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