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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   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 }
 307 EXPORT_SYMBOL_GPL(kvm_lmsw);
 308
 309 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 310 {
 311         if (cr4 & CR4_RESERVED_BITS) {
 312                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 313                 kvm_inject_gp(vcpu, 0);
 314                 return;
 315         }
 316
 317         if (is_long_mode(vcpu)) {
 318                 if (!(cr4 & X86_CR4_PAE)) {
 319                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 320                                "in long mode\n");
 321                         kvm_inject_gp(vcpu, 0);
 322                         return;
 323                 }
 324         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 325                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 326                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 327                 kvm_inject_gp(vcpu, 0);
 328                 return;
 329         }
 330
 331         if (cr4 & X86_CR4_VMXE) {
 332                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 333                 kvm_inject_gp(vcpu, 0);
 334                 return;
 335         }
 336         kvm_x86_ops->set_cr4(vcpu, cr4);
 337         vcpu->arch.cr4 = cr4;
 338         kvm_mmu_reset_context(vcpu);
 339 }
 340 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 341
 342 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 343 {
 344         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 345                 kvm_mmu_flush_tlb(vcpu);
 346                 return;
 347         }
 348
 349         if (is_long_mode(vcpu)) {
 350                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 351                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 352                         kvm_inject_gp(vcpu, 0);
 353                         return;
 354                 }
 355         } else {
 356                 if (is_pae(vcpu)) {
 357                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 358                                 printk(KERN_DEBUG
 359                                        "set_cr3: #GP, reserved bits\n");
 360                                 kvm_inject_gp(vcpu, 0);
 361                                 return;
 362                         }
 363                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 364                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 365                                        "reserved bits\n");
 366                                 kvm_inject_gp(vcpu, 0);
 367                                 return;
 368                         }
 369                 }
 370                 /*
 371                  * We don't check reserved bits in nonpae mode, because
 372                  * this isn't enforced, and VMware depends on this.
 373                  */
 374         }
 375
 376         /*
 377          * Does the new cr3 value map to physical memory? (Note, we
 378          * catch an invalid cr3 even in real-mode, because it would
 379          * cause trouble later on when we turn on paging anyway.)
 380          *
 381          * A real CPU would silently accept an invalid cr3 and would
 382          * attempt to use it - with largely undefined (and often hard
 383          * to debug) behavior on the guest side.
 384          */
 385         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 386                 kvm_inject_gp(vcpu, 0);
 387         else {
 388                 vcpu->arch.cr3 = cr3;
 389                 vcpu->arch.mmu.new_cr3(vcpu);
 390         }
 391 }
 392 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 393
 394 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 395 {
 396         if (cr8 & CR8_RESERVED_BITS) {
 397                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 398                 kvm_inject_gp(vcpu, 0);
 399                 return;
 400         }
 401         if (irqchip_in_kernel(vcpu->kvm))
 402                 kvm_lapic_set_tpr(vcpu, cr8);
 403         else
 404                 vcpu->arch.cr8 = cr8;
 405 }
 406 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 407
 408 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 409 {
 410         if (irqchip_in_kernel(vcpu->kvm))
 411                 return kvm_lapic_get_cr8(vcpu);
 412         else
 413                 return vcpu->arch.cr8;
 414 }
 415 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 416
 417 /*
 418  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 419  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 420  *
 421  * This list is modified at module load time to reflect the
 422  * capabilities of the host cpu.
 423  */
 424 static u32 msrs_to_save[] = {
 425         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 426         MSR_K6_STAR,
 427 #ifdef CONFIG_X86_64
 428         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 429 #endif
 430         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 431         MSR_IA32_PERF_STATUS,
 432 };
 433
 434 static unsigned num_msrs_to_save;
 435
 436 static u32 emulated_msrs[] = {
 437         MSR_IA32_MISC_ENABLE,
 438 };
 439
 440 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 441 {
 442         if (efer & efer_reserved_bits) {
 443                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 444                        efer);
 445                 kvm_inject_gp(vcpu, 0);
 446                 return;
 447         }
 448
 449         if (is_paging(vcpu)
 450             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 451                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 452                 kvm_inject_gp(vcpu, 0);
 453                 return;
 454         }
 455
 456         kvm_x86_ops->set_efer(vcpu, efer);
 457
 458         efer &= ~EFER_LMA;
 459         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 460
 461         vcpu->arch.shadow_efer = efer;
 462 }
 463
 464 void kvm_enable_efer_bits(u64 mask)
 465 {
 466        efer_reserved_bits &= ~mask;
 467 }
 468 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 469
 470
 471 /*
 472  * Writes msr value into into the appropriate "register".
 473  * Returns 0 on success, non-0 otherwise.
 474  * Assumes vcpu_load() was already called.
 475  */
 476 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 477 {
 478         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 479 }
 480
 481 /*
 482  * Adapt set_msr() to msr_io()'s calling convention
 483  */
 484 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 485 {
 486         return kvm_set_msr(vcpu, index, *data);
 487 }
 488
 489 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 490 {
 491         static int version;
 492         struct kvm_wall_clock wc;
 493         struct timespec wc_ts;
 494
 495         if (!wall_clock)
 496                 return;
 497
 498         version++;
 499
 500         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 501
 502         wc_ts = current_kernel_time();
 503         wc.wc_sec = wc_ts.tv_sec;
 504         wc.wc_nsec = wc_ts.tv_nsec;
 505         wc.wc_version = version;
 506
 507         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 508
 509         version++;
 510         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 511 }
 512
 513 static void kvm_write_guest_time(struct kvm_vcpu *v)
 514 {
 515         struct timespec ts;
 516         unsigned long flags;
 517         struct kvm_vcpu_arch *vcpu = &v->arch;
 518         void *shared_kaddr;
 519
 520         if ((!vcpu->time_page))
 521                 return;
 522
 523         /* Keep irq disabled to prevent changes to the clock */
 524         local_irq_save(flags);
 525         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 526                           &vcpu->hv_clock.tsc_timestamp);
 527         ktime_get_ts(&ts);
 528         local_irq_restore(flags);
 529
 530         /* With all the info we got, fill in the values */
 531
 532         vcpu->hv_clock.system_time = ts.tv_nsec +
 533                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 534         /*
 535          * The interface expects us to write an even number signaling that the
 536          * update is finished. Since the guest won't see the intermediate
 537          * state, we just write "2" at the end
 538          */
 539         vcpu->hv_clock.version = 2;
 540
 541         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 542
 543         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 544                 sizeof(vcpu->hv_clock));
 545
 546         kunmap_atomic(shared_kaddr, KM_USER0);
 547
 548         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 549 }
 550
 551
 552 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 553 {
 554         switch (msr) {
 555         case MSR_EFER:
 556                 set_efer(vcpu, data);
 557                 break;
 558         case MSR_IA32_MC0_STATUS:
 559                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 560                        __func__, data);
 561                 break;
 562         case MSR_IA32_MCG_STATUS:
 563                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 564                         __func__, data);
 565                 break;
 566         case MSR_IA32_MCG_CTL:
 567                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 568                         __func__, data);
 569                 break;
 570         case MSR_IA32_UCODE_REV:
 571         case MSR_IA32_UCODE_WRITE:
 572         case 0x200 ... 0x2ff: /* MTRRs */
 573                 break;
 574         case MSR_IA32_APICBASE:
 575                 kvm_set_apic_base(vcpu, data);
 576                 break;
 577         case MSR_IA32_MISC_ENABLE:
 578                 vcpu->arch.ia32_misc_enable_msr = data;
 579                 break;
 580         case MSR_KVM_WALL_CLOCK:
 581                 vcpu->kvm->arch.wall_clock = data;
 582                 kvm_write_wall_clock(vcpu->kvm, data);
 583                 break;
 584         case MSR_KVM_SYSTEM_TIME: {
 585                 if (vcpu->arch.time_page) {
 586                         kvm_release_page_dirty(vcpu->arch.time_page);
 587                         vcpu->arch.time_page = NULL;
 588                 }
 589
 590                 vcpu->arch.time = data;
 591
 592                 /* we verify if the enable bit is set... */
 593                 if (!(data & 1))
 594                         break;
 595
 596                 /* ...but clean it before doing the actual write */
 597                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 598
 599                 vcpu->arch.hv_clock.tsc_to_system_mul =
 600                                         clocksource_khz2mult(tsc_khz, 22);
 601                 vcpu->arch.hv_clock.tsc_shift = 22;
 602
 603                 down_read(&current->mm->mmap_sem);
 604                 vcpu->arch.time_page =
 605                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 606                 up_read(&current->mm->mmap_sem);
 607
 608                 if (is_error_page(vcpu->arch.time_page)) {
 609                         kvm_release_page_clean(vcpu->arch.time_page);
 610                         vcpu->arch.time_page = NULL;
 611                 }
 612
 613                 kvm_write_guest_time(vcpu);
 614                 break;
 615         }
 616         default:
 617                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 618                 return 1;
 619         }
 620         return 0;
 621 }
 622 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 623
 624
 625 /*
 626  * Reads an msr value (of 'msr_index') into 'pdata'.
 627  * Returns 0 on success, non-0 otherwise.
 628  * Assumes vcpu_load() was already called.
 629  */
 630 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 631 {
 632         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 633 }
 634
 635 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 636 {
 637         u64 data;
 638
 639         switch (msr) {
 640         case 0xc0010010: /* SYSCFG */
 641         case 0xc0010015: /* HWCR */
 642         case MSR_IA32_PLATFORM_ID:
 643         case MSR_IA32_P5_MC_ADDR:
 644         case MSR_IA32_P5_MC_TYPE:
 645         case MSR_IA32_MC0_CTL:
 646         case MSR_IA32_MCG_STATUS:
 647         case MSR_IA32_MCG_CAP:
 648         case MSR_IA32_MCG_CTL:
 649         case MSR_IA32_MC0_MISC:
 650         case MSR_IA32_MC0_MISC+4:
 651         case MSR_IA32_MC0_MISC+8:
 652         case MSR_IA32_MC0_MISC+12:
 653         case MSR_IA32_MC0_MISC+16:
 654         case MSR_IA32_UCODE_REV:
 655         case MSR_IA32_EBL_CR_POWERON:
 656                 /* MTRR registers */
 657         case 0xfe:
 658         case 0x200 ... 0x2ff:
 659                 data = 0;
 660                 break;
 661         case 0xcd: /* fsb frequency */
 662                 data = 3;
 663                 break;
 664         case MSR_IA32_APICBASE:
 665                 data = kvm_get_apic_base(vcpu);
 666                 break;
 667         case MSR_IA32_MISC_ENABLE:
 668                 data = vcpu->arch.ia32_misc_enable_msr;
 669                 break;
 670         case MSR_IA32_PERF_STATUS:
 671                 /* TSC increment by tick */
 672                 data = 1000ULL;
 673                 /* CPU multiplier */
 674                 data |= (((uint64_t)4ULL) << 40);
 675                 break;
 676         case MSR_EFER:
 677                 data = vcpu->arch.shadow_efer;
 678                 break;
 679         case MSR_KVM_WALL_CLOCK:
 680                 data = vcpu->kvm->arch.wall_clock;
 681                 break;
 682         case MSR_KVM_SYSTEM_TIME:
 683                 data = vcpu->arch.time;
 684                 break;
 685         default:
 686                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 687                 return 1;
 688         }
 689         *pdata = data;
 690         return 0;
 691 }
 692 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 693
 694 /*
 695  * Read or write a bunch of msrs. All parameters are kernel addresses.
 696  *
 697  * @return number of msrs set successfully.
 698  */
 699 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 700                     struct kvm_msr_entry *entries,
 701                     int (*do_msr)(struct kvm_vcpu *vcpu,
 702                                   unsigned index, u64 *data))
 703 {
 704         int i;
 705
 706         vcpu_load(vcpu);
 707
 708         down_read(&vcpu->kvm->slots_lock);
 709         for (i = 0; i < msrs->nmsrs; ++i)
 710                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 711                         break;
 712         up_read(&vcpu->kvm->slots_lock);
 713
 714         vcpu_put(vcpu);
 715
 716         return i;
 717 }
 718
 719 /*
 720  * Read or write a bunch of msrs. Parameters are user addresses.
 721  *
 722  * @return number of msrs set successfully.
 723  */
 724 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 725                   int (*do_msr)(struct kvm_vcpu *vcpu,
 726                                 unsigned index, u64 *data),
 727                   int writeback)
 728 {
 729         struct kvm_msrs msrs;
 730         struct kvm_msr_entry *entries;
 731         int r, n;
 732         unsigned size;
 733
 734         r = -EFAULT;
 735         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 736                 goto out;
 737
 738         r = -E2BIG;
 739         if (msrs.nmsrs >= MAX_IO_MSRS)
 740                 goto out;
 741
 742         r = -ENOMEM;
 743         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 744         entries = vmalloc(size);
 745         if (!entries)
 746                 goto out;
 747
 748         r = -EFAULT;
 749         if (copy_from_user(entries, user_msrs->entries, size))
 750                 goto out_free;
 751
 752         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 753         if (r < 0)
 754                 goto out_free;
 755
 756         r = -EFAULT;
 757         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 758                 goto out_free;
 759
 760         r = n;
 761
 762 out_free:
 763         vfree(entries);
 764 out:
 765         return r;
 766 }
 767
 768 /*
 769  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 770  * cached on it.
 771  */
 772 void decache_vcpus_on_cpu(int cpu)
 773 {
 774         struct kvm *vm;
 775         struct kvm_vcpu *vcpu;
 776         int i;
 777
 778         spin_lock(&kvm_lock);
 779         list_for_each_entry(vm, &vm_list, vm_list)
 780                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 781                         vcpu = vm->vcpus[i];
 782                         if (!vcpu)
 783                                 continue;
 784                         /*
 785                          * If the vcpu is locked, then it is running on some
 786                          * other cpu and therefore it is not cached on the
 787                          * cpu in question.
 788                          *
 789                          * If it's not locked, check the last cpu it executed
 790                          * on.
 791                          */
 792                         if (mutex_trylock(&vcpu->mutex)) {
 793                                 if (vcpu->cpu == cpu) {
 794                                         kvm_x86_ops->vcpu_decache(vcpu);
 795                                         vcpu->cpu = -1;
 796                                 }
 797                                 mutex_unlock(&vcpu->mutex);
 798                         }
 799                 }
 800         spin_unlock(&kvm_lock);
 801 }
 802
 803 int kvm_dev_ioctl_check_extension(long ext)
 804 {
 805         int r;
 806
 807         switch (ext) {
 808         case KVM_CAP_IRQCHIP:
 809         case KVM_CAP_HLT:
 810         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 811         case KVM_CAP_USER_MEMORY:
 812         case KVM_CAP_SET_TSS_ADDR:
 813         case KVM_CAP_EXT_CPUID:
 814         case KVM_CAP_CLOCKSOURCE:
 815         case KVM_CAP_PIT:
 816         case KVM_CAP_NOP_IO_DELAY:
 817                 r = 1;
 818                 break;
 819         case KVM_CAP_VAPIC:
 820                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 821                 break;
 822         case KVM_CAP_NR_VCPUS:
 823                 r = KVM_MAX_VCPUS;
 824                 break;
 825         case KVM_CAP_NR_MEMSLOTS:
 826                 r = KVM_MEMORY_SLOTS;
 827                 break;
 828         case KVM_CAP_PV_MMU:
 829                 r = !tdp_enabled;
 830                 break;
 831         default:
 832                 r = 0;
 833                 break;
 834         }
 835         return r;
 836
 837 }
 838
 839 long kvm_arch_dev_ioctl(struct file *filp,
 840                         unsigned int ioctl, unsigned long arg)
 841 {
 842         void __user *argp = (void __user *)arg;
 843         long r;
 844
 845         switch (ioctl) {
 846         case KVM_GET_MSR_INDEX_LIST: {
 847                 struct kvm_msr_list __user *user_msr_list = argp;
 848                 struct kvm_msr_list msr_list;
 849                 unsigned n;
 850
 851                 r = -EFAULT;
 852                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 853                         goto out;
 854                 n = msr_list.nmsrs;
 855                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 856                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 857                         goto out;
 858                 r = -E2BIG;
 859                 if (n < num_msrs_to_save)
 860                         goto out;
 861                 r = -EFAULT;
 862                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 863                                  num_msrs_to_save * sizeof(u32)))
 864                         goto out;
 865                 if (copy_to_user(user_msr_list->indices
 866                                  + num_msrs_to_save * sizeof(u32),
 867                                  &emulated_msrs,
 868                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 869                         goto out;
 870                 r = 0;
 871                 break;
 872         }
 873         case KVM_GET_SUPPORTED_CPUID: {
 874                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 875                 struct kvm_cpuid2 cpuid;
 876
 877                 r = -EFAULT;
 878                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 879                         goto out;
 880                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 881                         cpuid_arg->entries);
 882                 if (r)
 883                         goto out;
 884
 885                 r = -EFAULT;
 886                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 887                         goto out;
 888                 r = 0;
 889                 break;
 890         }
 891         default:
 892                 r = -EINVAL;
 893         }
 894 out:
 895         return r;
 896 }
 897
 898 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 899 {
 900         kvm_x86_ops->vcpu_load(vcpu, cpu);
 901         kvm_write_guest_time(vcpu);
 902 }
 903
 904 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 905 {
 906         kvm_x86_ops->vcpu_put(vcpu);
 907         kvm_put_guest_fpu(vcpu);
 908 }
 909
 910 static int is_efer_nx(void)
 911 {
 912         u64 efer;
 913
 914         rdmsrl(MSR_EFER, efer);
 915         return efer & EFER_NX;
 916 }
 917
 918 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 919 {
 920         int i;
 921         struct kvm_cpuid_entry2 *e, *entry;
 922
 923         entry = NULL;
 924         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 925                 e = &vcpu->arch.cpuid_entries[i];
 926                 if (e->function == 0x80000001) {
 927                         entry = e;
 928                         break;
 929                 }
 930         }
 931         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 932                 entry->edx &= ~(1 << 20);
 933                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 934         }
 935 }
 936
 937 /* when an old userspace process fills a new kernel module */
 938 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 939                                     struct kvm_cpuid *cpuid,
 940                                     struct kvm_cpuid_entry __user *entries)
 941 {
 942         int r, i;
 943         struct kvm_cpuid_entry *cpuid_entries;
 944
 945         r = -E2BIG;
 946         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 947                 goto out;
 948         r = -ENOMEM;
 949         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 950         if (!cpuid_entries)
 951                 goto out;
 952         r = -EFAULT;
 953         if (copy_from_user(cpuid_entries, entries,
 954                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 955                 goto out_free;
 956         for (i = 0; i < cpuid->nent; i++) {
 957                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 958                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 959                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 960                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 961                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 962                 vcpu->arch.cpuid_entries[i].index = 0;
 963                 vcpu->arch.cpuid_entries[i].flags = 0;
 964                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 965                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 966                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 967         }
 968         vcpu->arch.cpuid_nent = cpuid->nent;
 969         cpuid_fix_nx_cap(vcpu);
 970         r = 0;
 971
 972 out_free:
 973         vfree(cpuid_entries);
 974 out:
 975         return r;
 976 }
 977
 978 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 979                                     struct kvm_cpuid2 *cpuid,
 980                                     struct kvm_cpuid_entry2 __user *entries)
 981 {
 982         int r;
 983
 984         r = -E2BIG;
 985         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 986                 goto out;
 987         r = -EFAULT;
 988         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 989                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 990                 goto out;
 991         vcpu->arch.cpuid_nent = cpuid->nent;
 992         return 0;
 993
 994 out:
 995         return r;
 996 }
 997
 998 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 999                                     struct kvm_cpuid2 *cpuid,
1000                                     struct kvm_cpuid_entry2 __user *entries)
1001 {
1002         int r;
1003
1004         r = -E2BIG;
1005         if (cpuid->nent < vcpu->arch.cpuid_nent)
1006                 goto out;
1007         r = -EFAULT;
1008         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1009                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1010                 goto out;
1011         return 0;
1012
1013 out:
1014         cpuid->nent = vcpu->arch.cpuid_nent;
1015         return r;
1016 }
1017
1018 static inline u32 bit(int bitno)
1019 {
1020         return 1 << (bitno & 31);
1021 }
1022
1023 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1024                           u32 index)
1025 {
1026         entry->function = function;
1027         entry->index = index;
1028         cpuid_count(entry->function, entry->index,
1029                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1030         entry->flags = 0;
1031 }
1032
1033 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1034                          u32 index, int *nent, int maxnent)
1035 {
1036         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1037                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1038                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1039                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1040                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1041                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1042                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1043                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1044                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1045                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1046         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1047                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1048                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1049                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1050                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1051                 bit(X86_FEATURE_PGE) |
1052                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1053                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1054                 bit(X86_FEATURE_SYSCALL) |
1055                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1056 #ifdef CONFIG_X86_64
1057                 bit(X86_FEATURE_LM) |
1058 #endif
1059                 bit(X86_FEATURE_MMXEXT) |
1060                 bit(X86_FEATURE_3DNOWEXT) |
1061                 bit(X86_FEATURE_3DNOW);
1062         const u32 kvm_supported_word3_x86_features =
1063                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1064         const u32 kvm_supported_word6_x86_features =
1065                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1066
1067         /* all func 2 cpuid_count() should be called on the same cpu */
1068         get_cpu();
1069         do_cpuid_1_ent(entry, function, index);
1070         ++*nent;
1071
1072         switch (function) {
1073         case 0:
1074                 entry->eax = min(entry->eax, (u32)0xb);
1075                 break;
1076         case 1:
1077                 entry->edx &= kvm_supported_word0_x86_features;
1078                 entry->ecx &= kvm_supported_word3_x86_features;
1079                 break;
1080         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1081          * may return different values. This forces us to get_cpu() before
1082          * issuing the first command, and also to emulate this annoying behavior
1083          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1084         case 2: {
1085                 int t, times = entry->eax & 0xff;
1086
1087                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1088                 for (t = 1; t < times && *nent < maxnent; ++t) {
1089                         do_cpuid_1_ent(&entry[t], function, 0);
1090                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1091                         ++*nent;
1092                 }
1093                 break;
1094         }
1095         /* function 4 and 0xb have additional index. */
1096         case 4: {
1097                 int i, cache_type;
1098
1099                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1100                 /* read more entries until cache_type is zero */
1101                 for (i = 1; *nent < maxnent; ++i) {
1102                         cache_type = entry[i - 1].eax & 0x1f;
1103                         if (!cache_type)
1104                                 break;
1105                         do_cpuid_1_ent(&entry[i], function, i);
1106                         entry[i].flags |=
1107                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1108                         ++*nent;
1109                 }
1110                 break;
1111         }
1112         case 0xb: {
1113                 int i, level_type;
1114
1115                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1116                 /* read more entries until level_type is zero */
1117                 for (i = 1; *nent < maxnent; ++i) {
1118                         level_type = entry[i - 1].ecx & 0xff;
1119                         if (!level_type)
1120                                 break;
1121                         do_cpuid_1_ent(&entry[i], function, i);
1122                         entry[i].flags |=
1123                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1124                         ++*nent;
1125                 }
1126                 break;
1127         }
1128         case 0x80000000:
1129                 entry->eax = min(entry->eax, 0x8000001a);
1130                 break;
1131         case 0x80000001:
1132                 entry->edx &= kvm_supported_word1_x86_features;
1133                 entry->ecx &= kvm_supported_word6_x86_features;
1134                 break;
1135         }
1136         put_cpu();
1137 }
1138
1139 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1140                                     struct kvm_cpuid_entry2 __user *entries)
1141 {
1142         struct kvm_cpuid_entry2 *cpuid_entries;
1143         int limit, nent = 0, r = -E2BIG;
1144         u32 func;
1145
1146         if (cpuid->nent < 1)
1147                 goto out;
1148         r = -ENOMEM;
1149         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1150         if (!cpuid_entries)
1151                 goto out;
1152
1153         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1154         limit = cpuid_entries[0].eax;
1155         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1156                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1157                                 &nent, cpuid->nent);
1158         r = -E2BIG;
1159         if (nent >= cpuid->nent)
1160                 goto out_free;
1161
1162         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1163         limit = cpuid_entries[nent - 1].eax;
1164         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1165                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1166                                &nent, cpuid->nent);
1167         r = -EFAULT;
1168         if (copy_to_user(entries, cpuid_entries,
1169                         nent * sizeof(struct kvm_cpuid_entry2)))
1170                 goto out_free;
1171         cpuid->nent = nent;
1172         r = 0;
1173
1174 out_free:
1175         vfree(cpuid_entries);
1176 out:
1177         return r;
1178 }
1179
1180 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1181                                     struct kvm_lapic_state *s)
1182 {
1183         vcpu_load(vcpu);
1184         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1185         vcpu_put(vcpu);
1186
1187         return 0;
1188 }
1189
1190 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1191                                     struct kvm_lapic_state *s)
1192 {
1193         vcpu_load(vcpu);
1194         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1195         kvm_apic_post_state_restore(vcpu);
1196         vcpu_put(vcpu);
1197
1198         return 0;
1199 }
1200
1201 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1202                                     struct kvm_interrupt *irq)
1203 {
1204         if (irq->irq < 0 || irq->irq >= 256)
1205                 return -EINVAL;
1206         if (irqchip_in_kernel(vcpu->kvm))
1207                 return -ENXIO;
1208         vcpu_load(vcpu);
1209
1210         set_bit(irq->irq, vcpu->arch.irq_pending);
1211         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1212
1213         vcpu_put(vcpu);
1214
1215         return 0;
1216 }
1217
1218 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1219                                            struct kvm_tpr_access_ctl *tac)
1220 {
1221         if (tac->flags)
1222                 return -EINVAL;
1223         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1224         return 0;
1225 }
1226
1227 long kvm_arch_vcpu_ioctl(struct file *filp,
1228                          unsigned int ioctl, unsigned long arg)
1229 {
1230         struct kvm_vcpu *vcpu = filp->private_data;
1231         void __user *argp = (void __user *)arg;
1232         int r;
1233
1234         switch (ioctl) {
1235         case KVM_GET_LAPIC: {
1236                 struct kvm_lapic_state lapic;
1237
1238                 memset(&lapic, 0, sizeof lapic);
1239                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1240                 if (r)
1241                         goto out;
1242                 r = -EFAULT;
1243                 if (copy_to_user(argp, &lapic, sizeof lapic))
1244                         goto out;
1245                 r = 0;
1246                 break;
1247         }
1248         case KVM_SET_LAPIC: {
1249                 struct kvm_lapic_state lapic;
1250
1251                 r = -EFAULT;
1252                 if (copy_from_user(&lapic, argp, sizeof lapic))
1253                         goto out;
1254                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1255                 if (r)
1256                         goto out;
1257                 r = 0;
1258                 break;
1259         }
1260         case KVM_INTERRUPT: {
1261                 struct kvm_interrupt irq;
1262
1263                 r = -EFAULT;
1264                 if (copy_from_user(&irq, argp, sizeof irq))
1265                         goto out;
1266                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1267                 if (r)
1268                         goto out;
1269                 r = 0;
1270                 break;
1271         }
1272         case KVM_SET_CPUID: {
1273                 struct kvm_cpuid __user *cpuid_arg = argp;
1274                 struct kvm_cpuid cpuid;
1275
1276                 r = -EFAULT;
1277                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1278                         goto out;
1279                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1280                 if (r)
1281                         goto out;
1282                 break;
1283         }
1284         case KVM_SET_CPUID2: {
1285                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1286                 struct kvm_cpuid2 cpuid;
1287
1288                 r = -EFAULT;
1289                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1290                         goto out;
1291                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1292                                 cpuid_arg->entries);
1293                 if (r)
1294                         goto out;
1295                 break;
1296         }
1297         case KVM_GET_CPUID2: {
1298                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1299                 struct kvm_cpuid2 cpuid;
1300
1301                 r = -EFAULT;
1302                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1303                         goto out;
1304                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1305                                 cpuid_arg->entries);
1306                 if (r)
1307                         goto out;
1308                 r = -EFAULT;
1309                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1310                         goto out;
1311                 r = 0;
1312                 break;
1313         }
1314         case KVM_GET_MSRS:
1315                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1316                 break;
1317         case KVM_SET_MSRS:
1318                 r = msr_io(vcpu, argp, do_set_msr, 0);
1319                 break;
1320         case KVM_TPR_ACCESS_REPORTING: {
1321                 struct kvm_tpr_access_ctl tac;
1322
1323                 r = -EFAULT;
1324                 if (copy_from_user(&tac, argp, sizeof tac))
1325                         goto out;
1326                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1327                 if (r)
1328                         goto out;
1329                 r = -EFAULT;
1330                 if (copy_to_user(argp, &tac, sizeof tac))
1331                         goto out;
1332                 r = 0;
1333                 break;
1334         };
1335         case KVM_SET_VAPIC_ADDR: {
1336                 struct kvm_vapic_addr va;
1337
1338                 r = -EINVAL;
1339                 if (!irqchip_in_kernel(vcpu->kvm))
1340                         goto out;
1341                 r = -EFAULT;
1342                 if (copy_from_user(&va, argp, sizeof va))
1343                         goto out;
1344                 r = 0;
1345                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1346                 break;
1347         }
1348         default:
1349                 r = -EINVAL;
1350         }
1351 out:
1352         return r;
1353 }
1354
1355 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1356 {
1357         int ret;
1358
1359         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1360                 return -1;
1361         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1362         return ret;
1363 }
1364
1365 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1366                                           u32 kvm_nr_mmu_pages)
1367 {
1368         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1369                 return -EINVAL;
1370
1371         down_write(&kvm->slots_lock);
1372
1373         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1374         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1375
1376         up_write(&kvm->slots_lock);
1377         return 0;
1378 }
1379
1380 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1381 {
1382         return kvm->arch.n_alloc_mmu_pages;
1383 }
1384
1385 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1386 {
1387         int i;
1388         struct kvm_mem_alias *alias;
1389
1390         for (i = 0; i < kvm->arch.naliases; ++i) {
1391                 alias = &kvm->arch.aliases[i];
1392                 if (gfn >= alias->base_gfn
1393                     && gfn < alias->base_gfn + alias->npages)
1394                         return alias->target_gfn + gfn - alias->base_gfn;
1395         }
1396         return gfn;
1397 }
1398
1399 /*
1400  * Set a new alias region.  Aliases map a portion of physical memory into
1401  * another portion.  This is useful for memory windows, for example the PC
1402  * VGA region.
1403  */
1404 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1405                                          struct kvm_memory_alias *alias)
1406 {
1407         int r, n;
1408         struct kvm_mem_alias *p;
1409
1410         r = -EINVAL;
1411         /* General sanity checks */
1412         if (alias->memory_size & (PAGE_SIZE - 1))
1413                 goto out;
1414         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1415                 goto out;
1416         if (alias->slot >= KVM_ALIAS_SLOTS)
1417                 goto out;
1418         if (alias->guest_phys_addr + alias->memory_size
1419             < alias->guest_phys_addr)
1420                 goto out;
1421         if (alias->target_phys_addr + alias->memory_size
1422             < alias->target_phys_addr)
1423                 goto out;
1424
1425         down_write(&kvm->slots_lock);
1426
1427         p = &kvm->arch.aliases[alias->slot];
1428         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1429         p->npages = alias->memory_size >> PAGE_SHIFT;
1430         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1431
1432         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1433                 if (kvm->arch.aliases[n - 1].npages)
1434                         break;
1435         kvm->arch.naliases = n;
1436
1437         kvm_mmu_zap_all(kvm);
1438
1439         up_write(&kvm->slots_lock);
1440
1441         return 0;
1442
1443 out:
1444         return r;
1445 }
1446
1447 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1448 {
1449         int r;
1450
1451         r = 0;
1452         switch (chip->chip_id) {
1453         case KVM_IRQCHIP_PIC_MASTER:
1454                 memcpy(&chip->chip.pic,
1455                         &pic_irqchip(kvm)->pics[0],
1456                         sizeof(struct kvm_pic_state));
1457                 break;
1458         case KVM_IRQCHIP_PIC_SLAVE:
1459                 memcpy(&chip->chip.pic,
1460                         &pic_irqchip(kvm)->pics[1],
1461                         sizeof(struct kvm_pic_state));
1462                 break;
1463         case KVM_IRQCHIP_IOAPIC:
1464                 memcpy(&chip->chip.ioapic,
1465                         ioapic_irqchip(kvm),
1466                         sizeof(struct kvm_ioapic_state));
1467                 break;
1468         default:
1469                 r = -EINVAL;
1470                 break;
1471         }
1472         return r;
1473 }
1474
1475 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1476 {
1477         int r;
1478
1479         r = 0;
1480         switch (chip->chip_id) {
1481         case KVM_IRQCHIP_PIC_MASTER:
1482                 memcpy(&pic_irqchip(kvm)->pics[0],
1483                         &chip->chip.pic,
1484                         sizeof(struct kvm_pic_state));
1485                 break;
1486         case KVM_IRQCHIP_PIC_SLAVE:
1487                 memcpy(&pic_irqchip(kvm)->pics[1],
1488                         &chip->chip.pic,
1489                         sizeof(struct kvm_pic_state));
1490                 break;
1491         case KVM_IRQCHIP_IOAPIC:
1492                 memcpy(ioapic_irqchip(kvm),
1493                         &chip->chip.ioapic,
1494                         sizeof(struct kvm_ioapic_state));
1495                 break;
1496         default:
1497                 r = -EINVAL;
1498                 break;
1499         }
1500         kvm_pic_update_irq(pic_irqchip(kvm));
1501         return r;
1502 }
1503
1504 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1505 {
1506         int r = 0;
1507
1508         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1509         return r;
1510 }
1511
1512 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1513 {
1514         int r = 0;
1515
1516         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1517         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1518         return r;
1519 }
1520
1521 /*
1522  * Get (and clear) the dirty memory log for a memory slot.
1523  */
1524 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1525                                       struct kvm_dirty_log *log)
1526 {
1527         int r;
1528         int n;
1529         struct kvm_memory_slot *memslot;
1530         int is_dirty = 0;
1531
1532         down_write(&kvm->slots_lock);
1533
1534         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1535         if (r)
1536                 goto out;
1537
1538         /* If nothing is dirty, don't bother messing with page tables. */
1539         if (is_dirty) {
1540                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1541                 kvm_flush_remote_tlbs(kvm);
1542                 memslot = &kvm->memslots[log->slot];
1543                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1544                 memset(memslot->dirty_bitmap, 0, n);
1545         }
1546         r = 0;
1547 out:
1548         up_write(&kvm->slots_lock);
1549         return r;
1550 }
1551
1552 long kvm_arch_vm_ioctl(struct file *filp,
1553                        unsigned int ioctl, unsigned long arg)
1554 {
1555         struct kvm *kvm = filp->private_data;
1556         void __user *argp = (void __user *)arg;
1557         int r = -EINVAL;
1558
1559         switch (ioctl) {
1560         case KVM_SET_TSS_ADDR:
1561                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1562                 if (r < 0)
1563                         goto out;
1564                 break;
1565         case KVM_SET_MEMORY_REGION: {
1566                 struct kvm_memory_region kvm_mem;
1567                 struct kvm_userspace_memory_region kvm_userspace_mem;
1568
1569                 r = -EFAULT;
1570                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1571                         goto out;
1572                 kvm_userspace_mem.slot = kvm_mem.slot;
1573                 kvm_userspace_mem.flags = kvm_mem.flags;
1574                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1575                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1576                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1577                 if (r)
1578                         goto out;
1579                 break;
1580         }
1581         case KVM_SET_NR_MMU_PAGES:
1582                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1583                 if (r)
1584                         goto out;
1585                 break;
1586         case KVM_GET_NR_MMU_PAGES:
1587                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1588                 break;
1589         case KVM_SET_MEMORY_ALIAS: {
1590                 struct kvm_memory_alias alias;
1591
1592                 r = -EFAULT;
1593                 if (copy_from_user(&alias, argp, sizeof alias))
1594                         goto out;
1595                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1596                 if (r)
1597                         goto out;
1598                 break;
1599         }
1600         case KVM_CREATE_IRQCHIP:
1601                 r = -ENOMEM;
1602                 kvm->arch.vpic = kvm_create_pic(kvm);
1603                 if (kvm->arch.vpic) {
1604                         r = kvm_ioapic_init(kvm);
1605                         if (r) {
1606                                 kfree(kvm->arch.vpic);
1607                                 kvm->arch.vpic = NULL;
1608                                 goto out;
1609                         }
1610                 } else
1611                         goto out;
1612                 break;
1613         case KVM_CREATE_PIT:
1614                 r = -ENOMEM;
1615                 kvm->arch.vpit = kvm_create_pit(kvm);
1616                 if (kvm->arch.vpit)
1617                         r = 0;
1618                 break;
1619         case KVM_IRQ_LINE: {
1620                 struct kvm_irq_level irq_event;
1621
1622                 r = -EFAULT;
1623                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1624                         goto out;
1625                 if (irqchip_in_kernel(kvm)) {
1626                         mutex_lock(&kvm->lock);
1627                         if (irq_event.irq < 16)
1628                                 kvm_pic_set_irq(pic_irqchip(kvm),
1629                                         irq_event.irq,
1630                                         irq_event.level);
1631                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1632                                         irq_event.irq,
1633                                         irq_event.level);
1634                         mutex_unlock(&kvm->lock);
1635                         r = 0;
1636                 }
1637                 break;
1638         }
1639         case KVM_GET_IRQCHIP: {
1640                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1641                 struct kvm_irqchip chip;
1642
1643                 r = -EFAULT;
1644                 if (copy_from_user(&chip, argp, sizeof chip))
1645                         goto out;
1646                 r = -ENXIO;
1647                 if (!irqchip_in_kernel(kvm))
1648                         goto out;
1649                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1650                 if (r)
1651                         goto out;
1652                 r = -EFAULT;
1653                 if (copy_to_user(argp, &chip, sizeof chip))
1654                         goto out;
1655                 r = 0;
1656                 break;
1657         }
1658         case KVM_SET_IRQCHIP: {
1659                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1660                 struct kvm_irqchip chip;
1661
1662                 r = -EFAULT;
1663                 if (copy_from_user(&chip, argp, sizeof chip))
1664                         goto out;
1665                 r = -ENXIO;
1666                 if (!irqchip_in_kernel(kvm))
1667                         goto out;
1668                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1669                 if (r)
1670                         goto out;
1671                 r = 0;
1672                 break;
1673         }
1674         case KVM_GET_PIT: {
1675                 struct kvm_pit_state ps;
1676                 r = -EFAULT;
1677                 if (copy_from_user(&ps, argp, sizeof ps))
1678                         goto out;
1679                 r = -ENXIO;
1680                 if (!kvm->arch.vpit)
1681                         goto out;
1682                 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1683                 if (r)
1684                         goto out;
1685                 r = -EFAULT;
1686                 if (copy_to_user(argp, &ps, sizeof ps))
1687                         goto out;
1688                 r = 0;
1689                 break;
1690         }
1691         case KVM_SET_PIT: {
1692                 struct kvm_pit_state ps;
1693                 r = -EFAULT;
1694                 if (copy_from_user(&ps, argp, sizeof ps))
1695                         goto out;
1696                 r = -ENXIO;
1697                 if (!kvm->arch.vpit)
1698                         goto out;
1699                 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1700                 if (r)
1701                         goto out;
1702                 r = 0;
1703                 break;
1704         }
1705         default:
1706                 ;
1707         }
1708 out:
1709         return r;
1710 }
1711
1712 static void kvm_init_msr_list(void)
1713 {
1714         u32 dummy[2];
1715         unsigned i, j;
1716
1717         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1718                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1719                         continue;
1720                 if (j < i)
1721                         msrs_to_save[j] = msrs_to_save[i];
1722                 j++;
1723         }
1724         num_msrs_to_save = j;
1725 }
1726
1727 /*
1728  * Only apic need an MMIO device hook, so shortcut now..
1729  */
1730 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1731                                                 gpa_t addr)
1732 {
1733         struct kvm_io_device *dev;
1734
1735         if (vcpu->arch.apic) {
1736                 dev = &vcpu->arch.apic->dev;
1737                 if (dev->in_range(dev, addr))
1738                         return dev;
1739         }
1740         return NULL;
1741 }
1742
1743
1744 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1745                                                 gpa_t addr)
1746 {
1747         struct kvm_io_device *dev;
1748
1749         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1750         if (dev == NULL)
1751                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1752         return dev;
1753 }
1754
1755 int emulator_read_std(unsigned long addr,
1756                              void *val,
1757                              unsigned int bytes,
1758                              struct kvm_vcpu *vcpu)
1759 {
1760         void *data = val;
1761         int r = X86EMUL_CONTINUE;
1762
1763         while (bytes) {
1764                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1765                 unsigned offset = addr & (PAGE_SIZE-1);
1766                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1767                 int ret;
1768
1769                 if (gpa == UNMAPPED_GVA) {
1770                         r = X86EMUL_PROPAGATE_FAULT;
1771                         goto out;
1772                 }
1773                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1774                 if (ret < 0) {
1775                         r = X86EMUL_UNHANDLEABLE;
1776                         goto out;
1777                 }
1778
1779                 bytes -= tocopy;
1780                 data += tocopy;
1781                 addr += tocopy;
1782         }
1783 out:
1784         return r;
1785 }
1786 EXPORT_SYMBOL_GPL(emulator_read_std);
1787
1788 static int emulator_read_emulated(unsigned long addr,
1789                                   void *val,
1790                                   unsigned int bytes,
1791                                   struct kvm_vcpu *vcpu)
1792 {
1793         struct kvm_io_device *mmio_dev;
1794         gpa_t                 gpa;
1795
1796         if (vcpu->mmio_read_completed) {
1797                 memcpy(val, vcpu->mmio_data, bytes);
1798                 vcpu->mmio_read_completed = 0;
1799                 return X86EMUL_CONTINUE;
1800         }
1801
1802         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1803
1804         /* For APIC access vmexit */
1805         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1806                 goto mmio;
1807
1808         if (emulator_read_std(addr, val, bytes, vcpu)
1809                         == X86EMUL_CONTINUE)
1810                 return X86EMUL_CONTINUE;
1811         if (gpa == UNMAPPED_GVA)
1812                 return X86EMUL_PROPAGATE_FAULT;
1813
1814 mmio:
1815         /*
1816          * Is this MMIO handled locally?
1817          */
1818         mutex_lock(&vcpu->kvm->lock);
1819         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1820         if (mmio_dev) {
1821                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1822                 mutex_unlock(&vcpu->kvm->lock);
1823                 return X86EMUL_CONTINUE;
1824         }
1825         mutex_unlock(&vcpu->kvm->lock);
1826
1827         vcpu->mmio_needed = 1;
1828         vcpu->mmio_phys_addr = gpa;
1829         vcpu->mmio_size = bytes;
1830         vcpu->mmio_is_write = 0;
1831
1832         return X86EMUL_UNHANDLEABLE;
1833 }
1834
1835 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1836                           const void *val, int bytes)
1837 {
1838         int ret;
1839
1840         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1841         if (ret < 0)
1842                 return 0;
1843         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1844         return 1;
1845 }
1846
1847 static int emulator_write_emulated_onepage(unsigned long addr,
1848                                            const void *val,
1849                                            unsigned int bytes,
1850                                            struct kvm_vcpu *vcpu)
1851 {
1852         struct kvm_io_device *mmio_dev;
1853         gpa_t                 gpa;
1854
1855         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1856
1857         if (gpa == UNMAPPED_GVA) {
1858                 kvm_inject_page_fault(vcpu, addr, 2);
1859                 return X86EMUL_PROPAGATE_FAULT;
1860         }
1861
1862         /* For APIC access vmexit */
1863         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1864                 goto mmio;
1865
1866         if (emulator_write_phys(vcpu, gpa, val, bytes))
1867                 return X86EMUL_CONTINUE;
1868
1869 mmio:
1870         /*
1871          * Is this MMIO handled locally?
1872          */
1873         mutex_lock(&vcpu->kvm->lock);
1874         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1875         if (mmio_dev) {
1876                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1877                 mutex_unlock(&vcpu->kvm->lock);
1878                 return X86EMUL_CONTINUE;
1879         }
1880         mutex_unlock(&vcpu->kvm->lock);
1881
1882         vcpu->mmio_needed = 1;
1883         vcpu->mmio_phys_addr = gpa;
1884         vcpu->mmio_size = bytes;
1885         vcpu->mmio_is_write = 1;
1886         memcpy(vcpu->mmio_data, val, bytes);
1887
1888         return X86EMUL_CONTINUE;
1889 }
1890
1891 int emulator_write_emulated(unsigned long addr,
1892                                    const void *val,
1893                                    unsigned int bytes,
1894                                    struct kvm_vcpu *vcpu)
1895 {
1896         /* Crossing a page boundary? */
1897         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1898                 int rc, now;
1899
1900                 now = -addr & ~PAGE_MASK;
1901                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1902                 if (rc != X86EMUL_CONTINUE)
1903                         return rc;
1904                 addr += now;
1905                 val += now;
1906                 bytes -= now;
1907         }
1908         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1909 }
1910 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1911
1912 static int emulator_cmpxchg_emulated(unsigned long addr,
1913                                      const void *old,
1914                                      const void *new,
1915                                      unsigned int bytes,
1916                                      struct kvm_vcpu *vcpu)
1917 {
1918         static int reported;
1919
1920         if (!reported) {
1921                 reported = 1;
1922                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1923         }
1924 #ifndef CONFIG_X86_64
1925         /* guests cmpxchg8b have to be emulated atomically */
1926         if (bytes == 8) {
1927                 gpa_t gpa;
1928                 struct page *page;
1929                 char *kaddr;
1930                 u64 val;
1931
1932                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1933
1934                 if (gpa == UNMAPPED_GVA ||
1935                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1936                         goto emul_write;
1937
1938                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1939                         goto emul_write;
1940
1941                 val = *(u64 *)new;
1942
1943                 down_read(&current->mm->mmap_sem);
1944                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1945                 up_read(&current->mm->mmap_sem);
1946
1947                 kaddr = kmap_atomic(page, KM_USER0);
1948                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1949                 kunmap_atomic(kaddr, KM_USER0);
1950                 kvm_release_page_dirty(page);
1951         }
1952 emul_write:
1953 #endif
1954
1955         return emulator_write_emulated(addr, new, bytes, vcpu);
1956 }
1957
1958 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1959 {
1960         return kvm_x86_ops->get_segment_base(vcpu, seg);
1961 }
1962
1963 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1964 {
1965         return X86EMUL_CONTINUE;
1966 }
1967
1968 int emulate_clts(struct kvm_vcpu *vcpu)
1969 {
1970         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1971         return X86EMUL_CONTINUE;
1972 }
1973
1974 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1975 {
1976         struct kvm_vcpu *vcpu = ctxt->vcpu;
1977
1978         switch (dr) {
1979         case 0 ... 3:
1980                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1981                 return X86EMUL_CONTINUE;
1982         default:
1983                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
1984                 return X86EMUL_UNHANDLEABLE;
1985         }
1986 }
1987
1988 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1989 {
1990         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1991         int exception;
1992
1993         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1994         if (exception) {
1995                 /* FIXME: better handling */
1996                 return X86EMUL_UNHANDLEABLE;
1997         }
1998         return X86EMUL_CONTINUE;
1999 }
2000
2001 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2002 {
2003         static int reported;
2004         u8 opcodes[4];
2005         unsigned long rip = vcpu->arch.rip;
2006         unsigned long rip_linear;
2007
2008         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2009
2010         if (reported)
2011                 return;
2012
2013         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2014
2015         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2016                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2017         reported = 1;
2018 }
2019 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2020
2021 static struct x86_emulate_ops emulate_ops = {
2022         .read_std            = emulator_read_std,
2023         .read_emulated       = emulator_read_emulated,
2024         .write_emulated      = emulator_write_emulated,
2025         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2026 };
2027
2028 int emulate_instruction(struct kvm_vcpu *vcpu,
2029                         struct kvm_run *run,
2030                         unsigned long cr2,
2031                         u16 error_code,
2032                         int emulation_type)
2033 {
2034         int r;
2035         struct decode_cache *c;
2036
2037         vcpu->arch.mmio_fault_cr2 = cr2;
2038         kvm_x86_ops->cache_regs(vcpu);
2039
2040         vcpu->mmio_is_write = 0;
2041         vcpu->arch.pio.string = 0;
2042
2043         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2044                 int cs_db, cs_l;
2045                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2046
2047                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2048                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2049                 vcpu->arch.emulate_ctxt.mode =
2050                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2051                         ? X86EMUL_MODE_REAL : cs_l
2052                         ? X86EMUL_MODE_PROT64 : cs_db
2053                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2054
2055                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2056                         vcpu->arch.emulate_ctxt.cs_base = 0;
2057                         vcpu->arch.emulate_ctxt.ds_base = 0;
2058                         vcpu->arch.emulate_ctxt.es_base = 0;
2059                         vcpu->arch.emulate_ctxt.ss_base = 0;
2060                 } else {
2061                         vcpu->arch.emulate_ctxt.cs_base =
2062                                         get_segment_base(vcpu, VCPU_SREG_CS);
2063                         vcpu->arch.emulate_ctxt.ds_base =
2064                                         get_segment_base(vcpu, VCPU_SREG_DS);
2065                         vcpu->arch.emulate_ctxt.es_base =
2066                                         get_segment_base(vcpu, VCPU_SREG_ES);
2067                         vcpu->arch.emulate_ctxt.ss_base =
2068                                         get_segment_base(vcpu, VCPU_SREG_SS);
2069                 }
2070
2071                 vcpu->arch.emulate_ctxt.gs_base =
2072                                         get_segment_base(vcpu, VCPU_SREG_GS);
2073                 vcpu->arch.emulate_ctxt.fs_base =
2074                                         get_segment_base(vcpu, VCPU_SREG_FS);
2075
2076                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2077
2078                 /* Reject the instructions other than VMCALL/VMMCALL when
2079                  * try to emulate invalid opcode */
2080                 c = &vcpu->arch.emulate_ctxt.decode;
2081                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2082                     (!(c->twobyte && c->b == 0x01 &&
2083                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2084                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2085                         return EMULATE_FAIL;
2086
2087                 ++vcpu->stat.insn_emulation;
2088                 if (r)  {
2089                         ++vcpu->stat.insn_emulation_fail;
2090                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2091                                 return EMULATE_DONE;
2092                         return EMULATE_FAIL;
2093                 }
2094         }
2095
2096         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2097
2098         if (vcpu->arch.pio.string)
2099                 return EMULATE_DO_MMIO;
2100
2101         if ((r || vcpu->mmio_is_write) && run) {
2102                 run->exit_reason = KVM_EXIT_MMIO;
2103                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2104                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2105                 run->mmio.len = vcpu->mmio_size;
2106                 run->mmio.is_write = vcpu->mmio_is_write;
2107         }
2108
2109         if (r) {
2110                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2111                         return EMULATE_DONE;
2112                 if (!vcpu->mmio_needed) {
2113                         kvm_report_emulation_failure(vcpu, "mmio");
2114                         return EMULATE_FAIL;
2115                 }
2116                 return EMULATE_DO_MMIO;
2117         }
2118
2119         kvm_x86_ops->decache_regs(vcpu);
2120         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2121
2122         if (vcpu->mmio_is_write) {
2123                 vcpu->mmio_needed = 0;
2124                 return EMULATE_DO_MMIO;
2125         }
2126
2127         return EMULATE_DONE;
2128 }
2129 EXPORT_SYMBOL_GPL(emulate_instruction);
2130
2131 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2132 {
2133         int i;
2134
2135         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2136                 if (vcpu->arch.pio.guest_pages[i]) {
2137                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2138                         vcpu->arch.pio.guest_pages[i] = NULL;
2139                 }
2140 }
2141
2142 static int pio_copy_data(struct kvm_vcpu *vcpu)
2143 {
2144         void *p = vcpu->arch.pio_data;
2145         void *q;
2146         unsigned bytes;
2147         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2148
2149         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2150                  PAGE_KERNEL);
2151         if (!q) {
2152                 free_pio_guest_pages(vcpu);
2153                 return -ENOMEM;
2154         }
2155         q += vcpu->arch.pio.guest_page_offset;
2156         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2157         if (vcpu->arch.pio.in)
2158                 memcpy(q, p, bytes);
2159         else
2160                 memcpy(p, q, bytes);
2161         q -= vcpu->arch.pio.guest_page_offset;
2162         vunmap(q);
2163         free_pio_guest_pages(vcpu);
2164         return 0;
2165 }
2166
2167 int complete_pio(struct kvm_vcpu *vcpu)
2168 {
2169         struct kvm_pio_request *io = &vcpu->arch.pio;
2170         long delta;
2171         int r;
2172
2173         kvm_x86_ops->cache_regs(vcpu);
2174
2175         if (!io->string) {
2176                 if (io->in)
2177                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2178                                io->size);
2179         } else {
2180                 if (io->in) {
2181                         r = pio_copy_data(vcpu);
2182                         if (r) {
2183                                 kvm_x86_ops->cache_regs(vcpu);
2184                                 return r;
2185                         }
2186                 }
2187
2188                 delta = 1;
2189                 if (io->rep) {
2190                         delta *= io->cur_count;
2191                         /*
2192                          * The size of the register should really depend on
2193                          * current address size.
2194                          */
2195                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2196                 }
2197                 if (io->down)
2198                         delta = -delta;
2199                 delta *= io->size;
2200                 if (io->in)
2201                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2202                 else
2203                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2204         }
2205
2206         kvm_x86_ops->decache_regs(vcpu);
2207
2208         io->count -= io->cur_count;
2209         io->cur_count = 0;
2210
2211         return 0;
2212 }
2213
2214 static void kernel_pio(struct kvm_io_device *pio_dev,
2215                        struct kvm_vcpu *vcpu,
2216                        void *pd)
2217 {
2218         /* TODO: String I/O for in kernel device */
2219
2220         mutex_lock(&vcpu->kvm->lock);
2221         if (vcpu->arch.pio.in)
2222                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2223                                   vcpu->arch.pio.size,
2224                                   pd);
2225         else
2226                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2227                                    vcpu->arch.pio.size,
2228                                    pd);
2229         mutex_unlock(&vcpu->kvm->lock);
2230 }
2231
2232 static void pio_string_write(struct kvm_io_device *pio_dev,
2233                              struct kvm_vcpu *vcpu)
2234 {
2235         struct kvm_pio_request *io = &vcpu->arch.pio;
2236         void *pd = vcpu->arch.pio_data;
2237         int i;
2238
2239         mutex_lock(&vcpu->kvm->lock);
2240         for (i = 0; i < io->cur_count; i++) {
2241                 kvm_iodevice_write(pio_dev, io->port,
2242                                    io->size,
2243                                    pd);
2244                 pd += io->size;
2245         }
2246         mutex_unlock(&vcpu->kvm->lock);
2247 }
2248
2249 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2250                                                gpa_t addr)
2251 {
2252         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2253 }
2254
2255 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2256                   int size, unsigned port)
2257 {
2258         struct kvm_io_device *pio_dev;
2259
2260         vcpu->run->exit_reason = KVM_EXIT_IO;
2261         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2262         vcpu->run->io.size = vcpu->arch.pio.size = size;
2263         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2264         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2265         vcpu->run->io.port = vcpu->arch.pio.port = port;
2266         vcpu->arch.pio.in = in;
2267         vcpu->arch.pio.string = 0;
2268         vcpu->arch.pio.down = 0;
2269         vcpu->arch.pio.guest_page_offset = 0;
2270         vcpu->arch.pio.rep = 0;
2271
2272         kvm_x86_ops->cache_regs(vcpu);
2273         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2274         kvm_x86_ops->decache_regs(vcpu);
2275
2276         kvm_x86_ops->skip_emulated_instruction(vcpu);
2277
2278         pio_dev = vcpu_find_pio_dev(vcpu, port);
2279         if (pio_dev) {
2280                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2281                 complete_pio(vcpu);
2282                 return 1;
2283         }
2284         return 0;
2285 }
2286 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2287
2288 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2289                   int size, unsigned long count, int down,
2290                   gva_t address, int rep, unsigned port)
2291 {
2292         unsigned now, in_page;
2293         int i, ret = 0;
2294         int nr_pages = 1;
2295         struct page *page;
2296         struct kvm_io_device *pio_dev;
2297
2298         vcpu->run->exit_reason = KVM_EXIT_IO;
2299         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2300         vcpu->run->io.size = vcpu->arch.pio.size = size;
2301         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2302         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2303         vcpu->run->io.port = vcpu->arch.pio.port = port;
2304         vcpu->arch.pio.in = in;
2305         vcpu->arch.pio.string = 1;
2306         vcpu->arch.pio.down = down;
2307         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2308         vcpu->arch.pio.rep = rep;
2309
2310         if (!count) {
2311                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2312                 return 1;
2313         }
2314
2315         if (!down)
2316                 in_page = PAGE_SIZE - offset_in_page(address);
2317         else
2318                 in_page = offset_in_page(address) + size;
2319         now = min(count, (unsigned long)in_page / size);
2320         if (!now) {
2321                 /*
2322                  * String I/O straddles page boundary.  Pin two guest pages
2323                  * so that we satisfy atomicity constraints.  Do just one
2324                  * transaction to avoid complexity.
2325                  */
2326                 nr_pages = 2;
2327                 now = 1;
2328         }
2329         if (down) {
2330                 /*
2331                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2332                  */
2333                 pr_unimpl(vcpu, "guest string pio down\n");
2334                 kvm_inject_gp(vcpu, 0);
2335                 return 1;
2336         }
2337         vcpu->run->io.count = now;
2338         vcpu->arch.pio.cur_count = now;
2339
2340         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2341                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2342
2343         for (i = 0; i < nr_pages; ++i) {
2344                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2345                 vcpu->arch.pio.guest_pages[i] = page;
2346                 if (!page) {
2347                         kvm_inject_gp(vcpu, 0);
2348                         free_pio_guest_pages(vcpu);
2349                         return 1;
2350                 }
2351         }
2352
2353         pio_dev = vcpu_find_pio_dev(vcpu, port);
2354         if (!vcpu->arch.pio.in) {
2355                 /* string PIO write */
2356                 ret = pio_copy_data(vcpu);
2357                 if (ret >= 0 && pio_dev) {
2358                         pio_string_write(pio_dev, vcpu);
2359                         complete_pio(vcpu);
2360                         if (vcpu->arch.pio.count == 0)
2361                                 ret = 1;
2362                 }
2363         } else if (pio_dev)
2364                 pr_unimpl(vcpu, "no string pio read support yet, "
2365                        "port %x size %d count %ld\n",
2366                         port, size, count);
2367
2368         return ret;
2369 }
2370 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2371
2372 int kvm_arch_init(void *opaque)
2373 {
2374         int r;
2375         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2376
2377         if (kvm_x86_ops) {
2378                 printk(KERN_ERR "kvm: already loaded the other module\n");
2379                 r = -EEXIST;
2380                 goto out;
2381         }
2382
2383         if (!ops->cpu_has_kvm_support()) {
2384                 printk(KERN_ERR "kvm: no hardware support\n");
2385                 r = -EOPNOTSUPP;
2386                 goto out;
2387         }
2388         if (ops->disabled_by_bios()) {
2389                 printk(KERN_ERR "kvm: disabled by bios\n");
2390                 r = -EOPNOTSUPP;
2391                 goto out;
2392         }
2393
2394         r = kvm_mmu_module_init();
2395         if (r)
2396                 goto out;
2397
2398         kvm_init_msr_list();
2399
2400         kvm_x86_ops = ops;
2401         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2402         return 0;
2403
2404 out:
2405         return r;
2406 }
2407
2408 void kvm_arch_exit(void)
2409 {
2410         kvm_x86_ops = NULL;
2411         kvm_mmu_module_exit();
2412 }
2413
2414 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2415 {
2416         ++vcpu->stat.halt_exits;
2417         if (irqchip_in_kernel(vcpu->kvm)) {
2418                 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2419                 up_read(&vcpu->kvm->slots_lock);
2420                 kvm_vcpu_block(vcpu);
2421                 down_read(&vcpu->kvm->slots_lock);
2422                 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2423                         return -EINTR;
2424                 return 1;
2425         } else {
2426                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2427                 return 0;
2428         }
2429 }
2430 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2431
2432 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2433                            unsigned long a1)
2434 {
2435         if (is_long_mode(vcpu))
2436                 return a0;
2437         else
2438                 return a0 | ((gpa_t)a1 << 32);
2439 }
2440
2441 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2442 {
2443         unsigned long nr, a0, a1, a2, a3, ret;
2444         int r = 1;
2445
2446         kvm_x86_ops->cache_regs(vcpu);
2447
2448         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2449         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2450         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2451         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2452         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2453
2454         if (!is_long_mode(vcpu)) {
2455                 nr &= 0xFFFFFFFF;
2456                 a0 &= 0xFFFFFFFF;
2457                 a1 &= 0xFFFFFFFF;
2458                 a2 &= 0xFFFFFFFF;
2459                 a3 &= 0xFFFFFFFF;
2460         }
2461
2462         switch (nr) {
2463         case KVM_HC_VAPIC_POLL_IRQ:
2464                 ret = 0;
2465                 break;
2466         case KVM_HC_MMU_OP:
2467                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2468                 break;
2469         default:
2470                 ret = -KVM_ENOSYS;
2471                 break;
2472         }
2473         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2474         kvm_x86_ops->decache_regs(vcpu);
2475         ++vcpu->stat.hypercalls;
2476         return r;
2477 }
2478 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2479
2480 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2481 {
2482         char instruction[3];
2483         int ret = 0;
2484
2485
2486         /*
2487          * Blow out the MMU to ensure that no other VCPU has an active mapping
2488          * to ensure that the updated hypercall appears atomically across all
2489          * VCPUs.
2490          */
2491         kvm_mmu_zap_all(vcpu->kvm);
2492
2493         kvm_x86_ops->cache_regs(vcpu);
2494         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2495         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2496             != X86EMUL_CONTINUE)
2497                 ret = -EFAULT;
2498
2499         return ret;
2500 }
2501
2502 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2503 {
2504         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2505 }
2506
2507 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2508 {
2509         struct descriptor_table dt = { limit, base };
2510
2511         kvm_x86_ops->set_gdt(vcpu, &dt);
2512 }
2513
2514 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2515 {
2516         struct descriptor_table dt = { limit, base };
2517
2518         kvm_x86_ops->set_idt(vcpu, &dt);
2519 }
2520
2521 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2522                    unsigned long *rflags)
2523 {
2524         kvm_lmsw(vcpu, msw);
2525         *rflags = kvm_x86_ops->get_rflags(vcpu);
2526 }
2527
2528 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2529 {
2530         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2531         switch (cr) {
2532         case 0:
2533                 return vcpu->arch.cr0;
2534         case 2:
2535                 return vcpu->arch.cr2;
2536         case 3:
2537                 return vcpu->arch.cr3;
2538         case 4:
2539                 return vcpu->arch.cr4;
2540         case 8:
2541                 return kvm_get_cr8(vcpu);
2542         default:
2543                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2544                 return 0;
2545         }
2546 }
2547
2548 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2549                      unsigned long *rflags)
2550 {
2551         switch (cr) {
2552         case 0:
2553                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2554                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2555                 break;
2556         case 2:
2557                 vcpu->arch.cr2 = val;
2558                 break;
2559         case 3:
2560                 kvm_set_cr3(vcpu, val);
2561                 break;
2562         case 4:
2563                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2564                 break;
2565         case 8:
2566                 kvm_set_cr8(vcpu, val & 0xfUL);
2567                 break;
2568         default:
2569                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2570         }
2571 }
2572
2573 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2574 {
2575         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2576         int j, nent = vcpu->arch.cpuid_nent;
2577
2578         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2579         /* when no next entry is found, the current entry[i] is reselected */
2580         for (j = i + 1; j == i; j = (j + 1) % nent) {
2581                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2582                 if (ej->function == e->function) {
2583                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2584                         return j;
2585                 }
2586         }
2587         return 0; /* silence gcc, even though control never reaches here */
2588 }
2589
2590 /* find an entry with matching function, matching index (if needed), and that
2591  * should be read next (if it's stateful) */
2592 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2593         u32 function, u32 index)
2594 {
2595         if (e->function != function)
2596                 return 0;
2597         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2598                 return 0;
2599         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2600                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2601                 return 0;
2602         return 1;
2603 }
2604
2605 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2606 {
2607         int i;
2608         u32 function, index;
2609         struct kvm_cpuid_entry2 *e, *best;
2610
2611         kvm_x86_ops->cache_regs(vcpu);
2612         function = vcpu->arch.regs[VCPU_REGS_RAX];
2613         index = vcpu->arch.regs[VCPU_REGS_RCX];
2614         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2615         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2616         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2617         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2618         best = NULL;
2619         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2620                 e = &vcpu->arch.cpuid_entries[i];
2621                 if (is_matching_cpuid_entry(e, function, index)) {
2622                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2623                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2624                         best = e;
2625                         break;
2626                 }
2627                 /*
2628                  * Both basic or both extended?
2629                  */
2630                 if (((e->function ^ function) & 0x80000000) == 0)
2631                         if (!best || e->function > best->function)
2632                                 best = e;
2633         }
2634         if (best) {
2635                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2636                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2637                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2638                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2639         }
2640         kvm_x86_ops->decache_regs(vcpu);
2641         kvm_x86_ops->skip_emulated_instruction(vcpu);
2642 }
2643 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2644
2645 /*
2646  * Check if userspace requested an interrupt window, and that the
2647  * interrupt window is open.
2648  *
2649  * No need to exit to userspace if we already have an interrupt queued.
2650  */
2651 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2652                                           struct kvm_run *kvm_run)
2653 {
2654         return (!vcpu->arch.irq_summary &&
2655                 kvm_run->request_interrupt_window &&
2656                 vcpu->arch.interrupt_window_open &&
2657                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2658 }
2659
2660 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2661                               struct kvm_run *kvm_run)
2662 {
2663         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2664         kvm_run->cr8 = kvm_get_cr8(vcpu);
2665         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2666         if (irqchip_in_kernel(vcpu->kvm))
2667                 kvm_run->ready_for_interrupt_injection = 1;
2668         else
2669                 kvm_run->ready_for_interrupt_injection =
2670                                         (vcpu->arch.interrupt_window_open &&
2671                                          vcpu->arch.irq_summary == 0);
2672 }
2673
2674 static void vapic_enter(struct kvm_vcpu *vcpu)
2675 {
2676         struct kvm_lapic *apic = vcpu->arch.apic;
2677         struct page *page;
2678
2679         if (!apic || !apic->vapic_addr)
2680                 return;
2681
2682         down_read(&current->mm->mmap_sem);
2683         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2684         up_read(&current->mm->mmap_sem);
2685
2686         vcpu->arch.apic->vapic_page = page;
2687 }
2688
2689 static void vapic_exit(struct kvm_vcpu *vcpu)
2690 {
2691         struct kvm_lapic *apic = vcpu->arch.apic;
2692
2693         if (!apic || !apic->vapic_addr)
2694                 return;
2695
2696         kvm_release_page_dirty(apic->vapic_page);
2697         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2698 }
2699
2700 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2701 {
2702         int r;
2703
2704         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2705                 pr_debug("vcpu %d received sipi with vector # %x\n",
2706                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2707                 kvm_lapic_reset(vcpu);
2708                 r = kvm_x86_ops->vcpu_reset(vcpu);
2709                 if (r)
2710                         return r;
2711                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2712         }
2713
2714         down_read(&vcpu->kvm->slots_lock);
2715         vapic_enter(vcpu);
2716
2717 preempted:
2718         if (vcpu->guest_debug.enabled)
2719                 kvm_x86_ops->guest_debug_pre(vcpu);
2720
2721 again:
2722         if (vcpu->requests)
2723                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2724                         kvm_mmu_unload(vcpu);
2725
2726         r = kvm_mmu_reload(vcpu);
2727         if (unlikely(r))
2728                 goto out;
2729
2730         if (vcpu->requests) {
2731                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2732                         __kvm_migrate_apic_timer(vcpu);
2733                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2734                                        &vcpu->requests)) {
2735                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2736                         r = 0;
2737                         goto out;
2738                 }
2739                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2740                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2741                         r = 0;
2742                         goto out;
2743                 }
2744         }
2745
2746         kvm_inject_pending_timer_irqs(vcpu);
2747
2748         preempt_disable();
2749
2750         kvm_x86_ops->prepare_guest_switch(vcpu);
2751         kvm_load_guest_fpu(vcpu);
2752
2753         local_irq_disable();
2754
2755         if (need_resched()) {
2756                 local_irq_enable();
2757                 preempt_enable();
2758                 r = 1;
2759                 goto out;
2760         }
2761
2762         if (vcpu->requests)
2763                 if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
2764                         local_irq_enable();
2765                         preempt_enable();
2766                         r = 1;
2767                         goto out;
2768                 }
2769
2770         if (signal_pending(current)) {
2771                 local_irq_enable();
2772                 preempt_enable();
2773                 r = -EINTR;
2774                 kvm_run->exit_reason = KVM_EXIT_INTR;
2775                 ++vcpu->stat.signal_exits;
2776                 goto out;
2777         }
2778
2779         if (vcpu->arch.exception.pending)
2780                 __queue_exception(vcpu);
2781         else if (irqchip_in_kernel(vcpu->kvm))
2782                 kvm_x86_ops->inject_pending_irq(vcpu);
2783         else
2784                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2785
2786         kvm_lapic_sync_to_vapic(vcpu);
2787
2788         up_read(&vcpu->kvm->slots_lock);
2789
2790         vcpu->guest_mode = 1;
2791         kvm_guest_enter();
2792
2793         if (vcpu->requests)
2794                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2795                         kvm_x86_ops->tlb_flush(vcpu);
2796
2797         kvm_x86_ops->run(vcpu, kvm_run);
2798
2799         vcpu->guest_mode = 0;
2800         local_irq_enable();
2801
2802         ++vcpu->stat.exits;
2803
2804         /*
2805          * We must have an instruction between local_irq_enable() and
2806          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2807          * the interrupt shadow.  The stat.exits increment will do nicely.
2808          * But we need to prevent reordering, hence this barrier():
2809          */
2810         barrier();
2811
2812         kvm_guest_exit();
2813
2814         preempt_enable();
2815
2816         down_read(&vcpu->kvm->slots_lock);
2817
2818         /*
2819          * Profile KVM exit RIPs:
2820          */
2821         if (unlikely(prof_on == KVM_PROFILING)) {
2822                 kvm_x86_ops->cache_regs(vcpu);
2823                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2824         }
2825
2826         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2827                 vcpu->arch.exception.pending = false;
2828
2829         kvm_lapic_sync_from_vapic(vcpu);
2830
2831         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2832
2833         if (r > 0) {
2834                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2835                         r = -EINTR;
2836                         kvm_run->exit_reason = KVM_EXIT_INTR;
2837                         ++vcpu->stat.request_irq_exits;
2838                         goto out;
2839                 }
2840                 if (!need_resched())
2841                         goto again;
2842         }
2843
2844 out:
2845         up_read(&vcpu->kvm->slots_lock);
2846         if (r > 0) {
2847                 kvm_resched(vcpu);
2848                 down_read(&vcpu->kvm->slots_lock);
2849                 goto preempted;
2850         }
2851
2852         post_kvm_run_save(vcpu, kvm_run);
2853
2854         down_read(&vcpu->kvm->slots_lock);
2855         vapic_exit(vcpu);
2856         up_read(&vcpu->kvm->slots_lock);
2857
2858         return r;
2859 }
2860
2861 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2862 {
2863         int r;
2864         sigset_t sigsaved;
2865
2866         vcpu_load(vcpu);
2867
2868         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2869                 kvm_vcpu_block(vcpu);
2870                 vcpu_put(vcpu);
2871                 return -EAGAIN;
2872         }
2873
2874         if (vcpu->sigset_active)
2875                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2876
2877         /* re-sync apic's tpr */
2878         if (!irqchip_in_kernel(vcpu->kvm))
2879                 kvm_set_cr8(vcpu, kvm_run->cr8);
2880
2881         if (vcpu->arch.pio.cur_count) {
2882                 r = complete_pio(vcpu);
2883                 if (r)
2884                         goto out;
2885         }
2886 #if CONFIG_HAS_IOMEM
2887         if (vcpu->mmio_needed) {
2888                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2889                 vcpu->mmio_read_completed = 1;
2890                 vcpu->mmio_needed = 0;
2891
2892                 down_read(&vcpu->kvm->slots_lock);
2893                 r = emulate_instruction(vcpu, kvm_run,
2894                                         vcpu->arch.mmio_fault_cr2, 0,
2895                                         EMULTYPE_NO_DECODE);
2896                 up_read(&vcpu->kvm->slots_lock);
2897                 if (r == EMULATE_DO_MMIO) {
2898                         /*
2899                          * Read-modify-write.  Back to userspace.
2900                          */
2901                         r = 0;
2902                         goto out;
2903                 }
2904         }
2905 #endif
2906         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2907                 kvm_x86_ops->cache_regs(vcpu);
2908                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2909                 kvm_x86_ops->decache_regs(vcpu);
2910         }
2911
2912         r = __vcpu_run(vcpu, kvm_run);
2913
2914 out:
2915         if (vcpu->sigset_active)
2916                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2917
2918         vcpu_put(vcpu);
2919         return r;
2920 }
2921
2922 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2923 {
2924         vcpu_load(vcpu);
2925
2926         kvm_x86_ops->cache_regs(vcpu);
2927
2928         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2929         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2930         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2931         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2932         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2933         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2934         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2935         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2936 #ifdef CONFIG_X86_64
2937         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2938         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2939         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2940         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2941         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2942         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2943         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2944         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2945 #endif
2946
2947         regs->rip = vcpu->arch.rip;
2948         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2949
2950         /*
2951          * Don't leak debug flags in case they were set for guest debugging
2952          */
2953         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2954                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2955
2956         vcpu_put(vcpu);
2957
2958         return 0;
2959 }
2960
2961 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2962 {
2963         vcpu_load(vcpu);
2964
2965         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2966         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2967         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2968         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2969         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2970         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2971         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2972         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2973 #ifdef CONFIG_X86_64
2974         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2975         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2976         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2977         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2978         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2979         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2980         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2981         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2982 #endif
2983
2984         vcpu->arch.rip = regs->rip;
2985         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2986
2987         kvm_x86_ops->decache_regs(vcpu);
2988
2989         vcpu_put(vcpu);
2990
2991         return 0;
2992 }
2993
2994 static void get_segment(struct kvm_vcpu *vcpu,
2995                         struct kvm_segment *var, int seg)
2996 {
2997         kvm_x86_ops->get_segment(vcpu, var, seg);
2998 }
2999
3000 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3001 {
3002         struct kvm_segment cs;
3003
3004         get_segment(vcpu, &cs, VCPU_SREG_CS);
3005         *db = cs.db;
3006         *l = cs.l;
3007 }
3008 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3009
3010 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3011                                   struct kvm_sregs *sregs)
3012 {
3013         struct descriptor_table dt;
3014         int pending_vec;
3015
3016         vcpu_load(vcpu);
3017
3018         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3019         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3020         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3021         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3022         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3023         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3024
3025         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3026         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3027
3028         kvm_x86_ops->get_idt(vcpu, &dt);
3029         sregs->idt.limit = dt.limit;
3030         sregs->idt.base = dt.base;
3031         kvm_x86_ops->get_gdt(vcpu, &dt);
3032         sregs->gdt.limit = dt.limit;
3033         sregs->gdt.base = dt.base;
3034
3035         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3036         sregs->cr0 = vcpu->arch.cr0;
3037         sregs->cr2 = vcpu->arch.cr2;
3038         sregs->cr3 = vcpu->arch.cr3;
3039         sregs->cr4 = vcpu->arch.cr4;
3040         sregs->cr8 = kvm_get_cr8(vcpu);
3041         sregs->efer = vcpu->arch.shadow_efer;
3042         sregs->apic_base = kvm_get_apic_base(vcpu);
3043
3044         if (irqchip_in_kernel(vcpu->kvm)) {
3045                 memset(sregs->interrupt_bitmap, 0,
3046                        sizeof sregs->interrupt_bitmap);
3047                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3048                 if (pending_vec >= 0)
3049                         set_bit(pending_vec,
3050                                 (unsigned long *)sregs->interrupt_bitmap);
3051         } else
3052                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3053                        sizeof sregs->interrupt_bitmap);
3054
3055         vcpu_put(vcpu);
3056
3057         return 0;
3058 }
3059
3060 static void set_segment(struct kvm_vcpu *vcpu,
3061                         struct kvm_segment *var, int seg)
3062 {
3063         kvm_x86_ops->set_segment(vcpu, var, seg);
3064 }
3065
3066 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3067                                    struct kvm_segment *kvm_desct)
3068 {
3069         kvm_desct->base = seg_desc->base0;
3070         kvm_desct->base |= seg_desc->base1 << 16;
3071         kvm_desct->base |= seg_desc->base2 << 24;
3072         kvm_desct->limit = seg_desc->limit0;
3073         kvm_desct->limit |= seg_desc->limit << 16;
3074         kvm_desct->selector = selector;
3075         kvm_desct->type = seg_desc->type;
3076         kvm_desct->present = seg_desc->p;
3077         kvm_desct->dpl = seg_desc->dpl;
3078         kvm_desct->db = seg_desc->d;
3079         kvm_desct->s = seg_desc->s;
3080         kvm_desct->l = seg_desc->l;
3081         kvm_desct->g = seg_desc->g;
3082         kvm_desct->avl = seg_desc->avl;
3083         if (!selector)
3084                 kvm_desct->unusable = 1;
3085         else
3086                 kvm_desct->unusable = 0;
3087         kvm_desct->padding = 0;
3088 }
3089
3090 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3091                                            u16 selector,
3092                                            struct descriptor_table *dtable)
3093 {
3094         if (selector & 1 << 2) {
3095                 struct kvm_segment kvm_seg;
3096
3097                 get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3098
3099                 if (kvm_seg.unusable)
3100                         dtable->limit = 0;
3101                 else
3102                         dtable->limit = kvm_seg.limit;
3103                 dtable->base = kvm_seg.base;
3104         }
3105         else
3106                 kvm_x86_ops->get_gdt(vcpu, dtable);
3107 }
3108
3109 /* allowed just for 8 bytes segments */
3110 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3111                                          struct desc_struct *seg_desc)
3112 {
3113         struct descriptor_table dtable;
3114         u16 index = selector >> 3;
3115
3116         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3117
3118         if (dtable.limit < index * 8 + 7) {
3119                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3120                 return 1;
3121         }
3122         return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3123 }
3124
3125 /* allowed just for 8 bytes segments */
3126 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3127                                          struct desc_struct *seg_desc)
3128 {
3129         struct descriptor_table dtable;
3130         u16 index = selector >> 3;
3131
3132         get_segment_descritptor_dtable(vcpu, selector, &dtable);
3133
3134         if (dtable.limit < index * 8 + 7)
3135                 return 1;
3136         return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3137 }
3138
3139 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3140                              struct desc_struct *seg_desc)
3141 {
3142         u32 base_addr;
3143
3144         base_addr = seg_desc->base0;
3145         base_addr |= (seg_desc->base1 << 16);
3146         base_addr |= (seg_desc->base2 << 24);
3147
3148         return base_addr;
3149 }
3150
3151 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3152                               struct desc_struct *seg_desc,
3153                               struct tss_segment_32 *tss)
3154 {
3155         u32 base_addr;
3156
3157         base_addr = get_tss_base_addr(vcpu, seg_desc);
3158
3159         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3160                               sizeof(struct tss_segment_32));
3161 }
3162
3163 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3164                               struct desc_struct *seg_desc,
3165                               struct tss_segment_32 *tss)
3166 {
3167         u32 base_addr;
3168
3169         base_addr = get_tss_base_addr(vcpu, seg_desc);
3170
3171         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3172                                sizeof(struct tss_segment_32));
3173 }
3174
3175 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3176                               struct desc_struct *seg_desc,
3177                               struct tss_segment_16 *tss)
3178 {
3179         u32 base_addr;
3180
3181         base_addr = get_tss_base_addr(vcpu, seg_desc);
3182
3183         return kvm_read_guest(vcpu->kvm, base_addr, tss,
3184                               sizeof(struct tss_segment_16));
3185 }
3186
3187 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3188                               struct desc_struct *seg_desc,
3189                               struct tss_segment_16 *tss)
3190 {
3191         u32 base_addr;
3192
3193         base_addr = get_tss_base_addr(vcpu, seg_desc);
3194
3195         return kvm_write_guest(vcpu->kvm, base_addr, tss,
3196                                sizeof(struct tss_segment_16));
3197 }
3198
3199 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3200 {
3201         struct kvm_segment kvm_seg;
3202
3203         get_segment(vcpu, &kvm_seg, seg);
3204         return kvm_seg.selector;
3205 }
3206
3207 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3208                                                 u16 selector,
3209                                                 struct kvm_segment *kvm_seg)
3210 {
3211         struct desc_struct seg_desc;
3212
3213         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3214                 return 1;
3215         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3216         return 0;
3217 }
3218
3219 static int load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3220                                    int type_bits, int seg)
3221 {
3222         struct kvm_segment kvm_seg;
3223
3224         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3225                 return 1;
3226         kvm_seg.type |= type_bits;
3227
3228         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3229             seg != VCPU_SREG_LDTR)
3230                 if (!kvm_seg.s)
3231                         kvm_seg.unusable = 1;
3232
3233         set_segment(vcpu, &kvm_seg, seg);
3234         return 0;
3235 }
3236
3237 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3238                                 struct tss_segment_32 *tss)
3239 {
3240         tss->cr3 = vcpu->arch.cr3;
3241         tss->eip = vcpu->arch.rip;
3242         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3243         tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3244         tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3245         tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3246         tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3247         tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3248         tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3249         tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3250         tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3251
3252         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3253         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3254         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3255         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3256         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3257         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3258         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3259         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3260 }
3261
3262 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3263                                   struct tss_segment_32 *tss)
3264 {
3265         kvm_set_cr3(vcpu, tss->cr3);
3266
3267         vcpu->arch.rip = tss->eip;
3268         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3269
3270         vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3271         vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3272         vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3273         vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3274         vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3275         vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3276         vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3277         vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3278
3279         if (load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3280                 return 1;
3281
3282         if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3283                 return 1;
3284
3285         if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3286                 return 1;
3287
3288         if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3289                 return 1;
3290
3291         if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3292                 return 1;
3293
3294         if (load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3295                 return 1;
3296
3297         if (load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3298                 return 1;
3299         return 0;
3300 }
3301
3302 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3303                                 struct tss_segment_16 *tss)
3304 {
3305         tss->ip = vcpu->arch.rip;
3306         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3307         tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3308         tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3309         tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3310         tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3311         tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3312         tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3313         tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3314         tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3315
3316         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3317         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3318         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3319         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3320         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3321         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3322 }
3323
3324 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3325                                  struct tss_segment_16 *tss)
3326 {
3327         vcpu->arch.rip = tss->ip;
3328         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3329         vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3330         vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3331         vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3332         vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3333         vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3334         vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3335         vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3336         vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3337
3338         if (load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3339                 return 1;
3340
3341         if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3342                 return 1;
3343
3344         if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3345                 return 1;
3346
3347         if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3348                 return 1;
3349
3350         if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3351                 return 1;
3352         return 0;
3353 }
3354
3355 int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3356                        struct desc_struct *cseg_desc,
3357                        struct desc_struct *nseg_desc)
3358 {
3359         struct tss_segment_16 tss_segment_16;
3360         int ret = 0;
3361
3362         if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3363                 goto out;
3364
3365         save_state_to_tss16(vcpu, &tss_segment_16);
3366         save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3367
3368         if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3369                 goto out;
3370         if (load_state_from_tss16(vcpu, &tss_segment_16))
3371                 goto out;
3372
3373         ret = 1;
3374 out:
3375         return ret;
3376 }
3377
3378 int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3379                        struct desc_struct *cseg_desc,
3380                        struct desc_struct *nseg_desc)
3381 {
3382         struct tss_segment_32 tss_segment_32;
3383         int ret = 0;
3384
3385         if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3386                 goto out;
3387
3388         save_state_to_tss32(vcpu, &tss_segment_32);
3389         save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3390
3391         if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3392                 goto out;
3393         if (load_state_from_tss32(vcpu, &tss_segment_32))
3394                 goto out;
3395
3396         ret = 1;
3397 out:
3398         return ret;
3399 }
3400
3401 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3402 {
3403         struct kvm_segment tr_seg;
3404         struct desc_struct cseg_desc;
3405         struct desc_struct nseg_desc;
3406         int ret = 0;
3407
3408         get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3409
3410         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3411                 goto out;
3412
3413         if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3414                 goto out;
3415
3416
3417         if (reason != TASK_SWITCH_IRET) {
3418                 int cpl;
3419
3420                 cpl = kvm_x86_ops->get_cpl(vcpu);
3421                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3422                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3423                         return 1;
3424                 }
3425         }
3426
3427         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3428                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3429                 return 1;
3430         }
3431
3432         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3433                 cseg_desc.type &= ~(1 << 8); //clear the B flag
3434                 save_guest_segment_descriptor(vcpu, tr_seg.selector,
3435                                               &cseg_desc);
3436         }
3437
3438         if (reason == TASK_SWITCH_IRET) {
3439                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3440                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3441         }
3442
3443         kvm_x86_ops->skip_emulated_instruction(vcpu);
3444         kvm_x86_ops->cache_regs(vcpu);
3445
3446         if (nseg_desc.type & 8)
3447                 ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3448                                          &nseg_desc);
3449         else
3450                 ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3451                                          &nseg_desc);
3452
3453         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3454                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3455                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3456         }
3457
3458         if (reason != TASK_SWITCH_IRET) {
3459                 nseg_desc.type |= (1 << 8);
3460                 save_guest_segment_descriptor(vcpu, tss_selector,
3461                                               &nseg_desc);
3462         }
3463
3464         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3465         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3466         tr_seg.type = 11;
3467         set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3468 out:
3469         kvm_x86_ops->decache_regs(vcpu);
3470         return ret;
3471 }
3472 EXPORT_SYMBOL_GPL(kvm_task_switch);
3473
3474 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3475                                   struct kvm_sregs *sregs)
3476 {
3477         int mmu_reset_needed = 0;
3478         int i, pending_vec, max_bits;
3479         struct descriptor_table dt;
3480
3481         vcpu_load(vcpu);
3482
3483         dt.limit = sregs->idt.limit;
3484         dt.base = sregs->idt.base;
3485         kvm_x86_ops->set_idt(vcpu, &dt);
3486         dt.limit = sregs->gdt.limit;
3487         dt.base = sregs->gdt.base;
3488         kvm_x86_ops->set_gdt(vcpu, &dt);
3489
3490         vcpu->arch.cr2 = sregs->cr2;
3491         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3492         vcpu->arch.cr3 = sregs->cr3;
3493
3494         kvm_set_cr8(vcpu, sregs->cr8);
3495
3496         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3497         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3498         kvm_set_apic_base(vcpu, sregs->apic_base);
3499
3500         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3501
3502         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3503         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3504         vcpu->arch.cr0 = sregs->cr0;
3505
3506         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3507         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3508         if (!is_long_mode(vcpu) && is_pae(vcpu))
3509                 load_pdptrs(vcpu, vcpu->arch.cr3);
3510
3511         if (mmu_reset_needed)
3512                 kvm_mmu_reset_context(vcpu);
3513
3514         if (!irqchip_in_kernel(vcpu->kvm)) {
3515                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3516                        sizeof vcpu->arch.irq_pending);
3517                 vcpu->arch.irq_summary = 0;
3518                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3519                         if (vcpu->arch.irq_pending[i])
3520                                 __set_bit(i, &vcpu->arch.irq_summary);
3521         } else {
3522                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3523                 pending_vec = find_first_bit(
3524                         (const unsigned long *)sregs->interrupt_bitmap,
3525                         max_bits);
3526                 /* Only pending external irq is handled here */
3527                 if (pending_vec < max_bits) {
3528                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3529                         pr_debug("Set back pending irq %d\n",
3530                                  pending_vec);
3531                 }
3532         }
3533
3534         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3535         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3536         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3537         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3538         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3539         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3540
3541         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3542         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3543
3544         vcpu_put(vcpu);
3545
3546         return 0;
3547 }
3548
3549 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3550                                     struct kvm_debug_guest *dbg)
3551 {
3552         int r;
3553
3554         vcpu_load(vcpu);
3555
3556         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3557
3558         vcpu_put(vcpu);
3559
3560         return r;
3561 }
3562
3563 /*
3564  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3565  * we have asm/x86/processor.h
3566  */
3567 struct fxsave {
3568         u16     cwd;
3569         u16     swd;
3570         u16     twd;
3571         u16     fop;
3572         u64     rip;
3573         u64     rdp;
3574         u32     mxcsr;
3575         u32     mxcsr_mask;
3576         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3577 #ifdef CONFIG_X86_64
3578         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3579 #else
3580         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3581 #endif
3582 };
3583
3584 /*
3585  * Translate a guest virtual address to a guest physical address.
3586  */
3587 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3588                                     struct kvm_translation *tr)
3589 {
3590         unsigned long vaddr = tr->linear_address;
3591         gpa_t gpa;
3592
3593         vcpu_load(vcpu);
3594         down_read(&vcpu->kvm->slots_lock);
3595         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3596         up_read(&vcpu->kvm->slots_lock);
3597         tr->physical_address = gpa;
3598         tr->valid = gpa != UNMAPPED_GVA;
3599         tr->writeable = 1;
3600         tr->usermode = 0;
3601         vcpu_put(vcpu);
3602
3603         return 0;
3604 }
3605
3606 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3607 {
3608         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3609
3610         vcpu_load(vcpu);
3611
3612         memcpy(fpu->fpr, fxsave->st_space, 128);
3613         fpu->fcw = fxsave->cwd;
3614         fpu->fsw = fxsave->swd;
3615         fpu->ftwx = fxsave->twd;
3616         fpu->last_opcode = fxsave->fop;
3617         fpu->last_ip = fxsave->rip;
3618         fpu->last_dp = fxsave->rdp;
3619         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3620
3621         vcpu_put(vcpu);
3622
3623         return 0;
3624 }
3625
3626 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3627 {
3628         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3629
3630         vcpu_load(vcpu);
3631
3632         memcpy(fxsave->st_space, fpu->fpr, 128);
3633         fxsave->cwd = fpu->fcw;
3634         fxsave->swd = fpu->fsw;
3635         fxsave->twd = fpu->ftwx;
3636         fxsave->fop = fpu->last_opcode;
3637         fxsave->rip = fpu->last_ip;
3638         fxsave->rdp = fpu->last_dp;
3639         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3640
3641         vcpu_put(vcpu);
3642
3643         return 0;
3644 }
3645
3646 void fx_init(struct kvm_vcpu *vcpu)
3647 {
3648         unsigned after_mxcsr_mask;
3649
3650         /* Initialize guest FPU by resetting ours and saving into guest's */
3651         preempt_disable();
3652         fx_save(&vcpu->arch.host_fx_image);
3653         fpu_init();
3654         fx_save(&vcpu->arch.guest_fx_image);
3655         fx_restore(&vcpu->arch.host_fx_image);
3656         preempt_enable();
3657
3658         vcpu->arch.cr0 |= X86_CR0_ET;
3659         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3660         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3661         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3662                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3663 }
3664 EXPORT_SYMBOL_GPL(fx_init);
3665
3666 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3667 {
3668         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3669                 return;
3670
3671         vcpu->guest_fpu_loaded = 1;
3672         fx_save(&vcpu->arch.host_fx_image);
3673         fx_restore(&vcpu->arch.guest_fx_image);
3674 }
3675 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3676
3677 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3678 {
3679         if (!vcpu->guest_fpu_loaded)
3680                 return;
3681
3682         vcpu->guest_fpu_loaded = 0;
3683         fx_save(&vcpu->arch.guest_fx_image);
3684         fx_restore(&vcpu->arch.host_fx_image);
3685         ++vcpu->stat.fpu_reload;
3686 }
3687 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3688
3689 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3690 {
3691         kvm_x86_ops->vcpu_free(vcpu);
3692 }
3693
3694 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3695                                                 unsigned int id)
3696 {
3697         return kvm_x86_ops->vcpu_create(kvm, id);
3698 }
3699
3700 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3701 {
3702         int r;
3703
3704         /* We do fxsave: this must be aligned. */
3705         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3706
3707         vcpu_load(vcpu);
3708         r = kvm_arch_vcpu_reset(vcpu);
3709         if (r == 0)
3710                 r = kvm_mmu_setup(vcpu);
3711         vcpu_put(vcpu);
3712         if (r < 0)
3713                 goto free_vcpu;
3714
3715         return 0;
3716 free_vcpu:
3717         kvm_x86_ops->vcpu_free(vcpu);
3718         return r;
3719 }
3720
3721 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3722 {
3723         vcpu_load(vcpu);
3724         kvm_mmu_unload(vcpu);
3725         vcpu_put(vcpu);
3726
3727         kvm_x86_ops->vcpu_free(vcpu);
3728 }
3729
3730 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3731 {
3732         return kvm_x86_ops->vcpu_reset(vcpu);
3733 }
3734
3735 void kvm_arch_hardware_enable(void *garbage)
3736 {
3737         kvm_x86_ops->hardware_enable(garbage);
3738 }
3739
3740 void kvm_arch_hardware_disable(void *garbage)
3741 {
3742         kvm_x86_ops->hardware_disable(garbage);
3743 }
3744
3745 int kvm_arch_hardware_setup(void)
3746 {
3747         return kvm_x86_ops->hardware_setup();
3748 }
3749
3750 void kvm_arch_hardware_unsetup(void)
3751 {
3752         kvm_x86_ops->hardware_unsetup();
3753 }
3754
3755 void kvm_arch_check_processor_compat(void *rtn)
3756 {
3757         kvm_x86_ops->check_processor_compatibility(rtn);
3758 }
3759
3760 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3761 {
3762         struct page *page;
3763         struct kvm *kvm;
3764         int r;
3765
3766         BUG_ON(vcpu->kvm == NULL);
3767         kvm = vcpu->kvm;
3768
3769         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3770         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3771                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3772         else
3773                 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3774
3775         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3776         if (!page) {
3777                 r = -ENOMEM;
3778                 goto fail;
3779         }
3780         vcpu->arch.pio_data = page_address(page);
3781
3782         r = kvm_mmu_create(vcpu);
3783         if (r < 0)
3784                 goto fail_free_pio_data;
3785
3786         if (irqchip_in_kernel(kvm)) {
3787                 r = kvm_create_lapic(vcpu);
3788                 if (r < 0)
3789                         goto fail_mmu_destroy;
3790         }
3791
3792         return 0;
3793
3794 fail_mmu_destroy:
3795         kvm_mmu_destroy(vcpu);
3796 fail_free_pio_data:
3797         free_page((unsigned long)vcpu->arch.pio_data);
3798 fail:
3799         return r;
3800 }
3801
3802 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3803 {
3804         kvm_free_lapic(vcpu);
3805         down_read(&vcpu->kvm->slots_lock);
3806         kvm_mmu_destroy(vcpu);
3807         up_read(&vcpu->kvm->slots_lock);
3808         free_page((unsigned long)vcpu->arch.pio_data);
3809 }
3810
3811 struct  kvm *kvm_arch_create_vm(void)
3812 {
3813         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3814
3815         if (!kvm)
3816                 return ERR_PTR(-ENOMEM);
3817
3818         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3819
3820         return kvm;
3821 }
3822
3823 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3824 {
3825         vcpu_load(vcpu);
3826         kvm_mmu_unload(vcpu);
3827         vcpu_put(vcpu);
3828 }
3829
3830 static void kvm_free_vcpus(struct kvm *kvm)
3831 {
3832         unsigned int i;
3833
3834         /*
3835          * Unpin any mmu pages first.
3836          */
3837         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3838                 if (kvm->vcpus[i])
3839                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3840         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3841                 if (kvm->vcpus[i]) {
3842                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3843                         kvm->vcpus[i] = NULL;
3844                 }
3845         }
3846
3847 }
3848
3849 void kvm_arch_destroy_vm(struct kvm *kvm)
3850 {
3851         kvm_free_pit(kvm);
3852         kfree(kvm->arch.vpic);
3853         kfree(kvm->arch.vioapic);
3854         kvm_free_vcpus(kvm);
3855         kvm_free_physmem(kvm);
3856         kfree(kvm);
3857 }
3858
3859 int kvm_arch_set_memory_region(struct kvm *kvm,
3860                                 struct kvm_userspace_memory_region *mem,
3861                                 struct kvm_memory_slot old,
3862                                 int user_alloc)
3863 {
3864         int npages = mem->memory_size >> PAGE_SHIFT;
3865         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3866
3867         /*To keep backward compatibility with older userspace,
3868          *x86 needs to hanlde !user_alloc case.
3869          */
3870         if (!user_alloc) {
3871                 if (npages && !old.rmap) {
3872                         down_write(&current->mm->mmap_sem);
3873                         memslot->userspace_addr = do_mmap(NULL, 0,
3874                                                      npages * PAGE_SIZE,
3875                                                      PROT_READ | PROT_WRITE,
3876                                                      MAP_SHARED | MAP_ANONYMOUS,
3877                                                      0);
3878                         up_write(&current->mm->mmap_sem);
3879
3880                         if (IS_ERR((void *)memslot->userspace_addr))
3881                                 return PTR_ERR((void *)memslot->userspace_addr);
3882                 } else {
3883                         if (!old.user_alloc && old.rmap) {
3884                                 int ret;
3885
3886                                 down_write(&current->mm->mmap_sem);
3887                                 ret = do_munmap(current->mm, old.userspace_addr,
3888                                                 old.npages * PAGE_SIZE);
3889                                 up_write(&current->mm->mmap_sem);
3890                                 if (ret < 0)
3891                                         printk(KERN_WARNING
3892                                        "kvm_vm_ioctl_set_memory_region: "
3893                                        "failed to munmap memory\n");
3894                         }
3895                 }
3896         }
3897
3898         if (!kvm->arch.n_requested_mmu_pages) {
3899                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3900                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3901         }
3902
3903         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3904         kvm_flush_remote_tlbs(kvm);
3905
3906         return 0;
3907 }
3908
3909 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3910 {
3911         return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3912                || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3913 }
3914
3915 static void vcpu_kick_intr(void *info)
3916 {
3917 #ifdef DEBUG
3918         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3919         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3920 #endif
3921 }
3922
3923 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3924 {
3925         int ipi_pcpu = vcpu->cpu;
3926
3927         if (waitqueue_active(&vcpu->wq)) {
3928                 wake_up_interruptible(&vcpu->wq);
3929                 ++vcpu->stat.halt_wakeup;
3930         }
3931         if (vcpu->guest_mode)
3932                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
3933 }