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         case KVM_CAP_SYNC_MMU:
 887                 r = 1;
 888                 break;
 889         case KVM_CAP_COALESCED_MMIO:
 890                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 891                 break;
 892         case KVM_CAP_VAPIC:
 893                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 894                 break;
 895         case KVM_CAP_NR_VCPUS:
 896                 r = KVM_MAX_VCPUS;
 897                 break;
 898         case KVM_CAP_NR_MEMSLOTS:
 899                 r = KVM_MEMORY_SLOTS;
 900                 break;
 901         case KVM_CAP_PV_MMU:
 902                 r = !tdp_enabled;
 903                 break;
 904         default:
 905                 r = 0;
 906                 break;
 907         }
 908         return r;
 909
 910 }
 911
 912 long kvm_arch_dev_ioctl(struct file *filp,
 913                         unsigned int ioctl, unsigned long arg)
 914 {
 915         void __user *argp = (void __user *)arg;
 916         long r;
 917
 918         switch (ioctl) {
 919         case KVM_GET_MSR_INDEX_LIST: {
 920                 struct kvm_msr_list __user *user_msr_list = argp;
 921                 struct kvm_msr_list msr_list;
 922                 unsigned n;
 923
 924                 r = -EFAULT;
 925                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 926                         goto out;
 927                 n = msr_list.nmsrs;
 928                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 929                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 930                         goto out;
 931                 r = -E2BIG;
 932                 if (n < num_msrs_to_save)
 933                         goto out;
 934                 r = -EFAULT;
 935                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 936                                  num_msrs_to_save * sizeof(u32)))
 937                         goto out;
 938                 if (copy_to_user(user_msr_list->indices
 939                                  + num_msrs_to_save * sizeof(u32),
 940                                  &emulated_msrs,
 941                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 942                         goto out;
 943                 r = 0;
 944                 break;
 945         }
 946         case KVM_GET_SUPPORTED_CPUID: {
 947                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 948                 struct kvm_cpuid2 cpuid;
 949
 950                 r = -EFAULT;
 951                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 952                         goto out;
 953                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 954                         cpuid_arg->entries);
 955                 if (r)
 956                         goto out;
 957
 958                 r = -EFAULT;
 959                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 960                         goto out;
 961                 r = 0;
 962                 break;
 963         }
 964         default:
 965                 r = -EINVAL;
 966         }
 967 out:
 968         return r;
 969 }
 970
 971 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 972 {
 973         kvm_x86_ops->vcpu_load(vcpu, cpu);
 974         kvm_write_guest_time(vcpu);
 975 }
 976
 977 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 978 {
 979         kvm_x86_ops->vcpu_put(vcpu);
 980         kvm_put_guest_fpu(vcpu);
 981 }
 982
 983 static int is_efer_nx(void)
 984 {
 985         u64 efer;
 986
 987         rdmsrl(MSR_EFER, efer);
 988         return efer & EFER_NX;
 989 }
 990
 991 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 992 {
 993         int i;
 994         struct kvm_cpuid_entry2 *e, *entry;
 995
 996         entry = NULL;
 997         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 998                 e = &vcpu->arch.cpuid_entries[i];
 999                 if (e->function == 0x80000001) {
1000                         entry = e;
1001                         break;
1002                 }
1003         }
1004         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1005                 entry->edx &= ~(1 << 20);
1006                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1007         }
1008 }
1009
1010 /* when an old userspace process fills a new kernel module */
1011 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1012                                     struct kvm_cpuid *cpuid,
1013                                     struct kvm_cpuid_entry __user *entries)
1014 {
1015         int r, i;
1016         struct kvm_cpuid_entry *cpuid_entries;
1017
1018         r = -E2BIG;
1019         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1020                 goto out;
1021         r = -ENOMEM;
1022         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1023         if (!cpuid_entries)
1024                 goto out;
1025         r = -EFAULT;
1026         if (copy_from_user(cpuid_entries, entries,
1027                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1028                 goto out_free;
1029         for (i = 0; i < cpuid->nent; i++) {
1030                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1031                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1032                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1033                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1034                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1035                 vcpu->arch.cpuid_entries[i].index = 0;
1036                 vcpu->arch.cpuid_entries[i].flags = 0;
1037                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1038                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1039                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1040         }
1041         vcpu->arch.cpuid_nent = cpuid->nent;
1042         cpuid_fix_nx_cap(vcpu);
1043         r = 0;
1044
1045 out_free:
1046         vfree(cpuid_entries);
1047 out:
1048         return r;
1049 }
1050
1051 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1052                                     struct kvm_cpuid2 *cpuid,
1053                                     struct kvm_cpuid_entry2 __user *entries)
1054 {
1055         int r;
1056
1057         r = -E2BIG;
1058         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1059                 goto out;
1060         r = -EFAULT;
1061         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1062                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1063                 goto out;
1064         vcpu->arch.cpuid_nent = cpuid->nent;
1065         return 0;
1066
1067 out:
1068         return r;
1069 }
1070
1071 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1072                                     struct kvm_cpuid2 *cpuid,
1073                                     struct kvm_cpuid_entry2 __user *entries)
1074 {
1075         int r;
1076
1077         r = -E2BIG;
1078         if (cpuid->nent < vcpu->arch.cpuid_nent)
1079                 goto out;
1080         r = -EFAULT;
1081         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1082                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1083                 goto out;
1084         return 0;
1085
1086 out:
1087         cpuid->nent = vcpu->arch.cpuid_nent;
1088         return r;
1089 }
1090
1091 static inline u32 bit(int bitno)
1092 {
1093         return 1 << (bitno & 31);
1094 }
1095
1096 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1097                           u32 index)
1098 {
1099         entry->function = function;
1100         entry->index = index;
1101         cpuid_count(entry->function, entry->index,
1102                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1103         entry->flags = 0;
1104 }
1105
1106 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1107                          u32 index, int *nent, int maxnent)
1108 {
1109         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1110                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1111                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1112                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1113                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1114                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1115                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1116                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1117                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1118                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1119         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1120                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1121                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1122                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1123                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1124                 bit(X86_FEATURE_PGE) |
1125                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1126                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1127                 bit(X86_FEATURE_SYSCALL) |
1128                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1129 #ifdef CONFIG_X86_64
1130                 bit(X86_FEATURE_LM) |
1131 #endif
1132                 bit(X86_FEATURE_MMXEXT) |
1133                 bit(X86_FEATURE_3DNOWEXT) |
1134                 bit(X86_FEATURE_3DNOW);
1135         const u32 kvm_supported_word3_x86_features =
1136                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1137         const u32 kvm_supported_word6_x86_features =
1138                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1139
1140         /* all func 2 cpuid_count() should be called on the same cpu */
1141         get_cpu();
1142         do_cpuid_1_ent(entry, function, index);
1143         ++*nent;
1144
1145         switch (function) {
1146         case 0:
1147                 entry->eax = min(entry->eax, (u32)0xb);
1148                 break;
1149         case 1:
1150                 entry->edx &= kvm_supported_word0_x86_features;
1151                 entry->ecx &= kvm_supported_word3_x86_features;
1152                 break;
1153         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1154          * may return different values. This forces us to get_cpu() before
1155          * issuing the first command, and also to emulate this annoying behavior
1156          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1157         case 2: {
1158                 int t, times = entry->eax & 0xff;
1159
1160                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1161                 for (t = 1; t < times && *nent < maxnent; ++t) {
1162                         do_cpuid_1_ent(&entry[t], function, 0);
1163                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1164                         ++*nent;
1165                 }
1166                 break;
1167         }
1168         /* function 4 and 0xb have additional index. */
1169         case 4: {
1170                 int i, cache_type;
1171
1172                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1173                 /* read more entries until cache_type is zero */
1174                 for (i = 1; *nent < maxnent; ++i) {
1175                         cache_type = entry[i - 1].eax & 0x1f;
1176                         if (!cache_type)
1177                                 break;
1178                         do_cpuid_1_ent(&entry[i], function, i);
1179                         entry[i].flags |=
1180                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1181                         ++*nent;
1182                 }
1183                 break;
1184         }
1185         case 0xb: {
1186                 int i, level_type;
1187
1188                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1189                 /* read more entries until level_type is zero */
1190                 for (i = 1; *nent < maxnent; ++i) {
1191                         level_type = entry[i - 1].ecx & 0xff;
1192                         if (!level_type)
1193                                 break;
1194                         do_cpuid_1_ent(&entry[i], function, i);
1195                         entry[i].flags |=
1196                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1197                         ++*nent;
1198                 }
1199                 break;
1200         }
1201         case 0x80000000:
1202                 entry->eax = min(entry->eax, 0x8000001a);
1203                 break;
1204         case 0x80000001:
1205                 entry->edx &= kvm_supported_word1_x86_features;
1206                 entry->ecx &= kvm_supported_word6_x86_features;
1207                 break;
1208         }
1209         put_cpu();
1210 }
1211
1212 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1213                                     struct kvm_cpuid_entry2 __user *entries)
1214 {
1215         struct kvm_cpuid_entry2 *cpuid_entries;
1216         int limit, nent = 0, r = -E2BIG;
1217         u32 func;
1218
1219         if (cpuid->nent < 1)
1220                 goto out;
1221         r = -ENOMEM;
1222         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1223         if (!cpuid_entries)
1224                 goto out;
1225
1226         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1227         limit = cpuid_entries[0].eax;
1228         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1229                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1230                                 &nent, cpuid->nent);
1231         r = -E2BIG;
1232         if (nent >= cpuid->nent)
1233                 goto out_free;
1234
1235         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1236         limit = cpuid_entries[nent - 1].eax;
1237         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1238                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1239                                &nent, cpuid->nent);
1240         r = -EFAULT;
1241         if (copy_to_user(entries, cpuid_entries,
1242                         nent * sizeof(struct kvm_cpuid_entry2)))
1243                 goto out_free;
1244         cpuid->nent = nent;
1245         r = 0;
1246
1247 out_free:
1248         vfree(cpuid_entries);
1249 out:
1250         return r;
1251 }
1252
1253 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1254                                     struct kvm_lapic_state *s)
1255 {
1256         vcpu_load(vcpu);
1257         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1258         vcpu_put(vcpu);
1259
1260         return 0;
1261 }
1262
1263 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1264                                     struct kvm_lapic_state *s)
1265 {
1266         vcpu_load(vcpu);
1267         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1268         kvm_apic_post_state_restore(vcpu);
1269         vcpu_put(vcpu);
1270
1271         return 0;
1272 }
1273
1274 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1275                                     struct kvm_interrupt *irq)
1276 {
1277         if (irq->irq < 0 || irq->irq >= 256)
1278                 return -EINVAL;
1279         if (irqchip_in_kernel(vcpu->kvm))
1280                 return -ENXIO;
1281         vcpu_load(vcpu);
1282
1283         set_bit(irq->irq, vcpu->arch.irq_pending);
1284         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1285
1286         vcpu_put(vcpu);
1287
1288         return 0;
1289 }
1290
1291 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1292                                            struct kvm_tpr_access_ctl *tac)
1293 {
1294         if (tac->flags)
1295                 return -EINVAL;
1296         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1297         return 0;
1298 }
1299
1300 long kvm_arch_vcpu_ioctl(struct file *filp,
1301                          unsigned int ioctl, unsigned long arg)
1302 {
1303         struct kvm_vcpu *vcpu = filp->private_data;
1304         void __user *argp = (void __user *)arg;
1305         int r;
1306         struct kvm_lapic_state *lapic = NULL;
1307
1308         switch (ioctl) {
1309         case KVM_GET_LAPIC: {
1310                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1311
1312                 r = -ENOMEM;
1313                 if (!lapic)
1314                         goto out;
1315                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1316                 if (r)
1317                         goto out;
1318                 r = -EFAULT;
1319                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1320                         goto out;
1321                 r = 0;
1322                 break;
1323         }
1324         case KVM_SET_LAPIC: {
1325                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1326                 r = -ENOMEM;
1327                 if (!lapic)
1328                         goto out;
1329                 r = -EFAULT;
1330                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1331                         goto out;
1332                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1333                 if (r)
1334                         goto out;
1335                 r = 0;
1336                 break;
1337         }
1338         case KVM_INTERRUPT: {
1339                 struct kvm_interrupt irq;
1340
1341                 r = -EFAULT;
1342                 if (copy_from_user(&irq, argp, sizeof irq))
1343                         goto out;
1344                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1345                 if (r)
1346                         goto out;
1347                 r = 0;
1348                 break;
1349         }
1350         case KVM_SET_CPUID: {
1351                 struct kvm_cpuid __user *cpuid_arg = argp;
1352                 struct kvm_cpuid cpuid;
1353
1354                 r = -EFAULT;
1355                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1356                         goto out;
1357                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1358                 if (r)
1359                         goto out;
1360                 break;
1361         }
1362         case KVM_SET_CPUID2: {
1363                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1364                 struct kvm_cpuid2 cpuid;
1365
1366                 r = -EFAULT;
1367                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1368                         goto out;
1369                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1370                                 cpuid_arg->entries);
1371                 if (r)
1372                         goto out;
1373                 break;
1374         }
1375         case KVM_GET_CPUID2: {
1376                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1377                 struct kvm_cpuid2 cpuid;
1378
1379                 r = -EFAULT;
1380                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1381                         goto out;
1382                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1383                                 cpuid_arg->entries);
1384                 if (r)
1385                         goto out;
1386                 r = -EFAULT;
1387                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1388                         goto out;
1389                 r = 0;
1390                 break;
1391         }
1392         case KVM_GET_MSRS:
1393                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1394                 break;
1395         case KVM_SET_MSRS:
1396                 r = msr_io(vcpu, argp, do_set_msr, 0);
1397                 break;
1398         case KVM_TPR_ACCESS_REPORTING: {
1399                 struct kvm_tpr_access_ctl tac;
1400
1401                 r = -EFAULT;
1402                 if (copy_from_user(&tac, argp, sizeof tac))
1403                         goto out;
1404                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1405                 if (r)
1406                         goto out;
1407                 r = -EFAULT;
1408                 if (copy_to_user(argp, &tac, sizeof tac))
1409                         goto out;
1410                 r = 0;
1411                 break;
1412         };
1413         case KVM_SET_VAPIC_ADDR: {
1414                 struct kvm_vapic_addr va;
1415
1416                 r = -EINVAL;
1417                 if (!irqchip_in_kernel(vcpu->kvm))
1418                         goto out;
1419                 r = -EFAULT;
1420                 if (copy_from_user(&va, argp, sizeof va))
1421                         goto out;
1422                 r = 0;
1423                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1424                 break;
1425         }
1426         default:
1427                 r = -EINVAL;
1428         }
1429 out:
1430         if (lapic)
1431                 kfree(lapic);
1432         return r;
1433 }
1434
1435 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1436 {
1437         int ret;
1438
1439         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1440                 return -1;
1441         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1442         return ret;
1443 }
1444
1445 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1446                                           u32 kvm_nr_mmu_pages)
1447 {
1448         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1449                 return -EINVAL;
1450
1451         down_write(&kvm->slots_lock);
1452
1453         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1454         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1455
1456         up_write(&kvm->slots_lock);
1457         return 0;
1458 }
1459
1460 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1461 {
1462         return kvm->arch.n_alloc_mmu_pages;
1463 }
1464
1465 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1466 {
1467         int i;
1468         struct kvm_mem_alias *alias;
1469
1470         for (i = 0; i < kvm->arch.naliases; ++i) {
1471                 alias = &kvm->arch.aliases[i];
1472                 if (gfn >= alias->base_gfn
1473                     && gfn < alias->base_gfn + alias->npages)
1474                         return alias->target_gfn + gfn - alias->base_gfn;
1475         }
1476         return gfn;
1477 }
1478
1479 /*
1480  * Set a new alias region.  Aliases map a portion of physical memory into
1481  * another portion.  This is useful for memory windows, for example the PC
1482  * VGA region.
1483  */
1484 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1485                                          struct kvm_memory_alias *alias)
1486 {
1487         int r, n;
1488         struct kvm_mem_alias *p;
1489
1490         r = -EINVAL;
1491         /* General sanity checks */
1492         if (alias->memory_size & (PAGE_SIZE - 1))
1493                 goto out;
1494         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1495                 goto out;
1496         if (alias->slot >= KVM_ALIAS_SLOTS)
1497                 goto out;
1498         if (alias->guest_phys_addr + alias->memory_size
1499             < alias->guest_phys_addr)
1500                 goto out;
1501         if (alias->target_phys_addr + alias->memory_size
1502             < alias->target_phys_addr)
1503                 goto out;
1504
1505         down_write(&kvm->slots_lock);
1506         spin_lock(&kvm->mmu_lock);
1507
1508         p = &kvm->arch.aliases[alias->slot];
1509         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1510         p->npages = alias->memory_size >> PAGE_SHIFT;
1511         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1512
1513         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1514                 if (kvm->arch.aliases[n - 1].npages)
1515                         break;
1516         kvm->arch.naliases = n;
1517
1518         spin_unlock(&kvm->mmu_lock);
1519         kvm_mmu_zap_all(kvm);
1520
1521         up_write(&kvm->slots_lock);
1522
1523         return 0;
1524
1525 out:
1526         return r;
1527 }
1528
1529 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1530 {
1531         int r;
1532
1533         r = 0;
1534         switch (chip->chip_id) {
1535         case KVM_IRQCHIP_PIC_MASTER:
1536                 memcpy(&chip->chip.pic,
1537                         &pic_irqchip(kvm)->pics[0],
1538                         sizeof(struct kvm_pic_state));
1539                 break;
1540         case KVM_IRQCHIP_PIC_SLAVE:
1541                 memcpy(&chip->chip.pic,
1542                         &pic_irqchip(kvm)->pics[1],
1543                         sizeof(struct kvm_pic_state));
1544                 break;
1545         case KVM_IRQCHIP_IOAPIC:
1546                 memcpy(&chip->chip.ioapic,
1547                         ioapic_irqchip(kvm),
1548                         sizeof(struct kvm_ioapic_state));
1549                 break;
1550         default:
1551                 r = -EINVAL;
1552                 break;
1553         }
1554         return r;
1555 }
1556
1557 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1558 {
1559         int r;
1560
1561         r = 0;
1562         switch (chip->chip_id) {
1563         case KVM_IRQCHIP_PIC_MASTER:
1564                 memcpy(&pic_irqchip(kvm)->pics[0],
1565                         &chip->chip.pic,
1566                         sizeof(struct kvm_pic_state));
1567                 break;
1568         case KVM_IRQCHIP_PIC_SLAVE:
1569                 memcpy(&pic_irqchip(kvm)->pics[1],
1570                         &chip->chip.pic,
1571                         sizeof(struct kvm_pic_state));
1572                 break;
1573         case KVM_IRQCHIP_IOAPIC:
1574                 memcpy(ioapic_irqchip(kvm),
1575                         &chip->chip.ioapic,
1576                         sizeof(struct kvm_ioapic_state));
1577                 break;
1578         default:
1579                 r = -EINVAL;
1580                 break;
1581         }
1582         kvm_pic_update_irq(pic_irqchip(kvm));
1583         return r;
1584 }
1585
1586 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1587 {
1588         int r = 0;
1589
1590         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1591         return r;
1592 }
1593
1594 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1595 {
1596         int r = 0;
1597
1598         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1599         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1600         return r;
1601 }
1602
1603 /*
1604  * Get (and clear) the dirty memory log for a memory slot.
1605  */
1606 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1607                                       struct kvm_dirty_log *log)
1608 {
1609         int r;
1610         int n;
1611         struct kvm_memory_slot *memslot;
1612         int is_dirty = 0;
1613
1614         down_write(&kvm->slots_lock);
1615
1616         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1617         if (r)
1618                 goto out;
1619
1620         /* If nothing is dirty, don't bother messing with page tables. */
1621         if (is_dirty) {
1622                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1623                 kvm_flush_remote_tlbs(kvm);
1624                 memslot = &kvm->memslots[log->slot];
1625                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1626                 memset(memslot->dirty_bitmap, 0, n);
1627         }
1628         r = 0;
1629 out:
1630         up_write(&kvm->slots_lock);
1631         return r;
1632 }
1633
1634 long kvm_arch_vm_ioctl(struct file *filp,
1635                        unsigned int ioctl, unsigned long arg)
1636 {
1637         struct kvm *kvm = filp->private_data;
1638         void __user *argp = (void __user *)arg;
1639         int r = -EINVAL;
1640         /*
1641          * This union makes it completely explicit to gcc-3.x
1642          * that these two variables' stack usage should be
1643          * combined, not added together.
1644          */
1645         union {
1646                 struct kvm_pit_state ps;
1647                 struct kvm_memory_alias alias;
1648         } u;
1649
1650         switch (ioctl) {
1651         case KVM_SET_TSS_ADDR:
1652                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1653                 if (r < 0)
1654                         goto out;
1655                 break;
1656         case KVM_SET_MEMORY_REGION: {
1657                 struct kvm_memory_region kvm_mem;
1658                 struct kvm_userspace_memory_region kvm_userspace_mem;
1659
1660                 r = -EFAULT;
1661                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1662                         goto out;
1663                 kvm_userspace_mem.slot = kvm_mem.slot;
1664                 kvm_userspace_mem.flags = kvm_mem.flags;
1665                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1666                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1667                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1668                 if (r)
1669                         goto out;
1670                 break;
1671         }
1672         case KVM_SET_NR_MMU_PAGES:
1673                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1674                 if (r)
1675                         goto out;
1676                 break;
1677         case KVM_GET_NR_MMU_PAGES:
1678                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1679                 break;
1680         case KVM_SET_MEMORY_ALIAS:
1681                 r = -EFAULT;
1682                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1683                         goto out;
1684                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1685                 if (r)
1686                         goto out;
1687                 break;
1688         case KVM_CREATE_IRQCHIP:
1689                 r = -ENOMEM;
1690                 kvm->arch.vpic = kvm_create_pic(kvm);
1691                 if (kvm->arch.vpic) {
1692                         r = kvm_ioapic_init(kvm);
1693                         if (r) {
1694                                 kfree(kvm->arch.vpic);
1695                                 kvm->arch.vpic = NULL;
1696                                 goto out;
1697                         }
1698                 } else
1699                         goto out;
1700                 break;
1701         case KVM_CREATE_PIT:
1702                 r = -ENOMEM;
1703                 kvm->arch.vpit = kvm_create_pit(kvm);
1704                 if (kvm->arch.vpit)
1705                         r = 0;
1706                 break;
1707         case KVM_IRQ_LINE: {
1708                 struct kvm_irq_level irq_event;
1709
1710                 r = -EFAULT;
1711                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1712                         goto out;
1713                 if (irqchip_in_kernel(kvm)) {
1714                         mutex_lock(&kvm->lock);
1715                         if (irq_event.irq < 16)
1716                                 kvm_pic_set_irq(pic_irqchip(kvm),
1717                                         irq_event.irq,
1718                                         irq_event.level);
1719                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1720                                         irq_event.irq,
1721                                         irq_event.level);
1722                         mutex_unlock(&kvm->lock);
1723                         r = 0;
1724                 }
1725                 break;
1726         }
1727         case KVM_GET_IRQCHIP: {
1728                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1729                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1730
1731                 r = -ENOMEM;
1732                 if (!chip)
1733                         goto out;
1734                 r = -EFAULT;
1735                 if (copy_from_user(chip, argp, sizeof *chip))
1736                         goto get_irqchip_out;
1737                 r = -ENXIO;
1738                 if (!irqchip_in_kernel(kvm))
1739                         goto get_irqchip_out;
1740                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1741                 if (r)
1742                         goto get_irqchip_out;
1743                 r = -EFAULT;
1744                 if (copy_to_user(argp, chip, sizeof *chip))
1745                         goto get_irqchip_out;
1746                 r = 0;
1747         get_irqchip_out:
1748                 kfree(chip);
1749                 if (r)
1750                         goto out;
1751                 break;
1752         }
1753         case KVM_SET_IRQCHIP: {
1754                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1755                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1756
1757                 r = -ENOMEM;
1758                 if (!chip)
1759                         goto out;
1760                 r = -EFAULT;
1761                 if (copy_from_user(chip, argp, sizeof *chip))
1762                         goto set_irqchip_out;
1763                 r = -ENXIO;
1764                 if (!irqchip_in_kernel(kvm))
1765                         goto set_irqchip_out;
1766                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1767                 if (r)
1768                         goto set_irqchip_out;
1769                 r = 0;
1770         set_irqchip_out:
1771                 kfree(chip);
1772                 if (r)
1773                         goto out;
1774                 break;
1775         }
1776         case KVM_GET_PIT: {
1777                 r = -EFAULT;
1778                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1779                         goto out;
1780                 r = -ENXIO;
1781                 if (!kvm->arch.vpit)
1782                         goto out;
1783                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1784                 if (r)
1785                         goto out;
1786                 r = -EFAULT;
1787                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1788                         goto out;
1789                 r = 0;
1790                 break;
1791         }
1792         case KVM_SET_PIT: {
1793                 r = -EFAULT;
1794                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1795                         goto out;
1796                 r = -ENXIO;
1797                 if (!kvm->arch.vpit)
1798                         goto out;
1799                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1800                 if (r)
1801                         goto out;
1802                 r = 0;
1803                 break;
1804         }
1805         default:
1806                 ;
1807         }
1808 out:
1809         return r;
1810 }
1811
1812 static void kvm_init_msr_list(void)
1813 {
1814         u32 dummy[2];
1815         unsigned i, j;
1816
1817         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1818                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1819                         continue;
1820                 if (j < i)
1821                         msrs_to_save[j] = msrs_to_save[i];
1822                 j++;
1823         }
1824         num_msrs_to_save = j;
1825 }
1826
1827 /*
1828  * Only apic need an MMIO device hook, so shortcut now..
1829  */
1830 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1831                                                 gpa_t addr, int len,
1832                                                 int is_write)
1833 {
1834         struct kvm_io_device *dev;
1835
1836         if (vcpu->arch.apic) {
1837                 dev = &vcpu->arch.apic->dev;
1838                 if (dev->in_range(dev, addr, len, is_write))
1839                         return dev;
1840         }
1841         return NULL;
1842 }
1843
1844
1845 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1846                                                 gpa_t addr, int len,
1847                                                 int is_write)
1848 {
1849         struct kvm_io_device *dev;
1850
1851         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1852         if (dev == NULL)
1853                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1854                                           is_write);
1855         return dev;
1856 }
1857
1858 int emulator_read_std(unsigned long addr,
1859                              void *val,
1860                              unsigned int bytes,
1861                              struct kvm_vcpu *vcpu)
1862 {
1863         void *data = val;
1864         int r = X86EMUL_CONTINUE;
1865
1866         while (bytes) {
1867                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1868                 unsigned offset = addr & (PAGE_SIZE-1);
1869                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1870                 int ret;
1871
1872                 if (gpa == UNMAPPED_GVA) {
1873                         r = X86EMUL_PROPAGATE_FAULT;
1874                         goto out;
1875                 }
1876                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1877                 if (ret < 0) {
1878                         r = X86EMUL_UNHANDLEABLE;
1879                         goto out;
1880                 }
1881
1882                 bytes -= tocopy;
1883                 data += tocopy;
1884                 addr += tocopy;
1885         }
1886 out:
1887         return r;
1888 }
1889 EXPORT_SYMBOL_GPL(emulator_read_std);
1890
1891 static int emulator_read_emulated(unsigned long addr,
1892                                   void *val,
1893                                   unsigned int bytes,
1894                                   struct kvm_vcpu *vcpu)
1895 {
1896         struct kvm_io_device *mmio_dev;
1897         gpa_t                 gpa;
1898
1899         if (vcpu->mmio_read_completed) {
1900                 memcpy(val, vcpu->mmio_data, bytes);
1901                 vcpu->mmio_read_completed = 0;
1902                 return X86EMUL_CONTINUE;
1903         }
1904
1905         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1906
1907         /* For APIC access vmexit */
1908         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1909                 goto mmio;
1910
1911         if (emulator_read_std(addr, val, bytes, vcpu)
1912                         == X86EMUL_CONTINUE)
1913                 return X86EMUL_CONTINUE;
1914         if (gpa == UNMAPPED_GVA)
1915                 return X86EMUL_PROPAGATE_FAULT;
1916
1917 mmio:
1918         /*
1919          * Is this MMIO handled locally?
1920          */
1921         mutex_lock(&vcpu->kvm->lock);
1922         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1923         if (mmio_dev) {
1924                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1925                 mutex_unlock(&vcpu->kvm->lock);
1926                 return X86EMUL_CONTINUE;
1927         }
1928         mutex_unlock(&vcpu->kvm->lock);
1929
1930         vcpu->mmio_needed = 1;
1931         vcpu->mmio_phys_addr = gpa;
1932         vcpu->mmio_size = bytes;
1933         vcpu->mmio_is_write = 0;
1934
1935         return X86EMUL_UNHANDLEABLE;
1936 }
1937
1938 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1939                           const void *val, int bytes)
1940 {
1941         int ret;
1942
1943         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1944         if (ret < 0)
1945                 return 0;
1946         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1947         return 1;
1948 }
1949
1950 static int emulator_write_emulated_onepage(unsigned long addr,
1951                                            const void *val,
1952                                            unsigned int bytes,
1953                                            struct kvm_vcpu *vcpu)
1954 {
1955         struct kvm_io_device *mmio_dev;
1956         gpa_t                 gpa;
1957
1958         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1959
1960         if (gpa == UNMAPPED_GVA) {
1961                 kvm_inject_page_fault(vcpu, addr, 2);
1962                 return X86EMUL_PROPAGATE_FAULT;
1963         }
1964
1965         /* For APIC access vmexit */
1966         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1967                 goto mmio;
1968
1969         if (emulator_write_phys(vcpu, gpa, val, bytes))
1970                 return X86EMUL_CONTINUE;
1971
1972 mmio:
1973         /*
1974          * Is this MMIO handled locally?
1975          */
1976         mutex_lock(&vcpu->kvm->lock);
1977         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
1978         if (mmio_dev) {
1979                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1980                 mutex_unlock(&vcpu->kvm->lock);
1981                 return X86EMUL_CONTINUE;
1982         }
1983         mutex_unlock(&vcpu->kvm->lock);
1984
1985         vcpu->mmio_needed = 1;
1986         vcpu->mmio_phys_addr = gpa;
1987         vcpu->mmio_size = bytes;
1988         vcpu->mmio_is_write = 1;
1989         memcpy(vcpu->mmio_data, val, bytes);
1990
1991         return X86EMUL_CONTINUE;
1992 }
1993
1994 int emulator_write_emulated(unsigned long addr,
1995                                    const void *val,
1996                                    unsigned int bytes,
1997                                    struct kvm_vcpu *vcpu)
1998 {
1999         /* Crossing a page boundary? */
2000         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2001                 int rc, now;
2002
2003                 now = -addr & ~PAGE_MASK;
2004                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2005                 if (rc != X86EMUL_CONTINUE)
2006                         return rc;
2007                 addr += now;
2008                 val += now;
2009                 bytes -= now;
2010         }
2011         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2012 }
2013 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2014
2015 static int emulator_cmpxchg_emulated(unsigned long addr,
2016                                      const void *old,
2017                                      const void *new,
2018                                      unsigned int bytes,
2019                                      struct kvm_vcpu *vcpu)
2020 {
2021         static int reported;
2022
2023         if (!reported) {
2024                 reported = 1;
2025                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2026         }
2027 #ifndef CONFIG_X86_64
2028         /* guests cmpxchg8b have to be emulated atomically */
2029         if (bytes == 8) {
2030                 gpa_t gpa;
2031                 struct page *page;
2032                 char *kaddr;
2033                 u64 val;
2034
2035                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2036
2037                 if (gpa == UNMAPPED_GVA ||
2038                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2039                         goto emul_write;
2040
2041                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2042                         goto emul_write;
2043
2044                 val = *(u64 *)new;
2045
2046                 down_read(&current->mm->mmap_sem);
2047                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2048                 up_read(&current->mm->mmap_sem);
2049
2050                 kaddr = kmap_atomic(page, KM_USER0);
2051                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2052                 kunmap_atomic(kaddr, KM_USER0);
2053                 kvm_release_page_dirty(page);
2054         }
2055 emul_write:
2056 #endif
2057
2058         return emulator_write_emulated(addr, new, bytes, vcpu);
2059 }
2060
2061 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2062 {
2063         return kvm_x86_ops->get_segment_base(vcpu, seg);
2064 }
2065
2066 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2067 {
2068         return X86EMUL_CONTINUE;
2069 }
2070
2071 int emulate_clts(struct kvm_vcpu *vcpu)
2072 {
2073         KVMTRACE_0D(CLTS, vcpu, handler);
2074         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2075         return X86EMUL_CONTINUE;
2076 }
2077
2078 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2079 {
2080         struct kvm_vcpu *vcpu = ctxt->vcpu;
2081
2082         switch (dr) {
2083         case 0 ... 3:
2084                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2085                 return X86EMUL_CONTINUE;
2086         default:
2087                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2088                 return X86EMUL_UNHANDLEABLE;
2089         }
2090 }
2091
2092 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2093 {
2094         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2095         int exception;
2096
2097         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2098         if (exception) {
2099                 /* FIXME: better handling */
2100                 return X86EMUL_UNHANDLEABLE;
2101         }
2102         return X86EMUL_CONTINUE;
2103 }
2104
2105 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2106 {
2107         u8 opcodes[4];
2108         unsigned long rip = vcpu->arch.rip;
2109         unsigned long rip_linear;
2110
2111         if (!printk_ratelimit())
2112                 return;
2113
2114         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2115
2116         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2117
2118         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2119                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2120 }
2121 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2122
2123 static struct x86_emulate_ops emulate_ops = {
2124         .read_std            = emulator_read_std,
2125         .read_emulated       = emulator_read_emulated,
2126         .write_emulated      = emulator_write_emulated,
2127         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2128 };
2129
2130 int emulate_instruction(struct kvm_vcpu *vcpu,
2131                         struct kvm_run *run,
2132                         unsigned long cr2,
2133                         u16 error_code,
2134                         int emulation_type)
2135 {
2136         int r;
2137         struct decode_cache *c;
2138
2139         vcpu->arch.mmio_fault_cr2 = cr2;
2140         kvm_x86_ops->cache_regs(vcpu);
2141
2142         vcpu->mmio_is_write = 0;
2143         vcpu->arch.pio.string = 0;
2144
2145         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2146                 int cs_db, cs_l;
2147                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2148
2149                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2150                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2151                 vcpu->arch.emulate_ctxt.mode =
2152                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2153                         ? X86EMUL_MODE_REAL : cs_l
2154                         ? X86EMUL_MODE_PROT64 : cs_db
2155                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2156
2157                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2158
2159                 /* Reject the instructions other than VMCALL/VMMCALL when
2160                  * try to emulate invalid opcode */
2161                 c = &vcpu->arch.emulate_ctxt.decode;
2162                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2163                     (!(c->twobyte && c->b == 0x01 &&
2164                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2165                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2166                         return EMULATE_FAIL;
2167
2168                 ++vcpu->stat.insn_emulation;
2169                 if (r)  {
2170                         ++vcpu->stat.insn_emulation_fail;
2171                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2172                                 return EMULATE_DONE;
2173                         return EMULATE_FAIL;
2174                 }
2175         }
2176
2177         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2178
2179         if (vcpu->arch.pio.string)
2180                 return EMULATE_DO_MMIO;
2181
2182         if ((r || vcpu->mmio_is_write) && run) {
2183                 run->exit_reason = KVM_EXIT_MMIO;
2184                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2185                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2186                 run->mmio.len = vcpu->mmio_size;
2187                 run->mmio.is_write = vcpu->mmio_is_write;
2188         }
2189
2190         if (r) {
2191                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2192                         return EMULATE_DONE;
2193                 if (!vcpu->mmio_needed) {
2194                         kvm_report_emulation_failure(vcpu, "mmio");
2195                         return EMULATE_FAIL;
2196                 }
2197                 return EMULATE_DO_MMIO;
2198         }
2199
2200         kvm_x86_ops->decache_regs(vcpu);
2201         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2202
2203         if (vcpu->mmio_is_write) {
2204                 vcpu->mmio_needed = 0;
2205                 return EMULATE_DO_MMIO;
2206         }
2207
2208         return EMULATE_DONE;
2209 }
2210 EXPORT_SYMBOL_GPL(emulate_instruction);
2211
2212 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2213 {
2214         int i;
2215
2216         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2217                 if (vcpu->arch.pio.guest_pages[i]) {
2218                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2219                         vcpu->arch.pio.guest_pages[i] = NULL;
2220                 }
2221 }
2222
2223 static int pio_copy_data(struct kvm_vcpu *vcpu)
2224 {
2225         void *p = vcpu->arch.pio_data;
2226         void *q;
2227         unsigned bytes;
2228         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2229
2230         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2231                  PAGE_KERNEL);
2232         if (!q) {
2233                 free_pio_guest_pages(vcpu);
2234                 return -ENOMEM;
2235         }
2236         q += vcpu->arch.pio.guest_page_offset;
2237         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2238         if (vcpu->arch.pio.in)
2239                 memcpy(q, p, bytes);
2240         else
2241                 memcpy(p, q, bytes);
2242         q -= vcpu->arch.pio.guest_page_offset;
2243         vunmap(q);
2244         free_pio_guest_pages(vcpu);
2245         return 0;
2246 }
2247
2248 int complete_pio(struct kvm_vcpu *vcpu)
2249 {
2250         struct kvm_pio_request *io = &vcpu->arch.pio;
2251         long delta;
2252         int r;
2253
2254         kvm_x86_ops->cache_regs(vcpu);
2255
2256         if (!io->string) {
2257                 if (io->in)
2258                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2259                                io->size);
2260         } else {
2261                 if (io->in) {
2262                         r = pio_copy_data(vcpu);
2263                         if (r) {
2264                                 kvm_x86_ops->cache_regs(vcpu);
2265                                 return r;
2266                         }
2267                 }
2268
2269                 delta = 1;
2270                 if (io->rep) {
2271                         delta *= io->cur_count;
2272                         /*
2273                          * The size of the register should really depend on
2274                          * current address size.
2275                          */
2276                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2277                 }
2278                 if (io->down)
2279                         delta = -delta;
2280                 delta *= io->size;
2281                 if (io->in)
2282                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2283                 else
2284                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2285         }
2286
2287         kvm_x86_ops->decache_regs(vcpu);
2288
2289         io->count -= io->cur_count;
2290         io->cur_count = 0;
2291
2292         return 0;
2293 }
2294
2295 static void kernel_pio(struct kvm_io_device *pio_dev,
2296                        struct kvm_vcpu *vcpu,
2297                        void *pd)
2298 {
2299         /* TODO: String I/O for in kernel device */
2300
2301         mutex_lock(&vcpu->kvm->lock);
2302         if (vcpu->arch.pio.in)
2303                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2304                                   vcpu->arch.pio.size,
2305                                   pd);
2306         else
2307                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2308                                    vcpu->arch.pio.size,
2309                                    pd);
2310         mutex_unlock(&vcpu->kvm->lock);
2311 }
2312
2313 static void pio_string_write(struct kvm_io_device *pio_dev,
2314                              struct kvm_vcpu *vcpu)
2315 {
2316         struct kvm_pio_request *io = &vcpu->arch.pio;
2317         void *pd = vcpu->arch.pio_data;
2318         int i;
2319
2320         mutex_lock(&vcpu->kvm->lock);
2321         for (i = 0; i < io->cur_count; i++) {
2322                 kvm_iodevice_write(pio_dev, io->port,
2323                                    io->size,
2324                                    pd);
2325                 pd += io->size;
2326         }
2327         mutex_unlock(&vcpu->kvm->lock);
2328 }
2329
2330 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2331                                                gpa_t addr, int len,
2332                                                int is_write)
2333 {
2334         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2335 }
2336
2337 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2338                   int size, unsigned port)
2339 {
2340         struct kvm_io_device *pio_dev;
2341
2342         vcpu->run->exit_reason = KVM_EXIT_IO;
2343         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2344         vcpu->run->io.size = vcpu->arch.pio.size = size;
2345         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2346         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2347         vcpu->run->io.port = vcpu->arch.pio.port = port;
2348         vcpu->arch.pio.in = in;
2349         vcpu->arch.pio.string = 0;
2350         vcpu->arch.pio.down = 0;
2351         vcpu->arch.pio.guest_page_offset = 0;
2352         vcpu->arch.pio.rep = 0;
2353
2354         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2355                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2356                             handler);
2357         else
2358                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2359                             handler);
2360
2361         kvm_x86_ops->cache_regs(vcpu);
2362         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2363
2364         kvm_x86_ops->skip_emulated_instruction(vcpu);
2365
2366         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2367         if (pio_dev) {
2368                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2369                 complete_pio(vcpu);
2370                 return 1;
2371         }
2372         return 0;
2373 }
2374 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2375
2376 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2377                   int size, unsigned long count, int down,
2378                   gva_t address, int rep, unsigned port)
2379 {
2380         unsigned now, in_page;
2381         int i, ret = 0;
2382         int nr_pages = 1;
2383         struct page *page;
2384         struct kvm_io_device *pio_dev;
2385
2386         vcpu->run->exit_reason = KVM_EXIT_IO;
2387         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2388         vcpu->run->io.size = vcpu->arch.pio.size = size;
2389         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2390         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2391         vcpu->run->io.port = vcpu->arch.pio.port = port;
2392         vcpu->arch.pio.in = in;
2393         vcpu->arch.pio.string = 1;
2394         vcpu->arch.pio.down = down;
2395         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2396         vcpu->arch.pio.rep = rep;
2397
2398         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2399                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2400                             handler);
2401         else
2402                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2403                             handler);
2404
2405         if (!count) {
2406                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2407                 return 1;
2408         }
2409
2410         if (!down)
2411                 in_page = PAGE_SIZE - offset_in_page(address);
2412         else
2413                 in_page = offset_in_page(address) + size;
2414         now = min(count, (unsigned long)in_page / size);
2415         if (!now) {
2416                 /*
2417                  * String I/O straddles page boundary.  Pin two guest pages
2418                  * so that we satisfy atomicity constraints.  Do just one
2419                  * transaction to avoid complexity.
2420                  */
2421                 nr_pages = 2;
2422                 now = 1;
2423         }
2424         if (down) {
2425                 /*
2426                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2427                  */
2428                 pr_unimpl(vcpu, "guest string pio down\n");
2429                 kvm_inject_gp(vcpu, 0);
2430                 return 1;
2431         }
2432         vcpu->run->io.count = now;
2433         vcpu->arch.pio.cur_count = now;
2434
2435         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2436                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2437
2438         for (i = 0; i < nr_pages; ++i) {
2439                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2440                 vcpu->arch.pio.guest_pages[i] = page;
2441                 if (!page) {
2442                         kvm_inject_gp(vcpu, 0);
2443                         free_pio_guest_pages(vcpu);
2444                         return 1;
2445                 }
2446         }
2447
2448         pio_dev = vcpu_find_pio_dev(vcpu, port,
2449                                     vcpu->arch.pio.cur_count,
2450                                     !vcpu->arch.pio.in);
2451         if (!vcpu->arch.pio.in) {
2452                 /* string PIO write */
2453                 ret = pio_copy_data(vcpu);
2454                 if (ret >= 0 && pio_dev) {
2455                         pio_string_write(pio_dev, vcpu);
2456                         complete_pio(vcpu);
2457                         if (vcpu->arch.pio.count == 0)
2458                                 ret = 1;
2459                 }
2460         } else if (pio_dev)
2461                 pr_unimpl(vcpu, "no string pio read support yet, "
2462                        "port %x size %d count %ld\n",
2463                         port, size, count);
2464
2465         return ret;
2466 }
2467 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2468
2469 int kvm_arch_init(void *opaque)
2470 {
2471         int r;
2472         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2473
2474         if (kvm_x86_ops) {
2475                 printk(KERN_ERR "kvm: already loaded the other module\n");
2476                 r = -EEXIST;
2477                 goto out;
2478         }
2479
2480         if (!ops->cpu_has_kvm_support()) {
2481                 printk(KERN_ERR "kvm: no hardware support\n");
2482                 r = -EOPNOTSUPP;
2483                 goto out;
2484         }
2485         if (ops->disabled_by_bios()) {
2486                 printk(KERN_ERR "kvm: disabled by bios\n");
2487                 r = -EOPNOTSUPP;
2488                 goto out;
2489         }
2490
2491         r = kvm_mmu_module_init();
2492         if (r)
2493                 goto out;
2494
2495         kvm_init_msr_list();
2496
2497         kvm_x86_ops = ops;
2498         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2499         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2500         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2501                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
2502         return 0;
2503
2504 out:
2505         return r;
2506 }
2507
2508 void kvm_arch_exit(void)
2509 {
2510         kvm_x86_ops = NULL;
2511         kvm_mmu_module_exit();
2512 }
2513
2514 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2515 {
2516         ++vcpu->stat.halt_exits;
2517         KVMTRACE_0D(HLT, vcpu, handler);
2518         if (irqchip_in_kernel(vcpu->kvm)) {
2519                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2520                 up_read(&vcpu->kvm->slots_lock);
2521                 kvm_vcpu_block(vcpu);
2522                 down_read(&vcpu->kvm->slots_lock);
2523                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2524                         return -EINTR;
2525                 return 1;
2526         } else {
2527                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2528                 return 0;
2529         }
2530 }
2531 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2532
2533 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2534                            unsigned long a1)
2535 {
2536         if (is_long_mode(vcpu))
2537                 return a0;
2538         else
2539                 return a0 | ((gpa_t)a1 << 32);
2540 }
2541
2542 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2543 {
2544         unsigned long nr, a0, a1, a2, a3, ret;
2545         int r = 1;
2546
2547         kvm_x86_ops->cache_regs(vcpu);
2548
2549         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2550         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2551         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2552         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2553         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2554
2555         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2556
2557         if (!is_long_mode(vcpu)) {
2558                 nr &= 0xFFFFFFFF;
2559                 a0 &= 0xFFFFFFFF;
2560                 a1 &= 0xFFFFFFFF;
2561                 a2 &= 0xFFFFFFFF;
2562                 a3 &= 0xFFFFFFFF;
2563         }
2564
2565         switch (nr) {
2566         case KVM_HC_VAPIC_POLL_IRQ:
2567                 ret = 0;
2568                 break;
2569         case KVM_HC_MMU_OP:
2570                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2571                 break;
2572         default:
2573                 ret = -KVM_ENOSYS;
2574                 break;
2575         }
2576         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2577         kvm_x86_ops->decache_regs(vcpu);
2578         ++vcpu->stat.hypercalls;
2579         return r;
2580 }
2581 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2582
2583 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2584 {
2585         char instruction[3];
2586         int ret = 0;
2587
2588
2589         /*
2590          * Blow out the MMU to ensure that no other VCPU has an active mapping
2591          * to ensure that the updated hypercall appears atomically across all
2592          * VCPUs.
2593          */
2594         kvm_mmu_zap_all(vcpu->kvm);
2595
2596         kvm_x86_ops->cache_regs(vcpu);
2597         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2598         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2599             != X86EMUL_CONTINUE)
2600                 ret = -EFAULT;
2601
2602         return ret;
2603 }
2604
2605 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2606 {
2607         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2608 }
2609
2610 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2611 {
2612         struct descriptor_table dt = { limit, base };
2613
2614         kvm_x86_ops->set_gdt(vcpu, &dt);
2615 }
2616
2617 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2618 {
2619         struct descriptor_table dt = { limit, base };
2620
2621         kvm_x86_ops->set_idt(vcpu, &dt);
2622 }
2623
2624 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2625                    unsigned long *rflags)
2626 {
2627         kvm_lmsw(vcpu, msw);
2628         *rflags = kvm_x86_ops->get_rflags(vcpu);
2629 }
2630
2631 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2632 {
2633         unsigned long value;
2634
2635         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2636         switch (cr) {
2637         case 0:
2638                 value = vcpu->arch.cr0;
2639                 break;
2640         case 2:
2641                 value = vcpu->arch.cr2;
2642                 break;
2643         case 3:
2644                 value = vcpu->arch.cr3;
2645                 break;
2646         case 4:
2647                 value = vcpu->arch.cr4;
2648                 break;
2649         case 8:
2650                 value = kvm_get_cr8(vcpu);
2651                 break;
2652         default:
2653                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2654                 return 0;
2655         }
2656         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2657                     (u32)((u64)value >> 32), handler);
2658
2659         return value;
2660 }
2661
2662 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2663                      unsigned long *rflags)
2664 {
2665         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2666                     (u32)((u64)val >> 32), handler);
2667
2668         switch (cr) {
2669         case 0:
2670                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2671                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2672                 break;
2673         case 2:
2674                 vcpu->arch.cr2 = val;
2675                 break;
2676         case 3:
2677                 kvm_set_cr3(vcpu, val);
2678                 break;
2679         case 4:
2680                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2681                 break;
2682         case 8:
2683                 kvm_set_cr8(vcpu, val & 0xfUL);
2684                 break;
2685         default:
2686                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2687         }
2688 }
2689
2690 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2691 {
2692         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2693         int j, nent = vcpu->arch.cpuid_nent;
2694
2695         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2696         /* when no next entry is found, the current entry[i] is reselected */
2697         for (j = i + 1; j == i; j = (j + 1) % nent) {
2698                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2699                 if (ej->function == e->function) {
2700                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2701                         return j;
2702                 }
2703         }
2704         return 0; /* silence gcc, even though control never reaches here */
2705 }
2706
2707 /* find an entry with matching function, matching index (if needed), and that
2708  * should be read next (if it's stateful) */
2709 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2710         u32 function, u32 index)
2711 {
2712         if (e->function != function)
2713                 return 0;
2714         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2715                 return 0;
2716         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2717                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2718                 return 0;
2719         return 1;
2720 }
2721
2722 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2723 {
2724         int i;
2725         u32 function, index;
2726         struct kvm_cpuid_entry2 *e, *best;
2727
2728         kvm_x86_ops->cache_regs(vcpu);
2729         function = vcpu->arch.regs[VCPU_REGS_RAX];
2730         index = vcpu->arch.regs[VCPU_REGS_RCX];
2731         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2732         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2733         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2734         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2735         best = NULL;
2736         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2737                 e = &vcpu->arch.cpuid_entries[i];
2738                 if (is_matching_cpuid_entry(e, function, index)) {
2739                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2740                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2741                         best = e;
2742                         break;
2743                 }
2744                 /*
2745                  * Both basic or both extended?
2746                  */
2747                 if (((e->function ^ function) & 0x80000000) == 0)
2748                         if (!best || e->function > best->function)
2749                                 best = e;
2750         }
2751         if (best) {
2752                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2753                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2754                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2755                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2756         }
2757         kvm_x86_ops->decache_regs(vcpu);
2758         kvm_x86_ops->skip_emulated_instruction(vcpu);
2759         KVMTRACE_5D(CPUID, vcpu, function,
2760                     (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2761                     (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2762                     (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2763                     (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2764 }
2765 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2766
2767 /*
2768  * Check if userspace requested an interrupt window, and that the
2769  * interrupt window is open.
2770  *
2771  * No need to exit to userspace if we already have an interrupt queued.
2772  */
2773 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2774                                           struct kvm_run *kvm_run)
2775 {
2776         return (!vcpu->arch.irq_summary &&
2777                 kvm_run->request_interrupt_window &&
2778                 vcpu->arch.interrupt_window_open &&
2779                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2780 }
2781
2782 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2783                               struct kvm_run *kvm_run)
2784 {
2785         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2786         kvm_run->cr8 = kvm_get_cr8(vcpu);
2787         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2788         if (irqchip_in_kernel(vcpu->kvm))
2789                 kvm_run->ready_for_interrupt_injection = 1;
2790         else
2791                 kvm_run->ready_for_interrupt_injection =
2792                                         (vcpu->arch.interrupt_window_open &&
2793                                          vcpu->arch.irq_summary == 0);
2794 }
2795
2796 static void vapic_enter(struct kvm_vcpu *vcpu)
2797 {
2798         struct kvm_lapic *apic = vcpu->arch.apic;
2799         struct page *page;
2800
2801         if (!apic || !apic->vapic_addr)
2802                 return;
2803
2804         down_read(&current->mm->mmap_sem);
2805         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2806         up_read(&current->mm->mmap_sem);
2807
2808         vcpu->arch.apic->vapic_page = page;
2809 }
2810
2811 static void vapic_exit(struct kvm_vcpu *vcpu)
2812 {
2813         struct kvm_lapic *apic = vcpu->arch.apic;
2814
2815         if (!apic || !apic->vapic_addr)
2816                 return;
2817
2818         down_read(&vcpu->kvm->slots_lock);
2819         kvm_release_page_dirty(apic->vapic_page);
2820         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2821         up_read(&vcpu->kvm->slots_lock);
2822 }
2823
2824 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2825 {
2826         int r;
2827
2828         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2829                 pr_debug("vcpu %d received sipi with vector # %x\n",
2830                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2831                 kvm_lapic_reset(vcpu);
2832                 r = kvm_x86_ops->vcpu_reset(vcpu);
2833                 if (r)
2834                         return r;
2835                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2836         }
2837
2838         down_read(&vcpu->kvm->slots_lock);
2839         vapic_enter(vcpu);
2840
2841 preempted:
2842         if (vcpu->guest_debug.enabled)
2843                 kvm_x86_ops->guest_debug_pre(vcpu);
2844
2845 again:
2846         if (vcpu->requests)
2847                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2848                         kvm_mmu_unload(vcpu);
2849
2850         r = kvm_mmu_reload(vcpu);
2851         if (unlikely(r))
2852                 goto out;
2853
2854         if (vcpu->requests) {
2855                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2856                         __kvm_migrate_timers(vcpu);
2857                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2858                         kvm_x86_ops->tlb_flush(vcpu);
2859                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2860                                        &vcpu->requests)) {
2861                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2862                         r = 0;
2863                         goto out;
2864                 }
2865                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2866                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2867                         r = 0;
2868                         goto out;
2869                 }
2870         }
2871
2872         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2873         kvm_inject_pending_timer_irqs(vcpu);
2874
2875         preempt_disable();
2876
2877         kvm_x86_ops->prepare_guest_switch(vcpu);
2878         kvm_load_guest_fpu(vcpu);
2879
2880         local_irq_disable();
2881
2882         if (vcpu->requests || need_resched()) {
2883                 local_irq_enable();
2884                 preempt_enable();
2885                 r = 1;
2886                 goto out;
2887         }
2888
2889         if (signal_pending(current)) {
2890                 local_irq_enable();
2891                 preempt_enable();
2892                 r = -EINTR;
2893                 kvm_run->exit_reason = KVM_EXIT_INTR;
2894                 ++vcpu->stat.signal_exits;
2895                 goto out;
2896         }
2897
2898         vcpu->guest_mode = 1;
2899         /*
2900          * Make sure that guest_mode assignment won't happen after
2901          * testing the pending IRQ vector bitmap.
2902          */
2903         smp_wmb();
2904
2905         if (vcpu->arch.exception.pending)
2906                 __queue_exception(vcpu);
2907         else if (irqchip_in_kernel(vcpu->kvm))
2908                 kvm_x86_ops->inject_pending_irq(vcpu);
2909         else
2910                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2911
2912         kvm_lapic_sync_to_vapic(vcpu);
2913
2914         up_read(&vcpu->kvm->slots_lock);
2915
2916         kvm_guest_enter();
2917
2918
2919         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2920         kvm_x86_ops->run(vcpu, kvm_run);
2921
2922         vcpu->guest_mode = 0;
2923         local_irq_enable();
2924
2925         ++vcpu->stat.exits;
2926
2927         /*
2928          * We must have an instruction between local_irq_enable() and
2929          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2930          * the interrupt shadow.  The stat.exits increment will do nicely.
2931          * But we need to prevent reordering, hence this barrier():
2932          */
2933         barrier();
2934
2935         kvm_guest_exit();
2936
2937         preempt_enable();
2938
2939         down_read(&vcpu->kvm->slots_lock);
2940
2941         /*
2942          * Profile KVM exit RIPs:
2943          */
2944         if (unlikely(prof_on == KVM_PROFILING)) {
2945                 kvm_x86_ops->cache_regs(vcpu);
2946                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2947         }
2948
2949         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2950                 vcpu->arch.exception.pending = false;
2951
2952         kvm_lapic_sync_from_vapic(vcpu);
2953
2954         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2955
2956         if (r > 0) {
2957                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2958                         r = -EINTR;
2959                         kvm_run->exit_reason = KVM_EXIT_INTR;
2960                         ++vcpu->stat.request_irq_exits;
2961                         goto out;
2962                 }
2963                 if (!need_resched())
2964                         goto again;
2965         }
2966
2967 out:
2968         up_read(&vcpu->kvm->slots_lock);
2969         if (r > 0) {
2970                 kvm_resched(vcpu);
2971                 down_read(&vcpu->kvm->slots_lock);
2972                 goto preempted;
2973         }
2974
2975         post_kvm_run_save(vcpu, kvm_run);
2976
2977         vapic_exit(vcpu);
2978
2979         return r;
2980 }
2981
2982 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2983 {
2984         int r;
2985         sigset_t sigsaved;
2986
2987         vcpu_load(vcpu);
2988
2989         if (vcpu->sigset_active)
2990                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2991
2992         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2993                 kvm_vcpu_block(vcpu);
2994                 r = -EAGAIN;
2995                 goto out;
2996         }
2997
2998         /* re-sync apic's tpr */
2999         if (!irqchip_in_kernel(vcpu->kvm))
3000                 kvm_set_cr8(vcpu, kvm_run->cr8);
3001
3002         if (vcpu->arch.pio.cur_count) {
3003                 r = complete_pio(vcpu);
3004                 if (r)
3005                         goto out;
3006         }
3007 #if CONFIG_HAS_IOMEM
3008         if (vcpu->mmio_needed) {
3009                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3010                 vcpu->mmio_read_completed = 1;
3011                 vcpu->mmio_needed = 0;
3012
3013                 down_read(&vcpu->kvm->slots_lock);
3014                 r = emulate_instruction(vcpu, kvm_run,
3015                                         vcpu->arch.mmio_fault_cr2, 0,
3016                                         EMULTYPE_NO_DECODE);
3017                 up_read(&vcpu->kvm->slots_lock);
3018                 if (r == EMULATE_DO_MMIO) {
3019                         /*
3020                          * Read-modify-write.  Back to userspace.
3021                          */
3022                         r = 0;
3023                         goto out;
3024                 }
3025         }
3026 #endif
3027         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
3028                 kvm_x86_ops->cache_regs(vcpu);
3029                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
3030                 kvm_x86_ops->decache_regs(vcpu);
3031         }
3032
3033         r = __vcpu_run(vcpu, kvm_run);
3034
3035 out:
3036         if (vcpu->sigset_active)
3037                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3038
3039         vcpu_put(vcpu);
3040         return r;
3041 }
3042
3043 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3044 {
3045         vcpu_load(vcpu);
3046
3047         kvm_x86_ops->cache_regs(vcpu);
3048
3049         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
3050         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
3051         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
3052         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
3053         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
3054         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
3055         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3056         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3057 #ifdef CONFIG_X86_64
3058         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3059         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3060         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3061         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3062         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3063         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3064         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3065         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3066 #endif
3067
3068         regs->rip = vcpu->arch.rip;
3069         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3070
3071         /*
3072          * Don't leak debug flags in case they were set for guest debugging
3073          */
3074         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3075                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3076
3077         vcpu_put(vcpu);
3078
3079         return 0;
3080 }
3081
3082 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3083 {
3084         vcpu_load(vcpu);
3085
3086         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3087         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3088         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3089         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3090         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3091         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3092         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3093         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3094 #ifdef CONFIG_X86_64
3095         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3096         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3097         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3098         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3099         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3100         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3101         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3102         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3103 #endif
3104
3105         vcpu->arch.rip = regs->rip;
3106         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3107
3108         kvm_x86_ops->decache_regs(vcpu);
3109
3110         vcpu->arch.exception.pending = false;
3111
3112         vcpu_put(vcpu);
3113
3114         return 0;
3115 }
3116
3117 void kvm_get_segment(struct kvm_vcpu *vcpu,
3118                      struct kvm_segment *var, int seg)
3119 {
3120         kvm_x86_ops->get_segment(vcpu, var, seg);
3121 }
3122
3123 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3124 {
3125         struct kvm_segment cs;
3126
3127         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3128         *db = cs.db;
3129         *l = cs.l;
3130 }
3131 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3132
3133 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3134                                   struct kvm_sregs *sregs)
3135 {
3136         struct descriptor_table dt;
3137         int pending_vec;
3138
3139         vcpu_load(vcpu);
3140
3141         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3142         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3143         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3144         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3145         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3146         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3147
3148         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3149         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3150
3151         kvm_x86_ops->get_idt(vcpu, &dt);
3152         sregs->idt.limit = dt.limit;
3153         sregs->idt.base = dt.base;
3154         kvm_x86_ops->get_gdt(vcpu, &dt);
3155         sregs->gdt.limit = dt.limit;
3156         sregs->gdt.base = dt.base;
3157
3158         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3159         sregs->cr0 = vcpu->arch.cr0;
3160         sregs->cr2 = vcpu->arch.cr2;
3161         sregs->cr3 = vcpu->arch.cr3;
3162         sregs->cr4 = vcpu->arch.cr4;
3163         sregs->cr8 = kvm_get_cr8(vcpu);
3164         sregs->efer = vcpu->arch.shadow_efer;
3165         sregs->apic_base = kvm_get_apic_base(vcpu);
3166
3167         if (irqchip_in_kernel(vcpu->kvm)) {
3168                 memset(sregs->interrupt_bitmap, 0,
3169                        sizeof sregs->interrupt_bitmap);
3170                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3171                 if (pending_vec >= 0)
3172                         set_bit(pending_vec,
3173                                 (unsigned long *)sregs->interrupt_bitmap);
3174         } else
3175                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3176                        sizeof sregs->interrupt_bitmap);
3177
3178         vcpu_put(vcpu);
3179
3180         return 0;
3181 }
3182
3183 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3184                                     struct kvm_mp_state *mp_state)
3185 {
3186         vcpu_load(vcpu);
3187         mp_state->mp_state = vcpu->arch.mp_state;
3188         vcpu_put(vcpu);
3189         return 0;
3190 }
3191
3192 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3193                                     struct kvm_mp_state *mp_state)
3194 {
3195         vcpu_load(vcpu);
3196         vcpu->arch.mp_state = mp_state->mp_state;
3197         vcpu_put(vcpu);
3198         return 0;
3199 }
3200
3201 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3202                         struct kvm_segment *var, int seg)
3203 {
3204         kvm_x86_ops->set_segment(vcpu, var, seg);
3205 }
3206
3207 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3208                                    struct kvm_segment *kvm_desct)
3209 {
3210         kvm_desct->base = seg_desc->base0;
3211         kvm_desct->base |= seg_desc->base1 << 16;
3212         kvm_desct->base |= seg_desc->base2 << 24;
3213         kvm_desct->limit = seg_desc->limit0;
3214         kvm_desct->limit |= seg_desc->limit << 16;
3215         if (seg_desc->g) {
3216                 kvm_desct->limit <<= 12;
3217                 kvm_desct->limit |= 0xfff;
3218         }
3219         kvm_desct->selector = selector;
3220         kvm_desct->type = seg_desc->type;
3221         kvm_desct->present = seg_desc->p;
3222         kvm_desct->dpl = seg_desc->dpl;
3223         kvm_desct->db = seg_desc->d;
3224         kvm_desct->s = seg_desc->s;
3225         kvm_desct->l = seg_desc->l;
3226         kvm_desct->g = seg_desc->g;
3227         kvm_desct->avl = seg_desc->avl;
3228         if (!selector)
3229                 kvm_desct->unusable = 1;
3230         else
3231                 kvm_desct->unusable = 0;
3232         kvm_desct->padding = 0;
3233 }
3234
3235 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3236                                            u16 selector,
3237                                            struct descriptor_table *dtable)
3238 {
3239         if (selector & 1 << 2) {
3240                 struct kvm_segment kvm_seg;
3241
3242                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3243
3244                 if (kvm_seg.unusable)
3245                         dtable->limit = 0;
3246                 else
3247                         dtable->limit = kvm_seg.limit;
3248                 dtable->base = kvm_seg.base;
3249         }
3250         else
3251                 kvm_x86_ops->get_gdt(vcpu, dtable);
3252 }
3253
3254 /* allowed just for 8 bytes segments */
3255 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3256                                          struct desc_struct *seg_desc)
3257 {
3258         gpa_t gpa;
3259         struct descriptor_table dtable;
3260         u16 index = selector >> 3;
3261
3262         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3263
3264         if (dtable.limit < index * 8 + 7) {
3265                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3266                 return 1;
3267         }
3268         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3269         gpa += index * 8;
3270         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3271 }
3272
3273 /* allowed just for 8 bytes segments */
3274 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3275                                          struct desc_struct *seg_desc)
3276 {
3277         gpa_t gpa;
3278         struct descriptor_table dtable;
3279         u16 index = selector >> 3;
3280
3281         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3282
3283         if (dtable.limit < index * 8 + 7)
3284                 return 1;
3285         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3286         gpa += index * 8;
3287         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3288 }
3289
3290 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3291                              struct desc_struct *seg_desc)
3292 {
3293         u32 base_addr;
3294
3295         base_addr = seg_desc->base0;
3296         base_addr |= (seg_desc->base1 << 16);
3297         base_addr |= (seg_desc->base2 << 24);
3298
3299         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3300 }
3301
3302 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3303 {
3304         struct kvm_segment kvm_seg;
3305
3306         kvm_get_segment(vcpu, &kvm_seg, seg);
3307         return kvm_seg.selector;
3308 }
3309
3310 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3311                                                 u16 selector,
3312                                                 struct kvm_segment *kvm_seg)
3313 {
3314         struct desc_struct seg_desc;
3315
3316         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3317                 return 1;
3318         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3319         return 0;
3320 }
3321
3322 int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3323 {
3324         struct kvm_segment segvar = {
3325                 .base = selector << 4,
3326                 .limit = 0xffff,
3327                 .selector = selector,
3328                 .type = 3,
3329                 .present = 1,
3330                 .dpl = 3,
3331                 .db = 0,
3332                 .s = 1,
3333                 .l = 0,
3334                 .g = 0,
3335                 .avl = 0,
3336                 .unusable = 0,
3337         };
3338         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3339         return 0;
3340 }
3341
3342 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3343                                 int type_bits, int seg)
3344 {
3345         struct kvm_segment kvm_seg;
3346
3347         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3348                 return kvm_load_realmode_segment(vcpu, selector, seg);
3349         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3350                 return 1;
3351         kvm_seg.type |= type_bits;
3352
3353         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3354             seg != VCPU_SREG_LDTR)
3355                 if (!kvm_seg.s)
3356                         kvm_seg.unusable = 1;
3357
3358         kvm_set_segment(vcpu, &kvm_seg, seg);
3359         return 0;
3360 }
3361
3362 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3363                                 struct tss_segment_32 *tss)
3364 {
3365         tss->cr3 = vcpu->arch.cr3;
3366         tss->eip = vcpu->arch.rip;
3367         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3368         tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3369         tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3370         tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3371         tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3372         tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3373         tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3374         tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3375         tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3376
3377         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3378         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3379         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3380         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3381         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3382         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3383         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3384         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3385 }
3386
3387 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3388                                   struct tss_segment_32 *tss)
3389 {
3390         kvm_set_cr3(vcpu, tss->cr3);
3391
3392         vcpu->arch.rip = tss->eip;
3393         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3394
3395         vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3396         vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3397         vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3398         vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3399         vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3400         vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3401         vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3402         vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3403
3404         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3405                 return 1;
3406
3407         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3408                 return 1;
3409
3410         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3411                 return 1;
3412
3413         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3414                 return 1;
3415
3416         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3417                 return 1;
3418
3419         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3420                 return 1;
3421
3422         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3423                 return 1;
3424         return 0;
3425 }
3426
3427 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3428                                 struct tss_segment_16 *tss)
3429 {
3430         tss->ip = vcpu->arch.rip;
3431         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3432         tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3433         tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3434         tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3435         tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3436         tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3437         tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3438         tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3439         tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3440
3441         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3442         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3443         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3444         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3445         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3446         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3447 }
3448
3449 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3450                                  struct tss_segment_16 *tss)
3451 {
3452         vcpu->arch.rip = tss->ip;
3453         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3454         vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3455         vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3456         vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3457         vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3458         vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3459         vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3460         vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3461         vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3462
3463         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3464                 return 1;
3465
3466         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3467                 return 1;
3468
3469         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3470                 return 1;
3471
3472         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3473                 return 1;
3474
3475         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3476                 return 1;
3477         return 0;
3478 }
3479
3480 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3481                        u32 old_tss_base,
3482                        struct desc_struct *nseg_desc)
3483 {
3484         struct tss_segment_16 tss_segment_16;
3485         int ret = 0;
3486
3487         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3488                            sizeof tss_segment_16))
3489                 goto out;
3490
3491         save_state_to_tss16(vcpu, &tss_segment_16);
3492
3493         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3494                             sizeof tss_segment_16))
3495                 goto out;
3496
3497         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3498                            &tss_segment_16, sizeof tss_segment_16))
3499                 goto out;
3500
3501         if (load_state_from_tss16(vcpu, &tss_segment_16))
3502                 goto out;
3503
3504         ret = 1;
3505 out:
3506         return ret;
3507 }
3508
3509 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3510                        u32 old_tss_base,
3511                        struct desc_struct *nseg_desc)
3512 {
3513         struct tss_segment_32 tss_segment_32;
3514         int ret = 0;
3515
3516         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3517                            sizeof tss_segment_32))
3518                 goto out;
3519
3520         save_state_to_tss32(vcpu, &tss_segment_32);
3521
3522         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3523                             sizeof tss_segment_32))
3524                 goto out;
3525
3526         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3527                            &tss_segment_32, sizeof tss_segment_32))
3528                 goto out;
3529
3530         if (load_state_from_tss32(vcpu, &tss_segment_32))
3531                 goto out;
3532
3533         ret = 1;
3534 out:
3535         return ret;
3536 }
3537
3538 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3539 {
3540         struct kvm_segment tr_seg;
3541         struct desc_struct cseg_desc;
3542         struct desc_struct nseg_desc;
3543         int ret = 0;
3544         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3545         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3546
3547         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3548
3549         /* FIXME: Handle errors. Failure to read either TSS or their
3550          * descriptors should generate a pagefault.
3551          */
3552         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3553                 goto out;
3554
3555         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3556                 goto out;
3557
3558         if (reason != TASK_SWITCH_IRET) {
3559                 int cpl;
3560
3561                 cpl = kvm_x86_ops->get_cpl(vcpu);
3562                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3563                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3564                         return 1;
3565                 }
3566         }
3567
3568         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3569                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3570                 return 1;
3571         }
3572
3573         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3574                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3575                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3576         }
3577
3578         if (reason == TASK_SWITCH_IRET) {
3579                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3580                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3581         }
3582
3583         kvm_x86_ops->skip_emulated_instruction(vcpu);
3584         kvm_x86_ops->cache_regs(vcpu);
3585
3586         if (nseg_desc.type & 8)
3587                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3588                                          &nseg_desc);
3589         else
3590                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3591                                          &nseg_desc);
3592
3593         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3594                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3595                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3596         }
3597
3598         if (reason != TASK_SWITCH_IRET) {
3599                 nseg_desc.type |= (1 << 1);
3600                 save_guest_segment_descriptor(vcpu, tss_selector,
3601                                               &nseg_desc);
3602         }
3603
3604         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3605         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3606         tr_seg.type = 11;
3607         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3608 out:
3609         kvm_x86_ops->decache_regs(vcpu);
3610         return ret;
3611 }
3612 EXPORT_SYMBOL_GPL(kvm_task_switch);
3613
3614 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3615                                   struct kvm_sregs *sregs)
3616 {
3617         int mmu_reset_needed = 0;
3618         int i, pending_vec, max_bits;
3619         struct descriptor_table dt;
3620
3621         vcpu_load(vcpu);
3622
3623         dt.limit = sregs->idt.limit;
3624         dt.base = sregs->idt.base;
3625         kvm_x86_ops->set_idt(vcpu, &dt);
3626         dt.limit = sregs->gdt.limit;
3627         dt.base = sregs->gdt.base;
3628         kvm_x86_ops->set_gdt(vcpu, &dt);
3629
3630         vcpu->arch.cr2 = sregs->cr2;
3631         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3632         vcpu->arch.cr3 = sregs->cr3;
3633
3634         kvm_set_cr8(vcpu, sregs->cr8);
3635
3636         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3637         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3638         kvm_set_apic_base(vcpu, sregs->apic_base);
3639
3640         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3641
3642         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3643         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3644         vcpu->arch.cr0 = sregs->cr0;
3645
3646         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3647         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3648         if (!is_long_mode(vcpu) && is_pae(vcpu))
3649                 load_pdptrs(vcpu, vcpu->arch.cr3);
3650
3651         if (mmu_reset_needed)
3652                 kvm_mmu_reset_context(vcpu);
3653
3654         if (!irqchip_in_kernel(vcpu->kvm)) {
3655                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3656                        sizeof vcpu->arch.irq_pending);
3657                 vcpu->arch.irq_summary = 0;
3658                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3659                         if (vcpu->arch.irq_pending[i])
3660                                 __set_bit(i, &vcpu->arch.irq_summary);
3661         } else {
3662                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3663                 pending_vec = find_first_bit(
3664                         (const unsigned long *)sregs->interrupt_bitmap,
3665                         max_bits);
3666                 /* Only pending external irq is handled here */
3667                 if (pending_vec < max_bits) {
3668                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3669                         pr_debug("Set back pending irq %d\n",
3670                                  pending_vec);
3671                 }
3672         }
3673
3674         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3675         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3676         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3677         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3678         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3679         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3680
3681         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3682         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3683
3684         vcpu_put(vcpu);
3685
3686         return 0;
3687 }
3688
3689 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3690                                     struct kvm_debug_guest *dbg)
3691 {
3692         int r;
3693
3694         vcpu_load(vcpu);
3695
3696         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3697
3698         vcpu_put(vcpu);
3699
3700         return r;
3701 }
3702
3703 /*
3704  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3705  * we have asm/x86/processor.h
3706  */
3707 struct fxsave {
3708         u16     cwd;
3709         u16     swd;
3710         u16     twd;
3711         u16     fop;
3712         u64     rip;
3713         u64     rdp;
3714         u32     mxcsr;
3715         u32     mxcsr_mask;
3716         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3717 #ifdef CONFIG_X86_64
3718         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3719 #else
3720         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3721 #endif
3722 };
3723
3724 /*
3725  * Translate a guest virtual address to a guest physical address.
3726  */
3727 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3728                                     struct kvm_translation *tr)
3729 {
3730         unsigned long vaddr = tr->linear_address;
3731         gpa_t gpa;
3732
3733         vcpu_load(vcpu);
3734         down_read(&vcpu->kvm->slots_lock);
3735         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3736         up_read(&vcpu->kvm->slots_lock);
3737         tr->physical_address = gpa;
3738         tr->valid = gpa != UNMAPPED_GVA;
3739         tr->writeable = 1;
3740         tr->usermode = 0;
3741         vcpu_put(vcpu);
3742
3743         return 0;
3744 }
3745
3746 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3747 {
3748         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3749
3750         vcpu_load(vcpu);
3751
3752         memcpy(fpu->fpr, fxsave->st_space, 128);
3753         fpu->fcw = fxsave->cwd;
3754         fpu->fsw = fxsave->swd;
3755         fpu->ftwx = fxsave->twd;
3756         fpu->last_opcode = fxsave->fop;
3757         fpu->last_ip = fxsave->rip;
3758         fpu->last_dp = fxsave->rdp;
3759         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3760
3761         vcpu_put(vcpu);
3762
3763         return 0;
3764 }
3765
3766 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3767 {
3768         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3769
3770         vcpu_load(vcpu);
3771
3772         memcpy(fxsave->st_space, fpu->fpr, 128);
3773         fxsave->cwd = fpu->fcw;
3774         fxsave->swd = fpu->fsw;
3775         fxsave->twd = fpu->ftwx;
3776         fxsave->fop = fpu->last_opcode;
3777         fxsave->rip = fpu->last_ip;
3778         fxsave->rdp = fpu->last_dp;
3779         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3780
3781         vcpu_put(vcpu);
3782
3783         return 0;
3784 }
3785
3786 void fx_init(struct kvm_vcpu *vcpu)
3787 {
3788         unsigned after_mxcsr_mask;
3789
3790         /*
3791          * Touch the fpu the first time in non atomic context as if
3792          * this is the first fpu instruction the exception handler
3793          * will fire before the instruction returns and it'll have to
3794          * allocate ram with GFP_KERNEL.
3795          */
3796         if (!used_math())
3797                 kvm_fx_save(&vcpu->arch.host_fx_image);
3798
3799         /* Initialize guest FPU by resetting ours and saving into guest's */
3800         preempt_disable();
3801         kvm_fx_save(&vcpu->arch.host_fx_image);
3802         kvm_fx_finit();
3803         kvm_fx_save(&vcpu->arch.guest_fx_image);
3804         kvm_fx_restore(&vcpu->arch.host_fx_image);
3805         preempt_enable();
3806
3807         vcpu->arch.cr0 |= X86_CR0_ET;
3808         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3809         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3810         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3811                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3812 }
3813 EXPORT_SYMBOL_GPL(fx_init);
3814
3815 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3816 {
3817         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3818                 return;
3819
3820         vcpu->guest_fpu_loaded = 1;
3821         kvm_fx_save(&vcpu->arch.host_fx_image);
3822         kvm_fx_restore(&vcpu->arch.guest_fx_image);
3823 }
3824 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3825
3826 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3827 {
3828         if (!vcpu->guest_fpu_loaded)
3829                 return;
3830
3831         vcpu->guest_fpu_loaded = 0;
3832         kvm_fx_save(&vcpu->arch.guest_fx_image);
3833         kvm_fx_restore(&vcpu->arch.host_fx_image);
3834         ++vcpu->stat.fpu_reload;
3835 }
3836 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3837
3838 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3839 {
3840         kvm_x86_ops->vcpu_free(vcpu);
3841 }
3842
3843 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3844                                                 unsigned int id)
3845 {
3846         return kvm_x86_ops->vcpu_create(kvm, id);
3847 }
3848
3849 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3850 {
3851         int r;
3852
3853         /* We do fxsave: this must be aligned. */
3854         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3855
3856         vcpu_load(vcpu);
3857         r = kvm_arch_vcpu_reset(vcpu);
3858         if (r == 0)
3859                 r = kvm_mmu_setup(vcpu);
3860         vcpu_put(vcpu);
3861         if (r < 0)
3862                 goto free_vcpu;
3863
3864         return 0;
3865 free_vcpu:
3866         kvm_x86_ops->vcpu_free(vcpu);
3867         return r;
3868 }
3869
3870 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3871 {
3872         vcpu_load(vcpu);
3873         kvm_mmu_unload(vcpu);
3874         vcpu_put(vcpu);
3875
3876         kvm_x86_ops->vcpu_free(vcpu);
3877 }
3878
3879 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3880 {
3881         return kvm_x86_ops->vcpu_reset(vcpu);
3882 }
3883
3884 void kvm_arch_hardware_enable(void *garbage)
3885 {
3886         kvm_x86_ops->hardware_enable(garbage);
3887 }
3888
3889 void kvm_arch_hardware_disable(void *garbage)
3890 {
3891         kvm_x86_ops->hardware_disable(garbage);
3892 }
3893
3894 int kvm_arch_hardware_setup(void)
3895 {
3896         return kvm_x86_ops->hardware_setup();
3897 }
3898
3899 void kvm_arch_hardware_unsetup(void)
3900 {
3901         kvm_x86_ops->hardware_unsetup();
3902 }
3903
3904 void kvm_arch_check_processor_compat(void *rtn)
3905 {
3906         kvm_x86_ops->check_processor_compatibility(rtn);
3907 }
3908
3909 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3910 {
3911         struct page *page;
3912         struct kvm *kvm;
3913         int r;
3914
3915         BUG_ON(vcpu->kvm == NULL);
3916         kvm = vcpu->kvm;
3917
3918         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3919         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3920                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3921         else
3922                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3923
3924         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3925         if (!page) {
3926                 r = -ENOMEM;
3927                 goto fail;
3928         }
3929         vcpu->arch.pio_data = page_address(page);
3930
3931         r = kvm_mmu_create(vcpu);
3932         if (r < 0)
3933                 goto fail_free_pio_data;
3934
3935         if (irqchip_in_kernel(kvm)) {
3936                 r = kvm_create_lapic(vcpu);
3937                 if (r < 0)
3938                         goto fail_mmu_destroy;
3939         }
3940
3941         return 0;
3942
3943 fail_mmu_destroy:
3944         kvm_mmu_destroy(vcpu);
3945 fail_free_pio_data:
3946         free_page((unsigned long)vcpu->arch.pio_data);
3947 fail:
3948         return r;
3949 }
3950
3951 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3952 {
3953         kvm_free_lapic(vcpu);
3954         down_read(&vcpu->kvm->slots_lock);
3955         kvm_mmu_destroy(vcpu);
3956         up_read(&vcpu->kvm->slots_lock);
3957         free_page((unsigned long)vcpu->arch.pio_data);
3958 }
3959
3960 struct  kvm *kvm_arch_create_vm(void)
3961 {
3962         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3963
3964         if (!kvm)
3965                 return ERR_PTR(-ENOMEM);
3966
3967         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3968
3969         return kvm;
3970 }
3971
3972 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3973 {
3974         vcpu_load(vcpu);
3975         kvm_mmu_unload(vcpu);
3976         vcpu_put(vcpu);
3977 }
3978
3979 static void kvm_free_vcpus(struct kvm *kvm)
3980 {
3981         unsigned int i;
3982
3983         /*
3984          * Unpin any mmu pages first.
3985          */
3986         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3987                 if (kvm->vcpus[i])
3988                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3989         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3990                 if (kvm->vcpus[i]) {
3991                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3992                         kvm->vcpus[i] = NULL;
3993                 }
3994         }
3995
3996 }
3997
3998 void kvm_arch_destroy_vm(struct kvm *kvm)
3999 {
4000         kvm_free_pit(kvm);
4001         kfree(kvm->arch.vpic);
4002         kfree(kvm->arch.vioapic);
4003         kvm_free_vcpus(kvm);
4004         kvm_free_physmem(kvm);
4005         if (kvm->arch.apic_access_page)
4006                 put_page(kvm->arch.apic_access_page);
4007         if (kvm->arch.ept_identity_pagetable)
4008                 put_page(kvm->arch.ept_identity_pagetable);
4009         kfree(kvm);
4010 }
4011
4012 int kvm_arch_set_memory_region(struct kvm *kvm,
4013                                 struct kvm_userspace_memory_region *mem,
4014                                 struct kvm_memory_slot old,
4015                                 int user_alloc)
4016 {
4017         int npages = mem->memory_size >> PAGE_SHIFT;
4018         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4019
4020         /*To keep backward compatibility with older userspace,
4021          *x86 needs to hanlde !user_alloc case.
4022          */
4023         if (!user_alloc) {
4024                 if (npages && !old.rmap) {
4025                         unsigned long userspace_addr;
4026
4027                         down_write(&current->mm->mmap_sem);
4028                         userspace_addr = do_mmap(NULL, 0,
4029                                                  npages * PAGE_SIZE,
4030                                                  PROT_READ | PROT_WRITE,
4031                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4032                                                  0);
4033                         up_write(&current->mm->mmap_sem);
4034
4035                         if (IS_ERR((void *)userspace_addr))
4036                                 return PTR_ERR((void *)userspace_addr);
4037
4038                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4039                         spin_lock(&kvm->mmu_lock);
4040                         memslot->userspace_addr = userspace_addr;
4041                         spin_unlock(&kvm->mmu_lock);
4042                 } else {
4043                         if (!old.user_alloc && old.rmap) {
4044                                 int ret;
4045
4046                                 down_write(&current->mm->mmap_sem);
4047                                 ret = do_munmap(current->mm, old.userspace_addr,
4048                                                 old.npages * PAGE_SIZE);
4049                                 up_write(&current->mm->mmap_sem);
4050                                 if (ret < 0)
4051                                         printk(KERN_WARNING
4052                                        "kvm_vm_ioctl_set_memory_region: "
4053                                        "failed to munmap memory\n");
4054                         }
4055                 }
4056         }
4057
4058         if (!kvm->arch.n_requested_mmu_pages) {
4059                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4060                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4061         }
4062
4063         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4064         kvm_flush_remote_tlbs(kvm);
4065
4066         return 0;
4067 }
4068
4069 void kvm_arch_flush_shadow(struct kvm *kvm)
4070 {
4071         kvm_mmu_zap_all(kvm);
4072 }
4073
4074 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4075 {
4076         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4077                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4078 }
4079
4080 static void vcpu_kick_intr(void *info)
4081 {
4082 #ifdef DEBUG
4083         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4084         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4085 #endif
4086 }
4087
4088 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4089 {
4090         int ipi_pcpu = vcpu->cpu;
4091         int cpu = get_cpu();
4092
4093         if (waitqueue_active(&vcpu->wq)) {
4094                 wake_up_interruptible(&vcpu->wq);
4095                 ++vcpu->stat.halt_wakeup;
4096         }
4097         /*
4098          * We may be called synchronously with irqs disabled in guest mode,
4099          * So need not to call smp_call_function_single() in that case.
4100          */
4101         if (vcpu->guest_mode && vcpu->cpu != cpu)
4102                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4103         put_cpu();
4104 }