git.oblomov.eu Git - linux-2.6/blob - arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   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_VAPIC:
 889                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 890                 break;
 891         case KVM_CAP_NR_VCPUS:
 892                 r = KVM_MAX_VCPUS;
 893                 break;
 894         case KVM_CAP_NR_MEMSLOTS:
 895                 r = KVM_MEMORY_SLOTS;
 896                 break;
 897         case KVM_CAP_PV_MMU:
 898                 r = !tdp_enabled;
 899                 break;
 900         default:
 901                 r = 0;
 902                 break;
 903         }
 904         return r;
 905
 906 }
 907
 908 long kvm_arch_dev_ioctl(struct file *filp,
 909                         unsigned int ioctl, unsigned long arg)
 910 {
 911         void __user *argp = (void __user *)arg;
 912         long r;
 913
 914         switch (ioctl) {
 915         case KVM_GET_MSR_INDEX_LIST: {
 916                 struct kvm_msr_list __user *user_msr_list = argp;
 917                 struct kvm_msr_list msr_list;
 918                 unsigned n;
 919
 920                 r = -EFAULT;
 921                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 922                         goto out;
 923                 n = msr_list.nmsrs;
 924                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 925                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 926                         goto out;
 927                 r = -E2BIG;
 928                 if (n < num_msrs_to_save)
 929                         goto out;
 930                 r = -EFAULT;
 931                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 932                                  num_msrs_to_save * sizeof(u32)))
 933                         goto out;
 934                 if (copy_to_user(user_msr_list->indices
 935                                  + num_msrs_to_save * sizeof(u32),
 936                                  &emulated_msrs,
 937                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 938                         goto out;
 939                 r = 0;
 940                 break;
 941         }
 942         case KVM_GET_SUPPORTED_CPUID: {
 943                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 944                 struct kvm_cpuid2 cpuid;
 945
 946                 r = -EFAULT;
 947                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 948                         goto out;
 949                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 950                         cpuid_arg->entries);
 951                 if (r)
 952                         goto out;
 953
 954                 r = -EFAULT;
 955                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 956                         goto out;
 957                 r = 0;
 958                 break;
 959         }
 960         default:
 961                 r = -EINVAL;
 962         }
 963 out:
 964         return r;
 965 }
 966
 967 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 968 {
 969         kvm_x86_ops->vcpu_load(vcpu, cpu);
 970         kvm_write_guest_time(vcpu);
 971 }
 972
 973 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 974 {
 975         kvm_x86_ops->vcpu_put(vcpu);
 976         kvm_put_guest_fpu(vcpu);
 977 }
 978
 979 static int is_efer_nx(void)
 980 {
 981         u64 efer;
 982
 983         rdmsrl(MSR_EFER, efer);
 984         return efer & EFER_NX;
 985 }
 986
 987 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 988 {
 989         int i;
 990         struct kvm_cpuid_entry2 *e, *entry;
 991
 992         entry = NULL;
 993         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 994                 e = &vcpu->arch.cpuid_entries[i];
 995                 if (e->function == 0x80000001) {
 996                         entry = e;
 997                         break;
 998                 }
 999         }
1000         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1001                 entry->edx &= ~(1 << 20);
1002                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1003         }
1004 }
1005
1006 /* when an old userspace process fills a new kernel module */
1007 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1008                                     struct kvm_cpuid *cpuid,
1009                                     struct kvm_cpuid_entry __user *entries)
1010 {
1011         int r, i;
1012         struct kvm_cpuid_entry *cpuid_entries;
1013
1014         r = -E2BIG;
1015         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1016                 goto out;
1017         r = -ENOMEM;
1018         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1019         if (!cpuid_entries)
1020                 goto out;
1021         r = -EFAULT;
1022         if (copy_from_user(cpuid_entries, entries,
1023                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1024                 goto out_free;
1025         for (i = 0; i < cpuid->nent; i++) {
1026                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1027                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1028                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1029                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1030                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1031                 vcpu->arch.cpuid_entries[i].index = 0;
1032                 vcpu->arch.cpuid_entries[i].flags = 0;
1033                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1034                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1035                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1036         }
1037         vcpu->arch.cpuid_nent = cpuid->nent;
1038         cpuid_fix_nx_cap(vcpu);
1039         r = 0;
1040
1041 out_free:
1042         vfree(cpuid_entries);
1043 out:
1044         return r;
1045 }
1046
1047 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1048                                     struct kvm_cpuid2 *cpuid,
1049                                     struct kvm_cpuid_entry2 __user *entries)
1050 {
1051         int r;
1052
1053         r = -E2BIG;
1054         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1055                 goto out;
1056         r = -EFAULT;
1057         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1058                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1059                 goto out;
1060         vcpu->arch.cpuid_nent = cpuid->nent;
1061         return 0;
1062
1063 out:
1064         return r;
1065 }
1066
1067 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1068                                     struct kvm_cpuid2 *cpuid,
1069                                     struct kvm_cpuid_entry2 __user *entries)
1070 {
1071         int r;
1072
1073         r = -E2BIG;
1074         if (cpuid->nent < vcpu->arch.cpuid_nent)
1075                 goto out;
1076         r = -EFAULT;
1077         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1078                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1079                 goto out;
1080         return 0;
1081
1082 out:
1083         cpuid->nent = vcpu->arch.cpuid_nent;
1084         return r;
1085 }
1086
1087 static inline u32 bit(int bitno)
1088 {
1089         return 1 << (bitno & 31);
1090 }
1091
1092 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1093                           u32 index)
1094 {
1095         entry->function = function;
1096         entry->index = index;
1097         cpuid_count(entry->function, entry->index,
1098                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1099         entry->flags = 0;
1100 }
1101
1102 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1103                          u32 index, int *nent, int maxnent)
1104 {
1105         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1106                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1107                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1108                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1109                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1110                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1111                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1112                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1113                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1114                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1115         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1116                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1117                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1118                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1119                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1120                 bit(X86_FEATURE_PGE) |
1121                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1122                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1123                 bit(X86_FEATURE_SYSCALL) |
1124                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1125 #ifdef CONFIG_X86_64
1126                 bit(X86_FEATURE_LM) |
1127 #endif
1128                 bit(X86_FEATURE_MMXEXT) |
1129                 bit(X86_FEATURE_3DNOWEXT) |
1130                 bit(X86_FEATURE_3DNOW);
1131         const u32 kvm_supported_word3_x86_features =
1132                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1133         const u32 kvm_supported_word6_x86_features =
1134                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1135
1136         /* all func 2 cpuid_count() should be called on the same cpu */
1137         get_cpu();
1138         do_cpuid_1_ent(entry, function, index);
1139         ++*nent;
1140
1141         switch (function) {
1142         case 0:
1143                 entry->eax = min(entry->eax, (u32)0xb);
1144                 break;
1145         case 1:
1146                 entry->edx &= kvm_supported_word0_x86_features;
1147                 entry->ecx &= kvm_supported_word3_x86_features;
1148                 break;
1149         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1150          * may return different values. This forces us to get_cpu() before
1151          * issuing the first command, and also to emulate this annoying behavior
1152          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1153         case 2: {
1154                 int t, times = entry->eax & 0xff;
1155
1156                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1157                 for (t = 1; t < times && *nent < maxnent; ++t) {
1158                         do_cpuid_1_ent(&entry[t], function, 0);
1159                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1160                         ++*nent;
1161                 }
1162                 break;
1163         }
1164         /* function 4 and 0xb have additional index. */
1165         case 4: {
1166                 int i, cache_type;
1167
1168                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1169                 /* read more entries until cache_type is zero */
1170                 for (i = 1; *nent < maxnent; ++i) {
1171                         cache_type = entry[i - 1].eax & 0x1f;
1172                         if (!cache_type)
1173                                 break;
1174                         do_cpuid_1_ent(&entry[i], function, i);
1175                         entry[i].flags |=
1176                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1177                         ++*nent;
1178                 }
1179                 break;
1180         }
1181         case 0xb: {
1182                 int i, level_type;
1183
1184                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1185                 /* read more entries until level_type is zero */
1186                 for (i = 1; *nent < maxnent; ++i) {
1187                         level_type = entry[i - 1].ecx & 0xff;
1188                         if (!level_type)
1189                                 break;
1190                         do_cpuid_1_ent(&entry[i], function, i);
1191                         entry[i].flags |=
1192                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1193                         ++*nent;
1194                 }
1195                 break;
1196         }
1197         case 0x80000000:
1198                 entry->eax = min(entry->eax, 0x8000001a);
1199                 break;
1200         case 0x80000001:
1201                 entry->edx &= kvm_supported_word1_x86_features;
1202                 entry->ecx &= kvm_supported_word6_x86_features;
1203                 break;
1204         }
1205         put_cpu();
1206 }
1207
1208 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1209                                     struct kvm_cpuid_entry2 __user *entries)
1210 {
1211         struct kvm_cpuid_entry2 *cpuid_entries;
1212         int limit, nent = 0, r = -E2BIG;
1213         u32 func;
1214
1215         if (cpuid->nent < 1)
1216                 goto out;
1217         r = -ENOMEM;
1218         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1219         if (!cpuid_entries)
1220                 goto out;
1221
1222         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1223         limit = cpuid_entries[0].eax;
1224         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1225                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1226                                 &nent, cpuid->nent);
1227         r = -E2BIG;
1228         if (nent >= cpuid->nent)
1229                 goto out_free;
1230
1231         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1232         limit = cpuid_entries[nent - 1].eax;
1233         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1234                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1235                                &nent, cpuid->nent);
1236         r = -EFAULT;
1237         if (copy_to_user(entries, cpuid_entries,
1238                         nent * sizeof(struct kvm_cpuid_entry2)))
1239                 goto out_free;
1240         cpuid->nent = nent;
1241         r = 0;
1242
1243 out_free:
1244         vfree(cpuid_entries);
1245 out:
1246         return r;
1247 }
1248
1249 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1250                                     struct kvm_lapic_state *s)
1251 {
1252         vcpu_load(vcpu);
1253         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1254         vcpu_put(vcpu);
1255
1256         return 0;
1257 }
1258
1259 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1260                                     struct kvm_lapic_state *s)
1261 {
1262         vcpu_load(vcpu);
1263         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1264         kvm_apic_post_state_restore(vcpu);
1265         vcpu_put(vcpu);
1266
1267         return 0;
1268 }
1269
1270 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1271                                     struct kvm_interrupt *irq)
1272 {
1273         if (irq->irq < 0 || irq->irq >= 256)
1274                 return -EINVAL;
1275         if (irqchip_in_kernel(vcpu->kvm))
1276                 return -ENXIO;
1277         vcpu_load(vcpu);
1278
1279         set_bit(irq->irq, vcpu->arch.irq_pending);
1280         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1281
1282         vcpu_put(vcpu);
1283
1284         return 0;
1285 }
1286
1287 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1288                                            struct kvm_tpr_access_ctl *tac)
1289 {
1290         if (tac->flags)
1291                 return -EINVAL;
1292         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1293         return 0;
1294 }
1295
1296 long kvm_arch_vcpu_ioctl(struct file *filp,
1297                          unsigned int ioctl, unsigned long arg)
1298 {
1299         struct kvm_vcpu *vcpu = filp->private_data;
1300         void __user *argp = (void __user *)arg;
1301         int r;
1302
1303         switch (ioctl) {
1304         case KVM_GET_LAPIC: {
1305                 struct kvm_lapic_state lapic;
1306
1307                 memset(&lapic, 0, sizeof lapic);
1308                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1309                 if (r)
1310                         goto out;
1311                 r = -EFAULT;
1312                 if (copy_to_user(argp, &lapic, sizeof lapic))
1313                         goto out;
1314                 r = 0;
1315                 break;
1316         }
1317         case KVM_SET_LAPIC: {
1318                 struct kvm_lapic_state lapic;
1319
1320                 r = -EFAULT;
1321                 if (copy_from_user(&lapic, argp, sizeof lapic))
1322                         goto out;
1323                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1324                 if (r)
1325                         goto out;
1326                 r = 0;
1327                 break;
1328         }
1329         case KVM_INTERRUPT: {
1330                 struct kvm_interrupt irq;
1331
1332                 r = -EFAULT;
1333                 if (copy_from_user(&irq, argp, sizeof irq))
1334                         goto out;
1335                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1336                 if (r)
1337                         goto out;
1338                 r = 0;
1339                 break;
1340         }
1341         case KVM_SET_CPUID: {
1342                 struct kvm_cpuid __user *cpuid_arg = argp;
1343                 struct kvm_cpuid cpuid;
1344
1345                 r = -EFAULT;
1346                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1347                         goto out;
1348                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1349                 if (r)
1350                         goto out;
1351                 break;
1352         }
1353         case KVM_SET_CPUID2: {
1354                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1355                 struct kvm_cpuid2 cpuid;
1356
1357                 r = -EFAULT;
1358                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1359                         goto out;
1360                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1361                                 cpuid_arg->entries);
1362                 if (r)
1363                         goto out;
1364                 break;
1365         }
1366         case KVM_GET_CPUID2: {
1367                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1368                 struct kvm_cpuid2 cpuid;
1369
1370                 r = -EFAULT;
1371                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1372                         goto out;
1373                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1374                                 cpuid_arg->entries);
1375                 if (r)
1376                         goto out;
1377                 r = -EFAULT;
1378                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1379                         goto out;
1380                 r = 0;
1381                 break;
1382         }
1383         case KVM_GET_MSRS:
1384                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1385                 break;
1386         case KVM_SET_MSRS:
1387                 r = msr_io(vcpu, argp, do_set_msr, 0);
1388                 break;
1389         case KVM_TPR_ACCESS_REPORTING: {
1390                 struct kvm_tpr_access_ctl tac;
1391
1392                 r = -EFAULT;
1393                 if (copy_from_user(&tac, argp, sizeof tac))
1394                         goto out;
1395                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1396                 if (r)
1397                         goto out;
1398                 r = -EFAULT;
1399                 if (copy_to_user(argp, &tac, sizeof tac))
1400                         goto out;
1401                 r = 0;
1402                 break;
1403         };
1404         case KVM_SET_VAPIC_ADDR: {
1405                 struct kvm_vapic_addr va;
1406
1407                 r = -EINVAL;
1408                 if (!irqchip_in_kernel(vcpu->kvm))
1409                         goto out;
1410                 r = -EFAULT;
1411                 if (copy_from_user(&va, argp, sizeof va))
1412                         goto out;
1413                 r = 0;
1414                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1415                 break;
1416         }
1417         default:
1418                 r = -EINVAL;
1419         }
1420 out:
1421         return r;
1422 }
1423
1424 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1425 {
1426         int ret;
1427
1428         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1429                 return -1;
1430         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1431         return ret;
1432 }
1433
1434 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1435                                           u32 kvm_nr_mmu_pages)
1436 {
1437         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1438                 return -EINVAL;
1439
1440         down_write(&kvm->slots_lock);
1441
1442         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1443         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1444
1445         up_write(&kvm->slots_lock);
1446         return 0;
1447 }
1448
1449 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1450 {
1451         return kvm->arch.n_alloc_mmu_pages;
1452 }
1453
1454 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1455 {
1456         int i;
1457         struct kvm_mem_alias *alias;
1458
1459         for (i = 0; i < kvm->arch.naliases; ++i) {
1460                 alias = &kvm->arch.aliases[i];
1461                 if (gfn >= alias->base_gfn
1462                     && gfn < alias->base_gfn + alias->npages)
1463                         return alias->target_gfn + gfn - alias->base_gfn;
1464         }
1465         return gfn;
1466 }
1467
1468 /*
1469  * Set a new alias region.  Aliases map a portion of physical memory into
1470  * another portion.  This is useful for memory windows, for example the PC
1471  * VGA region.
1472  */
1473 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1474                                          struct kvm_memory_alias *alias)
1475 {
1476         int r, n;
1477         struct kvm_mem_alias *p;
1478
1479         r = -EINVAL;
1480         /* General sanity checks */
1481         if (alias->memory_size & (PAGE_SIZE - 1))
1482                 goto out;
1483         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1484                 goto out;
1485         if (alias->slot >= KVM_ALIAS_SLOTS)
1486                 goto out;
1487         if (alias->guest_phys_addr + alias->memory_size
1488             < alias->guest_phys_addr)
1489                 goto out;
1490         if (alias->target_phys_addr + alias->memory_size
1491             < alias->target_phys_addr)
1492                 goto out;
1493
1494         down_write(&kvm->slots_lock);
1495
1496         p = &kvm->arch.aliases[alias->slot];
1497         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1498         p->npages = alias->memory_size >> PAGE_SHIFT;
1499         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1500
1501         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1502                 if (kvm->arch.aliases[n - 1].npages)
1503                         break;
1504         kvm->arch.naliases = n;
1505
1506         kvm_mmu_zap_all(kvm);
1507
1508         up_write(&kvm->slots_lock);
1509
1510         return 0;
1511
1512 out:
1513         return r;
1514 }
1515
1516 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1517 {
1518         int r;
1519
1520         r = 0;
1521         switch (chip->chip_id) {
1522         case KVM_IRQCHIP_PIC_MASTER:
1523                 memcpy(&chip->chip.pic,
1524                         &pic_irqchip(kvm)->pics[0],
1525                         sizeof(struct kvm_pic_state));
1526                 break;
1527         case KVM_IRQCHIP_PIC_SLAVE:
1528                 memcpy(&chip->chip.pic,
1529                         &pic_irqchip(kvm)->pics[1],
1530                         sizeof(struct kvm_pic_state));
1531                 break;
1532         case KVM_IRQCHIP_IOAPIC:
1533                 memcpy(&chip->chip.ioapic,
1534                         ioapic_irqchip(kvm),
1535                         sizeof(struct kvm_ioapic_state));
1536                 break;
1537         default:
1538                 r = -EINVAL;
1539                 break;
1540         }
1541         return r;
1542 }
1543
1544 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1545 {
1546         int r;
1547
1548         r = 0;
1549         switch (chip->chip_id) {
1550         case KVM_IRQCHIP_PIC_MASTER:
1551                 memcpy(&pic_irqchip(kvm)->pics[0],
1552                         &chip->chip.pic,
1553                         sizeof(struct kvm_pic_state));
1554                 break;
1555         case KVM_IRQCHIP_PIC_SLAVE:
1556                 memcpy(&pic_irqchip(kvm)->pics[1],
1557                         &chip->chip.pic,
1558                         sizeof(struct kvm_pic_state));
1559                 break;
1560         case KVM_IRQCHIP_IOAPIC:
1561                 memcpy(ioapic_irqchip(kvm),
1562                         &chip->chip.ioapic,
1563                         sizeof(struct kvm_ioapic_state));
1564                 break;
1565         default:
1566                 r = -EINVAL;
1567                 break;
1568         }
1569         kvm_pic_update_irq(pic_irqchip(kvm));
1570         return r;
1571 }
1572
1573 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1574 {
1575         int r = 0;
1576
1577         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1578         return r;
1579 }
1580
1581 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1582 {
1583         int r = 0;
1584
1585         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1586         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1587         return r;
1588 }
1589
1590 /*
1591  * Get (and clear) the dirty memory log for a memory slot.
1592  */
1593 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1594                                       struct kvm_dirty_log *log)
1595 {
1596         int r;
1597         int n;
1598         struct kvm_memory_slot *memslot;
1599         int is_dirty = 0;
1600
1601         down_write(&kvm->slots_lock);
1602
1603         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1604         if (r)
1605                 goto out;
1606
1607         /* If nothing is dirty, don't bother messing with page tables. */
1608         if (is_dirty) {
1609                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1610                 kvm_flush_remote_tlbs(kvm);
1611                 memslot = &kvm->memslots[log->slot];
1612                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1613                 memset(memslot->dirty_bitmap, 0, n);
1614         }
1615         r = 0;
1616 out:
1617         up_write(&kvm->slots_lock);
1618         return r;
1619 }
1620
1621 long kvm_arch_vm_ioctl(struct file *filp,
1622                        unsigned int ioctl, unsigned long arg)
1623 {
1624         struct kvm *kvm = filp->private_data;
1625         void __user *argp = (void __user *)arg;
1626         int r = -EINVAL;
1627
1628         switch (ioctl) {
1629         case KVM_SET_TSS_ADDR:
1630                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1631                 if (r < 0)
1632                         goto out;
1633                 break;
1634         case KVM_SET_MEMORY_REGION: {
1635                 struct kvm_memory_region kvm_mem;
1636                 struct kvm_userspace_memory_region kvm_userspace_mem;
1637
1638                 r = -EFAULT;
1639                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1640                         goto out;
1641                 kvm_userspace_mem.slot = kvm_mem.slot;
1642                 kvm_userspace_mem.flags = kvm_mem.flags;
1643                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1644                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1645                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1646                 if (r)
1647                         goto out;
1648                 break;
1649         }
1650         case KVM_SET_NR_MMU_PAGES:
1651                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1652                 if (r)
1653                         goto out;
1654                 break;
1655         case KVM_GET_NR_MMU_PAGES:
1656                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1657                 break;
1658         case KVM_SET_MEMORY_ALIAS: {
1659                 struct kvm_memory_alias alias;
1660
1661                 r = -EFAULT;
1662                 if (copy_from_user(&alias, argp, sizeof alias))
1663                         goto out;
1664                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1665                 if (r)
1666                         goto out;
1667                 break;
1668         }
1669         case KVM_CREATE_IRQCHIP:
1670                 r = -ENOMEM;
1671                 kvm->arch.vpic = kvm_create_pic(kvm);
1672                 if (kvm->arch.vpic) {
1673                         r = kvm_ioapic_init(kvm);
1674                         if (r) {
1675                                 kfree(kvm->arch.vpic);
1676                                 kvm->arch.vpic = NULL;
1677                                 goto out;
1678                         }
1679                 } else
1680                         goto out;
1681                 break;
1682         case KVM_CREATE_PIT:
1683                 r = -ENOMEM;
1684                 kvm->arch.vpit = kvm_create_pit(kvm);
1685                 if (kvm->arch.vpit)
1686                         r = 0;
1687                 break;
1688         case KVM_IRQ_LINE: {
1689                 struct kvm_irq_level irq_event;
1690
1691                 r = -EFAULT;
1692                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1693                         goto out;
1694                 if (irqchip_in_kernel(kvm)) {
1695                         mutex_lock(&kvm->lock);
1696                         if (irq_event.irq < 16)
1697                                 kvm_pic_set_irq(pic_irqchip(kvm),
1698                                         irq_event.irq,
1699                                         irq_event.level);
1700                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1701                                         irq_event.irq,
1702                                         irq_event.level);
1703                         mutex_unlock(&kvm->lock);
1704                         r = 0;
1705                 }
1706                 break;
1707         }
1708         case KVM_GET_IRQCHIP: {
1709                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1710                 struct kvm_irqchip chip;
1711
1712                 r = -EFAULT;
1713                 if (copy_from_user(&chip, argp, sizeof chip))
1714                         goto out;
1715                 r = -ENXIO;
1716                 if (!irqchip_in_kernel(kvm))
1717                         goto out;
1718                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1719                 if (r)
1720                         goto out;
1721                 r = -EFAULT;
1722                 if (copy_to_user(argp, &chip, sizeof chip))
1723                         goto out;
1724                 r = 0;
1725                 break;
1726         }
1727         case KVM_SET_IRQCHIP: {
1728                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1729                 struct kvm_irqchip chip;
1730
1731                 r = -EFAULT;
1732                 if (copy_from_user(&chip, argp, sizeof chip))
1733                         goto out;
1734                 r = -ENXIO;
1735                 if (!irqchip_in_kernel(kvm))
1736                         goto out;
1737                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1738                 if (r)
1739                         goto out;
1740                 r = 0;
1741                 break;
1742         }
1743         case KVM_GET_PIT: {
1744                 struct kvm_pit_state ps;
1745                 r = -EFAULT;
1746                 if (copy_from_user(&ps, argp, sizeof ps))
1747                         goto out;
1748                 r = -ENXIO;
1749                 if (!kvm->arch.vpit)
1750                         goto out;
1751                 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1752                 if (r)
1753                         goto out;
1754                 r = -EFAULT;
1755                 if (copy_to_user(argp, &ps, sizeof ps))
1756                         goto out;
1757                 r = 0;
1758                 break;
1759         }
1760         case KVM_SET_PIT: {
1761                 struct kvm_pit_state ps;
1762                 r = -EFAULT;
1763                 if (copy_from_user(&ps, argp, sizeof ps))
1764                         goto out;
1765                 r = -ENXIO;
1766                 if (!kvm->arch.vpit)
1767                         goto out;
1768                 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1769                 if (r)
1770                         goto out;
1771                 r = 0;
1772                 break;
1773         }
1774         default:
1775                 ;
1776         }
1777 out:
1778         return r;
1779 }
1780
1781 static void kvm_init_msr_list(void)
1782 {
1783         u32 dummy[2];
1784         unsigned i, j;
1785
1786         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1787                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1788                         continue;
1789                 if (j < i)
1790                         msrs_to_save[j] = msrs_to_save[i];
1791                 j++;
1792         }
1793         num_msrs_to_save = j;
1794 }
1795
1796 /*
1797  * Only apic need an MMIO device hook, so shortcut now..
1798  */
1799 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1800                                                 gpa_t addr, int len,
1801                                                 int is_write)
1802 {
1803         struct kvm_io_device *dev;
1804
1805         if (vcpu->arch.apic) {
1806                 dev = &vcpu->arch.apic->dev;
1807                 if (dev->in_range(dev, addr, len, is_write))
1808                         return dev;
1809         }
1810         return NULL;
1811 }
1812
1813
1814 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1815                                                 gpa_t addr, int len,
1816                                                 int is_write)
1817 {
1818         struct kvm_io_device *dev;
1819
1820         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1821         if (dev == NULL)
1822                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1823                                           is_write);
1824         return dev;
1825 }
1826
1827 int emulator_read_std(unsigned long addr,
1828                              void *val,
1829                              unsigned int bytes,
1830                              struct kvm_vcpu *vcpu)
1831 {
1832         void *data = val;
1833         int r = X86EMUL_CONTINUE;
1834
1835         while (bytes) {
1836                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1837                 unsigned offset = addr & (PAGE_SIZE-1);
1838                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1839                 int ret;
1840
1841                 if (gpa == UNMAPPED_GVA) {
1842                         r = X86EMUL_PROPAGATE_FAULT;
1843                         goto out;
1844                 }
1845                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1846                 if (ret < 0) {
1847                         r = X86EMUL_UNHANDLEABLE;
1848                         goto out;
1849                 }
1850
1851                 bytes -= tocopy;
1852                 data += tocopy;
1853                 addr += tocopy;
1854         }
1855 out:
1856         return r;
1857 }
1858 EXPORT_SYMBOL_GPL(emulator_read_std);
1859
1860 static int emulator_read_emulated(unsigned long addr,
1861                                   void *val,
1862                                   unsigned int bytes,
1863                                   struct kvm_vcpu *vcpu)
1864 {
1865         struct kvm_io_device *mmio_dev;
1866         gpa_t                 gpa;
1867
1868         if (vcpu->mmio_read_completed) {
1869                 memcpy(val, vcpu->mmio_data, bytes);
1870                 vcpu->mmio_read_completed = 0;
1871                 return X86EMUL_CONTINUE;
1872         }
1873
1874         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1875
1876         /* For APIC access vmexit */
1877         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1878                 goto mmio;
1879
1880         if (emulator_read_std(addr, val, bytes, vcpu)
1881                         == X86EMUL_CONTINUE)
1882                 return X86EMUL_CONTINUE;
1883         if (gpa == UNMAPPED_GVA)
1884                 return X86EMUL_PROPAGATE_FAULT;
1885
1886 mmio:
1887         /*
1888          * Is this MMIO handled locally?
1889          */
1890         mutex_lock(&vcpu->kvm->lock);
1891         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1892         if (mmio_dev) {
1893                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1894                 mutex_unlock(&vcpu->kvm->lock);
1895                 return X86EMUL_CONTINUE;
1896         }
1897         mutex_unlock(&vcpu->kvm->lock);
1898
1899         vcpu->mmio_needed = 1;
1900         vcpu->mmio_phys_addr = gpa;
1901         vcpu->mmio_size = bytes;
1902         vcpu->mmio_is_write = 0;
1903
1904         return X86EMUL_UNHANDLEABLE;
1905 }
1906
1907 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1908                           const void *val, int bytes)
1909 {
1910         int ret;
1911
1912         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1913         if (ret < 0)
1914                 return 0;
1915         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1916         return 1;
1917 }
1918
1919 static int emulator_write_emulated_onepage(unsigned long addr,
1920                                            const void *val,
1921                                            unsigned int bytes,
1922                                            struct kvm_vcpu *vcpu)
1923 {
1924         struct kvm_io_device *mmio_dev;
1925         gpa_t                 gpa;
1926
1927         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1928
1929         if (gpa == UNMAPPED_GVA) {
1930                 kvm_inject_page_fault(vcpu, addr, 2);
1931                 return X86EMUL_PROPAGATE_FAULT;
1932         }
1933
1934         /* For APIC access vmexit */
1935         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1936                 goto mmio;
1937
1938         if (emulator_write_phys(vcpu, gpa, val, bytes))
1939                 return X86EMUL_CONTINUE;
1940
1941 mmio:
1942         /*
1943          * Is this MMIO handled locally?
1944          */
1945         mutex_lock(&vcpu->kvm->lock);
1946         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
1947         if (mmio_dev) {
1948                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1949                 mutex_unlock(&vcpu->kvm->lock);
1950                 return X86EMUL_CONTINUE;
1951         }
1952         mutex_unlock(&vcpu->kvm->lock);
1953
1954         vcpu->mmio_needed = 1;
1955         vcpu->mmio_phys_addr = gpa;
1956         vcpu->mmio_size = bytes;
1957         vcpu->mmio_is_write = 1;
1958         memcpy(vcpu->mmio_data, val, bytes);
1959
1960         return X86EMUL_CONTINUE;
1961 }
1962
1963 int emulator_write_emulated(unsigned long addr,
1964                                    const void *val,
1965                                    unsigned int bytes,
1966                                    struct kvm_vcpu *vcpu)
1967 {
1968         /* Crossing a page boundary? */
1969         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1970                 int rc, now;
1971
1972                 now = -addr & ~PAGE_MASK;
1973                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1974                 if (rc != X86EMUL_CONTINUE)
1975                         return rc;
1976                 addr += now;
1977                 val += now;
1978                 bytes -= now;
1979         }
1980         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1981 }
1982 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1983
1984 static int emulator_cmpxchg_emulated(unsigned long addr,
1985                                      const void *old,
1986                                      const void *new,
1987                                      unsigned int bytes,
1988                                      struct kvm_vcpu *vcpu)
1989 {
1990         static int reported;
1991
1992         if (!reported) {
1993                 reported = 1;
1994                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1995         }
1996 #ifndef CONFIG_X86_64
1997         /* guests cmpxchg8b have to be emulated atomically */
1998         if (bytes == 8) {
1999                 gpa_t gpa;
2000                 struct page *page;
2001                 char *kaddr;
2002                 u64 val;
2003
2004                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2005
2006                 if (gpa == UNMAPPED_GVA ||
2007                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2008                         goto emul_write;
2009
2010                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2011                         goto emul_write;
2012
2013                 val = *(u64 *)new;
2014
2015                 down_read(&current->mm->mmap_sem);
2016                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2017                 up_read(&current->mm->mmap_sem);
2018
2019                 kaddr = kmap_atomic(page, KM_USER0);
2020                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2021                 kunmap_atomic(kaddr, KM_USER0);
2022                 kvm_release_page_dirty(page);
2023         }
2024 emul_write:
2025 #endif
2026
2027         return emulator_write_emulated(addr, new, bytes, vcpu);
2028 }
2029
2030 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2031 {
2032         return kvm_x86_ops->get_segment_base(vcpu, seg);
2033 }
2034
2035 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2036 {
2037         return X86EMUL_CONTINUE;
2038 }
2039
2040 int emulate_clts(struct kvm_vcpu *vcpu)
2041 {
2042         KVMTRACE_0D(CLTS, vcpu, handler);
2043         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2044         return X86EMUL_CONTINUE;
2045 }
2046
2047 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2048 {
2049         struct kvm_vcpu *vcpu = ctxt->vcpu;
2050
2051         switch (dr) {
2052         case 0 ... 3:
2053                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2054                 return X86EMUL_CONTINUE;
2055         default:
2056                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2057                 return X86EMUL_UNHANDLEABLE;
2058         }
2059 }
2060
2061 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2062 {
2063         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2064         int exception;
2065
2066         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2067         if (exception) {
2068                 /* FIXME: better handling */
2069                 return X86EMUL_UNHANDLEABLE;
2070         }
2071         return X86EMUL_CONTINUE;
2072 }
2073
2074 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2075 {
2076         static int reported;
2077         u8 opcodes[4];
2078         unsigned long rip = vcpu->arch.rip;
2079         unsigned long rip_linear;
2080
2081         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2082
2083         if (reported)
2084                 return;
2085
2086         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2087
2088         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2089                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2090         reported = 1;
2091 }
2092 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2093
2094 static struct x86_emulate_ops emulate_ops = {
2095         .read_std            = emulator_read_std,
2096         .read_emulated       = emulator_read_emulated,
2097         .write_emulated      = emulator_write_emulated,
2098         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2099 };
2100
2101 int emulate_instruction(struct kvm_vcpu *vcpu,
2102                         struct kvm_run *run,
2103                         unsigned long cr2,
2104                         u16 error_code,
2105                         int emulation_type)
2106 {
2107         int r;
2108         struct decode_cache *c;
2109
2110         vcpu->arch.mmio_fault_cr2 = cr2;
2111         kvm_x86_ops->cache_regs(vcpu);
2112
2113         vcpu->mmio_is_write = 0;
2114         vcpu->arch.pio.string = 0;
2115
2116         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2117                 int cs_db, cs_l;
2118                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2119
2120                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2121                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2122                 vcpu->arch.emulate_ctxt.mode =
2123                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2124                         ? X86EMUL_MODE_REAL : cs_l
2125                         ? X86EMUL_MODE_PROT64 : cs_db
2126                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2127
2128                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2129                         vcpu->arch.emulate_ctxt.cs_base = 0;
2130                         vcpu->arch.emulate_ctxt.ds_base = 0;
2131                         vcpu->arch.emulate_ctxt.es_base = 0;
2132                         vcpu->arch.emulate_ctxt.ss_base = 0;
2133                 } else {
2134                         vcpu->arch.emulate_ctxt.cs_base =
2135                                         get_segment_base(vcpu, VCPU_SREG_CS);
2136                         vcpu->arch.emulate_ctxt.ds_base =
2137                                         get_segment_base(vcpu, VCPU_SREG_DS);
2138                         vcpu->arch.emulate_ctxt.es_base =
2139                                         get_segment_base(vcpu, VCPU_SREG_ES);
2140                         vcpu->arch.emulate_ctxt.ss_base =
2141                                         get_segment_base(vcpu, VCPU_SREG_SS);
2142                 }
2143
2144                 vcpu->arch.emulate_ctxt.gs_base =
2145                                         get_segment_base(vcpu, VCPU_SREG_GS);
2146                 vcpu->arch.emulate_ctxt.fs_base =
2147                                         get_segment_base(vcpu, VCPU_SREG_FS);
2148
2149                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2150
2151                 /* Reject the instructions other than VMCALL/VMMCALL when
2152                  * try to emulate invalid opcode */
2153                 c = &vcpu->arch.emulate_ctxt.decode;
2154                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2155                     (!(c->twobyte && c->b == 0x01 &&
2156                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2157                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2158                         return EMULATE_FAIL;
2159
2160                 ++vcpu->stat.insn_emulation;
2161                 if (r)  {
2162                         ++vcpu->stat.insn_emulation_fail;
2163                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2164                                 return EMULATE_DONE;
2165                         return EMULATE_FAIL;
2166                 }
2167         }
2168
2169         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2170
2171         if (vcpu->arch.pio.string)
2172                 return EMULATE_DO_MMIO;
2173
2174         if ((r || vcpu->mmio_is_write) && run) {
2175                 run->exit_reason = KVM_EXIT_MMIO;
2176                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2177                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2178                 run->mmio.len = vcpu->mmio_size;
2179                 run->mmio.is_write = vcpu->mmio_is_write;
2180         }
2181
2182         if (r) {
2183                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2184                         return EMULATE_DONE;
2185                 if (!vcpu->mmio_needed) {
2186                         kvm_report_emulation_failure(vcpu, "mmio");
2187                         return EMULATE_FAIL;
2188                 }
2189                 return EMULATE_DO_MMIO;
2190         }
2191
2192         kvm_x86_ops->decache_regs(vcpu);
2193         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2194
2195         if (vcpu->mmio_is_write) {
2196                 vcpu->mmio_needed = 0;
2197                 return EMULATE_DO_MMIO;
2198         }
2199
2200         return EMULATE_DONE;
2201 }
2202 EXPORT_SYMBOL_GPL(emulate_instruction);
2203
2204 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2205 {
2206         int i;
2207
2208         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2209                 if (vcpu->arch.pio.guest_pages[i]) {
2210                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2211                         vcpu->arch.pio.guest_pages[i] = NULL;
2212                 }
2213 }
2214
2215 static int pio_copy_data(struct kvm_vcpu *vcpu)
2216 {
2217         void *p = vcpu->arch.pio_data;
2218         void *q;
2219         unsigned bytes;
2220         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2221
2222         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2223                  PAGE_KERNEL);
2224         if (!q) {
2225                 free_pio_guest_pages(vcpu);
2226                 return -ENOMEM;
2227         }
2228         q += vcpu->arch.pio.guest_page_offset;
2229         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2230         if (vcpu->arch.pio.in)
2231                 memcpy(q, p, bytes);
2232         else
2233                 memcpy(p, q, bytes);
2234         q -= vcpu->arch.pio.guest_page_offset;
2235         vunmap(q);
2236         free_pio_guest_pages(vcpu);
2237         return 0;
2238 }
2239
2240 int complete_pio(struct kvm_vcpu *vcpu)
2241 {
2242         struct kvm_pio_request *io = &vcpu->arch.pio;
2243         long delta;
2244         int r;
2245
2246         kvm_x86_ops->cache_regs(vcpu);
2247
2248         if (!io->string) {
2249                 if (io->in)
2250                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2251                                io->size);
2252         } else {
2253                 if (io->in) {
2254                         r = pio_copy_data(vcpu);
2255                         if (r) {
2256                                 kvm_x86_ops->cache_regs(vcpu);
2257                                 return r;
2258                         }
2259                 }
2260
2261                 delta = 1;
2262                 if (io->rep) {
2263                         delta *= io->cur_count;
2264                         /*
2265                          * The size of the register should really depend on
2266                          * current address size.
2267                          */
2268                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2269                 }
2270                 if (io->down)
2271                         delta = -delta;
2272                 delta *= io->size;
2273                 if (io->in)
2274                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2275                 else
2276                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2277         }
2278
2279         kvm_x86_ops->decache_regs(vcpu);
2280
2281         io->count -= io->cur_count;
2282         io->cur_count = 0;
2283
2284         return 0;
2285 }
2286
2287 static void kernel_pio(struct kvm_io_device *pio_dev,
2288                        struct kvm_vcpu *vcpu,
2289                        void *pd)
2290 {
2291         /* TODO: String I/O for in kernel device */
2292
2293         mutex_lock(&vcpu->kvm->lock);
2294         if (vcpu->arch.pio.in)
2295                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2296                                   vcpu->arch.pio.size,
2297                                   pd);
2298         else
2299                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2300                                    vcpu->arch.pio.size,
2301                                    pd);
2302         mutex_unlock(&vcpu->kvm->lock);
2303 }
2304
2305 static void pio_string_write(struct kvm_io_device *pio_dev,
2306                              struct kvm_vcpu *vcpu)
2307 {
2308         struct kvm_pio_request *io = &vcpu->arch.pio;
2309         void *pd = vcpu->arch.pio_data;
2310         int i;
2311
2312         mutex_lock(&vcpu->kvm->lock);
2313         for (i = 0; i < io->cur_count; i++) {
2314                 kvm_iodevice_write(pio_dev, io->port,
2315                                    io->size,
2316                                    pd);
2317                 pd += io->size;
2318         }
2319         mutex_unlock(&vcpu->kvm->lock);
2320 }
2321
2322 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2323                                                gpa_t addr, int len,
2324                                                int is_write)
2325 {
2326         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2327 }
2328
2329 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2330                   int size, unsigned port)
2331 {
2332         struct kvm_io_device *pio_dev;
2333
2334         vcpu->run->exit_reason = KVM_EXIT_IO;
2335         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2336         vcpu->run->io.size = vcpu->arch.pio.size = size;
2337         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2338         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2339         vcpu->run->io.port = vcpu->arch.pio.port = port;
2340         vcpu->arch.pio.in = in;
2341         vcpu->arch.pio.string = 0;
2342         vcpu->arch.pio.down = 0;
2343         vcpu->arch.pio.guest_page_offset = 0;
2344         vcpu->arch.pio.rep = 0;
2345
2346         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2347                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2348                             handler);
2349         else
2350                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2351                             handler);
2352
2353         kvm_x86_ops->cache_regs(vcpu);
2354         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2355
2356         kvm_x86_ops->skip_emulated_instruction(vcpu);
2357
2358         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2359         if (pio_dev) {
2360                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2361                 complete_pio(vcpu);
2362                 return 1;
2363         }
2364         return 0;
2365 }
2366 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2367
2368 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2369                   int size, unsigned long count, int down,
2370                   gva_t address, int rep, unsigned port)
2371 {
2372         unsigned now, in_page;
2373         int i, ret = 0;
2374         int nr_pages = 1;
2375         struct page *page;
2376         struct kvm_io_device *pio_dev;
2377
2378         vcpu->run->exit_reason = KVM_EXIT_IO;
2379         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2380         vcpu->run->io.size = vcpu->arch.pio.size = size;
2381         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2382         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2383         vcpu->run->io.port = vcpu->arch.pio.port = port;
2384         vcpu->arch.pio.in = in;
2385         vcpu->arch.pio.string = 1;
2386         vcpu->arch.pio.down = down;
2387         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2388         vcpu->arch.pio.rep = rep;
2389
2390         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2391                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2392                             handler);
2393         else
2394                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2395                             handler);
2396
2397         if (!count) {
2398                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2399                 return 1;
2400         }
2401
2402         if (!down)
2403                 in_page = PAGE_SIZE - offset_in_page(address);
2404         else
2405                 in_page = offset_in_page(address) + size;
2406         now = min(count, (unsigned long)in_page / size);
2407         if (!now) {
2408                 /*
2409                  * String I/O straddles page boundary.  Pin two guest pages
2410                  * so that we satisfy atomicity constraints.  Do just one
2411                  * transaction to avoid complexity.
2412                  */
2413                 nr_pages = 2;
2414                 now = 1;
2415         }
2416         if (down) {
2417                 /*
2418                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2419                  */
2420                 pr_unimpl(vcpu, "guest string pio down\n");
2421                 kvm_inject_gp(vcpu, 0);
2422                 return 1;
2423         }
2424         vcpu->run->io.count = now;
2425         vcpu->arch.pio.cur_count = now;
2426
2427         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2428                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2429
2430         for (i = 0; i < nr_pages; ++i) {
2431                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2432                 vcpu->arch.pio.guest_pages[i] = page;
2433                 if (!page) {
2434                         kvm_inject_gp(vcpu, 0);
2435                         free_pio_guest_pages(vcpu);
2436                         return 1;
2437                 }
2438         }
2439
2440         pio_dev = vcpu_find_pio_dev(vcpu, port,
2441                                     vcpu->arch.pio.cur_count,
2442                                     !vcpu->arch.pio.in);
2443         if (!vcpu->arch.pio.in) {
2444                 /* string PIO write */
2445                 ret = pio_copy_data(vcpu);
2446                 if (ret >= 0 && pio_dev) {
2447                         pio_string_write(pio_dev, vcpu);
2448                         complete_pio(vcpu);
2449                         if (vcpu->arch.pio.count == 0)
2450                                 ret = 1;
2451                 }
2452         } else if (pio_dev)
2453                 pr_unimpl(vcpu, "no string pio read support yet, "
2454                        "port %x size %d count %ld\n",
2455                         port, size, count);
2456
2457         return ret;
2458 }
2459 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2460
2461 int kvm_arch_init(void *opaque)
2462 {
2463         int r;
2464         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2465
2466         if (kvm_x86_ops) {
2467                 printk(KERN_ERR "kvm: already loaded the other module\n");
2468                 r = -EEXIST;
2469                 goto out;
2470         }
2471
2472         if (!ops->cpu_has_kvm_support()) {
2473                 printk(KERN_ERR "kvm: no hardware support\n");
2474                 r = -EOPNOTSUPP;
2475                 goto out;
2476         }
2477         if (ops->disabled_by_bios()) {
2478                 printk(KERN_ERR "kvm: disabled by bios\n");
2479                 r = -EOPNOTSUPP;
2480                 goto out;
2481         }
2482
2483         r = kvm_mmu_module_init();
2484         if (r)
2485                 goto out;
2486
2487         kvm_init_msr_list();
2488
2489         kvm_x86_ops = ops;
2490         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2491         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2492         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2493                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2494         return 0;
2495
2496 out:
2497         return r;
2498 }
2499
2500 void kvm_arch_exit(void)
2501 {
2502         kvm_x86_ops = NULL;
2503         kvm_mmu_module_exit();
2504 }
2505
2506 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2507 {
2508         ++vcpu->stat.halt_exits;
2509         KVMTRACE_0D(HLT, vcpu, handler);
2510         if (irqchip_in_kernel(vcpu->kvm)) {
2511                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2512                 up_read(&vcpu->kvm->slots_lock);
2513                 kvm_vcpu_block(vcpu);
2514                 down_read(&vcpu->kvm->slots_lock);
2515                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2516                         return -EINTR;
2517                 return 1;
2518         } else {
2519                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2520                 return 0;
2521         }
2522 }
2523 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2524
2525 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2526                            unsigned long a1)
2527 {
2528         if (is_long_mode(vcpu))
2529                 return a0;
2530         else
2531                 return a0 | ((gpa_t)a1 << 32);
2532 }
2533
2534 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2535 {
2536         unsigned long nr, a0, a1, a2, a3, ret;
2537         int r = 1;
2538
2539         kvm_x86_ops->cache_regs(vcpu);
2540
2541         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2542         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2543         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2544         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2545         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2546
2547         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2548
2549         if (!is_long_mode(vcpu)) {
2550                 nr &= 0xFFFFFFFF;
2551                 a0 &= 0xFFFFFFFF;
2552                 a1 &= 0xFFFFFFFF;
2553                 a2 &= 0xFFFFFFFF;
2554                 a3 &= 0xFFFFFFFF;
2555         }
2556
2557         switch (nr) {
2558         case KVM_HC_VAPIC_POLL_IRQ:
2559                 ret = 0;
2560                 break;
2561         case KVM_HC_MMU_OP:
2562                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2563                 break;
2564         default:
2565                 ret = -KVM_ENOSYS;
2566                 break;
2567         }
2568         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2569         kvm_x86_ops->decache_regs(vcpu);
2570         ++vcpu->stat.hypercalls;
2571         return r;
2572 }
2573 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2574
2575 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2576 {
2577         char instruction[3];
2578         int ret = 0;
2579
2580
2581         /*
2582          * Blow out the MMU to ensure that no other VCPU has an active mapping
2583          * to ensure that the updated hypercall appears atomically across all
2584          * VCPUs.
2585          */
2586         kvm_mmu_zap_all(vcpu->kvm);
2587
2588         kvm_x86_ops->cache_regs(vcpu);
2589         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2590         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2591             != X86EMUL_CONTINUE)
2592                 ret = -EFAULT;
2593
2594         return ret;
2595 }
2596
2597 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2598 {
2599         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2600 }
2601
2602 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2603 {
2604         struct descriptor_table dt = { limit, base };
2605
2606         kvm_x86_ops->set_gdt(vcpu, &dt);
2607 }
2608
2609 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2610 {
2611         struct descriptor_table dt = { limit, base };
2612
2613         kvm_x86_ops->set_idt(vcpu, &dt);
2614 }
2615
2616 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2617                    unsigned long *rflags)
2618 {
2619         kvm_lmsw(vcpu, msw);
2620         *rflags = kvm_x86_ops->get_rflags(vcpu);
2621 }
2622
2623 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2624 {
2625         unsigned long value;
2626
2627         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2628         switch (cr) {
2629         case 0:
2630                 value = vcpu->arch.cr0;
2631                 break;
2632         case 2:
2633                 value = vcpu->arch.cr2;
2634                 break;
2635         case 3:
2636                 value = vcpu->arch.cr3;
2637                 break;
2638         case 4:
2639                 value = vcpu->arch.cr4;
2640                 break;
2641         case 8:
2642                 value = kvm_get_cr8(vcpu);
2643                 break;
2644         default:
2645                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2646                 return 0;
2647         }
2648         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2649                     (u32)((u64)value >> 32), handler);
2650
2651         return value;
2652 }
2653
2654 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2655                      unsigned long *rflags)
2656 {
2657         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2658                     (u32)((u64)val >> 32), handler);
2659
2660         switch (cr) {
2661         case 0:
2662                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2663                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2664                 break;
2665         case 2:
2666                 vcpu->arch.cr2 = val;
2667                 break;
2668         case 3:
2669                 kvm_set_cr3(vcpu, val);
2670                 break;
2671         case 4:
2672                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2673                 break;
2674         case 8:
2675                 kvm_set_cr8(vcpu, val & 0xfUL);
2676                 break;
2677         default:
2678                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2679         }
2680 }
2681
2682 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2683 {
2684         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2685         int j, nent = vcpu->arch.cpuid_nent;
2686
2687         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2688         /* when no next entry is found, the current entry[i] is reselected */
2689         for (j = i + 1; j == i; j = (j + 1) % nent) {
2690                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2691                 if (ej->function == e->function) {
2692                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2693                         return j;
2694                 }
2695         }
2696         return 0; /* silence gcc, even though control never reaches here */
2697 }
2698
2699 /* find an entry with matching function, matching index (if needed), and that
2700  * should be read next (if it's stateful) */
2701 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2702         u32 function, u32 index)
2703 {
2704         if (e->function != function)
2705                 return 0;
2706         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2707                 return 0;
2708         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2709                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2710                 return 0;
2711         return 1;
2712 }
2713
2714 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2715 {
2716         int i;
2717         u32 function, index;
2718         struct kvm_cpuid_entry2 *e, *best;
2719
2720         kvm_x86_ops->cache_regs(vcpu);
2721         function = vcpu->arch.regs[VCPU_REGS_RAX];
2722         index = vcpu->arch.regs[VCPU_REGS_RCX];
2723         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2724         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2725         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2726         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2727         best = NULL;
2728         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2729                 e = &vcpu->arch.cpuid_entries[i];
2730                 if (is_matching_cpuid_entry(e, function, index)) {
2731                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2732                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2733                         best = e;
2734                         break;
2735                 }
2736                 /*
2737                  * Both basic or both extended?
2738                  */
2739                 if (((e->function ^ function) & 0x80000000) == 0)
2740                         if (!best || e->function > best->function)
2741                                 best = e;
2742         }
2743         if (best) {
2744                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2745                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2746                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2747                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2748         }
2749         kvm_x86_ops->decache_regs(vcpu);
2750         kvm_x86_ops->skip_emulated_instruction(vcpu);
2751         KVMTRACE_5D(CPUID, vcpu, function,
2752                     (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2753                     (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2754                     (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2755                     (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2756 }
2757 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2758
2759 /*
2760  * Check if userspace requested an interrupt window, and that the
2761  * interrupt window is open.
2762  *
2763  * No need to exit to userspace if we already have an interrupt queued.
2764  */
2765 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2766                                           struct kvm_run *kvm_run)
2767 {
2768         return (!vcpu->arch.irq_summary &&
2769                 kvm_run->request_interrupt_window &&
2770                 vcpu->arch.interrupt_window_open &&
2771                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2772 }
2773
2774 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2775                               struct kvm_run *kvm_run)
2776 {
2777         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2778         kvm_run->cr8 = kvm_get_cr8(vcpu);
2779         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2780         if (irqchip_in_kernel(vcpu->kvm))
2781                 kvm_run->ready_for_interrupt_injection = 1;
2782         else
2783                 kvm_run->ready_for_interrupt_injection =
2784                                         (vcpu->arch.interrupt_window_open &&
2785                                          vcpu->arch.irq_summary == 0);
2786 }
2787
2788 static void vapic_enter(struct kvm_vcpu *vcpu)
2789 {
2790         struct kvm_lapic *apic = vcpu->arch.apic;
2791         struct page *page;
2792
2793         if (!apic || !apic->vapic_addr)
2794                 return;
2795
2796         down_read(&current->mm->mmap_sem);
2797         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2798         up_read(&current->mm->mmap_sem);
2799
2800         vcpu->arch.apic->vapic_page = page;
2801 }
2802
2803 static void vapic_exit(struct kvm_vcpu *vcpu)
2804 {
2805         struct kvm_lapic *apic = vcpu->arch.apic;
2806
2807         if (!apic || !apic->vapic_addr)
2808                 return;
2809
2810         kvm_release_page_dirty(apic->vapic_page);
2811         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2812 }
2813
2814 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2815 {
2816         int r;
2817
2818         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2819                 pr_debug("vcpu %d received sipi with vector # %x\n",
2820                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2821                 kvm_lapic_reset(vcpu);
2822                 r = kvm_x86_ops->vcpu_reset(vcpu);
2823                 if (r)
2824                         return r;
2825                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2826         }
2827
2828         down_read(&vcpu->kvm->slots_lock);
2829         vapic_enter(vcpu);
2830
2831 preempted:
2832         if (vcpu->guest_debug.enabled)
2833                 kvm_x86_ops->guest_debug_pre(vcpu);
2834
2835 again:
2836         if (vcpu->requests)
2837                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2838                         kvm_mmu_unload(vcpu);
2839
2840         r = kvm_mmu_reload(vcpu);
2841         if (unlikely(r))
2842                 goto out;
2843
2844         if (vcpu->requests) {
2845                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2846                         __kvm_migrate_timers(vcpu);
2847                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2848                         kvm_x86_ops->tlb_flush(vcpu);
2849                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2850                                        &vcpu->requests)) {
2851                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2852                         r = 0;
2853                         goto out;
2854                 }
2855                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2856                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2857                         r = 0;
2858                         goto out;
2859                 }
2860         }
2861
2862         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2863         kvm_inject_pending_timer_irqs(vcpu);
2864
2865         preempt_disable();
2866
2867         kvm_x86_ops->prepare_guest_switch(vcpu);
2868         kvm_load_guest_fpu(vcpu);
2869
2870         local_irq_disable();
2871
2872         if (vcpu->requests || need_resched()) {
2873                 local_irq_enable();
2874                 preempt_enable();
2875                 r = 1;
2876                 goto out;
2877         }
2878
2879         if (signal_pending(current)) {
2880                 local_irq_enable();
2881                 preempt_enable();
2882                 r = -EINTR;
2883                 kvm_run->exit_reason = KVM_EXIT_INTR;
2884                 ++vcpu->stat.signal_exits;
2885                 goto out;
2886         }
2887
2888         vcpu->guest_mode = 1;
2889         /*
2890          * Make sure that guest_mode assignment won't happen after
2891          * testing the pending IRQ vector bitmap.
2892          */
2893         smp_wmb();
2894
2895         if (vcpu->arch.exception.pending)
2896                 __queue_exception(vcpu);
2897         else if (irqchip_in_kernel(vcpu->kvm))
2898                 kvm_x86_ops->inject_pending_irq(vcpu);
2899         else
2900                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2901
2902         kvm_lapic_sync_to_vapic(vcpu);
2903
2904         up_read(&vcpu->kvm->slots_lock);
2905
2906         kvm_guest_enter();
2907
2908
2909         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2910         kvm_x86_ops->run(vcpu, kvm_run);
2911
2912         vcpu->guest_mode = 0;
2913         local_irq_enable();
2914
2915         ++vcpu->stat.exits;
2916
2917         /*
2918          * We must have an instruction between local_irq_enable() and
2919          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2920          * the interrupt shadow.  The stat.exits increment will do nicely.
2921          * But we need to prevent reordering, hence this barrier():
2922          */
2923         barrier();
2924
2925         kvm_guest_exit();
2926
2927         preempt_enable();
2928
2929         down_read(&vcpu->kvm->slots_lock);
2930
2931         /*
2932          * Profile KVM exit RIPs:
2933          */
2934         if (unlikely(prof_on == KVM_PROFILING)) {
2935                 kvm_x86_ops->cache_regs(vcpu);
2936                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2937         }
2938
2939         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2940                 vcpu->arch.exception.pending = false;
2941
2942         kvm_lapic_sync_from_vapic(vcpu);
2943
2944         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2945
2946         if (r > 0) {
2947                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2948                         r = -EINTR;
2949                         kvm_run->exit_reason = KVM_EXIT_INTR;
2950                         ++vcpu->stat.request_irq_exits;
2951                         goto out;
2952                 }
2953                 if (!need_resched())
2954                         goto again;
2955         }
2956
2957 out:
2958         up_read(&vcpu->kvm->slots_lock);
2959         if (r > 0) {
2960                 kvm_resched(vcpu);
2961                 down_read(&vcpu->kvm->slots_lock);
2962                 goto preempted;
2963         }
2964
2965         post_kvm_run_save(vcpu, kvm_run);
2966
2967         down_read(&vcpu->kvm->slots_lock);
2968         vapic_exit(vcpu);
2969         up_read(&vcpu->kvm->slots_lock);
2970
2971         return r;
2972 }
2973
2974 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2975 {
2976         int r;
2977         sigset_t sigsaved;
2978
2979         vcpu_load(vcpu);
2980
2981         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2982                 kvm_vcpu_block(vcpu);
2983                 vcpu_put(vcpu);
2984                 return -EAGAIN;
2985         }
2986
2987         if (vcpu->sigset_active)
2988                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2989
2990         /* re-sync apic's tpr */
2991         if (!irqchip_in_kernel(vcpu->kvm))
2992                 kvm_set_cr8(vcpu, kvm_run->cr8);
2993
2994         if (vcpu->arch.pio.cur_count) {
2995                 r = complete_pio(vcpu);
2996                 if (r)
2997                         goto out;
2998         }
2999 #if CONFIG_HAS_IOMEM
3000         if (vcpu->mmio_needed) {
3001                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3002                 vcpu->mmio_read_completed = 1;
3003                 vcpu->mmio_needed = 0;
3004
3005                 down_read(&vcpu->kvm->slots_lock);
3006                 r = emulate_instruction(vcpu, kvm_run,
3007                                         vcpu->arch.mmio_fault_cr2, 0,
3008                                         EMULTYPE_NO_DECODE);
3009                 up_read(&vcpu->kvm->slots_lock);
3010                 if (r == EMULATE_DO_MMIO) {
3011                         /*
3012                          * Read-modify-write.  Back to userspace.
3013                          */
3014                         r = 0;
3015                         goto out;
3016                 }
3017         }
3018 #endif
3019         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
3020                 kvm_x86_ops->cache_regs(vcpu);
3021                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
3022                 kvm_x86_ops->decache_regs(vcpu);
3023         }
3024
3025         r = __vcpu_run(vcpu, kvm_run);
3026
3027 out:
3028         if (vcpu->sigset_active)
3029                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3030
3031         vcpu_put(vcpu);
3032         return r;
3033 }
3034
3035 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3036 {
3037         vcpu_load(vcpu);
3038
3039         kvm_x86_ops->cache_regs(vcpu);
3040
3041         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
3042         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
3043         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
3044         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
3045         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
3046         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
3047         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3048         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3049 #ifdef CONFIG_X86_64
3050         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3051         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3052         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3053         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3054         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3055         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3056         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3057         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3058 #endif
3059
3060         regs->rip = vcpu->arch.rip;
3061         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3062
3063         /*
3064          * Don't leak debug flags in case they were set for guest debugging
3065          */
3066         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3067                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3068
3069         vcpu_put(vcpu);
3070
3071         return 0;
3072 }
3073
3074 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3075 {
3076         vcpu_load(vcpu);
3077
3078         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3079         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3080         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3081         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3082         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3083         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3084         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3085         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3086 #ifdef CONFIG_X86_64
3087         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3088         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3089         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3090         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3091         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3092         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3093         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3094         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3095 #endif
3096
3097         vcpu->arch.rip = regs->rip;
3098         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3099
3100         kvm_x86_ops->decache_regs(vcpu);
3101
3102         vcpu->arch.exception.pending = false;
3103
3104         vcpu_put(vcpu);
3105
3106         return 0;
3107 }
3108
3109 void kvm_get_segment(struct kvm_vcpu *vcpu,
3110                      struct kvm_segment *var, int seg)
3111 {
3112         kvm_x86_ops->get_segment(vcpu, var, seg);
3113 }
3114
3115 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3116 {
3117         struct kvm_segment cs;
3118
3119         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3120         *db = cs.db;
3121         *l = cs.l;
3122 }
3123 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3124
3125 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3126                                   struct kvm_sregs *sregs)
3127 {
3128         struct descriptor_table dt;
3129         int pending_vec;
3130
3131         vcpu_load(vcpu);
3132
3133         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3134         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3135         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3136         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3137         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3138         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3139
3140         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3141         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3142
3143         kvm_x86_ops->get_idt(vcpu, &dt);
3144         sregs->idt.limit = dt.limit;
3145         sregs->idt.base = dt.base;
3146         kvm_x86_ops->get_gdt(vcpu, &dt);
3147         sregs->gdt.limit = dt.limit;
3148         sregs->gdt.base = dt.base;
3149
3150         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3151         sregs->cr0 = vcpu->arch.cr0;
3152         sregs->cr2 = vcpu->arch.cr2;
3153         sregs->cr3 = vcpu->arch.cr3;
3154         sregs->cr4 = vcpu->arch.cr4;
3155         sregs->cr8 = kvm_get_cr8(vcpu);
3156         sregs->efer = vcpu->arch.shadow_efer;
3157         sregs->apic_base = kvm_get_apic_base(vcpu);
3158
3159         if (irqchip_in_kernel(vcpu->kvm)) {
3160                 memset(sregs->interrupt_bitmap, 0,
3161                        sizeof sregs->interrupt_bitmap);
3162                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3163                 if (pending_vec >= 0)
3164                         set_bit(pending_vec,
3165                                 (unsigned long *)sregs->interrupt_bitmap);
3166         } else
3167                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3168                        sizeof sregs->interrupt_bitmap);
3169
3170         vcpu_put(vcpu);
3171
3172         return 0;
3173 }
3174
3175 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3176                                     struct kvm_mp_state *mp_state)
3177 {
3178         vcpu_load(vcpu);
3179         mp_state->mp_state = vcpu->arch.mp_state;
3180         vcpu_put(vcpu);
3181         return 0;
3182 }
3183
3184 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3185                                     struct kvm_mp_state *mp_state)
3186 {
3187         vcpu_load(vcpu);
3188         vcpu->arch.mp_state = mp_state->mp_state;
3189         vcpu_put(vcpu);
3190         return 0;
3191 }
3192
3193 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3194                         struct kvm_segment *var, int seg)
3195 {
3196         kvm_x86_ops->set_segment(vcpu, var, seg);
3197 }
3198
3199 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3200                                    struct kvm_segment *kvm_desct)
3201 {
3202         kvm_desct->base = seg_desc->base0;
3203         kvm_desct->base |= seg_desc->base1 << 16;
3204         kvm_desct->base |= seg_desc->base2 << 24;
3205         kvm_desct->limit = seg_desc->limit0;
3206         kvm_desct->limit |= seg_desc->limit << 16;
3207         kvm_desct->selector = selector;
3208         kvm_desct->type = seg_desc->type;
3209         kvm_desct->present = seg_desc->p;
3210         kvm_desct->dpl = seg_desc->dpl;
3211         kvm_desct->db = seg_desc->d;
3212         kvm_desct->s = seg_desc->s;
3213         kvm_desct->l = seg_desc->l;
3214         kvm_desct->g = seg_desc->g;
3215         kvm_desct->avl = seg_desc->avl;
3216         if (!selector)
3217                 kvm_desct->unusable = 1;
3218         else
3219                 kvm_desct->unusable = 0;
3220         kvm_desct->padding = 0;
3221 }
3222
3223 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3224                                            u16 selector,
3225                                            struct descriptor_table *dtable)
3226 {
3227         if (selector & 1 << 2) {
3228                 struct kvm_segment kvm_seg;
3229
3230                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3231
3232                 if (kvm_seg.unusable)
3233                         dtable->limit = 0;
3234                 else
3235                         dtable->limit = kvm_seg.limit;
3236                 dtable->base = kvm_seg.base;
3237         }
3238         else
3239                 kvm_x86_ops->get_gdt(vcpu, dtable);
3240 }
3241
3242 /* allowed just for 8 bytes segments */
3243 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3244                                          struct desc_struct *seg_desc)
3245 {
3246         struct descriptor_table dtable;
3247         u16 index = selector >> 3;
3248
3249         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3250
3251         if (dtable.limit < index * 8 + 7) {
3252                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3253                 return 1;
3254         }
3255         return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3256 }
3257
3258 /* allowed just for 8 bytes segments */
3259 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3260                                          struct desc_struct *seg_desc)
3261 {
3262         struct descriptor_table dtable;
3263         u16 index = selector >> 3;
3264
3265         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3266
3267         if (dtable.limit < index * 8 + 7)
3268                 return 1;
3269         return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3270 }
3271
3272 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3273                              struct desc_struct *seg_desc)
3274 {
3275         u32 base_addr;
3276
3277         base_addr = seg_desc->base0;
3278         base_addr |= (seg_desc->base1 << 16);
3279         base_addr |= (seg_desc->base2 << 24);
3280
3281         return base_addr;
3282 }
3283
3284 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3285                               struct desc_struct *seg_desc,
3286                               struct tss_segment_32 *tss)
3287 {
3288         u32 base_addr;
3289
3290         base_addr = get_tss_base_addr(vcpu, seg_desc);
3291
3292         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3293                               sizeof(struct tss_segment_32));
3294 }
3295
3296 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3297                               struct desc_struct *seg_desc,
3298                               struct tss_segment_32 *tss)
3299 {
3300         u32 base_addr;
3301
3302         base_addr = get_tss_base_addr(vcpu, seg_desc);
3303
3304         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3305                                sizeof(struct tss_segment_32));
3306 }
3307
3308 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3309                               struct desc_struct *seg_desc,
3310                               struct tss_segment_16 *tss)
3311 {
3312         u32 base_addr;
3313
3314         base_addr = get_tss_base_addr(vcpu, seg_desc);
3315
3316         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3317                               sizeof(struct tss_segment_16));
3318 }
3319
3320 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3321                               struct desc_struct *seg_desc,
3322                               struct tss_segment_16 *tss)
3323 {
3324         u32 base_addr;
3325
3326         base_addr = get_tss_base_addr(vcpu, seg_desc);
3327
3328         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3329                                sizeof(struct tss_segment_16));
3330 }
3331
3332 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3333 {
3334         struct kvm_segment kvm_seg;
3335
3336         kvm_get_segment(vcpu, &kvm_seg, seg);
3337         return kvm_seg.selector;
3338 }
3339
3340 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3341                                                 u16 selector,
3342                                                 struct kvm_segment *kvm_seg)
3343 {
3344         struct desc_struct seg_desc;
3345
3346         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3347                 return 1;
3348         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3349         return 0;
3350 }
3351
3352 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3353                                 int type_bits, int seg)
3354 {
3355         struct kvm_segment kvm_seg;
3356
3357         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3358                 return 1;
3359         kvm_seg.type |= type_bits;
3360
3361         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3362             seg != VCPU_SREG_LDTR)
3363                 if (!kvm_seg.s)
3364                         kvm_seg.unusable = 1;
3365
3366         kvm_set_segment(vcpu, &kvm_seg, seg);
3367         return 0;
3368 }
3369
3370 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3371                                 struct tss_segment_32 *tss)
3372 {
3373         tss->cr3 = vcpu->arch.cr3;
3374         tss->eip = vcpu->arch.rip;
3375         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3376         tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3377         tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3378         tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3379         tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3380         tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3381         tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3382         tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3383         tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3384
3385         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3386         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3387         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3388         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3389         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3390         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3391         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3392         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3393 }
3394
3395 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3396                                   struct tss_segment_32 *tss)
3397 {
3398         kvm_set_cr3(vcpu, tss->cr3);
3399
3400         vcpu->arch.rip = tss->eip;
3401         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3402
3403         vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3404         vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3405         vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3406         vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3407         vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3408         vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3409         vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3410         vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3411
3412         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3413                 return 1;
3414
3415         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3416                 return 1;
3417
3418         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3419                 return 1;
3420
3421         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3422                 return 1;
3423
3424         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3425                 return 1;
3426
3427         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3428                 return 1;
3429
3430         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3431                 return 1;
3432         return 0;
3433 }
3434
3435 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3436                                 struct tss_segment_16 *tss)
3437 {
3438         tss->ip = vcpu->arch.rip;
3439         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3440         tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3441         tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3442         tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3443         tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3444         tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3445         tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3446         tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3447         tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3448
3449         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3450         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3451         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3452         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3453         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3454         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3455 }
3456
3457 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3458                                  struct tss_segment_16 *tss)
3459 {
3460         vcpu->arch.rip = tss->ip;
3461         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3462         vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3463         vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3464         vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3465         vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3466         vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3467         vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3468         vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3469         vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3470
3471         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3472                 return 1;
3473
3474         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3475                 return 1;
3476
3477         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3478                 return 1;
3479
3480         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3481                 return 1;
3482
3483         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3484                 return 1;
3485         return 0;
3486 }
3487
3488 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3489                        struct desc_struct *cseg_desc,
3490                        struct desc_struct *nseg_desc)
3491 {
3492         struct tss_segment_16 tss_segment_16;
3493         int ret = 0;
3494
3495         if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3496                 goto out;
3497
3498         save_state_to_tss16(vcpu, &tss_segment_16);
3499         save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3500
3501         if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3502                 goto out;
3503         if (load_state_from_tss16(vcpu, &tss_segment_16))
3504                 goto out;
3505
3506         ret = 1;
3507 out:
3508         return ret;
3509 }
3510
3511 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3512                        struct desc_struct *cseg_desc,
3513                        struct desc_struct *nseg_desc)
3514 {
3515         struct tss_segment_32 tss_segment_32;
3516         int ret = 0;
3517
3518         if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3519                 goto out;
3520
3521         save_state_to_tss32(vcpu, &tss_segment_32);
3522         save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3523
3524         if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3525                 goto out;
3526         if (load_state_from_tss32(vcpu, &tss_segment_32))
3527                 goto out;
3528
3529         ret = 1;
3530 out:
3531         return ret;
3532 }
3533
3534 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3535 {
3536         struct kvm_segment tr_seg;
3537         struct desc_struct cseg_desc;
3538         struct desc_struct nseg_desc;
3539         int ret = 0;
3540
3541         kvm_get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3542
3543         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3544                 goto out;
3545
3546         if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3547                 goto out;
3548
3549
3550         if (reason != TASK_SWITCH_IRET) {
3551                 int cpl;
3552
3553                 cpl = kvm_x86_ops->get_cpl(vcpu);
3554                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3555                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3556                         return 1;
3557                 }
3558         }
3559
3560         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3561                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3562                 return 1;
3563         }
3564
3565         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3566                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3567                 save_guest_segment_descriptor(vcpu, tr_seg.selector,
3568                                               &cseg_desc);
3569         }
3570
3571         if (reason == TASK_SWITCH_IRET) {
3572                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3573                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3574         }
3575
3576         kvm_x86_ops->skip_emulated_instruction(vcpu);
3577         kvm_x86_ops->cache_regs(vcpu);
3578
3579         if (nseg_desc.type & 8)
3580                 ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3581                                          &nseg_desc);
3582         else
3583                 ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3584                                          &nseg_desc);
3585
3586         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3587                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3588                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3589         }
3590
3591         if (reason != TASK_SWITCH_IRET) {
3592                 nseg_desc.type |= (1 << 1);
3593                 save_guest_segment_descriptor(vcpu, tss_selector,
3594                                               &nseg_desc);
3595         }
3596
3597         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3598         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3599         tr_seg.type = 11;
3600         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3601 out:
3602         kvm_x86_ops->decache_regs(vcpu);
3603         return ret;
3604 }
3605 EXPORT_SYMBOL_GPL(kvm_task_switch);
3606
3607 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3608                                   struct kvm_sregs *sregs)
3609 {
3610         int mmu_reset_needed = 0;
3611         int i, pending_vec, max_bits;
3612         struct descriptor_table dt;
3613
3614         vcpu_load(vcpu);
3615
3616         dt.limit = sregs->idt.limit;
3617         dt.base = sregs->idt.base;
3618         kvm_x86_ops->set_idt(vcpu, &dt);
3619         dt.limit = sregs->gdt.limit;
3620         dt.base = sregs->gdt.base;
3621         kvm_x86_ops->set_gdt(vcpu, &dt);
3622
3623         vcpu->arch.cr2 = sregs->cr2;
3624         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3625         vcpu->arch.cr3 = sregs->cr3;
3626
3627         kvm_set_cr8(vcpu, sregs->cr8);
3628
3629         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3630         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3631         kvm_set_apic_base(vcpu, sregs->apic_base);
3632
3633         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3634
3635         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3636         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3637         vcpu->arch.cr0 = sregs->cr0;
3638
3639         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3640         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3641         if (!is_long_mode(vcpu) && is_pae(vcpu))
3642                 load_pdptrs(vcpu, vcpu->arch.cr3);
3643
3644         if (mmu_reset_needed)
3645                 kvm_mmu_reset_context(vcpu);
3646
3647         if (!irqchip_in_kernel(vcpu->kvm)) {
3648                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3649                        sizeof vcpu->arch.irq_pending);
3650                 vcpu->arch.irq_summary = 0;
3651                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3652                         if (vcpu->arch.irq_pending[i])
3653                                 __set_bit(i, &vcpu->arch.irq_summary);
3654         } else {
3655                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3656                 pending_vec = find_first_bit(
3657                         (const unsigned long *)sregs->interrupt_bitmap,
3658                         max_bits);
3659                 /* Only pending external irq is handled here */
3660                 if (pending_vec < max_bits) {
3661                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3662                         pr_debug("Set back pending irq %d\n",
3663                                  pending_vec);
3664                 }
3665         }
3666
3667         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3668         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3669         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3670         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3671         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3672         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3673
3674         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3675         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3676
3677         vcpu_put(vcpu);
3678
3679         return 0;
3680 }
3681
3682 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3683                                     struct kvm_debug_guest *dbg)
3684 {
3685         int r;
3686
3687         vcpu_load(vcpu);
3688
3689         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3690
3691         vcpu_put(vcpu);
3692
3693         return r;
3694 }
3695
3696 /*
3697  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3698  * we have asm/x86/processor.h
3699  */
3700 struct fxsave {
3701         u16     cwd;
3702         u16     swd;
3703         u16     twd;
3704         u16     fop;
3705         u64     rip;
3706         u64     rdp;
3707         u32     mxcsr;
3708         u32     mxcsr_mask;
3709         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3710 #ifdef CONFIG_X86_64
3711         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3712 #else
3713         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3714 #endif
3715 };
3716
3717 /*
3718  * Translate a guest virtual address to a guest physical address.
3719  */
3720 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3721                                     struct kvm_translation *tr)
3722 {
3723         unsigned long vaddr = tr->linear_address;
3724         gpa_t gpa;
3725
3726         vcpu_load(vcpu);
3727         down_read(&vcpu->kvm->slots_lock);
3728         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3729         up_read(&vcpu->kvm->slots_lock);
3730         tr->physical_address = gpa;
3731         tr->valid = gpa != UNMAPPED_GVA;
3732         tr->writeable = 1;
3733         tr->usermode = 0;
3734         vcpu_put(vcpu);
3735
3736         return 0;
3737 }
3738
3739 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3740 {
3741         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3742
3743         vcpu_load(vcpu);
3744
3745         memcpy(fpu->fpr, fxsave->st_space, 128);
3746         fpu->fcw = fxsave->cwd;
3747         fpu->fsw = fxsave->swd;
3748         fpu->ftwx = fxsave->twd;
3749         fpu->last_opcode = fxsave->fop;
3750         fpu->last_ip = fxsave->rip;
3751         fpu->last_dp = fxsave->rdp;
3752         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3753
3754         vcpu_put(vcpu);
3755
3756         return 0;
3757 }
3758
3759 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3760 {
3761         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3762
3763         vcpu_load(vcpu);
3764
3765         memcpy(fxsave->st_space, fpu->fpr, 128);
3766         fxsave->cwd = fpu->fcw;
3767         fxsave->swd = fpu->fsw;
3768         fxsave->twd = fpu->ftwx;
3769         fxsave->fop = fpu->last_opcode;
3770         fxsave->rip = fpu->last_ip;
3771         fxsave->rdp = fpu->last_dp;
3772         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3773
3774         vcpu_put(vcpu);
3775
3776         return 0;
3777 }
3778
3779 void fx_init(struct kvm_vcpu *vcpu)
3780 {
3781         unsigned after_mxcsr_mask;
3782
3783         /*
3784          * Touch the fpu the first time in non atomic context as if
3785          * this is the first fpu instruction the exception handler
3786          * will fire before the instruction returns and it'll have to
3787          * allocate ram with GFP_KERNEL.
3788          */
3789         if (!used_math())
3790                 fx_save(&vcpu->arch.host_fx_image);
3791
3792         /* Initialize guest FPU by resetting ours and saving into guest's */
3793         preempt_disable();
3794         fx_save(&vcpu->arch.host_fx_image);
3795         fx_finit();
3796         fx_save(&vcpu->arch.guest_fx_image);
3797         fx_restore(&vcpu->arch.host_fx_image);
3798         preempt_enable();
3799
3800         vcpu->arch.cr0 |= X86_CR0_ET;
3801         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3802         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3803         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3804                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3805 }
3806 EXPORT_SYMBOL_GPL(fx_init);
3807
3808 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3809 {
3810         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3811                 return;
3812
3813         vcpu->guest_fpu_loaded = 1;
3814         fx_save(&vcpu->arch.host_fx_image);
3815         fx_restore(&vcpu->arch.guest_fx_image);
3816 }
3817 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3818
3819 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3820 {
3821         if (!vcpu->guest_fpu_loaded)
3822                 return;
3823
3824         vcpu->guest_fpu_loaded = 0;
3825         fx_save(&vcpu->arch.guest_fx_image);
3826         fx_restore(&vcpu->arch.host_fx_image);
3827         ++vcpu->stat.fpu_reload;
3828 }
3829 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3830
3831 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3832 {
3833         kvm_x86_ops->vcpu_free(vcpu);
3834 }
3835
3836 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3837                                                 unsigned int id)
3838 {
3839         return kvm_x86_ops->vcpu_create(kvm, id);
3840 }
3841
3842 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3843 {
3844         int r;
3845
3846         /* We do fxsave: this must be aligned. */
3847         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3848
3849         vcpu_load(vcpu);
3850         r = kvm_arch_vcpu_reset(vcpu);
3851         if (r == 0)
3852                 r = kvm_mmu_setup(vcpu);
3853         vcpu_put(vcpu);
3854         if (r < 0)
3855                 goto free_vcpu;
3856
3857         return 0;
3858 free_vcpu:
3859         kvm_x86_ops->vcpu_free(vcpu);
3860         return r;
3861 }
3862
3863 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3864 {
3865         vcpu_load(vcpu);
3866         kvm_mmu_unload(vcpu);
3867         vcpu_put(vcpu);
3868
3869         kvm_x86_ops->vcpu_free(vcpu);
3870 }
3871
3872 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3873 {
3874         return kvm_x86_ops->vcpu_reset(vcpu);
3875 }
3876
3877 void kvm_arch_hardware_enable(void *garbage)
3878 {
3879         kvm_x86_ops->hardware_enable(garbage);
3880 }
3881
3882 void kvm_arch_hardware_disable(void *garbage)
3883 {
3884         kvm_x86_ops->hardware_disable(garbage);
3885 }
3886
3887 int kvm_arch_hardware_setup(void)
3888 {
3889         return kvm_x86_ops->hardware_setup();
3890 }
3891
3892 void kvm_arch_hardware_unsetup(void)
3893 {
3894         kvm_x86_ops->hardware_unsetup();
3895 }
3896
3897 void kvm_arch_check_processor_compat(void *rtn)
3898 {
3899         kvm_x86_ops->check_processor_compatibility(rtn);
3900 }
3901
3902 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3903 {
3904         struct page *page;
3905         struct kvm *kvm;
3906         int r;
3907
3908         BUG_ON(vcpu->kvm == NULL);
3909         kvm = vcpu->kvm;
3910
3911         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3912         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3913                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3914         else
3915                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3916
3917         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3918         if (!page) {
3919                 r = -ENOMEM;
3920                 goto fail;
3921         }
3922         vcpu->arch.pio_data = page_address(page);
3923
3924         r = kvm_mmu_create(vcpu);
3925         if (r < 0)
3926                 goto fail_free_pio_data;
3927
3928         if (irqchip_in_kernel(kvm)) {
3929                 r = kvm_create_lapic(vcpu);
3930                 if (r < 0)
3931                         goto fail_mmu_destroy;
3932         }
3933
3934         return 0;
3935
3936 fail_mmu_destroy:
3937         kvm_mmu_destroy(vcpu);
3938 fail_free_pio_data:
3939         free_page((unsigned long)vcpu->arch.pio_data);
3940 fail:
3941         return r;
3942 }
3943
3944 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3945 {
3946         kvm_free_lapic(vcpu);
3947         down_read(&vcpu->kvm->slots_lock);
3948         kvm_mmu_destroy(vcpu);
3949         up_read(&vcpu->kvm->slots_lock);
3950         free_page((unsigned long)vcpu->arch.pio_data);
3951 }
3952
3953 struct  kvm *kvm_arch_create_vm(void)
3954 {
3955         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3956
3957         if (!kvm)
3958                 return ERR_PTR(-ENOMEM);
3959
3960         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3961
3962         return kvm;
3963 }
3964
3965 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3966 {
3967         vcpu_load(vcpu);
3968         kvm_mmu_unload(vcpu);
3969         vcpu_put(vcpu);
3970 }
3971
3972 static void kvm_free_vcpus(struct kvm *kvm)
3973 {
3974         unsigned int i;
3975
3976         /*
3977          * Unpin any mmu pages first.
3978          */
3979         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3980                 if (kvm->vcpus[i])
3981                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3982         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3983                 if (kvm->vcpus[i]) {
3984                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3985                         kvm->vcpus[i] = NULL;
3986                 }
3987         }
3988
3989 }
3990
3991 void kvm_arch_destroy_vm(struct kvm *kvm)
3992 {
3993         kvm_free_pit(kvm);
3994         kfree(kvm->arch.vpic);
3995         kfree(kvm->arch.vioapic);
3996         kvm_free_vcpus(kvm);
3997         kvm_free_physmem(kvm);
3998         if (kvm->arch.apic_access_page)
3999                 put_page(kvm->arch.apic_access_page);
4000         if (kvm->arch.ept_identity_pagetable)
4001                 put_page(kvm->arch.ept_identity_pagetable);
4002         kfree(kvm);
4003 }
4004
4005 int kvm_arch_set_memory_region(struct kvm *kvm,
4006                                 struct kvm_userspace_memory_region *mem,
4007                                 struct kvm_memory_slot old,
4008                                 int user_alloc)
4009 {
4010         int npages = mem->memory_size >> PAGE_SHIFT;
4011         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4012
4013         /*To keep backward compatibility with older userspace,
4014          *x86 needs to hanlde !user_alloc case.
4015          */
4016         if (!user_alloc) {
4017                 if (npages && !old.rmap) {
4018                         down_write(&current->mm->mmap_sem);
4019                         memslot->userspace_addr = do_mmap(NULL, 0,
4020                                                      npages * PAGE_SIZE,
4021                                                      PROT_READ | PROT_WRITE,
4022                                                      MAP_SHARED | MAP_ANONYMOUS,
4023                                                      0);
4024                         up_write(&current->mm->mmap_sem);
4025
4026                         if (IS_ERR((void *)memslot->userspace_addr))
4027                                 return PTR_ERR((void *)memslot->userspace_addr);
4028                 } else {
4029                         if (!old.user_alloc && old.rmap) {
4030                                 int ret;
4031
4032                                 down_write(&current->mm->mmap_sem);
4033                                 ret = do_munmap(current->mm, old.userspace_addr,
4034                                                 old.npages * PAGE_SIZE);
4035                                 up_write(&current->mm->mmap_sem);
4036                                 if (ret < 0)
4037                                         printk(KERN_WARNING
4038                                        "kvm_vm_ioctl_set_memory_region: "
4039                                        "failed to munmap memory\n");
4040                         }
4041                 }
4042         }
4043
4044         if (!kvm->arch.n_requested_mmu_pages) {
4045                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4046                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4047         }
4048
4049         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4050         kvm_flush_remote_tlbs(kvm);
4051
4052         return 0;
4053 }
4054
4055 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4056 {
4057         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4058                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4059 }
4060
4061 static void vcpu_kick_intr(void *info)
4062 {
4063 #ifdef DEBUG
4064         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4065         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4066 #endif
4067 }
4068
4069 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4070 {
4071         int ipi_pcpu = vcpu->cpu;
4072         int cpu = get_cpu();
4073
4074         if (waitqueue_active(&vcpu->wq)) {
4075                 wake_up_interruptible(&vcpu->wq);
4076                 ++vcpu->stat.halt_wakeup;
4077         }
4078         /*
4079          * We may be called synchronously with irqs disabled in guest mode,
4080          * So need not to call smp_call_function_single() in that case.
4081          */
4082         if (vcpu->guest_mode && vcpu->cpu != cpu)
4083                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4084         put_cpu();
4085 }