULT;
1642                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1643                         goto out;
1644                 kvm_userspace_mem.slot = kvm_mem.slot;
1645                 kvm_userspace_mem.flags = kvm_mem.flags;
1646                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1647                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1648                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1649                 if (r)
1650                         goto out;
1651                 break;
1652         }
1653         case KVM_SET_NR_MMU_PAGES:
1654                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1655                 if (r)
1656                         goto out;
1657                 break;
1658         case KVM_GET_NR_MMU_PAGES:
1659                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1660                 break;
1661         case KVM_SET_MEMORY_ALIAS: {
1662                 struct kvm_memory_alias alias;
1663
1664                 r = -EFAULT;
1665                 if (copy_from_user(&alias, argp, sizeof alias))
1666                         goto out;
1667                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1668                 if (r)
1669                         goto out;
1670                 break;
1671         }
1672         case KVM_CREATE_IRQCHIP:
1673                 r = -ENOMEM;
1674                 kvm->arch.vpic = kvm_create_pic(kvm);
1675                 if (kvm->arch.vpic) {
1676                         r = kvm_ioapic_init(kvm);
1677                         if (r) {
1678                                 kfree(kvm->arch.vpic);
1679                                 kvm->arch.vpic = NULL;
1680                                 goto out;
1681                         }
1682                 } else
1683                         goto out;
1684                 break;
1685         case KVM_CREATE_PIT:
1686                 r = -ENOMEM;
1687                 kvm->arch.vpit = kvm_create_pit(kvm);
1688                 if (kvm->arch.vpit)
1689                         r = 0;
1690                 break;
1691         case KVM_IRQ_LINE: {
1692                 struct kvm_irq_level irq_event;
1693
1694                 r = -EFAULT;
1695                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1696                         goto out;
1697                 if (irqchip_in_kernel(kvm)) {
1698                         mutex_lock(&kvm->lock);
1699                         if (irq_event.irq < 16)
1700                                 kvm_pic_set_irq(pic_irqchip(kvm),
1701                                         irq_event.irq,
1702                                         irq_event.level);
1703                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1704                                         irq_event.irq,
1705                                         irq_event.level);
1706                         mutex_unlock(&kvm->lock);
1707                         r = 0;
1708                 }
1709                 break;
1710         }
1711         case KVM_GET_IRQCHIP: {
1712                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1713                 struct kvm_irqchip chip;
1714
1715                 r = -EFAULT;
1716                 if (copy_from_user(&chip, argp, sizeof chip))
1717                         goto out;
1718                 r = -ENXIO;
1719                 if (!irqchip_in_kernel(kvm))
1720                         goto out;
1721                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1722                 if (r)
1723                         goto out;
1724                 r = -EFAULT;
1725                 if (copy_to_user(argp, &chip, sizeof chip))
1726                         goto out;
1727                 r = 0;
1728                 break;
1729         }
1730         case KVM_SET_IRQCHIP: {
1731                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1732                 struct kvm_irqchip chip;
1733
1734                 r = -EFAULT;
1735                 if (copy_from_user(&chip, argp, sizeof chip))
1736                         goto out;
1737                 r = -ENXIO;
1738                 if (!irqchip_in_kernel(kvm))
1739                         goto out;
1740                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1741                 if (r)
1742                         goto out;
1743                 r = 0;
1744                 break;
1745         }
1746         case KVM_GET_PIT: {
1747                 struct kvm_pit_state ps;
1748                 r = -EFAULT;
1749                 if (copy_from_user(&ps, argp, sizeof ps))
1750                         goto out;
1751                 r = -ENXIO;
1752                 if (!kvm->arch.vpit)
1753                         goto out;
1754                 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1755                 if (r)
1756                         goto out;
1757                 r = -EFAULT;
1758                 if (copy_to_user(argp, &ps, sizeof ps))
1759                         goto out;
1760                 r = 0;
1761                 break;
1762         }
1763         case KVM_SET_PIT: {
1764                 struct kvm_pit_state ps;
1765                 r = -EFAULT;
1766                 if (copy_from_user(&ps, argp, sizeof ps))
1767                         goto out;
1768                 r = -ENXIO;
1769                 if (!kvm->arch.vpit)
1770                         goto out;
1771                 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1772                 if (r)
1773                         goto out;
1774                 r = 0;
1775                 break;
1776         }
1777         default:
1778                 ;
1779         }
1780 out:
1781         return r;
1782 }
1783
1784 static void kvm_init_msr_list(void)
1785 {
1786         u32 dummy[2];
1787         unsigned i, j;
1788
1789         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1790                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1791                         continue;
1792                 if (j < i)
1793                         msrs_to_save[j] = msrs_to_save[i];
1794                 j++;
1795         }
1796         num_msrs_to_save = j;
1797 }
1798
1799 /*
1800  * Only apic need an MMIO device hook, so shortcut now..
1801  */
1802 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1803                                                 gpa_t addr, int len,
1804                                                 int is_write)
1805 {
1806         struct kvm_io_device *dev;
1807
1808         if (vcpu->arch.apic) {
1809                 dev = &vcpu->arch.apic->dev;
1810                 if (dev->in_range(dev, addr, len, is_write))
1811                         return dev;
1812         }
1813         return NULL;
1814 }
1815
1816
1817 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1818                                                 gpa_t addr, int len,
1819                                                 int is_write)
1820 {
1821         struct kvm_io_device *dev;
1822
1823         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1824         if (dev == NULL)
1825                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1826                                           is_write);
1827         return dev;
1828 }
1829
1830 int emulator_read_std(unsigned long addr,
1831                              void *val,
1832                              unsigned int bytes,
1833                              struct kvm_vcpu *vcpu)
1834 {
1835         void *data = val;
1836         int r = X86EMUL_CONTINUE;
1837
1838         while (bytes) {
1839                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1840                 unsigned offset = addr & (PAGE_SIZE-1);
1841                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1842                 int ret;
1843
1844                 if (gpa == UNMAPPED_GVA) {
1845                         r = X86EMUL_PROPAGATE_FAULT;
1846                         goto out;
1847                 }
1848                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1849                 if (ret < 0) {
1850                         r = X86EMUL_UNHANDLEABLE;
1851                         goto out;
1852                 }
1853
1854                 bytes -= tocopy;
1855                 data += tocopy;
1856                 addr += tocopy;
1857         }
1858 out:
1859         return r;
1860 }
1861 EXPORT_SYMBOL_GPL(emulator_read_std);
1862
1863 static int emulator_read_emulated(unsigned long addr,
1864                                   void *val,
1865                                   unsigned int bytes,
1866                                   struct kvm_vcpu *vcpu)
1867 {
1868         struct kvm_io_device *mmio_dev;
1869         gpa_t                 gpa;
1870
1871         if (vcpu->mmio_read_completed) {
1872                 memcpy(val, vcpu->mmio_data, bytes);
1873                 vcpu->mmio_read_completed = 0;
1874                 return X86EMUL_CONTINUE;
1875         }
1876
1877         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1878
1879         /* For APIC access vmexit */
1880         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1881                 goto mmio;
1882
1883         if (emulator_read_std(addr, val, bytes, vcpu)
1884                         == X86EMUL_CONTINUE)
1885                 return X86EMUL_CONTINUE;
1886         if (gpa == UNMAPPED_GVA)
1887                 return X86EMUL_PROPAGATE_FAULT;
1888
1889 mmio:
1890         /*
1891          * Is this MMIO handled locally?
1892          */
1893         mutex_lock(&vcpu->kvm->lock);
1894         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1895         if (mmio_dev) {
1896                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1897                 mutex_unlock(&vcpu->kvm->lock);
1898                 return X86EMUL_CONTINUE;
1899         }
1900         mutex_unlock(&vcpu->kvm->lock);
1901
1902         vcpu->mmio_needed = 1;
1903         vcpu->mmio_phys_addr = gpa;
1904         vcpu->mmio_size = bytes;
1905         vcpu->mmio_is_write = 0;
1906
1907         return X86EMUL_UNHANDLEABLE;
1908 }
1909
1910 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1911                           const void *val, int bytes)
1912 {
1913         int ret;
1914
1915         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1916         if (ret < 0)
1917                 return 0;
1918         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1919         return 1;
1920 }
1921
1922 static int emulator_write_emulated_onepage(unsigned long addr,
1923                                            const void *val,
1924                                            unsigned int bytes,
1925                                            struct kvm_vcpu *vcpu)
1926 {
1927         struct kvm_io_device *mmio_dev;
1928         gpa_t                 gpa;
1929
1930         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1931
1932         if (gpa == UNMAPPED_GVA) {
1933                 kvm_inject_page_fault(vcpu, addr, 2);
1934                 return X86EMUL_PROPAGATE_FAULT;
1935         }
1936
1937         /* For APIC access vmexit */
1938         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1939                 goto mmio;
1940
1941         if (emulator_write_phys(vcpu, gpa, val, bytes))
1942                 return X86EMUL_CONTINUE;
1943
1944 mmio:
1945         /*
1946          * Is this MMIO handled locally?
1947          */
1948         mutex_lock(&vcpu->kvm->lock);
1949         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
1950         if (mmio_dev) {
1951                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1952                 mutex_unlock(&vcpu->kvm->lock);
1953                 return X86EMUL_CONTINUE;
1954         }
1955         mutex_unlock(&vcpu->kvm->lock);
1956
1957         vcpu->mmio_needed = 1;
1958         vcpu->mmio_phys_addr = gpa;
1959         vcpu->mmio_size = bytes;
1960         vcpu->mmio_is_write = 1;
1961         memcpy(vcpu->mmio_data, val, bytes);
1962
1963         return X86EMUL_CONTINUE;
1964 }
1965
1966 int emulator_write_emulated(unsigned long addr,
1967                                    const void *val,
1968                                    unsigned int bytes,
1969                                    struct kvm_vcpu *vcpu)
1970 {
1971         /* Crossing a page boundary? */
1972         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1973                 int rc, now;
1974
1975                 now = -addr & ~PAGE_MASK;
1976                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1977                 if (rc != X86EMUL_CONTINUE)
1978                         return rc;
1979                 addr += now;
1980                 val += now;
1981                 bytes -= now;
1982         }
1983         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1984 }
1985 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1986
1987 static int emulator_cmpxchg_emulated(unsigned long addr,
1988                                      const void *old,
1989                                      const void *new,
1990                                      unsigned int bytes,
1991                                      struct kvm_vcpu *vcpu)
1992 {
1993         static int reported;
1994
1995         if (!reported) {
1996                 reported = 1;
1997                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1998         }
1999 #ifndef CONFIG_X86_64
2000         /* guests cmpxchg8b have to be emulated atomically */
2001         if (bytes == 8) {
2002                 gpa_t gpa;
2003                 struct page *page;
2004                 char *kaddr;
2005                 u64 val;
2006
2007                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2008
2009                 if (gpa == UNMAPPED_GVA ||
2010                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2011                         goto emul_write;
2012
2013                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2014                         goto emul_write;
2015
2016                 val = *(u64 *)new;
2017
2018                 down_read(&current->mm->mmap_sem);
2019                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2020                 up_read(&current->mm->mmap_sem);
2021
2022                 kaddr = kmap_atomic(page, KM_USER0);
2023                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2024                 kunmap_atomic(kaddr, KM_USER0);
2025                 kvm_release_page_dirty(page);
2026         }
2027 emul_write:
2028 #endif
2029
2030         return emulator_write_emulated(addr, new, bytes, vcpu);
2031 }
2032
2033 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2034 {
2035         return kvm_x86_ops->get_segment_base(vcpu, seg);
2036 }
2037
2038 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2039 {
2040         return X86EMUL_CONTINUE;
2041 }
2042
2043 int emulate_clts(struct kvm_vcpu *vcpu)
2044 {
2045         KVMTRACE_0D(CLTS, vcpu, handler);
2046         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2047         return X86EMUL_CONTINUE;
2048 }
2049
2050 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2051 {
2052         struct kvm_vcpu *vcpu = ctxt->vcpu;
2053
2054         switch (dr) {
2055         case 0 ... 3:
2056                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2057                 return X86EMUL_CONTINUE;
2058         default:
2059                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2060                 return X86EMUL_UNHANDLEABLE;
2061         }
2062 }
2063
2064 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2065 {
2066         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2067         int exception;
2068
2069         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2070         if (exception) {
2071                 /* FIXME: better handling */
2072                 return X86EMUL_UNHANDLEABLE;
2073         }
2074         return X86EMUL_CONTINUE;
2075 }
2076
2077 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2078 {
2079         u8 opcodes[4];
2080         unsigned long rip = vcpu->arch.rip;
2081         unsigned long rip_linear;
2082
2083         if (!printk_ratelimit())
2084                 return;
2085
2086         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2087
2088         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2089
2090         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2091                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2092 }
2093 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2094
2095 static struct x86_emulate_ops emulate_ops = {
2096         .read_std            = emulator_read_std,
2097         .read_emulated       = emulator_read_emulated,
2098         .write_emulated      = emulator_write_emulated,
2099         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2100 };
2101
2102 int emulate_instruction(struct kvm_vcpu *vcpu,
2103                         struct kvm_run *run,
2104                         unsigned long cr2,
2105                         u16 error_code,
2106                         int emulation_type)
2107 {
2108         int r;
2109         struct decode_cache *c;
2110
2111         vcpu->arch.mmio_fault_cr2 = cr2;
2112         kvm_x86_ops->cache_regs(vcpu);
2113
2114         vcpu->mmio_is_write = 0;
2115         vcpu->arch.pio.string = 0;
2116
2117         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2118                 int cs_db, cs_l;
2119                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2120
2121                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2122                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2123                 vcpu->arch.emulate_ctxt.mode =
2124                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2125                         ? X86EMUL_MODE_REAL : cs_l
2126                         ? X86EMUL_MODE_PROT64 : cs_db
2127                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2128
2129                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2130                         vcpu->arch.emulate_ctxt.cs_base = 0;
2131                         vcpu->arch.emulate_ctxt.ds_base = 0;
2132                         vcpu->arch.emulate_ctxt.es_base = 0;
2133                         vcpu->arch.emulate_ctxt.ss_base = 0;
2134                 } else {
2135                         vcpu->arch.emulate_ctxt.cs_base =
2136                                         get_segment_base(vcpu, VCPU_SREG_CS);
2137                         vcpu->arch.emulate_ctxt.ds_base =
2138                                         get_segment_base(vcpu, VCPU_SREG_DS);
2139                         vcpu->arch.emulate_ctxt.es_base =
2140                                         get_segment_base(vcpu, VCPU_SREG_ES);
2141                         vcpu->arch.emulate_ctxt.ss_base =
2142                                         get_segment_base(vcpu, VCPU_SREG_SS);
2143                 }
2144
2145                 vcpu->arch.emulate_ctxt.gs_base =
2146                                         get_segment_base(vcpu, VCPU_SREG_GS);
2147                 vcpu->arch.emulate_ctxt.fs_base =
2148                                         get_segment_base(vcpu, VCPU_SREG_FS);
2149
2150                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2151
2152                 /* Reject the instructions other than VMCALL/VMMCALL when
2153                  * try to emulate invalid opcode */
2154                 c = &vcpu->arch.emulate_ctxt.decode;
2155                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2156                     (!(c->twobyte && c->b == 0x01 &&
2157                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2158                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2159                         return EMULATE_FAIL;
2160
2161                 ++vcpu->stat.insn_emulation;
2162                 if (r)  {
2163                         ++vcpu->stat.insn_emulation_fail;
2164                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2165                                 return EMULATE_DONE;
2166                         return EMULATE_FAIL;
2167                 }
2168         }
2169
2170         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2171
2172         if (vcpu->arch.pio.string)
2173                 return EMULATE_DO_MMIO;
2174
2175         if ((r || vcpu->mmio_is_write) && run) {
2176                 run->exit_reason = KVM_EXIT_MMIO;
2177                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2178                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2179                 run->mmio.len = vcpu->mmio_size;
2180                 run->mmio.is_write = vcpu->mmio_is_write;
2181         }
2182
2183         if (r) {
2184                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2185                         return EMULATE_DONE;
2186                 if (!vcpu->mmio_needed) {
2187                         kvm_report_emulation_failure(vcpu, "mmio");
2188                         return EMULATE_FAIL;
2189                 }
2190                 return EMULATE_DO_MMIO;
2191         }
2192
2193         kvm_x86_ops->decache_regs(vcpu);
2194         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2195
2196         if (vcpu->mmio_is_write) {
2197                 vcpu->mmio_needed = 0;
2198                 return EMULATE_DO_MMIO;
2199         }
2200
2201         return EMULATE_DONE;
2202 }
2203 EXPORT_SYMBOL_GPL(emulate_instruction);
2204
2205 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2206 {
2207         int i;
2208
2209         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2210                 if (vcpu->arch.pio.guest_pages[i]) {
2211                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2212                         vcpu->arch.pio.guest_pages[i] = NULL;
2213                 }
2214 }
2215
2216 static int pio_copy_data(struct kvm_vcpu *vcpu)
2217 {
2218         void *p = vcpu->arch.pio_data;
2219         void *q;
2220         unsigned bytes;
2221         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2222
2223         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2224                  PAGE_KERNEL);
2225         if (!q) {
2226                 free_pio_guest_pages(vcpu);
2227                 return -ENOMEM;
2228         }
2229         q += vcpu->arch.pio.guest_page_offset;
2230         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2231         if (vcpu->arch.pio.in)
2232                 memcpy(q, p, bytes);
2233         else
2234                 memcpy(p, q, bytes);
2235         q -= vcpu->arch.pio.guest_page_offset;
2236         vunmap(q);
2237         free_pio_guest_pages(vcpu);
2238         return 0;
2239 }
2240
2241 int complete_pio(struct kvm_vcpu *vcpu)
2242 {
2243         struct kvm_pio_request *io = &vcpu->arch.pio;
2244         long delta;
2245         int r;
2246
2247         kvm_x86_ops->cache_regs(vcpu);
2248
2249         if (!io->string) {
2250                 if (io->in)
2251                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2252                                io->size);
2253         } else {
2254                 if (io->in) {
2255                         r = pio_copy_data(vcpu);
2256                         if (r) {
2257                                 kvm_x86_ops->cache_regs(vcpu);
2258                                 return r;
2259                         }
2260                 }
2261
2262                 delta = 1;
2263                 if (io->rep) {
2264                         delta *= io->cur_count;
2265                         /*
2266                          * The size of the register should really depend on
2267                          * current address size.
2268                          */
2269                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2270                 }
2271                 if (io->down)
2272                         delta = -delta;
2273                 delta *= io->size;
2274                 if (io->in)
2275                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2276                 else
2277                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2278         }
2279
2280         kvm_x86_ops->decache_regs(vcpu);
2281
2282         io->count -= io->cur_count;
2283         io->cur_count = 0;
2284
2285         return 0;
2286 }
2287
2288 static void kernel_pio(struct kvm_io_device *pio_dev,
2289                        struct kvm_vcpu *vcpu,
2290                        void *pd)
2291 {
2292         /* TODO: String I/O for in kernel device */
2293
2294         mutex_lock(&vcpu->kvm->lock);
2295         if (vcpu->arch.pio.in)
2296                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2297                                   vcpu->arch.pio.size,
2298                                   pd);
2299         else
2300                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2301                                    vcpu->arch.pio.size,
2302                                    pd);
2303         mutex_unlock(&vcpu->kvm->lock);
2304 }
2305
2306 static void pio_string_write(struct kvm_io_device *pio_dev,
2307                              struct kvm_vcpu *vcpu)
2308 {
2309         struct kvm_pio_request *io = &vcpu->arch.pio;
2310         void *pd = vcpu->arch.pio_data;
2311         int i;
2312
2313         mutex_lock(&vcpu->kvm->lock);
2314         for (i = 0; i < io->cur_count; i++) {
2315                 kvm_iodevice_write(pio_dev, io->port,
2316                                    io->size,
2317                                    pd);
2318                 pd += io->size;
2319         }
2320         mutex_unlock(&vcpu->kvm->lock);
2321 }
2322
2323 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2324                                                gpa_t addr, int len,
2325                                                int is_write)
2326 {
2327         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2328 }
2329
2330 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2331                   int size, unsigned port)
2332 {
2333         struct kvm_io_device *pio_dev;
2334
2335         vcpu->run->exit_reason = KVM_EXIT_IO;
2336         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2337         vcpu->run->io.size = vcpu->arch.pio.size = size;
2338         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2339         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2340         vcpu->run->io.port = vcpu->arch.pio.port = port;
2341         vcpu->arch.pio.in = in;
2342         vcpu->arch.pio.string = 0;
2343         vcpu->arch.pio.down = 0;
2344         vcpu->arch.pio.guest_page_offset = 0;
2345         vcpu->arch.pio.rep = 0;
2346
2347         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2348                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2349                             handler);
2350         else
2351                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2352                             handler);
2353
2354         kvm_x86_ops->cache_regs(vcpu);
2355         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2356
2357         kvm_x86_ops->skip_emulated_instruction(vcpu);
2358
2359         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2360         if (pio_dev) {
2361                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2362                 complete_pio(vcpu);
2363                 return 1;
2364         }
2365         return 0;
2366 }
2367 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2368
2369 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2370                   int size, unsigned long count, int down,
2371                   gva_t address, int rep, unsigned port)
2372 {
2373         unsigned now, in_page;
2374         int i, ret = 0;
2375         int nr_pages = 1;
2376         struct page *page;
2377         struct kvm_io_device *pio_dev;
2378
2379         vcpu->run->exit_reason = KVM_EXIT_IO;
2380         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2381         vcpu->run->io.size = vcpu->arch.pio.size = size;
2382         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2383         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2384         vcpu->run->io.port = vcpu->arch.pio.port = port;
2385         vcpu->arch.pio.in = in;
2386         vcpu->arch.pio.string = 1;
2387         vcpu->arch.pio.down = down;
2388         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2389         vcpu->arch.pio.rep = rep;
2390
2391         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2392                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2393                             handler);
2394         else
2395                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2396                             handler);
2397
2398         if (!count) {
2399                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2400                 return 1;
2401         }
2402
2403         if (!down)
2404                 in_page = PAGE_SIZE - offset_in_page(address);
2405         else
2406                 in_page = offset_in_page(address) + size;
2407         now = min(count, (unsigned long)in_page / size);
2408         if (!now) {
2409                 /*
2410                  * String I/O straddles page boundary.  Pin two guest pages
2411                  * so that we satisfy atomicity constraints.  Do just one
2412                  * transaction to avoid complexity.
2413                  */
2414                 nr_pages = 2;
2415                 now = 1;
2416         }
2417         if (down) {
2418                 /*
2419                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2420                  */
2421                 pr_unimpl(vcpu, "guest string pio down\n");
2422                 kvm_inject_gp(vcpu, 0);
2423                 return 1;
2424         }
2425         vcpu->run->io.count = now;
2426         vcpu->arch.pio.cur_count = now;
2427
2428         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2429                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2430
2431         for (i = 0; i < nr_pages; ++i) {
2432                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2433                 vcpu->arch.pio.guest_pages[i] = page;
2434                 if (!page) {
2435                         kvm_inject_gp(vcpu, 0);
2436                         free_pio_guest_pages(vcpu);
2437                         return 1;
2438                 }
2439         }
2440
2441         pio_dev = vcpu_find_pio_dev(vcpu, port,
2442                                     vcpu->arch.pio.cur_count,
2443                                     !vcpu->arch.pio.in);
2444         if (!vcpu->arch.pio.in) {
2445                 /* string PIO write */
2446                 ret = pio_copy_data(vcpu);
2447                 if (ret >= 0 && pio_dev) {
2448                         pio_string_write(pio_dev, vcpu);
2449                         complete_pio(vcpu);
2450                         if (vcpu->arch.pio.count == 0)
2451                                 ret = 1;
2452                 }
2453         } else if (pio_dev)
2454                 pr_unimpl(vcpu, "no string pio read support yet, "
2455                        "port %x size %d count %ld\n",
2456                         port, size, count);
2457
2458         return ret;
2459 }
2460 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2461
2462 int kvm_arch_init(void *opaque)
2463 {
2464         int r;
2465         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2466
2467         if (kvm_x86_ops) {
2468                 printk(KERN_ERR "kvm: already loaded the other module\n");
2469                 r = -EEXIST;
2470                 goto out;
2471         }
2472
2473         if (!ops->cpu_has_kvm_support()) {
2474                 printk(KERN_ERR "kvm: no hardware support\n");
2475                 r = -EOPNOTSUPP;
2476                 goto out;
2477         }
2478         if (ops->disabled_by_bios()) {
2479                 printk(KERN_ERR "kvm: disabled by bios\n");
2480                 r = -EOPNOTSUPP;
2481                 goto out;
2482         }
2483
2484         r = kvm_mmu_module_init();
2485         if (r)
2486                 goto out;
2487
2488         kvm_init_msr_list();
2489
2490         kvm_x86_ops = ops;
2491         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2492         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2493         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2494                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2495         return 0;
2496
2497 out:
2498         return r;
2499 }
2500
2501 void kvm_arch_exit(void)
2502 {
2503         kvm_x86_ops = NULL;
2504         kvm_mmu_module_exit();
2505 }
2506
2507 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2508 {
2509         ++vcpu->stat.halt_exits;
2510         KVMTRACE_0D(HLT, vcpu, handler);
2511         if (irqchip_in_kernel(vcpu->kvm)) {
2512                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2513                 up_read(&vcpu->kvm->slots_lock);
2514                 kvm_vcpu_block(vcpu);
2515                 down_read(&vcpu->kvm->slots_lock);
2516                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2517                         return -EINTR;
2518                 return 1;
2519         } else {
2520                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2521                 return 0;
2522         }
2523 }
2524 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2525
2526 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2527                            unsigned long a1)
2528 {
2529         if (is_long_mode(vcpu))
2530                 return a0;
2531         else
2532                 return a0 | ((gpa_t)a1 << 32);
2533 }
2534
2535 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2536 {
2537         unsigned long nr, a0, a1, a2, a3, ret;
2538         int r = 1;
2539
2540         kvm_x86_ops->cache_regs(vcpu);
2541
2542         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2543         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2544         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2545         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2546         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2547
2548         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2549
2550         if (!is_long_mode(vcpu)) {
2551                 nr &= 0xFFFFFFFF;
2552                 a0 &= 0xFFFFFFFF;
2553                 a1 &= 0xFFFFFFFF;
2554                 a2 &= 0xFFFFFFFF;
2555                 a3 &= 0xFFFFFFFF;
2556         }
2557
2558         switch (nr) {
2559         case KVM_HC_VAPIC_POLL_IRQ:
2560                 ret = 0;
2561                 break;
2562         case KVM_HC_MMU_OP:
2563                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2564                 break;
2565         default:
2566                 ret = -KVM_ENOSYS;
2567                 break;
2568         }
2569         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2570         kvm_x86_ops->decache_regs(vcpu);
2571         ++vcpu->stat.hypercalls;
2572         return r;
2573 }
2574 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2575
2576 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2577 {
2578         char instruction[3];
2579         int ret = 0;
2580
2581
2582         /*
2583          * Blow out the MMU to ensure that no other VCPU has an active mapping
2584          * to ensure that the updated hypercall appears atomically across all
2585          * VCPUs.
2586          */
2587         kvm_mmu_zap_all(vcpu->kvm);
2588
2589         kvm_x86_ops->cache_regs(vcpu);
2590         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2591         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2592             != X86EMUL_CONTINUE)
2593                 ret = -EFAULT;
2594
2595         return ret;
2596 }
2597
2598 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2599 {
2600         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2601 }
2602
2603 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2604 {
2605         struct descriptor_table dt = { limit, base };
2606
2607         kvm_x86_ops->set_gdt(vcpu, &dt);
2608 }
2609
2610 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2611 {
2612         struct descriptor_table dt = { limit, base };
2613
2614         kvm_x86_ops->set_idt(vcpu, &dt);
2615 }
2616
2617 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2618                    unsigned long *rflags)
2619 {
2620         kvm_lmsw(vcpu, msw);
2621         *rflags = kvm_x86_ops->get_rflags(vcpu);
2622 }
2623
2624 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2625 {
2626         unsigned long value;
2627
2628         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2629         switch (cr) {
2630         case 0:
2631                 value = vcpu->arch.cr0;
2632                 break;
2633         case 2:
2634                 value = vcpu->arch.cr2;
2635                 break;
2636         case 3:
2637                 value = vcpu->arch.cr3;
2638                 break;
2639         case 4:
2640                 value = vcpu->arch.cr4;
2641                 break;
2642         case 8:
2643                 value = kvm_get_cr8(vcpu);
2644                 break;
2645         default:
2646                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2647                 return 0;
2648         }
2649         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2650                     (u32)((u64)value >> 32), handler);
2651
2652         return value;
2653 }
2654
2655 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2656                      unsigned long *rflags)
2657 {
2658         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2659                     (u32)((u64)val >> 32), handler);
2660
2661         switch (cr) {
2662         case 0:
2663                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2664                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2665                 break;
2666         case 2:
2667                 vcpu->arch.cr2 = val;
2668                 break;
2669         case 3:
2670                 kvm_set_cr3(vcpu, val);
2671                 break;
2672         case 4:
2673                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2674                 break;
2675         case 8:
2676                 kvm_set_cr8(vcpu, val & 0xfUL);
2677                 break;
2678         default:
2679                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2680         }
2681 }
2682
2683 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2684 {
2685         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2686         int j, nent = vcpu->arch.cpuid_nent;
2687
2688         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2689         /* when no next entry is found, the current entry[i] is reselected */
2690         for (j = i + 1; j == i; j = (j + 1) % nent) {
2691                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2692                 if (ej->function == e->function) {
2693                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2694                         return j;
2695                 }
2696         }
2697         return 0; /* silence gcc, even though control never reaches here */
2698 }
2699
2700 /* find an entry with matching function, matching index (if needed), and that
2701  * should be read next (if it's stateful) */
2702 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2703         u32 function, u32 index)
2704 {
2705         if (e->function != function)
2706                 return 0;
2707         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2708                 return 0;
2709         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2710                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2711                 return 0;
2712         return 1;
2713 }
2714
2715 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2716 {
2717         int i;
2718         u32 function, index;
2719         struct kvm_cpuid_entry2 *e, *best;
2720
2721         kvm_x86_ops->cache_regs(vcpu);
2722         function = vcpu->arch.regs[VCPU_REGS_RAX];
2723         index = vcpu->arch.regs[VCPU_REGS_RCX];
2724         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2725         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2726         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2727         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2728         best = NULL;
2729         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2730                 e = &vcpu->arch.cpuid_entries[i];
2731                 if (is_matching_cpuid_entry(e, function, index)) {
2732                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2733                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2734                         best = e;
2735                         break;
2736                 }
2737                 /*
2738                  * Both basic or both extended?
2739                  */
2740                 if (((e->function ^ function) & 0x80000000) == 0)
2741                         if (!best || e->function > best->function)
2742                                 best = e;
2743         }
2744         if (best) {
2745                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2746                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2747                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2748                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2749         }
2750         kvm_x86_ops->decache_regs(vcpu);
2751         kvm_x86_ops->skip_emulated_instruction(vcpu);
2752         KVMTRACE_5D(CPUID, vcpu, function,
2753                     (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2754                     (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2755                     (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2756                     (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2757 }
2758 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2759
2760 /*
2761  * Check if userspace requested an interrupt window, and that the
2762  * interrupt window is open.
2763  *
2764  * No need to exit to userspace if we already have an interrupt queued.
2765  */
2766 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2767                                           struct kvm_run *kvm_run)
2768 {
2769         return (!vcpu->arch.irq_summary &&
2770                 kvm_run->request_interrupt_window &&
2771                 vcpu->arch.interrupt_window_open &&
2772                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2773 }
2774
2775 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2776                               struct kvm_run *kvm_run)
2777 {
2778         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2779         kvm_run->cr8 = kvm_get_cr8(vcpu);
2780         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2781         if (irqchip_in_kernel(vcpu->kvm))
2782                 kvm_run->ready_for_interrupt_injection = 1;
2783         else
2784                 kvm_run->ready_for_interrupt_injection =
2785                                         (vcpu->arch.interrupt_window_open &&
2786                                          vcpu->arch.irq_summary == 0);
2787 }
2788
2789 static void vapic_enter(struct kvm_vcpu *vcpu)
2790 {
2791         struct kvm_lapic *apic = vcpu->arch.apic;
2792         struct page *page;
2793
2794         if (!apic || !apic->vapic_addr)
2795                 return;
2796
2797         down_read(&current->mm->mmap_sem);
2798         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2799         up_read(&current->mm->mmap_sem);
2800
2801         vcpu->arch.apic->vapic_page = page;
2802 }
2803
2804 static void vapic_exit(struct kvm_vcpu *vcpu)
2805 {
2806         struct kvm_lapic *apic = vcpu->arch.apic;
2807
2808         if (!apic || !apic->vapic_addr)
2809                 return;
2810
2811         kvm_release_page_dirty(apic->vapic_page);
2812         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2813 }
2814
2815 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2816 {
2817         int r;
2818
2819         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2820                 pr_debug("vcpu %d received sipi with vector # %x\n",
2821                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2822                 kvm_lapic_reset(vcpu);
2823                 r = kvm_x86_ops->vcpu_reset(vcpu);
2824                 if (r)
2825                         return r;
2826                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2827         }
2828
2829         down_read(&vcpu->kvm->slots_lock);
2830         vapic_enter(vcpu);
2831
2832 preempted:
2833         if (vcpu->guest_debug.enabled)
2834                 kvm_x86_ops->guest_debug_pre(vcpu);
2835
2836 again:
2837         if (vcpu->requests)
2838                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2839                         kvm_mmu_unload(vcpu);
2840
2841         r = kvm_mmu_reload(vcpu);
2842         if (unlikely(r))
2843                 goto out;
2844
2845         if (vcpu->requests) {
2846                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2847                         __kvm_migrate_timers(vcpu);
2848                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2849                         kvm_x86_ops->tlb_flush(vcpu);
2850                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2851                                        &vcpu->requests)) {
2852                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2853                         r = 0;
2854                         goto out;
2855                 }
2856                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2857                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2858                         r = 0;
2859                         goto out;
2860                 }
2861         }
2862
2863         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2864         kvm_inject_pending_timer_irqs(vcpu);
2865
2866         preempt_disable();
2867
2868         kvm_x86_ops->prepare_guest_switch(vcpu);
2869         kvm_load_guest_fpu(vcpu);
2870
2871         local_irq_disable();
2872
2873         if (vcpu->requests || need_resched()) {
2874                 local_irq_enable();
2875                 preempt_enable();
2876                 r = 1;
2877                 goto out;
2878         }
2879
2880         if (signal_pending(current)) {
2881                 local_irq_enable();
2882                 preempt_enable();
2883                 r = -EINTR;
2884                 kvm_run->exit_reason = KVM_EXIT_INTR;
2885                 ++vcpu->stat.signal_exits;
2886                 goto out;
2887         }
2888
2889         vcpu->guest_mode = 1;
2890         /*
2891          * Make sure that guest_mode assignment won't happen after
2892          * testing the pending IRQ vector bitmap.
2893          */
2894         smp_wmb();
2895
2896         if (vcpu->arch.exception.pending)
2897                 __queue_exception(vcpu);
2898         else if (irqchip_in_kernel(vcpu->kvm))
2899                 kvm_x86_ops->inject_pending_irq(vcpu);
2900         else
2901                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2902
2903         kvm_lapic_sync_to_vapic(vcpu);
2904
2905         up_read(&vcpu->kvm->slots_lock);
2906
2907         kvm_guest_enter();
2908
2909
2910         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2911         kvm_x86_ops->run(vcpu, kvm_run);
2912
2913         vcpu->guest_mode = 0;
2914         local_irq_enable();
2915
2916         ++vcpu->stat.exits;
2917
2918         /*
2919          * We must have an instruction between local_irq_enable() and
2920          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2921          * the interrupt shadow.  The stat.exits increment will do nicely.
2922          * But we need to prevent reordering, hence this barrier():
2923          */
2924         barrier();
2925
2926         kvm_guest_exit();
2927
2928         preempt_enable();
2929
2930         down_read(&vcpu->kvm->slots_lock);
2931
2932         /*
2933          * Profile KVM exit RIPs:
2934          */
2935         if (unlikely(prof_on == KVM_PROFILING)) {
2936                 kvm_x86_ops->cache_regs(vcpu);
2937                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2938         }
2939
2940         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2941                 vcpu->arch.exception.pending = false;
2942
2943         kvm_lapic_sync_from_vapic(vcpu);
2944
2945         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2946
2947         if (r > 0) {
2948                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2949                         r = -EINTR;
2950                         kvm_run->exit_reason = KVM_EXIT_INTR;
2951                         ++vcpu->stat.request_irq_exits;
2952                         goto out;
2953                 }
2954                 if (!need_resched())
2955                         goto again;
2956         }
2957
2958 out:
2959         up_read(&vcpu->kvm->slots_lock);
2960         if (r > 0) {
2961                 kvm_resched(vcpu);
2962                 down_read(&vcpu->kvm->slots_lock);
2963                 goto preempted;
2964         }
2965
2966         post_kvm_run_save(vcpu, kvm_run);
2967
2968         down_read(&vcpu->kvm->slots_lock);
2969         vapic_exit(vcpu);
2970         up_read(&vcpu->kvm->slots_lock);
2971
2972         return r;
2973 }
2974
2975 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2976 {
2977         int r;
2978         sigset_t sigsaved;
2979
2980         vcpu_load(vcpu);
2981
2982         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2983                 kvm_vcpu_block(vcpu);
2984                 vcpu_put(vcpu);
2985                 return -EAGAIN;
2986         }
2987
2988         if (vcpu->sigset_active)
2989                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2990
2991         /* re-sync apic's tpr */
2992         if (!irqchip_in_kernel(vcpu->kvm))
2993                 kvm_set_cr8(vcpu, kvm_run->cr8);
2994
2995         if (vcpu->arch.pio.cur_count) {
2996                 r = complete_pio(vcpu);
2997                 if (r)
2998                         goto out;
2999         }
3000 #if CONFIG_HAS_IOMEM
3001         if (vcpu->mmio_needed) {
3002                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3003                 vcpu->mmio_read_completed = 1;
3004                 vcpu->mmio_needed = 0;
3005
3006                 down_read(&vcpu->kvm->slots_lock);
3007                 r = emulate_instruction(vcpu, kvm_run,
3008                                         vcpu->arch.mmio_fault_cr2, 0,
3009                                         EMULTYPE_NO_DECODE);
3010                 up_read(&vcpu->kvm->slots_lock);
3011                 if (r == EMULATE_DO_MMIO) {
3012                         /*
3013                          * Read-modify-write.  Back to userspace.
3014                          */
3015                         r = 0;
3016                         goto out;
3017                 }
3018         }
3019 #endif
3020         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
3021                 kvm_x86_ops->cache_regs(vcpu);
3022                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
3023                 kvm_x86_ops->decache_regs(vcpu);
3024         }
3025
3026         r = __vcpu_run(vcpu, kvm_run);
3027
3028 out:
3029         if (vcpu->sigset_active)
3030                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3031
3032         vcpu_put(vcpu);
3033         return r;
3034 }
3035
3036 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3037 {
3038         vcpu_load(vcpu);
3039
3040         kvm_x86_ops->cache_regs(vcpu);
3041
3042         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
3043         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
3044         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
3045         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
3046         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
3047         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
3048         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3049         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3050 #ifdef CONFIG_X86_64
3051         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3052         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3053         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3054         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3055         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3056         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3057         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3058         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3059 #endif
3060
3061         regs->rip = vcpu->arch.rip;
3062         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3063
3064         /*
3065          * Don't leak debug flags in case they were set for guest debugging
3066          */
3067         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3068                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3069
3070         vcpu_put(vcpu);
3071
3072         return 0;
3073 }
3074
3075 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3076 {
3077         vcpu_load(vcpu);
3078
3079         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3080         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3081         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3082         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3083         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3084         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3085         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3086         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3087 #ifdef CONFIG_X86_64
3088         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3089         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3090         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3091         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3092         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3093         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3094         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3095         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3096 #endif
3097
3098         vcpu->arch.rip = regs->rip;
3099         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3100
3101         kvm_x86_ops->decache_regs(vcpu);
3102
3103         vcpu->arch.exception.pending = false;
3104
3105         vcpu_put(vcpu);
3106
3107         return 0;
3108 }
3109
3110 void kvm_get_segment(struct kvm_vcpu *vcpu,
3111                      struct kvm_segment *var, int seg)
3112 {
3113         kvm_x86_ops->get_segment(vcpu, var, seg);
3114 }
3115
3116 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3117 {
3118         struct kvm_segment cs;
3119
3120         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3121         *db = cs.db;
3122         *l = cs.l;
3123 }
3124 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3125
3126 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3127                                   struct kvm_sregs *sregs)
3128 {
3129         struct descriptor_table dt;
3130         int pending_vec;
3131
3132         vcpu_load(vcpu);
3133
3134         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3135         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3136         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3137         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3138         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3139         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3140
3141         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3142         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3143
3144         kvm_x86_ops->get_idt(vcpu, &dt);
3145         sregs->idt.limit = dt.limit;
3146         sregs->idt.base = dt.base;
3147         kvm_x86_ops->get_gdt(vcpu, &dt);
3148         sregs->gdt.limit = dt.limit;
3149         sregs->gdt.base = dt.base;
3150
3151         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3152         sregs->cr0 = vcpu->arch.cr0;
3153         sregs->cr2 = vcpu->arch.cr2;
3154         sregs->cr3 = vcpu->arch.cr3;
3155         sregs->cr4 = vcpu->arch.cr4;
3156         sregs->cr8 = kvm_get_cr8(vcpu);
3157         sregs->efer = vcpu->arch.shadow_efer;
3158         sregs->apic_base = kvm_get_apic_base(vcpu);
3159
3160         if (irqchip_in_kernel(vcpu->kvm)) {
3161                 memset(sregs->interrupt_bitmap, 0,
3162                        sizeof sregs->interrupt_bitmap);
3163                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3164                 if (pending_vec >= 0)
3165                         set_bit(pending_vec,
3166                                 (unsigned long *)sregs->interrupt_bitmap);
3167         } else
3168                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3169                        sizeof sregs->interrupt_bitmap);
3170
3171         vcpu_put(vcpu);
3172
3173         return 0;
3174 }
3175
3176 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3177                                     struct kvm_mp_state *mp_state)
3178 {
3179         vcpu_load(vcpu);
3180         mp_state->mp_state = vcpu->arch.mp_state;
3181         vcpu_put(vcpu);
3182         return 0;
3183 }
3184
3185 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3186                                     struct kvm_mp_state *mp_state)
3187 {
3188         vcpu_load(vcpu);
3189         vcpu->arch.mp_state = mp_state->mp_state;
3190         vcpu_put(vcpu);
3191         return 0;
3192 }
3193
3194 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3195                         struct kvm_segment *var, int seg)
3196 {
3197         kvm_x86_ops->set_segment(vcpu, var, seg);
3198 }
3199
3200 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3201                                    struct kvm_segment *kvm_desct)
3202 {
3203         kvm_desct->base = seg_desc->base0;
3204         kvm_desct->base |= seg_desc->base1 << 16;
3205         kvm_desct->base |= seg_desc->base2 << 24;
3206         kvm_desct->limit = seg_desc->limit0;
3207         kvm_desct->limit |= seg_desc->limit << 16;
3208         kvm_desct->selector = selector;
3209         kvm_desct->type = seg_desc->type;
3210         kvm_desct->present = seg_desc->p;
3211         kvm_desct->dpl = seg_desc->dpl;
3212         kvm_desct->db = seg_desc->d;
3213         kvm_desct->s = seg_desc->s;
3214         kvm_desct->l = seg_desc->l;
3215         kvm_desct->g = seg_desc->g;
3216         kvm_desct->avl = seg_desc->avl;
3217         if (!selector)
3218                 kvm_desct->unusable = 1;
3219         else
3220                 kvm_desct->unusable = 0;
3221         kvm_desct->padding = 0;
3222 }
3223
3224 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3225                                            u16 selector,
3226                                            struct descriptor_table *dtable)
3227 {
3228         if (selector & 1 << 2) {
3229                 struct kvm_segment kvm_seg;
3230
3231                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3232
3233                 if (kvm_seg.unusable)
3234                         dtable->limit = 0;
3235                 else
3236                         dtable->limit = kvm_seg.limit;
3237                 dtable->base = kvm_seg.base;
3238         }
3239         else
3240                 kvm_x86_ops->get_gdt(vcpu, dtable);
3241 }
3242
3243 /* allowed just for 8 bytes segments */
3244 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3245                                          struct desc_struct *seg_desc)
3246 {
3247         struct descriptor_table dtable;
3248         u16 index = selector >> 3;
3249
3250         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3251
3252         if (dtable.limit < index * 8 + 7) {
3253                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3254                 return 1;
3255         }
3256         return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3257 }
3258
3259 /* allowed just for 8 bytes segments */
3260 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3261                                          struct desc_struct *seg_desc)
3262 {
3263         struct descriptor_table dtable;
3264         u16 index = selector >> 3;
3265
3266         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3267
3268         if (dtable.limit < index * 8 + 7)
3269                 return 1;
3270         return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3271 }
3272
3273 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3274                              struct desc_struct *seg_desc)
3275 {
3276         u32 base_addr;
3277
3278         base_addr = seg_desc->base0;
3279         base_addr |= (seg_desc->base1 << 16);
3280         base_addr |= (seg_desc->base2 << 24);
3281
3282         return base_addr;
3283 }
3284
3285 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3286                               struct desc_struct *seg_desc,
3287                               struct tss_segment_32 *tss)
3288 {
3289         u32 base_addr;
3290
3291         base_addr = get_tss_base_addr(vcpu, seg_desc);
3292
3293         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3294                               sizeof(struct tss_segment_32));
3295 }
3296
3297 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3298                               struct desc_struct *seg_desc,
3299                               struct tss_segment_32 *tss)
3300 {
3301         u32 base_addr;
3302
3303         base_addr = get_tss_base_addr(vcpu, seg_desc);
3304
3305         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3306                                sizeof(struct tss_segment_32));
3307 }
3308
3309 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3310                               struct desc_struct *seg_desc,
3311                               struct tss_segment_16 *tss)
3312 {
3313         u32 base_addr;
3314
3315         base_addr = get_tss_base_addr(vcpu, seg_desc);
3316
3317         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3318                               sizeof(struct tss_segment_16));
3319 }
3320
3321 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3322                               struct desc_struct *seg_desc,
3323                               struct tss_segment_16 *tss)
3324 {
3325         u32 base_addr;
3326
3327         base_addr = get_tss_base_addr(vcpu, seg_desc);
3328
3329         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3330                                sizeof(struct tss_segment_16));
3331 }
3332
3333 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3334 {
3335         struct kvm_segment kvm_seg;
3336
3337         kvm_get_segment(vcpu, &kvm_seg, seg);
3338         return kvm_seg.selector;
3339 }
3340
3341 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3342                                                 u16 selector,
3343                                                 struct kvm_segment *kvm_seg)
3344 {
3345         struct desc_struct seg_desc;
3346
3347         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3348                 return 1;
3349         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3350         return 0;
3351 }
3352
3353 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3354                                 int type_bits, int seg)
3355 {
3356         struct kvm_segment kvm_seg;
3357
3358         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3359                 return 1;
3360         kvm_seg.type |= type_bits;
3361
3362         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3363             seg != VCPU_SREG_LDTR)
3364                 if (!kvm_seg.s)
3365                         kvm_seg.unusable = 1;
3366
3367         kvm_set_segment(vcpu, &kvm_seg, seg);
3368         return 0;
3369 }
3370
3371 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3372                                 struct tss_segment_32 *tss)
3373 {
3374         tss->cr3 = vcpu->arch.cr3;
3375         tss->eip = vcpu->arch.rip;
3376         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3377         tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3378         tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3379         tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3380         tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3381         tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3382         tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3383         tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3384         tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3385
3386         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3387         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3388         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3389         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3390         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3391         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3392         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3393         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3394 }
3395
3396 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3397                                   struct tss_segment_32 *tss)
3398 {
3399         kvm_set_cr3(vcpu, tss->cr3);
3400
3401         vcpu->arch.rip = tss->eip;
3402         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3403
3404         vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3405         vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3406         vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3407         vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3408         vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3409         vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3410         vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3411         vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3412
3413         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3414                 return 1;
3415
3416         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3417                 return 1;
3418
3419         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3420                 return 1;
3421
3422         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3423                 return 1;
3424
3425         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3426                 return 1;
3427
3428         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3429                 return 1;
3430
3431         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3432                 return 1;
3433         return 0;
3434 }
3435
3436 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3437                                 struct tss_segment_16 *tss)
3438 {
3439         tss->ip = vcpu->arch.rip;
3440         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3441         tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3442         tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3443         tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3444         tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3445         tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3446         tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3447         tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3448         tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3449
3450         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3451         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3452         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3453         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3454         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3455         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3456 }
3457
3458 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3459                                  struct tss_segment_16 *tss)
3460 {
3461         vcpu->arch.rip = tss->ip;
3462         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3463         vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3464         vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3465         vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3466         vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3467         vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3468         vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3469         vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3470         vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3471
3472         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3473                 return 1;
3474
3475         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3476                 return 1;
3477
3478         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3479                 return 1;
3480
3481         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3482                 return 1;
3483
3484         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3485                 return 1;
3486         return 0;
3487 }
3488
3489 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3490                        struct desc_struct *cseg_desc,
3491                        struct desc_struct *nseg_desc)
3492 {
3493         struct tss_segment_16 tss_segment_16;
3494         int ret = 0;
3495
3496         if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3497                 goto out;
3498
3499         save_state_to_tss16(vcpu, &tss_segment_16);
3500         save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3501
3502         if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3503                 goto out;
3504         if (load_state_from_tss16(vcpu, &tss_segment_16))
3505                 goto out;
3506
3507         ret = 1;
3508 out:
3509         return ret;
3510 }
3511
3512 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3513                        struct desc_struct *cseg_desc,
3514                        struct desc_struct *nseg_desc)
3515 {
3516         struct tss_segment_32 tss_segment_32;
3517         int ret = 0;
3518
3519         if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3520                 goto out;
3521
3522         save_state_to_tss32(vcpu, &tss_segment_32);
3523         save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3524
3525         if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3526                 goto out;
3527         if (load_state_from_tss32(vcpu, &tss_segment_32))
3528                 goto out;
3529
3530         ret = 1;
3531 out:
3532         return ret;
3533 }
3534
3535 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3536 {
3537         struct kvm_segment tr_seg;
3538         struct desc_struct cseg_desc;
3539         struct desc_struct nseg_desc;
3540         int ret = 0;
3541
3542         kvm_get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3543
3544         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3545                 goto out;
3546
3547         if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3548                 goto out;
3549
3550
3551         if (reason != TASK_SWITCH_IRET) {
3552                 int cpl;
3553
3554                 cpl = kvm_x86_ops->get_cpl(vcpu);
3555                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3556                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3557                         return 1;
3558                 }
3559         }
3560
3561         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3562                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3563                 return 1;
3564         }
3565
3566         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3567                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3568                 save_guest_segment_descriptor(vcpu, tr_seg.selector,
3569                                               &cseg_desc);
3570         }
3571
3572         if (reason == TASK_SWITCH_IRET) {
3573                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3574                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3575         }
3576
3577         kvm_x86_ops->skip_emulated_instruction(vcpu);
3578         kvm_x86_ops->cache_regs(vcpu);
3579
3580         if (nseg_desc.type & 8)
3581                 ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3582                                          &nseg_desc);
3583         else
3584                 ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3585                                          &nseg_desc);
3586
3587         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3588                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3589                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3590         }
3591
3592         if (reason != TASK_SWITCH_IRET) {
3593                 nseg_desc.type |= (1 << 1);
3594                 save_guest_segment_descriptor(vcpu, tss_selector,
3595                                               &nseg_desc);
3596         }
3597
3598         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3599         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3600         tr_seg.type = 11;
3601         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3602 out:
3603         kvm_x86_ops->decache_regs(vcpu);
3604         return ret;
3605 }
3606 EXPORT_SYMBOL_GPL(kvm_task_switch);
3607
3608 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3609                                   struct kvm_sregs *sregs)
3610 {
3611         int mmu_reset_needed = 0;
3612         int i, pending_vec, max_bits;
3613         struct descriptor_table dt;
3614
3615         vcpu_load(vcpu);
3616
3617         dt.limit = sregs->idt.limit;
3618         dt.base = sregs->idt.base;
3619         kvm_x86_ops->set_idt(vcpu, &dt);
3620         dt.limit = sregs->gdt.limit;
3621         dt.base = sregs->gdt.base;
3622         kvm_x86_ops->set_gdt(vcpu, &dt);
3623
3624         vcpu->arch.cr2 = sregs->cr2;
3625         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3626         vcpu->arch.cr3 = sregs->cr3;
3627
3628         kvm_set_cr8(vcpu, sregs->cr8);
3629
3630         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3631         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3632         kvm_set_apic_base(vcpu, sregs->apic_base);
3633
3634         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3635
3636         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3637         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3638         vcpu->arch.cr0 = sregs->cr0;
3639
3640         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3641         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3642         if (!is_long_mode(vcpu) && is_pae(vcpu))
3643                 load_pdptrs(vcpu, vcpu->arch.cr3);
3644
3645         if (mmu_reset_needed)
3646                 kvm_mmu_reset_context(vcpu);
3647
3648         if (!irqchip_in_kernel(vcpu->kvm)) {
3649                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3650                        sizeof vcpu->arch.irq_pending);
3651                 vcpu->arch.irq_summary = 0;
3652                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3653                         if (vcpu->arch.irq_pending[i])
3654                                 __set_bit(i, &vcpu->arch.irq_summary);
3655         } else {
3656                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3657                 pending_vec = find_first_bit(
3658                         (const unsigned long *)sregs->interrupt_bitmap,
3659                         max_bits);
3660                 /* Only pending external irq is handled here */
3661                 if (pending_vec < max_bits) {
3662                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3663                         pr_debug("Set back pending irq %d\n",
3664                                  pending_vec);
3665                 }
3666         }
3667
3668         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3669         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3670         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3671         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3672         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3673         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3674
3675         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3676         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3677
3678         vcpu_put(vcpu);
3679
3680         return 0;
3681 }
3682
3683 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3684                                     struct kvm_debug_guest *dbg)
3685 {
3686         int r;
3687
3688         vcpu_load(vcpu);
3689
3690         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3691
3692         vcpu_put(vcpu);
3693
3694         return r;
3695 }
3696
3697 /*
3698  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3699  * we have asm/x86/processor.h
3700  */
3701 struct fxsave {
3702         u16     cwd;
3703         u16     swd;
3704         u16     twd;
3705         u16     fop;
3706         u64     rip;
3707         u64     rdp;
3708         u32     mxcsr;
3709         u32     mxcsr_mask;
3710         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3711 #ifdef CONFIG_X86_64
3712         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3713 #else
3714         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3715 #endif
3716 };
3717
3718 /*
3719  * Translate a guest virtual address to a guest physical address.
3720  */
3721 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3722                                     struct kvm_translation *tr)
3723 {
3724         unsigned long vaddr = tr->linear_address;
3725         gpa_t gpa;
3726
3727         vcpu_load(vcpu);
3728         down_read(&vcpu->kvm->slots_lock);
3729         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3730         up_read(&vcpu->kvm->slots_lock);
3731         tr->physical_address = gpa;
3732         tr->valid = gpa != UNMAPPED_GVA;
3733         tr->writeable = 1;
3734         tr->usermode = 0;
3735         vcpu_put(vcpu);
3736
3737         return 0;
3738 }
3739
3740 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3741 {
3742         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3743
3744         vcpu_load(vcpu);
3745
3746         memcpy(fpu->fpr, fxsave->st_space, 128);
3747         fpu->fcw = fxsave->cwd;
3748         fpu->fsw = fxsave->swd;
3749         fpu->ftwx = fxsave->twd;
3750         fpu->last_opcode = fxsave->fop;
3751         fpu->last_ip = fxsave->rip;
3752         fpu->last_dp = fxsave->rdp;
3753         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3754
3755         vcpu_put(vcpu);
3756
3757         return 0;
3758 }
3759
3760 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3761 {
3762         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3763
3764         vcpu_load(vcpu);
3765
3766         memcpy(fxsave->st_space, fpu->fpr, 128);
3767         fxsave->cwd = fpu->fcw;
3768         fxsave->swd = fpu->fsw;
3769         fxsave->twd = fpu->ftwx;
3770         fxsave->fop = fpu->last_opcode;
3771         fxsave->rip = fpu->last_ip;
3772         fxsave->rdp = fpu->last_dp;
3773         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3774
3775         vcpu_put(vcpu);
3776
3777         return 0;
3778 }
3779
3780 void fx_init(struct kvm_vcpu *vcpu)
3781 {
3782         unsigned after_mxcsr_mask;
3783
3784         /*
3785          * Touch the fpu the first time in non atomic context as if
3786          * this is the first fpu instruction the exception handler
3787          * will fire before the instruction returns and it'll have to
3788          * allocate ram with GFP_KERNEL.
3789          */
3790         if (!used_math())
3791                 fx_save(&vcpu->arch.host_fx_image);
3792
3793         /* Initialize guest FPU by resetting ours and saving into guest's */
3794         preempt_disable();
3795         fx_save(&vcpu->arch.host_fx_image);
3796         fx_finit();
3797         fx_save(&vcpu->arch.guest_fx_image);
3798         fx_restore(&vcpu->arch.host_fx_image);
3799         preempt_enable();
3800
3801         vcpu->arch.cr0 |= X86_CR0_ET;
3802         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3803         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3804         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3805                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3806 }
3807 EXPORT_SYMBOL_GPL(fx_init);
3808
3809 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3810 {
3811         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3812                 return;
3813
3814         vcpu->guest_fpu_loaded = 1;
3815         fx_save(&vcpu->arch.host_fx_image);
3816         fx_restore(&vcpu->arch.guest_fx_image);
3817 }
3818 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3819
3820 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3821 {
3822         if (!vcpu->guest_fpu_loaded)
3823                 return;
3824
3825         vcpu->guest_fpu_loaded = 0;
3826         fx_save(&vcpu->arch.guest_fx_image);
3827         fx_restore(&vcpu->arch.host_fx_image);
3828         ++vcpu->stat.fpu_reload;
3829 }
3830 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3831
3832 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3833 {
3834         kvm_x86_ops->vcpu_free(vcpu);
3835 }
3836
3837 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3838                                                 unsigned int id)
3839 {
3840         return kvm_x86_ops->vcpu_create(kvm, id);
3841 }
3842
3843 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3844 {
3845         int r;
3846
3847         /* We do fxsave: this must be aligned. */
3848         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3849
3850         vcpu_load(vcpu);
3851         r = kvm_arch_vcpu_reset(vcpu);
3852         if (r == 0)
3853                 r = kvm_mmu_setup(vcpu);
3854         vcpu_put(vcpu);
3855         if (r < 0)
3856                 goto free_vcpu;
3857
3858         return 0;
3859 free_vcpu:
3860         kvm_x86_ops->vcpu_free(vcpu);
3861         return r;
3862 }
3863
3864 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3865 {
3866         vcpu_load(vcpu);
3867         kvm_mmu_unload(vcpu);
3868         vcpu_put(vcpu);
3869
3870         kvm_x86_ops->vcpu_free(vcpu);
3871 }
3872
3873 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3874 {
3875         return kvm_x86_ops->vcpu_reset(vcpu);
3876 }
3877
3878 void kvm_arch_hardware_enable(void *garbage)
3879 {
3880         kvm_x86_ops->hardware_enable(garbage);
3881 }
3882
3883 void kvm_arch_hardware_disable(void *garbage)
3884 {
3885         kvm_x86_ops->hardware_disable(garbage);
3886 }
3887
3888 int kvm_arch_hardware_setup(void)
3889 {
3890         return kvm_x86_ops->hardware_setup();
3891 }
3892
3893 void kvm_arch_hardware_unsetup(void)
3894 {
3895         kvm_x86_ops->hardware_unsetup();
3896 }
3897
3898 void kvm_arch_check_processor_compat(void *rtn)
3899 {
3900         kvm_x86_ops->check_processor_compatibility(rtn);
3901 }
3902
3903 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3904 {
3905         struct page *page;
3906         struct kvm *kvm;
3907         int r;
3908
3909         BUG_ON(vcpu->kvm == NULL);
3910         kvm = vcpu->kvm;
3911
3912         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3913         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3914                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3915         else
3916                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3917
3918         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3919         if (!page) {
3920                 r = -ENOMEM;
3921                 goto fail;
3922         }
3923         vcpu->arch.pio_data = page_address(page);
3924
3925         r = kvm_mmu_create(vcpu);
3926         if (r < 0)
3927                 goto fail_free_pio_data;
3928
3929         if (irqchip_in_kernel(kvm)) {
3930                 r = kvm_create_lapic(vcpu);
3931                 if (r < 0)
3932                         goto fail_mmu_destroy;
3933         }
3934
3935         return 0;
3936
3937 fail_mmu_destroy:
3938         kvm_mmu_destroy(vcpu);
3939 fail_free_pio_data:
3940         free_page((unsigned long)vcpu->arch.pio_data);
3941 fail:
3942         return r;
3943 }
3944
3945 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3946 {
3947         kvm_free_lapic(vcpu);
3948         down_read(&vcpu->kvm->slots_lock);
3949         kvm_mmu_destroy(vcpu);
3950         up_read(&vcpu->kvm->slots_lock);
3951         free_page((unsigned long)vcpu->arch.pio_data);
3952 }
3953
3954 struct  kvm *kvm_arch_create_vm(void)
3955 {
3956         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3957
3958         if (!kvm)
3959                 return ERR_PTR(-ENOMEM);
3960
3961         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3962
3963         return kvm;
3964 }
3965
3966 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3967 {
3968         vcpu_load(vcpu);
3969         kvm_mmu_unload(vcpu);
3970         vcpu_put(vcpu);
3971 }
3972
3973 static void kvm_free_vcpus(struct kvm *kvm)
3974 {
3975         unsigned int i;
3976
3977         /*
3978          * Unpin any mmu pages first.
3979          */
3980         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3981                 if (kvm->vcpus[i])
3982                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3983         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3984                 if (kvm->vcpus[i]) {
3985                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3986                         kvm->vcpus[i] = NULL;
3987                 }
3988         }
3989
3990 }
3991
3992 void kvm_arch_destroy_vm(struct kvm *kvm)
3993 {
3994         kvm_free_pit(kvm);
3995         kfree(kvm->arch.vpic);
3996         kfree(kvm->arch.vioapic);
3997         kvm_free_vcpus(kvm);
3998         kvm_free_physmem(kvm);
3999         if (kvm->arch.apic_access_page)
4000                 put_page(kvm->arch.apic_access_page);
4001         if (kvm->arch.ept_identity_pagetable)
4002                 put_page(kvm->arch.ept_identity_pagetable);
4003         kfree(kvm);
4004 }
4005
4006 int kvm_arch_set_memory_region(struct kvm *kvm,
4007                                 struct kvm_userspace_memory_region *mem,
4008                                 struct kvm_memory_slot old,
4009                                 int user_alloc)
4010 {
4011         int npages = mem->memory_size >> PAGE_SHIFT;
4012         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4013
4014         /*To keep backward compatibility with older userspace,
4015          *x86 needs to hanlde !user_alloc case.
4016          */
4017         if (!user_alloc) {
4018                 if (npages && !old.rmap) {
4019                         down_write(&current->mm->mmap_sem);
4020                         memslot->userspace_addr = do_mmap(NULL, 0,
4021                                                      npages * PAGE_SIZE,
4022                                                      PROT_READ | PROT_WRITE,
4023                                                      MAP_SHARED | MAP_ANONYMOUS,
4024                                                      0);
4025                         up_write(&current->mm->mmap_sem);
4026
4027                         if (IS_ERR((void *)memslot->userspace_addr))
4028                                 return PTR_ERR((void *)memslot->userspace_addr);
4029                 } else {
4030                         if (!old.user_alloc && old.rmap) {
4031                                 int ret;
4032
4033                                 down_write(&current->mm->mmap_sem);
4034                                 ret = do_munmap(current->mm, old.userspace_addr,
4035                                                 old.npages * PAGE_SIZE);
4036                                 up_write(&current->mm->mmap_sem);
4037                                 if (ret < 0)
4038                                         printk(KERN_WARNING
4039                                        "kvm_vm_ioctl_set_memory_region: "
4040                                        "failed to munmap memory\n");
4041                         }
4042                 }
4043         }
4044
4045         if (!kvm->arch.n_requested_mmu_pages) {
4046                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4047                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4048         }
4049
4050         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4051         kvm_flush_remote_tlbs(kvm);
4052
4053         return 0;
4054 }
4055
4056 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4057 {
4058         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4059                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4060 }
4061
4062 static void vcpu_kick_intr(void *info)
4063 {
4064 #ifdef DEBUG
4065         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4066         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4067 #endif
4068 }
4069
4070 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4071 {
4072         int ipi_pcpu = vcpu->cpu;
4073         int cpu = get_cpu();
4074
4075         if (waitqueue_active(&vcpu->wq)) {
4076                 wake_up_interruptible(&vcpu->wq);
4077                 ++vcpu->stat.halt_wakeup;
4078         }
4079         /*
4080          * We may be called synchronously with irqs disabled in guest mode,
4081          * So need not to call smp_call_function_single() in that case.
4082          */
4083         if (vcpu->guest_mode && vcpu->cpu != cpu)
4084                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4085         put_cpu();
4086 }