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  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/intel-iommu.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/msr.h>
  41 #include <asm/desc.h>
  42 #include <asm/mtrr.h>
  43
  44 #define MAX_IO_MSRS 256
  45 #define CR0_RESERVED_BITS                                               \
  46         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  47                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  48                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  49 #define CR4_RESERVED_BITS                                               \
  50         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  51                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  52                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  53                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  54
  55 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  56 /* EFER defaults:
  57  * - enable syscall per default because its emulated by KVM
  58  * - enable LME and LMA per default on 64 bit KVM
  59  */
  60 #ifdef CONFIG_X86_64
  61 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  62 #else
  63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  64 #endif
  65
  66 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  67 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  68
  69 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  70                                     struct kvm_cpuid_entry2 __user *entries);
  71
  72 struct kvm_x86_ops *kvm_x86_ops;
  73 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  74
  75 struct kvm_stats_debugfs_item debugfs_entries[] = {
  76         { "pf_fixed", VCPU_STAT(pf_fixed) },
  77         { "pf_guest", VCPU_STAT(pf_guest) },
  78         { "tlb_flush", VCPU_STAT(tlb_flush) },
  79         { "invlpg", VCPU_STAT(invlpg) },
  80         { "exits", VCPU_STAT(exits) },
  81         { "io_exits", VCPU_STAT(io_exits) },
  82         { "mmio_exits", VCPU_STAT(mmio_exits) },
  83         { "signal_exits", VCPU_STAT(signal_exits) },
  84         { "irq_window", VCPU_STAT(irq_window_exits) },
  85         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  86         { "halt_exits", VCPU_STAT(halt_exits) },
  87         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  88         { "hypercalls", VCPU_STAT(hypercalls) },
  89         { "request_irq", VCPU_STAT(request_irq_exits) },
  90         { "request_nmi", VCPU_STAT(request_nmi_exits) },
  91         { "irq_exits", VCPU_STAT(irq_exits) },
  92         { "host_state_reload", VCPU_STAT(host_state_reload) },
  93         { "efer_reload", VCPU_STAT(efer_reload) },
  94         { "fpu_reload", VCPU_STAT(fpu_reload) },
  95         { "insn_emulation", VCPU_STAT(insn_emulation) },
  96         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  97         { "irq_injections", VCPU_STAT(irq_injections) },
  98         { "nmi_injections", VCPU_STAT(nmi_injections) },
  99         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 100         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 101         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 102         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 103         { "mmu_flooded", VM_STAT(mmu_flooded) },
 104         { "mmu_recycled", VM_STAT(mmu_recycled) },
 105         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 106         { "mmu_unsync", VM_STAT(mmu_unsync) },
 107         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 108         { "largepages", VM_STAT(lpages) },
 109         { NULL }
 110 };
 111
 112 unsigned long segment_base(u16 selector)
 113 {
 114         struct descriptor_table gdt;
 115         struct desc_struct *d;
 116         unsigned long table_base;
 117         unsigned long v;
 118
 119         if (selector == 0)
 120                 return 0;
 121
 122         asm("sgdt %0" : "=m"(gdt));
 123         table_base = gdt.base;
 124
 125         if (selector & 4) {           /* from ldt */
 126                 u16 ldt_selector;
 127
 128                 asm("sldt %0" : "=g"(ldt_selector));
 129                 table_base = segment_base(ldt_selector);
 130         }
 131         d = (struct desc_struct *)(table_base + (selector & ~7));
 132         v = d->base0 | ((unsigned long)d->base1 << 16) |
 133                 ((unsigned long)d->base2 << 24);
 134 #ifdef CONFIG_X86_64
 135         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 136                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 137 #endif
 138         return v;
 139 }
 140 EXPORT_SYMBOL_GPL(segment_base);
 141
 142 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 143 {
 144         if (irqchip_in_kernel(vcpu->kvm))
 145                 return vcpu->arch.apic_base;
 146         else
 147                 return vcpu->arch.apic_base;
 148 }
 149 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 150
 151 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 152 {
 153         /* TODO: reserve bits check */
 154         if (irqchip_in_kernel(vcpu->kvm))
 155                 kvm_lapic_set_base(vcpu, data);
 156         else
 157                 vcpu->arch.apic_base = data;
 158 }
 159 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 160
 161 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 162 {
 163         WARN_ON(vcpu->arch.exception.pending);
 164         vcpu->arch.exception.pending = true;
 165         vcpu->arch.exception.has_error_code = false;
 166         vcpu->arch.exception.nr = nr;
 167 }
 168 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 169
 170 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 171                            u32 error_code)
 172 {
 173         ++vcpu->stat.pf_guest;
 174         if (vcpu->arch.exception.pending) {
 175                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 176                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 177                                         " double fault 0x%lx\n", addr);
 178                         vcpu->arch.exception.nr = DF_VECTOR;
 179                         vcpu->arch.exception.error_code = 0;
 180                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 181                         /* triple fault -> shutdown */
 182                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 183                 }
 184                 return;
 185         }
 186         vcpu->arch.cr2 = addr;
 187         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 188 }
 189
 190 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 191 {
 192         vcpu->arch.nmi_pending = 1;
 193 }
 194 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 195
 196 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 197 {
 198         WARN_ON(vcpu->arch.exception.pending);
 199         vcpu->arch.exception.pending = true;
 200         vcpu->arch.exception.has_error_code = true;
 201         vcpu->arch.exception.nr = nr;
 202         vcpu->arch.exception.error_code = error_code;
 203 }
 204 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 205
 206 static void __queue_exception(struct kvm_vcpu *vcpu)
 207 {
 208         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 209                                      vcpu->arch.exception.has_error_code,
 210                                      vcpu->arch.exception.error_code);
 211 }
 212
 213 /*
 214  * Load the pae pdptrs.  Return true is they are all valid.
 215  */
 216 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 217 {
 218         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 219         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 220         int i;
 221         int ret;
 222         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 223
 224         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 225                                   offset * sizeof(u64), sizeof(pdpte));
 226         if (ret < 0) {
 227                 ret = 0;
 228                 goto out;
 229         }
 230         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 231                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 232                         ret = 0;
 233                         goto out;
 234                 }
 235         }
 236         ret = 1;
 237
 238         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 239 out:
 240
 241         return ret;
 242 }
 243 EXPORT_SYMBOL_GPL(load_pdptrs);
 244
 245 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 246 {
 247         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 248         bool changed = true;
 249         int r;
 250
 251         if (is_long_mode(vcpu) || !is_pae(vcpu))
 252                 return false;
 253
 254         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 255         if (r < 0)
 256                 goto out;
 257         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 258 out:
 259
 260         return changed;
 261 }
 262
 263 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 264 {
 265         if (cr0 & CR0_RESERVED_BITS) {
 266                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 267                        cr0, vcpu->arch.cr0);
 268                 kvm_inject_gp(vcpu, 0);
 269                 return;
 270         }
 271
 272         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 273                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 274                 kvm_inject_gp(vcpu, 0);
 275                 return;
 276         }
 277
 278         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 279                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 280                        "and a clear PE flag\n");
 281                 kvm_inject_gp(vcpu, 0);
 282                 return;
 283         }
 284
 285         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 286 #ifdef CONFIG_X86_64
 287                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 288                         int cs_db, cs_l;
 289
 290                         if (!is_pae(vcpu)) {
 291                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 292                                        "in long mode while PAE is disabled\n");
 293                                 kvm_inject_gp(vcpu, 0);
 294                                 return;
 295                         }
 296                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 297                         if (cs_l) {
 298                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 299                                        "in long mode while CS.L == 1\n");
 300                                 kvm_inject_gp(vcpu, 0);
 301                                 return;
 302
 303                         }
 304                 } else
 305 #endif
 306                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 307                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 308                                "reserved bits\n");
 309                         kvm_inject_gp(vcpu, 0);
 310                         return;
 311                 }
 312
 313         }
 314
 315         kvm_x86_ops->set_cr0(vcpu, cr0);
 316         vcpu->arch.cr0 = cr0;
 317
 318         kvm_mmu_reset_context(vcpu);
 319         return;
 320 }
 321 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 322
 323 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 324 {
 325         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 326         KVMTRACE_1D(LMSW, vcpu,
 327                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 328                     handler);
 329 }
 330 EXPORT_SYMBOL_GPL(kvm_lmsw);
 331
 332 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 333 {
 334         if (cr4 & CR4_RESERVED_BITS) {
 335                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 336                 kvm_inject_gp(vcpu, 0);
 337                 return;
 338         }
 339
 340         if (is_long_mode(vcpu)) {
 341                 if (!(cr4 & X86_CR4_PAE)) {
 342                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 343                                "in long mode\n");
 344                         kvm_inject_gp(vcpu, 0);
 345                         return;
 346                 }
 347         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 348                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 349                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 350                 kvm_inject_gp(vcpu, 0);
 351                 return;
 352         }
 353
 354         if (cr4 & X86_CR4_VMXE) {
 355                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 356                 kvm_inject_gp(vcpu, 0);
 357                 return;
 358         }
 359         kvm_x86_ops->set_cr4(vcpu, cr4);
 360         vcpu->arch.cr4 = cr4;
 361         kvm_mmu_reset_context(vcpu);
 362 }
 363 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 364
 365 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 366 {
 367         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 368                 kvm_mmu_sync_roots(vcpu);
 369                 kvm_mmu_flush_tlb(vcpu);
 370                 return;
 371         }
 372
 373         if (is_long_mode(vcpu)) {
 374                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 375                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 376                         kvm_inject_gp(vcpu, 0);
 377                         return;
 378                 }
 379         } else {
 380                 if (is_pae(vcpu)) {
 381                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 382                                 printk(KERN_DEBUG
 383                                        "set_cr3: #GP, reserved bits\n");
 384                                 kvm_inject_gp(vcpu, 0);
 385                                 return;
 386                         }
 387                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 388                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 389                                        "reserved bits\n");
 390                                 kvm_inject_gp(vcpu, 0);
 391                                 return;
 392                         }
 393                 }
 394                 /*
 395                  * We don't check reserved bits in nonpae mode, because
 396                  * this isn't enforced, and VMware depends on this.
 397                  */
 398         }
 399
 400         /*
 401          * Does the new cr3 value map to physical memory? (Note, we
 402          * catch an invalid cr3 even in real-mode, because it would
 403          * cause trouble later on when we turn on paging anyway.)
 404          *
 405          * A real CPU would silently accept an invalid cr3 and would
 406          * attempt to use it - with largely undefined (and often hard
 407          * to debug) behavior on the guest side.
 408          */
 409         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 410                 kvm_inject_gp(vcpu, 0);
 411         else {
 412                 vcpu->arch.cr3 = cr3;
 413                 vcpu->arch.mmu.new_cr3(vcpu);
 414         }
 415 }
 416 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 417
 418 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 419 {
 420         if (cr8 & CR8_RESERVED_BITS) {
 421                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 422                 kvm_inject_gp(vcpu, 0);
 423                 return;
 424         }
 425         if (irqchip_in_kernel(vcpu->kvm))
 426                 kvm_lapic_set_tpr(vcpu, cr8);
 427         else
 428                 vcpu->arch.cr8 = cr8;
 429 }
 430 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 431
 432 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 433 {
 434         if (irqchip_in_kernel(vcpu->kvm))
 435                 return kvm_lapic_get_cr8(vcpu);
 436         else
 437                 return vcpu->arch.cr8;
 438 }
 439 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 440
 441 /*
 442  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 443  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 444  *
 445  * This list is modified at module load time to reflect the
 446  * capabilities of the host cpu.
 447  */
 448 static u32 msrs_to_save[] = {
 449         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 450         MSR_K6_STAR,
 451 #ifdef CONFIG_X86_64
 452         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 453 #endif
 454         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 455         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT
 456 };
 457
 458 static unsigned num_msrs_to_save;
 459
 460 static u32 emulated_msrs[] = {
 461         MSR_IA32_MISC_ENABLE,
 462 };
 463
 464 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 465 {
 466         if (efer & efer_reserved_bits) {
 467                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 468                        efer);
 469                 kvm_inject_gp(vcpu, 0);
 470                 return;
 471         }
 472
 473         if (is_paging(vcpu)
 474             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 475                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 476                 kvm_inject_gp(vcpu, 0);
 477                 return;
 478         }
 479
 480         kvm_x86_ops->set_efer(vcpu, efer);
 481
 482         efer &= ~EFER_LMA;
 483         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 484
 485         vcpu->arch.shadow_efer = efer;
 486 }
 487
 488 void kvm_enable_efer_bits(u64 mask)
 489 {
 490        efer_reserved_bits &= ~mask;
 491 }
 492 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 493
 494
 495 /*
 496  * Writes msr value into into the appropriate "register".
 497  * Returns 0 on success, non-0 otherwise.
 498  * Assumes vcpu_load() was already called.
 499  */
 500 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 501 {
 502         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 503 }
 504
 505 /*
 506  * Adapt set_msr() to msr_io()'s calling convention
 507  */
 508 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 509 {
 510         return kvm_set_msr(vcpu, index, *data);
 511 }
 512
 513 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 514 {
 515         static int version;
 516         struct pvclock_wall_clock wc;
 517         struct timespec now, sys, boot;
 518
 519         if (!wall_clock)
 520                 return;
 521
 522         version++;
 523
 524         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 525
 526         /*
 527          * The guest calculates current wall clock time by adding
 528          * system time (updated by kvm_write_guest_time below) to the
 529          * wall clock specified here.  guest system time equals host
 530          * system time for us, thus we must fill in host boot time here.
 531          */
 532         now = current_kernel_time();
 533         ktime_get_ts(&sys);
 534         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 535
 536         wc.sec = boot.tv_sec;
 537         wc.nsec = boot.tv_nsec;
 538         wc.version = version;
 539
 540         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 541
 542         version++;
 543         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 544 }
 545
 546 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 547 {
 548         uint32_t quotient, remainder;
 549
 550         /* Don't try to replace with do_div(), this one calculates
 551          * "(dividend << 32) / divisor" */
 552         __asm__ ( "divl %4"
 553                   : "=a" (quotient), "=d" (remainder)
 554                   : "0" (0), "1" (dividend), "r" (divisor) );
 555         return quotient;
 556 }
 557
 558 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 559 {
 560         uint64_t nsecs = 1000000000LL;
 561         int32_t  shift = 0;
 562         uint64_t tps64;
 563         uint32_t tps32;
 564
 565         tps64 = tsc_khz * 1000LL;
 566         while (tps64 > nsecs*2) {
 567                 tps64 >>= 1;
 568                 shift--;
 569         }
 570
 571         tps32 = (uint32_t)tps64;
 572         while (tps32 <= (uint32_t)nsecs) {
 573                 tps32 <<= 1;
 574                 shift++;
 575         }
 576
 577         hv_clock->tsc_shift = shift;
 578         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 579
 580         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 581                  __func__, tsc_khz, hv_clock->tsc_shift,
 582                  hv_clock->tsc_to_system_mul);
 583 }
 584
 585 static void kvm_write_guest_time(struct kvm_vcpu *v)
 586 {
 587         struct timespec ts;
 588         unsigned long flags;
 589         struct kvm_vcpu_arch *vcpu = &v->arch;
 590         void *shared_kaddr;
 591
 592         if ((!vcpu->time_page))
 593                 return;
 594
 595         if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
 596                 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
 597                 vcpu->hv_clock_tsc_khz = tsc_khz;
 598         }
 599
 600         /* Keep irq disabled to prevent changes to the clock */
 601         local_irq_save(flags);
 602         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 603                           &vcpu->hv_clock.tsc_timestamp);
 604         ktime_get_ts(&ts);
 605         local_irq_restore(flags);
 606
 607         /* With all the info we got, fill in the values */
 608
 609         vcpu->hv_clock.system_time = ts.tv_nsec +
 610                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 611         /*
 612          * The interface expects us to write an even number signaling that the
 613          * update is finished. Since the guest won't see the intermediate
 614          * state, we just increase by 2 at the end.
 615          */
 616         vcpu->hv_clock.version += 2;
 617
 618         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 619
 620         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 621                sizeof(vcpu->hv_clock));
 622
 623         kunmap_atomic(shared_kaddr, KM_USER0);
 624
 625         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 626 }
 627
 628 static bool msr_mtrr_valid(unsigned msr)
 629 {
 630         switch (msr) {
 631         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 632         case MSR_MTRRfix64K_00000:
 633         case MSR_MTRRfix16K_80000:
 634         case MSR_MTRRfix16K_A0000:
 635         case MSR_MTRRfix4K_C0000:
 636         case MSR_MTRRfix4K_C8000:
 637         case MSR_MTRRfix4K_D0000:
 638         case MSR_MTRRfix4K_D8000:
 639         case MSR_MTRRfix4K_E0000:
 640         case MSR_MTRRfix4K_E8000:
 641         case MSR_MTRRfix4K_F0000:
 642         case MSR_MTRRfix4K_F8000:
 643         case MSR_MTRRdefType:
 644         case MSR_IA32_CR_PAT:
 645                 return true;
 646         case 0x2f8:
 647                 return true;
 648         }
 649         return false;
 650 }
 651
 652 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 653 {
 654         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 655
 656         if (!msr_mtrr_valid(msr))
 657                 return 1;
 658
 659         if (msr == MSR_MTRRdefType) {
 660                 vcpu->arch.mtrr_state.def_type = data;
 661                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 662         } else if (msr == MSR_MTRRfix64K_00000)
 663                 p[0] = data;
 664         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 665                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 666         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 667                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 668         else if (msr == MSR_IA32_CR_PAT)
 669                 vcpu->arch.pat = data;
 670         else {  /* Variable MTRRs */
 671                 int idx, is_mtrr_mask;
 672                 u64 *pt;
 673
 674                 idx = (msr - 0x200) / 2;
 675                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 676                 if (!is_mtrr_mask)
 677                         pt =
 678                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 679                 else
 680                         pt =
 681                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 682                 *pt = data;
 683         }
 684
 685         kvm_mmu_reset_context(vcpu);
 686         return 0;
 687 }
 688
 689 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 690 {
 691         switch (msr) {
 692         case MSR_EFER:
 693                 set_efer(vcpu, data);
 694                 break;
 695         case MSR_IA32_MC0_STATUS:
 696                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 697                        __func__, data);
 698                 break;
 699         case MSR_IA32_MCG_STATUS:
 700                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 701                         __func__, data);
 702                 break;
 703         case MSR_IA32_MCG_CTL:
 704                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 705                         __func__, data);
 706                 break;
 707         case MSR_IA32_DEBUGCTLMSR:
 708                 if (!data) {
 709                         /* We support the non-activated case already */
 710                         break;
 711                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 712                         /* Values other than LBR and BTF are vendor-specific,
 713                            thus reserved and should throw a #GP */
 714                         return 1;
 715                 }
 716                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 717                         __func__, data);
 718                 break;
 719         case MSR_IA32_UCODE_REV:
 720         case MSR_IA32_UCODE_WRITE:
 721                 break;
 722         case 0x200 ... 0x2ff:
 723                 return set_msr_mtrr(vcpu, msr, data);
 724         case MSR_IA32_APICBASE:
 725                 kvm_set_apic_base(vcpu, data);
 726                 break;
 727         case MSR_IA32_MISC_ENABLE:
 728                 vcpu->arch.ia32_misc_enable_msr = data;
 729                 break;
 730         case MSR_KVM_WALL_CLOCK:
 731                 vcpu->kvm->arch.wall_clock = data;
 732                 kvm_write_wall_clock(vcpu->kvm, data);
 733                 break;
 734         case MSR_KVM_SYSTEM_TIME: {
 735                 if (vcpu->arch.time_page) {
 736                         kvm_release_page_dirty(vcpu->arch.time_page);
 737                         vcpu->arch.time_page = NULL;
 738                 }
 739
 740                 vcpu->arch.time = data;
 741
 742                 /* we verify if the enable bit is set... */
 743                 if (!(data & 1))
 744                         break;
 745
 746                 /* ...but clean it before doing the actual write */
 747                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 748
 749                 vcpu->arch.time_page =
 750                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 751
 752                 if (is_error_page(vcpu->arch.time_page)) {
 753                         kvm_release_page_clean(vcpu->arch.time_page);
 754                         vcpu->arch.time_page = NULL;
 755                 }
 756
 757                 kvm_write_guest_time(vcpu);
 758                 break;
 759         }
 760         default:
 761                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 762                 return 1;
 763         }
 764         return 0;
 765 }
 766 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 767
 768
 769 /*
 770  * Reads an msr value (of 'msr_index') into 'pdata'.
 771  * Returns 0 on success, non-0 otherwise.
 772  * Assumes vcpu_load() was already called.
 773  */
 774 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 775 {
 776         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 777 }
 778
 779 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 780 {
 781         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 782
 783         if (!msr_mtrr_valid(msr))
 784                 return 1;
 785
 786         if (msr == MSR_MTRRdefType)
 787                 *pdata = vcpu->arch.mtrr_state.def_type +
 788                          (vcpu->arch.mtrr_state.enabled << 10);
 789         else if (msr == MSR_MTRRfix64K_00000)
 790                 *pdata = p[0];
 791         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 792                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 793         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 794                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 795         else if (msr == MSR_IA32_CR_PAT)
 796                 *pdata = vcpu->arch.pat;
 797         else {  /* Variable MTRRs */
 798                 int idx, is_mtrr_mask;
 799                 u64 *pt;
 800
 801                 idx = (msr - 0x200) / 2;
 802                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 803                 if (!is_mtrr_mask)
 804                         pt =
 805                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 806                 else
 807                         pt =
 808                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 809                 *pdata = *pt;
 810         }
 811
 812         return 0;
 813 }
 814
 815 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 816 {
 817         u64 data;
 818
 819         switch (msr) {
 820         case 0xc0010010: /* SYSCFG */
 821         case 0xc0010015: /* HWCR */
 822         case MSR_IA32_PLATFORM_ID:
 823         case MSR_IA32_P5_MC_ADDR:
 824         case MSR_IA32_P5_MC_TYPE:
 825         case MSR_IA32_MC0_CTL:
 826         case MSR_IA32_MCG_STATUS:
 827         case MSR_IA32_MCG_CAP:
 828         case MSR_IA32_MCG_CTL:
 829         case MSR_IA32_MC0_MISC:
 830         case MSR_IA32_MC0_MISC+4:
 831         case MSR_IA32_MC0_MISC+8:
 832         case MSR_IA32_MC0_MISC+12:
 833         case MSR_IA32_MC0_MISC+16:
 834         case MSR_IA32_MC0_MISC+20:
 835         case MSR_IA32_UCODE_REV:
 836         case MSR_IA32_EBL_CR_POWERON:
 837         case MSR_IA32_DEBUGCTLMSR:
 838         case MSR_IA32_LASTBRANCHFROMIP:
 839         case MSR_IA32_LASTBRANCHTOIP:
 840         case MSR_IA32_LASTINTFROMIP:
 841         case MSR_IA32_LASTINTTOIP:
 842                 data = 0;
 843                 break;
 844         case MSR_MTRRcap:
 845                 data = 0x500 | KVM_NR_VAR_MTRR;
 846                 break;
 847         case 0x200 ... 0x2ff:
 848                 return get_msr_mtrr(vcpu, msr, pdata);
 849         case 0xcd: /* fsb frequency */
 850                 data = 3;
 851                 break;
 852         case MSR_IA32_APICBASE:
 853                 data = kvm_get_apic_base(vcpu);
 854                 break;
 855         case MSR_IA32_MISC_ENABLE:
 856                 data = vcpu->arch.ia32_misc_enable_msr;
 857                 break;
 858         case MSR_IA32_PERF_STATUS:
 859                 /* TSC increment by tick */
 860                 data = 1000ULL;
 861                 /* CPU multiplier */
 862                 data |= (((uint64_t)4ULL) << 40);
 863                 break;
 864         case MSR_EFER:
 865                 data = vcpu->arch.shadow_efer;
 866                 break;
 867         case MSR_KVM_WALL_CLOCK:
 868                 data = vcpu->kvm->arch.wall_clock;
 869                 break;
 870         case MSR_KVM_SYSTEM_TIME:
 871                 data = vcpu->arch.time;
 872                 break;
 873         default:
 874                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 875                 return 1;
 876         }
 877         *pdata = data;
 878         return 0;
 879 }
 880 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 881
 882 /*
 883  * Read or write a bunch of msrs. All parameters are kernel addresses.
 884  *
 885  * @return number of msrs set successfully.
 886  */
 887 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 888                     struct kvm_msr_entry *entries,
 889                     int (*do_msr)(struct kvm_vcpu *vcpu,
 890                                   unsigned index, u64 *data))
 891 {
 892         int i;
 893
 894         vcpu_load(vcpu);
 895
 896         down_read(&vcpu->kvm->slots_lock);
 897         for (i = 0; i < msrs->nmsrs; ++i)
 898                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 899                         break;
 900         up_read(&vcpu->kvm->slots_lock);
 901
 902         vcpu_put(vcpu);
 903
 904         return i;
 905 }
 906
 907 /*
 908  * Read or write a bunch of msrs. Parameters are user addresses.
 909  *
 910  * @return number of msrs set successfully.
 911  */
 912 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 913                   int (*do_msr)(struct kvm_vcpu *vcpu,
 914                                 unsigned index, u64 *data),
 915                   int writeback)
 916 {
 917         struct kvm_msrs msrs;
 918         struct kvm_msr_entry *entries;
 919         int r, n;
 920         unsigned size;
 921
 922         r = -EFAULT;
 923         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 924                 goto out;
 925
 926         r = -E2BIG;
 927         if (msrs.nmsrs >= MAX_IO_MSRS)
 928                 goto out;
 929
 930         r = -ENOMEM;
 931         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 932         entries = vmalloc(size);
 933         if (!entries)
 934                 goto out;
 935
 936         r = -EFAULT;
 937         if (copy_from_user(entries, user_msrs->entries, size))
 938                 goto out_free;
 939
 940         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 941         if (r < 0)
 942                 goto out_free;
 943
 944         r = -EFAULT;
 945         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 946                 goto out_free;
 947
 948         r = n;
 949
 950 out_free:
 951         vfree(entries);
 952 out:
 953         return r;
 954 }
 955
 956 int kvm_dev_ioctl_check_extension(long ext)
 957 {
 958         int r;
 959
 960         switch (ext) {
 961         case KVM_CAP_IRQCHIP:
 962         case KVM_CAP_HLT:
 963         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 964         case KVM_CAP_USER_MEMORY:
 965         case KVM_CAP_SET_TSS_ADDR:
 966         case KVM_CAP_EXT_CPUID:
 967         case KVM_CAP_CLOCKSOURCE:
 968         case KVM_CAP_PIT:
 969         case KVM_CAP_NOP_IO_DELAY:
 970         case KVM_CAP_MP_STATE:
 971         case KVM_CAP_SYNC_MMU:
 972                 r = 1;
 973                 break;
 974         case KVM_CAP_COALESCED_MMIO:
 975                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 976                 break;
 977         case KVM_CAP_VAPIC:
 978                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 979                 break;
 980         case KVM_CAP_NR_VCPUS:
 981                 r = KVM_MAX_VCPUS;
 982                 break;
 983         case KVM_CAP_NR_MEMSLOTS:
 984                 r = KVM_MEMORY_SLOTS;
 985                 break;
 986         case KVM_CAP_PV_MMU:
 987                 r = !tdp_enabled;
 988                 break;
 989         case KVM_CAP_IOMMU:
 990                 r = intel_iommu_found();
 991                 break;
 992         default:
 993                 r = 0;
 994                 break;
 995         }
 996         return r;
 997
 998 }
 999
1000 long kvm_arch_dev_ioctl(struct file *filp,
1001                         unsigned int ioctl, unsigned long arg)
1002 {
1003         void __user *argp = (void __user *)arg;
1004         long r;
1005
1006         switch (ioctl) {
1007         case KVM_GET_MSR_INDEX_LIST: {
1008                 struct kvm_msr_list __user *user_msr_list = argp;
1009                 struct kvm_msr_list msr_list;
1010                 unsigned n;
1011
1012                 r = -EFAULT;
1013                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1014                         goto out;
1015                 n = msr_list.nmsrs;
1016                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1017                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1018                         goto out;
1019                 r = -E2BIG;
1020                 if (n < num_msrs_to_save)
1021                         goto out;
1022                 r = -EFAULT;
1023                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1024                                  num_msrs_to_save * sizeof(u32)))
1025                         goto out;
1026                 if (copy_to_user(user_msr_list->indices
1027                                  + num_msrs_to_save * sizeof(u32),
1028                                  &emulated_msrs,
1029                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1030                         goto out;
1031                 r = 0;
1032                 break;
1033         }
1034         case KVM_GET_SUPPORTED_CPUID: {
1035                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1036                 struct kvm_cpuid2 cpuid;
1037
1038                 r = -EFAULT;
1039                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1040                         goto out;
1041                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1042                         cpuid_arg->entries);
1043                 if (r)
1044                         goto out;
1045
1046                 r = -EFAULT;
1047                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1048                         goto out;
1049                 r = 0;
1050                 break;
1051         }
1052         default:
1053                 r = -EINVAL;
1054         }
1055 out:
1056         return r;
1057 }
1058
1059 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1060 {
1061         kvm_x86_ops->vcpu_load(vcpu, cpu);
1062         kvm_write_guest_time(vcpu);
1063 }
1064
1065 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1066 {
1067         kvm_x86_ops->vcpu_put(vcpu);
1068         kvm_put_guest_fpu(vcpu);
1069 }
1070
1071 static int is_efer_nx(void)
1072 {
1073         u64 efer;
1074
1075         rdmsrl(MSR_EFER, efer);
1076         return efer & EFER_NX;
1077 }
1078
1079 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1080 {
1081         int i;
1082         struct kvm_cpuid_entry2 *e, *entry;
1083
1084         entry = NULL;
1085         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1086                 e = &vcpu->arch.cpuid_entries[i];
1087                 if (e->function == 0x80000001) {
1088                         entry = e;
1089                         break;
1090                 }
1091         }
1092         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1093                 entry->edx &= ~(1 << 20);
1094                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1095         }
1096 }
1097
1098 /* when an old userspace process fills a new kernel module */
1099 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1100                                     struct kvm_cpuid *cpuid,
1101                                     struct kvm_cpuid_entry __user *entries)
1102 {
1103         int r, i;
1104         struct kvm_cpuid_entry *cpuid_entries;
1105
1106         r = -E2BIG;
1107         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1108                 goto out;
1109         r = -ENOMEM;
1110         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1111         if (!cpuid_entries)
1112                 goto out;
1113         r = -EFAULT;
1114         if (copy_from_user(cpuid_entries, entries,
1115                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1116                 goto out_free;
1117         for (i = 0; i < cpuid->nent; i++) {
1118                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1119                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1120                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1121                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1122                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1123                 vcpu->arch.cpuid_entries[i].index = 0;
1124                 vcpu->arch.cpuid_entries[i].flags = 0;
1125                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1126                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1127                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1128         }
1129         vcpu->arch.cpuid_nent = cpuid->nent;
1130         cpuid_fix_nx_cap(vcpu);
1131         r = 0;
1132
1133 out_free:
1134         vfree(cpuid_entries);
1135 out:
1136         return r;
1137 }
1138
1139 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1140                                     struct kvm_cpuid2 *cpuid,
1141                                     struct kvm_cpuid_entry2 __user *entries)
1142 {
1143         int r;
1144
1145         r = -E2BIG;
1146         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1147                 goto out;
1148         r = -EFAULT;
1149         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1150                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1151                 goto out;
1152         vcpu->arch.cpuid_nent = cpuid->nent;
1153         return 0;
1154
1155 out:
1156         return r;
1157 }
1158
1159 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1160                                     struct kvm_cpuid2 *cpuid,
1161                                     struct kvm_cpuid_entry2 __user *entries)
1162 {
1163         int r;
1164
1165         r = -E2BIG;
1166         if (cpuid->nent < vcpu->arch.cpuid_nent)
1167                 goto out;
1168         r = -EFAULT;
1169         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1170                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1171                 goto out;
1172         return 0;
1173
1174 out:
1175         cpuid->nent = vcpu->arch.cpuid_nent;
1176         return r;
1177 }
1178
1179 static inline u32 bit(int bitno)
1180 {
1181         return 1 << (bitno & 31);
1182 }
1183
1184 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1185                           u32 index)
1186 {
1187         entry->function = function;
1188         entry->index = index;
1189         cpuid_count(entry->function, entry->index,
1190                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1191         entry->flags = 0;
1192 }
1193
1194 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1195                          u32 index, int *nent, int maxnent)
1196 {
1197         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1198                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1199                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1200                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1201                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1202                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1203                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1204                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1205                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1206                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1207         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1208                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1209                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1210                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1211                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1212                 bit(X86_FEATURE_PGE) |
1213                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1214                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1215                 bit(X86_FEATURE_SYSCALL) |
1216                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1217 #ifdef CONFIG_X86_64
1218                 bit(X86_FEATURE_LM) |
1219 #endif
1220                 bit(X86_FEATURE_MMXEXT) |
1221                 bit(X86_FEATURE_3DNOWEXT) |
1222                 bit(X86_FEATURE_3DNOW);
1223         const u32 kvm_supported_word3_x86_features =
1224                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1225         const u32 kvm_supported_word6_x86_features =
1226                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1227
1228         /* all func 2 cpuid_count() should be called on the same cpu */
1229         get_cpu();
1230         do_cpuid_1_ent(entry, function, index);
1231         ++*nent;
1232
1233         switch (function) {
1234         case 0:
1235                 entry->eax = min(entry->eax, (u32)0xb);
1236                 break;
1237         case 1:
1238                 entry->edx &= kvm_supported_word0_x86_features;
1239                 entry->ecx &= kvm_supported_word3_x86_features;
1240                 break;
1241         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1242          * may return different values. This forces us to get_cpu() before
1243          * issuing the first command, and also to emulate this annoying behavior
1244          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1245         case 2: {
1246                 int t, times = entry->eax & 0xff;
1247
1248                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1249                 for (t = 1; t < times && *nent < maxnent; ++t) {
1250                         do_cpuid_1_ent(&entry[t], function, 0);
1251                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1252                         ++*nent;
1253                 }
1254                 break;
1255         }
1256         /* function 4 and 0xb have additional index. */
1257         case 4: {
1258                 int i, cache_type;
1259
1260                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1261                 /* read more entries until cache_type is zero */
1262                 for (i = 1; *nent < maxnent; ++i) {
1263                         cache_type = entry[i - 1].eax & 0x1f;
1264                         if (!cache_type)
1265                                 break;
1266                         do_cpuid_1_ent(&entry[i], function, i);
1267                         entry[i].flags |=
1268                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1269                         ++*nent;
1270                 }
1271                 break;
1272         }
1273         case 0xb: {
1274                 int i, level_type;
1275
1276                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1277                 /* read more entries until level_type is zero */
1278                 for (i = 1; *nent < maxnent; ++i) {
1279                         level_type = entry[i - 1].ecx & 0xff00;
1280                         if (!level_type)
1281                                 break;
1282                         do_cpuid_1_ent(&entry[i], function, i);
1283                         entry[i].flags |=
1284                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1285                         ++*nent;
1286                 }
1287                 break;
1288         }
1289         case 0x80000000:
1290                 entry->eax = min(entry->eax, 0x8000001a);
1291                 break;
1292         case 0x80000001:
1293                 entry->edx &= kvm_supported_word1_x86_features;
1294                 entry->ecx &= kvm_supported_word6_x86_features;
1295                 break;
1296         }
1297         put_cpu();
1298 }
1299
1300 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1301                                     struct kvm_cpuid_entry2 __user *entries)
1302 {
1303         struct kvm_cpuid_entry2 *cpuid_entries;
1304         int limit, nent = 0, r = -E2BIG;
1305         u32 func;
1306
1307         if (cpuid->nent < 1)
1308                 goto out;
1309         r = -ENOMEM;
1310         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1311         if (!cpuid_entries)
1312                 goto out;
1313
1314         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1315         limit = cpuid_entries[0].eax;
1316         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1317                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1318                                 &nent, cpuid->nent);
1319         r = -E2BIG;
1320         if (nent >= cpuid->nent)
1321                 goto out_free;
1322
1323         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1324         limit = cpuid_entries[nent - 1].eax;
1325         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1326                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1327                                &nent, cpuid->nent);
1328         r = -EFAULT;
1329         if (copy_to_user(entries, cpuid_entries,
1330                         nent * sizeof(struct kvm_cpuid_entry2)))
1331                 goto out_free;
1332         cpuid->nent = nent;
1333         r = 0;
1334
1335 out_free:
1336         vfree(cpuid_entries);
1337 out:
1338         return r;
1339 }
1340
1341 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1342                                     struct kvm_lapic_state *s)
1343 {
1344         vcpu_load(vcpu);
1345         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1346         vcpu_put(vcpu);
1347
1348         return 0;
1349 }
1350
1351 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1352                                     struct kvm_lapic_state *s)
1353 {
1354         vcpu_load(vcpu);
1355         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1356         kvm_apic_post_state_restore(vcpu);
1357         vcpu_put(vcpu);
1358
1359         return 0;
1360 }
1361
1362 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1363                                     struct kvm_interrupt *irq)
1364 {
1365         if (irq->irq < 0 || irq->irq >= 256)
1366                 return -EINVAL;
1367         if (irqchip_in_kernel(vcpu->kvm))
1368                 return -ENXIO;
1369         vcpu_load(vcpu);
1370
1371         set_bit(irq->irq, vcpu->arch.irq_pending);
1372         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1373
1374         vcpu_put(vcpu);
1375
1376         return 0;
1377 }
1378
1379 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1380 {
1381         vcpu_load(vcpu);
1382         kvm_inject_nmi(vcpu);
1383         vcpu_put(vcpu);
1384
1385         return 0;
1386 }
1387
1388 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1389                                            struct kvm_tpr_access_ctl *tac)
1390 {
1391         if (tac->flags)
1392                 return -EINVAL;
1393         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1394         return 0;
1395 }
1396
1397 long kvm_arch_vcpu_ioctl(struct file *filp,
1398                          unsigned int ioctl, unsigned long arg)
1399 {
1400         struct kvm_vcpu *vcpu = filp->private_data;
1401         void __user *argp = (void __user *)arg;
1402         int r;
1403         struct kvm_lapic_state *lapic = NULL;
1404
1405         switch (ioctl) {
1406         case KVM_GET_LAPIC: {
1407                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1408
1409                 r = -ENOMEM;
1410                 if (!lapic)
1411                         goto out;
1412                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1413                 if (r)
1414                         goto out;
1415                 r = -EFAULT;
1416                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1417                         goto out;
1418                 r = 0;
1419                 break;
1420         }
1421         case KVM_SET_LAPIC: {
1422                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1423                 r = -ENOMEM;
1424                 if (!lapic)
1425                         goto out;
1426                 r = -EFAULT;
1427                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1428                         goto out;
1429                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1430                 if (r)
1431                         goto out;
1432                 r = 0;
1433                 break;
1434         }
1435         case KVM_INTERRUPT: {
1436                 struct kvm_interrupt irq;
1437
1438                 r = -EFAULT;
1439                 if (copy_from_user(&irq, argp, sizeof irq))
1440                         goto out;
1441                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1442                 if (r)
1443                         goto out;
1444                 r = 0;
1445                 break;
1446         }
1447         case KVM_NMI: {
1448                 r = kvm_vcpu_ioctl_nmi(vcpu);
1449                 if (r)
1450                         goto out;
1451                 r = 0;
1452                 break;
1453         }
1454         case KVM_SET_CPUID: {
1455                 struct kvm_cpuid __user *cpuid_arg = argp;
1456                 struct kvm_cpuid cpuid;
1457
1458                 r = -EFAULT;
1459                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1460                         goto out;
1461                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1462                 if (r)
1463                         goto out;
1464                 break;
1465         }
1466         case KVM_SET_CPUID2: {
1467                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1468                 struct kvm_cpuid2 cpuid;
1469
1470                 r = -EFAULT;
1471                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1472                         goto out;
1473                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1474                                 cpuid_arg->entries);
1475                 if (r)
1476                         goto out;
1477                 break;
1478         }
1479         case KVM_GET_CPUID2: {
1480                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1481                 struct kvm_cpuid2 cpuid;
1482
1483                 r = -EFAULT;
1484                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1485                         goto out;
1486                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1487                                 cpuid_arg->entries);
1488                 if (r)
1489                         goto out;
1490                 r = -EFAULT;
1491                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1492                         goto out;
1493                 r = 0;
1494                 break;
1495         }
1496         case KVM_GET_MSRS:
1497                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1498                 break;
1499         case KVM_SET_MSRS:
1500                 r = msr_io(vcpu, argp, do_set_msr, 0);
1501                 break;
1502         case KVM_TPR_ACCESS_REPORTING: {
1503                 struct kvm_tpr_access_ctl tac;
1504
1505                 r = -EFAULT;
1506                 if (copy_from_user(&tac, argp, sizeof tac))
1507                         goto out;
1508                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1509                 if (r)
1510                         goto out;
1511                 r = -EFAULT;
1512                 if (copy_to_user(argp, &tac, sizeof tac))
1513                         goto out;
1514                 r = 0;
1515                 break;
1516         };
1517         case KVM_SET_VAPIC_ADDR: {
1518                 struct kvm_vapic_addr va;
1519
1520                 r = -EINVAL;
1521                 if (!irqchip_in_kernel(vcpu->kvm))
1522                         goto out;
1523                 r = -EFAULT;
1524                 if (copy_from_user(&va, argp, sizeof va))
1525                         goto out;
1526                 r = 0;
1527                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1528                 break;
1529         }
1530         default:
1531                 r = -EINVAL;
1532         }
1533 out:
1534         if (lapic)
1535                 kfree(lapic);
1536         return r;
1537 }
1538
1539 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1540 {
1541         int ret;
1542
1543         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1544                 return -1;
1545         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1546         return ret;
1547 }
1548
1549 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1550                                           u32 kvm_nr_mmu_pages)
1551 {
1552         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1553                 return -EINVAL;
1554
1555         down_write(&kvm->slots_lock);
1556
1557         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1558         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1559
1560         up_write(&kvm->slots_lock);
1561         return 0;
1562 }
1563
1564 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1565 {
1566         return kvm->arch.n_alloc_mmu_pages;
1567 }
1568
1569 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1570 {
1571         int i;
1572         struct kvm_mem_alias *alias;
1573
1574         for (i = 0; i < kvm->arch.naliases; ++i) {
1575                 alias = &kvm->arch.aliases[i];
1576                 if (gfn >= alias->base_gfn
1577                     && gfn < alias->base_gfn + alias->npages)
1578                         return alias->target_gfn + gfn - alias->base_gfn;
1579         }
1580         return gfn;
1581 }
1582
1583 /*
1584  * Set a new alias region.  Aliases map a portion of physical memory into
1585  * another portion.  This is useful for memory windows, for example the PC
1586  * VGA region.
1587  */
1588 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1589                                          struct kvm_memory_alias *alias)
1590 {
1591         int r, n;
1592         struct kvm_mem_alias *p;
1593
1594         r = -EINVAL;
1595         /* General sanity checks */
1596         if (alias->memory_size & (PAGE_SIZE - 1))
1597                 goto out;
1598         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1599                 goto out;
1600         if (alias->slot >= KVM_ALIAS_SLOTS)
1601                 goto out;
1602         if (alias->guest_phys_addr + alias->memory_size
1603             < alias->guest_phys_addr)
1604                 goto out;
1605         if (alias->target_phys_addr + alias->memory_size
1606             < alias->target_phys_addr)
1607                 goto out;
1608
1609         down_write(&kvm->slots_lock);
1610         spin_lock(&kvm->mmu_lock);
1611
1612         p = &kvm->arch.aliases[alias->slot];
1613         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1614         p->npages = alias->memory_size >> PAGE_SHIFT;
1615         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1616
1617         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1618                 if (kvm->arch.aliases[n - 1].npages)
1619                         break;
1620         kvm->arch.naliases = n;
1621
1622         spin_unlock(&kvm->mmu_lock);
1623         kvm_mmu_zap_all(kvm);
1624
1625         up_write(&kvm->slots_lock);
1626
1627         return 0;
1628
1629 out:
1630         return r;
1631 }
1632
1633 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1634 {
1635         int r;
1636
1637         r = 0;
1638         switch (chip->chip_id) {
1639         case KVM_IRQCHIP_PIC_MASTER:
1640                 memcpy(&chip->chip.pic,
1641                         &pic_irqchip(kvm)->pics[0],
1642                         sizeof(struct kvm_pic_state));
1643                 break;
1644         case KVM_IRQCHIP_PIC_SLAVE:
1645                 memcpy(&chip->chip.pic,
1646                         &pic_irqchip(kvm)->pics[1],
1647                         sizeof(struct kvm_pic_state));
1648                 break;
1649         case KVM_IRQCHIP_IOAPIC:
1650                 memcpy(&chip->chip.ioapic,
1651                         ioapic_irqchip(kvm),
1652                         sizeof(struct kvm_ioapic_state));
1653                 break;
1654         default:
1655                 r = -EINVAL;
1656                 break;
1657         }
1658         return r;
1659 }
1660
1661 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1662 {
1663         int r;
1664
1665         r = 0;
1666         switch (chip->chip_id) {
1667         case KVM_IRQCHIP_PIC_MASTER:
1668                 memcpy(&pic_irqchip(kvm)->pics[0],
1669                         &chip->chip.pic,
1670                         sizeof(struct kvm_pic_state));
1671                 break;
1672         case KVM_IRQCHIP_PIC_SLAVE:
1673                 memcpy(&pic_irqchip(kvm)->pics[1],
1674                         &chip->chip.pic,
1675                         sizeof(struct kvm_pic_state));
1676                 break;
1677         case KVM_IRQCHIP_IOAPIC:
1678                 memcpy(ioapic_irqchip(kvm),
1679                         &chip->chip.ioapic,
1680                         sizeof(struct kvm_ioapic_state));
1681                 break;
1682         default:
1683                 r = -EINVAL;
1684                 break;
1685         }
1686         kvm_pic_update_irq(pic_irqchip(kvm));
1687         return r;
1688 }
1689
1690 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1691 {
1692         int r = 0;
1693
1694         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1695         return r;
1696 }
1697
1698 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1699 {
1700         int r = 0;
1701
1702         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1703         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1704         return r;
1705 }
1706
1707 /*
1708  * Get (and clear) the dirty memory log for a memory slot.
1709  */
1710 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1711                                       struct kvm_dirty_log *log)
1712 {
1713         int r;
1714         int n;
1715         struct kvm_memory_slot *memslot;
1716         int is_dirty = 0;
1717
1718         down_write(&kvm->slots_lock);
1719
1720         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1721         if (r)
1722                 goto out;
1723
1724         /* If nothing is dirty, don't bother messing with page tables. */
1725         if (is_dirty) {
1726                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1727                 kvm_flush_remote_tlbs(kvm);
1728                 memslot = &kvm->memslots[log->slot];
1729                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1730                 memset(memslot->dirty_bitmap, 0, n);
1731         }
1732         r = 0;
1733 out:
1734         up_write(&kvm->slots_lock);
1735         return r;
1736 }
1737
1738 long kvm_arch_vm_ioctl(struct file *filp,
1739                        unsigned int ioctl, unsigned long arg)
1740 {
1741         struct kvm *kvm = filp->private_data;
1742         void __user *argp = (void __user *)arg;
1743         int r = -EINVAL;
1744         /*
1745          * This union makes it completely explicit to gcc-3.x
1746          * that these two variables' stack usage should be
1747          * combined, not added together.
1748          */
1749         union {
1750                 struct kvm_pit_state ps;
1751                 struct kvm_memory_alias alias;
1752         } u;
1753
1754         switch (ioctl) {
1755         case KVM_SET_TSS_ADDR:
1756                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1757                 if (r < 0)
1758                         goto out;
1759                 break;
1760         case KVM_SET_MEMORY_REGION: {
1761                 struct kvm_memory_region kvm_mem;
1762                 struct kvm_userspace_memory_region kvm_userspace_mem;
1763
1764                 r = -EFAULT;
1765                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1766                         goto out;
1767                 kvm_userspace_mem.slot = kvm_mem.slot;
1768                 kvm_userspace_mem.flags = kvm_mem.flags;
1769                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1770                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1771                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1772                 if (r)
1773                         goto out;
1774                 break;
1775         }
1776         case KVM_SET_NR_MMU_PAGES:
1777                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1778                 if (r)
1779                         goto out;
1780                 break;
1781         case KVM_GET_NR_MMU_PAGES:
1782                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1783                 break;
1784         case KVM_SET_MEMORY_ALIAS:
1785                 r = -EFAULT;
1786                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1787                         goto out;
1788                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1789                 if (r)
1790                         goto out;
1791                 break;
1792         case KVM_CREATE_IRQCHIP:
1793                 r = -ENOMEM;
1794                 kvm->arch.vpic = kvm_create_pic(kvm);
1795                 if (kvm->arch.vpic) {
1796                         r = kvm_ioapic_init(kvm);
1797                         if (r) {
1798                                 kfree(kvm->arch.vpic);
1799                                 kvm->arch.vpic = NULL;
1800                                 goto out;
1801                         }
1802                 } else
1803                         goto out;
1804                 break;
1805         case KVM_CREATE_PIT:
1806                 r = -ENOMEM;
1807                 kvm->arch.vpit = kvm_create_pit(kvm);
1808                 if (kvm->arch.vpit)
1809                         r = 0;
1810                 break;
1811         case KVM_IRQ_LINE: {
1812                 struct kvm_irq_level irq_event;
1813
1814                 r = -EFAULT;
1815                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1816                         goto out;
1817                 if (irqchip_in_kernel(kvm)) {
1818                         mutex_lock(&kvm->lock);
1819                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1820                                     irq_event.irq, irq_event.level);
1821                         mutex_unlock(&kvm->lock);
1822                         r = 0;
1823                 }
1824                 break;
1825         }
1826         case KVM_GET_IRQCHIP: {
1827                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1828                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1829
1830                 r = -ENOMEM;
1831                 if (!chip)
1832                         goto out;
1833                 r = -EFAULT;
1834                 if (copy_from_user(chip, argp, sizeof *chip))
1835                         goto get_irqchip_out;
1836                 r = -ENXIO;
1837                 if (!irqchip_in_kernel(kvm))
1838                         goto get_irqchip_out;
1839                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1840                 if (r)
1841                         goto get_irqchip_out;
1842                 r = -EFAULT;
1843                 if (copy_to_user(argp, chip, sizeof *chip))
1844                         goto get_irqchip_out;
1845                 r = 0;
1846         get_irqchip_out:
1847                 kfree(chip);
1848                 if (r)
1849                         goto out;
1850                 break;
1851         }
1852         case KVM_SET_IRQCHIP: {
1853                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1854                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1855
1856                 r = -ENOMEM;
1857                 if (!chip)
1858                         goto out;
1859                 r = -EFAULT;
1860                 if (copy_from_user(chip, argp, sizeof *chip))
1861                         goto set_irqchip_out;
1862                 r = -ENXIO;
1863                 if (!irqchip_in_kernel(kvm))
1864                         goto set_irqchip_out;
1865                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1866                 if (r)
1867                         goto set_irqchip_out;
1868                 r = 0;
1869         set_irqchip_out:
1870                 kfree(chip);
1871                 if (r)
1872                         goto out;
1873                 break;
1874         }
1875         case KVM_GET_PIT: {
1876                 r = -EFAULT;
1877                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1878                         goto out;
1879                 r = -ENXIO;
1880                 if (!kvm->arch.vpit)
1881                         goto out;
1882                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1883                 if (r)
1884                         goto out;
1885                 r = -EFAULT;
1886                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1887                         goto out;
1888                 r = 0;
1889                 break;
1890         }
1891         case KVM_SET_PIT: {
1892                 r = -EFAULT;
1893                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1894                         goto out;
1895                 r = -ENXIO;
1896                 if (!kvm->arch.vpit)
1897                         goto out;
1898                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1899                 if (r)
1900                         goto out;
1901                 r = 0;
1902                 break;
1903         }
1904         default:
1905                 ;
1906         }
1907 out:
1908         return r;
1909 }
1910
1911 static void kvm_init_msr_list(void)
1912 {
1913         u32 dummy[2];
1914         unsigned i, j;
1915
1916         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1917                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1918                         continue;
1919                 if (j < i)
1920                         msrs_to_save[j] = msrs_to_save[i];
1921                 j++;
1922         }
1923         num_msrs_to_save = j;
1924 }
1925
1926 /*
1927  * Only apic need an MMIO device hook, so shortcut now..
1928  */
1929 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1930                                                 gpa_t addr, int len,
1931                                                 int is_write)
1932 {
1933         struct kvm_io_device *dev;
1934
1935         if (vcpu->arch.apic) {
1936                 dev = &vcpu->arch.apic->dev;
1937                 if (dev->in_range(dev, addr, len, is_write))
1938                         return dev;
1939         }
1940         return NULL;
1941 }
1942
1943
1944 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1945                                                 gpa_t addr, int len,
1946                                                 int is_write)
1947 {
1948         struct kvm_io_device *dev;
1949
1950         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1951         if (dev == NULL)
1952                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1953                                           is_write);
1954         return dev;
1955 }
1956
1957 int emulator_read_std(unsigned long addr,
1958                              void *val,
1959                              unsigned int bytes,
1960                              struct kvm_vcpu *vcpu)
1961 {
1962         void *data = val;
1963         int r = X86EMUL_CONTINUE;
1964
1965         while (bytes) {
1966                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1967                 unsigned offset = addr & (PAGE_SIZE-1);
1968                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1969                 int ret;
1970
1971                 if (gpa == UNMAPPED_GVA) {
1972                         r = X86EMUL_PROPAGATE_FAULT;
1973                         goto out;
1974                 }
1975                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1976                 if (ret < 0) {
1977                         r = X86EMUL_UNHANDLEABLE;
1978                         goto out;
1979                 }
1980
1981                 bytes -= tocopy;
1982                 data += tocopy;
1983                 addr += tocopy;
1984         }
1985 out:
1986         return r;
1987 }
1988 EXPORT_SYMBOL_GPL(emulator_read_std);
1989
1990 static int emulator_read_emulated(unsigned long addr,
1991                                   void *val,
1992                                   unsigned int bytes,
1993                                   struct kvm_vcpu *vcpu)
1994 {
1995         struct kvm_io_device *mmio_dev;
1996         gpa_t                 gpa;
1997
1998         if (vcpu->mmio_read_completed) {
1999                 memcpy(val, vcpu->mmio_data, bytes);
2000                 vcpu->mmio_read_completed = 0;
2001                 return X86EMUL_CONTINUE;
2002         }
2003
2004         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2005
2006         /* For APIC access vmexit */
2007         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2008                 goto mmio;
2009
2010         if (emulator_read_std(addr, val, bytes, vcpu)
2011                         == X86EMUL_CONTINUE)
2012                 return X86EMUL_CONTINUE;
2013         if (gpa == UNMAPPED_GVA)
2014                 return X86EMUL_PROPAGATE_FAULT;
2015
2016 mmio:
2017         /*
2018          * Is this MMIO handled locally?
2019          */
2020         mutex_lock(&vcpu->kvm->lock);
2021         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2022         if (mmio_dev) {
2023                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2024                 mutex_unlock(&vcpu->kvm->lock);
2025                 return X86EMUL_CONTINUE;
2026         }
2027         mutex_unlock(&vcpu->kvm->lock);
2028
2029         vcpu->mmio_needed = 1;
2030         vcpu->mmio_phys_addr = gpa;
2031         vcpu->mmio_size = bytes;
2032         vcpu->mmio_is_write = 0;
2033
2034         return X86EMUL_UNHANDLEABLE;
2035 }
2036
2037 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2038                           const void *val, int bytes)
2039 {
2040         int ret;
2041
2042         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2043         if (ret < 0)
2044                 return 0;
2045         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2046         return 1;
2047 }
2048
2049 static int emulator_write_emulated_onepage(unsigned long addr,
2050                                            const void *val,
2051                                            unsigned int bytes,
2052                                            struct kvm_vcpu *vcpu)
2053 {
2054         struct kvm_io_device *mmio_dev;
2055         gpa_t                 gpa;
2056
2057         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2058
2059         if (gpa == UNMAPPED_GVA) {
2060                 kvm_inject_page_fault(vcpu, addr, 2);
2061                 return X86EMUL_PROPAGATE_FAULT;
2062         }
2063
2064         /* For APIC access vmexit */
2065         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2066                 goto mmio;
2067
2068         if (emulator_write_phys(vcpu, gpa, val, bytes))
2069                 return X86EMUL_CONTINUE;
2070
2071 mmio:
2072         /*
2073          * Is this MMIO handled locally?
2074          */
2075         mutex_lock(&vcpu->kvm->lock);
2076         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2077         if (mmio_dev) {
2078                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2079                 mutex_unlock(&vcpu->kvm->lock);
2080                 return X86EMUL_CONTINUE;
2081         }
2082         mutex_unlock(&vcpu->kvm->lock);
2083
2084         vcpu->mmio_needed = 1;
2085         vcpu->mmio_phys_addr = gpa;
2086         vcpu->mmio_size = bytes;
2087         vcpu->mmio_is_write = 1;
2088         memcpy(vcpu->mmio_data, val, bytes);
2089
2090         return X86EMUL_CONTINUE;
2091 }
2092
2093 int emulator_write_emulated(unsigned long addr,
2094                                    const void *val,
2095                                    unsigned int bytes,
2096                                    struct kvm_vcpu *vcpu)
2097 {
2098         /* Crossing a page boundary? */
2099         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2100                 int rc, now;
2101
2102                 now = -addr & ~PAGE_MASK;
2103                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2104                 if (rc != X86EMUL_CONTINUE)
2105                         return rc;
2106                 addr += now;
2107                 val += now;
2108                 bytes -= now;
2109         }
2110         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2111 }
2112 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2113
2114 static int emulator_cmpxchg_emulated(unsigned long addr,
2115                                      const void *old,
2116                                      const void *new,
2117                                      unsigned int bytes,
2118                                      struct kvm_vcpu *vcpu)
2119 {
2120         static int reported;
2121
2122         if (!reported) {
2123                 reported = 1;
2124                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2125         }
2126 #ifndef CONFIG_X86_64
2127         /* guests cmpxchg8b have to be emulated atomically */
2128         if (bytes == 8) {
2129                 gpa_t gpa;
2130                 struct page *page;
2131                 char *kaddr;
2132                 u64 val;
2133
2134                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2135
2136                 if (gpa == UNMAPPED_GVA ||
2137                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2138                         goto emul_write;
2139
2140                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2141                         goto emul_write;
2142
2143                 val = *(u64 *)new;
2144
2145                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2146
2147                 kaddr = kmap_atomic(page, KM_USER0);
2148                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2149                 kunmap_atomic(kaddr, KM_USER0);
2150                 kvm_release_page_dirty(page);
2151         }
2152 emul_write:
2153 #endif
2154
2155         return emulator_write_emulated(addr, new, bytes, vcpu);
2156 }
2157
2158 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2159 {
2160         return kvm_x86_ops->get_segment_base(vcpu, seg);
2161 }
2162
2163 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2164 {
2165         kvm_mmu_invlpg(vcpu, address);
2166         return X86EMUL_CONTINUE;
2167 }
2168
2169 int emulate_clts(struct kvm_vcpu *vcpu)
2170 {
2171         KVMTRACE_0D(CLTS, vcpu, handler);
2172         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2173         return X86EMUL_CONTINUE;
2174 }
2175
2176 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2177 {
2178         struct kvm_vcpu *vcpu = ctxt->vcpu;
2179
2180         switch (dr) {
2181         case 0 ... 3:
2182                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2183                 return X86EMUL_CONTINUE;
2184         default:
2185                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2186                 return X86EMUL_UNHANDLEABLE;
2187         }
2188 }
2189
2190 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2191 {
2192         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2193         int exception;
2194
2195         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2196         if (exception) {
2197                 /* FIXME: better handling */
2198                 return X86EMUL_UNHANDLEABLE;
2199         }
2200         return X86EMUL_CONTINUE;
2201 }
2202
2203 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2204 {
2205         u8 opcodes[4];
2206         unsigned long rip = kvm_rip_read(vcpu);
2207         unsigned long rip_linear;
2208
2209         if (!printk_ratelimit())
2210                 return;
2211
2212         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2213
2214         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2215
2216         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2217                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2218 }
2219 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2220
2221 static struct x86_emulate_ops emulate_ops = {
2222         .read_std            = emulator_read_std,
2223         .read_emulated       = emulator_read_emulated,
2224         .write_emulated      = emulator_write_emulated,
2225         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2226 };
2227
2228 static void cache_all_regs(struct kvm_vcpu *vcpu)
2229 {
2230         kvm_register_read(vcpu, VCPU_REGS_RAX);
2231         kvm_register_read(vcpu, VCPU_REGS_RSP);
2232         kvm_register_read(vcpu, VCPU_REGS_RIP);
2233         vcpu->arch.regs_dirty = ~0;
2234 }
2235
2236 int emulate_instruction(struct kvm_vcpu *vcpu,
2237                         struct kvm_run *run,
2238                         unsigned long cr2,
2239                         u16 error_code,
2240                         int emulation_type)
2241 {
2242         int r;
2243         struct decode_cache *c;
2244
2245         kvm_clear_exception_queue(vcpu);
2246         vcpu->arch.mmio_fault_cr2 = cr2;
2247         /*
2248          * TODO: fix x86_emulate.c to use guest_read/write_register
2249          * instead of direct ->regs accesses, can save hundred cycles
2250          * on Intel for instructions that don't read/change RSP, for
2251          * for example.
2252          */
2253         cache_all_regs(vcpu);
2254
2255         vcpu->mmio_is_write = 0;
2256         vcpu->arch.pio.string = 0;
2257
2258         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2259                 int cs_db, cs_l;
2260                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2261
2262                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2263                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2264                 vcpu->arch.emulate_ctxt.mode =
2265                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2266                         ? X86EMUL_MODE_REAL : cs_l
2267                         ? X86EMUL_MODE_PROT64 : cs_db
2268                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2269
2270                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2271
2272                 /* Reject the instructions other than VMCALL/VMMCALL when
2273                  * try to emulate invalid opcode */
2274                 c = &vcpu->arch.emulate_ctxt.decode;
2275                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2276                     (!(c->twobyte && c->b == 0x01 &&
2277                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2278                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2279                         return EMULATE_FAIL;
2280
2281                 ++vcpu->stat.insn_emulation;
2282                 if (r)  {
2283                         ++vcpu->stat.insn_emulation_fail;
2284                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2285                                 return EMULATE_DONE;
2286                         return EMULATE_FAIL;
2287                 }
2288         }
2289
2290         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2291
2292         if (vcpu->arch.pio.string)
2293                 return EMULATE_DO_MMIO;
2294
2295         if ((r || vcpu->mmio_is_write) && run) {
2296                 run->exit_reason = KVM_EXIT_MMIO;
2297                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2298                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2299                 run->mmio.len = vcpu->mmio_size;
2300                 run->mmio.is_write = vcpu->mmio_is_write;
2301         }
2302
2303         if (r) {
2304                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2305                         return EMULATE_DONE;
2306                 if (!vcpu->mmio_needed) {
2307                         kvm_report_emulation_failure(vcpu, "mmio");
2308                         return EMULATE_FAIL;
2309                 }
2310                 return EMULATE_DO_MMIO;
2311         }
2312
2313         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2314
2315         if (vcpu->mmio_is_write) {
2316                 vcpu->mmio_needed = 0;
2317                 return EMULATE_DO_MMIO;
2318         }
2319
2320         return EMULATE_DONE;
2321 }
2322 EXPORT_SYMBOL_GPL(emulate_instruction);
2323
2324 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2325 {
2326         int i;
2327
2328         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2329                 if (vcpu->arch.pio.guest_pages[i]) {
2330                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2331                         vcpu->arch.pio.guest_pages[i] = NULL;
2332                 }
2333 }
2334
2335 static int pio_copy_data(struct kvm_vcpu *vcpu)
2336 {
2337         void *p = vcpu->arch.pio_data;
2338         void *q;
2339         unsigned bytes;
2340         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2341
2342         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2343                  PAGE_KERNEL);
2344         if (!q) {
2345                 free_pio_guest_pages(vcpu);
2346                 return -ENOMEM;
2347         }
2348         q += vcpu->arch.pio.guest_page_offset;
2349         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2350         if (vcpu->arch.pio.in)
2351                 memcpy(q, p, bytes);
2352         else
2353                 memcpy(p, q, bytes);
2354         q -= vcpu->arch.pio.guest_page_offset;
2355         vunmap(q);
2356         free_pio_guest_pages(vcpu);
2357         return 0;
2358 }
2359
2360 int complete_pio(struct kvm_vcpu *vcpu)
2361 {
2362         struct kvm_pio_request *io = &vcpu->arch.pio;
2363         long delta;
2364         int r;
2365         unsigned long val;
2366
2367         if (!io->string) {
2368                 if (io->in) {
2369                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2370                         memcpy(&val, vcpu->arch.pio_data, io->size);
2371                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2372                 }
2373         } else {
2374                 if (io->in) {
2375                         r = pio_copy_data(vcpu);
2376                         if (r)
2377                                 return r;
2378                 }
2379
2380                 delta = 1;
2381                 if (io->rep) {
2382                         delta *= io->cur_count;
2383                         /*
2384                          * The size of the register should really depend on
2385                          * current address size.
2386                          */
2387                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2388                         val -= delta;
2389                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2390                 }
2391                 if (io->down)
2392                         delta = -delta;
2393                 delta *= io->size;
2394                 if (io->in) {
2395                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2396                         val += delta;
2397                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2398                 } else {
2399                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2400                         val += delta;
2401                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2402                 }
2403         }
2404
2405         io->count -= io->cur_count;
2406         io->cur_count = 0;
2407
2408         return 0;
2409 }
2410
2411 static void kernel_pio(struct kvm_io_device *pio_dev,
2412                        struct kvm_vcpu *vcpu,
2413                        void *pd)
2414 {
2415         /* TODO: String I/O for in kernel device */
2416
2417         mutex_lock(&vcpu->kvm->lock);
2418         if (vcpu->arch.pio.in)
2419                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2420                                   vcpu->arch.pio.size,
2421                                   pd);
2422         else
2423                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2424                                    vcpu->arch.pio.size,
2425                                    pd);
2426         mutex_unlock(&vcpu->kvm->lock);
2427 }
2428
2429 static void pio_string_write(struct kvm_io_device *pio_dev,
2430                              struct kvm_vcpu *vcpu)
2431 {
2432         struct kvm_pio_request *io = &vcpu->arch.pio;
2433         void *pd = vcpu->arch.pio_data;
2434         int i;
2435
2436         mutex_lock(&vcpu->kvm->lock);
2437         for (i = 0; i < io->cur_count; i++) {
2438                 kvm_iodevice_write(pio_dev, io->port,
2439                                    io->size,
2440                                    pd);
2441                 pd += io->size;
2442         }
2443         mutex_unlock(&vcpu->kvm->lock);
2444 }
2445
2446 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2447                                                gpa_t addr, int len,
2448                                                int is_write)
2449 {
2450         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2451 }
2452
2453 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2454                   int size, unsigned port)
2455 {
2456         struct kvm_io_device *pio_dev;
2457         unsigned long val;
2458
2459         vcpu->run->exit_reason = KVM_EXIT_IO;
2460         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2461         vcpu->run->io.size = vcpu->arch.pio.size = size;
2462         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2463         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2464         vcpu->run->io.port = vcpu->arch.pio.port = port;
2465         vcpu->arch.pio.in = in;
2466         vcpu->arch.pio.string = 0;
2467         vcpu->arch.pio.down = 0;
2468         vcpu->arch.pio.guest_page_offset = 0;
2469         vcpu->arch.pio.rep = 0;
2470
2471         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2472                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2473                             handler);
2474         else
2475                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2476                             handler);
2477
2478         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2479         memcpy(vcpu->arch.pio_data, &val, 4);
2480
2481         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2482         if (pio_dev) {
2483                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2484                 complete_pio(vcpu);
2485                 return 1;
2486         }
2487         return 0;
2488 }
2489 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2490
2491 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2492                   int size, unsigned long count, int down,
2493                   gva_t address, int rep, unsigned port)
2494 {
2495         unsigned now, in_page;
2496         int i, ret = 0;
2497         int nr_pages = 1;
2498         struct page *page;
2499         struct kvm_io_device *pio_dev;
2500
2501         vcpu->run->exit_reason = KVM_EXIT_IO;
2502         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2503         vcpu->run->io.size = vcpu->arch.pio.size = size;
2504         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2505         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2506         vcpu->run->io.port = vcpu->arch.pio.port = port;
2507         vcpu->arch.pio.in = in;
2508         vcpu->arch.pio.string = 1;
2509         vcpu->arch.pio.down = down;
2510         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2511         vcpu->arch.pio.rep = rep;
2512
2513         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2514                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2515                             handler);
2516         else
2517                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2518                             handler);
2519
2520         if (!count) {
2521                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2522                 return 1;
2523         }
2524
2525         if (!down)
2526                 in_page = PAGE_SIZE - offset_in_page(address);
2527         else
2528                 in_page = offset_in_page(address) + size;
2529         now = min(count, (unsigned long)in_page / size);
2530         if (!now) {
2531                 /*
2532                  * String I/O straddles page boundary.  Pin two guest pages
2533                  * so that we satisfy atomicity constraints.  Do just one
2534                  * transaction to avoid complexity.
2535                  */
2536                 nr_pages = 2;
2537                 now = 1;
2538         }
2539         if (down) {
2540                 /*
2541                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2542                  */
2543                 pr_unimpl(vcpu, "guest string pio down\n");
2544                 kvm_inject_gp(vcpu, 0);
2545                 return 1;
2546         }
2547         vcpu->run->io.count = now;
2548         vcpu->arch.pio.cur_count = now;
2549
2550         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2551                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2552
2553         for (i = 0; i < nr_pages; ++i) {
2554                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2555                 vcpu->arch.pio.guest_pages[i] = page;
2556                 if (!page) {
2557                         kvm_inject_gp(vcpu, 0);
2558                         free_pio_guest_pages(vcpu);
2559                         return 1;
2560                 }
2561         }
2562
2563         pio_dev = vcpu_find_pio_dev(vcpu, port,
2564                                     vcpu->arch.pio.cur_count,
2565                                     !vcpu->arch.pio.in);
2566         if (!vcpu->arch.pio.in) {
2567                 /* string PIO write */
2568                 ret = pio_copy_data(vcpu);
2569                 if (ret >= 0 && pio_dev) {
2570                         pio_string_write(pio_dev, vcpu);
2571                         complete_pio(vcpu);
2572                         if (vcpu->arch.pio.count == 0)
2573                                 ret = 1;
2574                 }
2575         } else if (pio_dev)
2576                 pr_unimpl(vcpu, "no string pio read support yet, "
2577                        "port %x size %d count %ld\n",
2578                         port, size, count);
2579
2580         return ret;
2581 }
2582 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2583
2584 int kvm_arch_init(void *opaque)
2585 {
2586         int r;
2587         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2588
2589         if (kvm_x86_ops) {
2590                 printk(KERN_ERR "kvm: already loaded the other module\n");
2591                 r = -EEXIST;
2592                 goto out;
2593         }
2594
2595         if (!ops->cpu_has_kvm_support()) {
2596                 printk(KERN_ERR "kvm: no hardware support\n");
2597                 r = -EOPNOTSUPP;
2598                 goto out;
2599         }
2600         if (ops->disabled_by_bios()) {
2601                 printk(KERN_ERR "kvm: disabled by bios\n");
2602                 r = -EOPNOTSUPP;
2603                 goto out;
2604         }
2605
2606         r = kvm_mmu_module_init();
2607         if (r)
2608                 goto out;
2609
2610         kvm_init_msr_list();
2611
2612         kvm_x86_ops = ops;
2613         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2614         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2615         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2616                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2617         return 0;
2618
2619 out:
2620         return r;
2621 }
2622
2623 void kvm_arch_exit(void)
2624 {
2625         kvm_x86_ops = NULL;
2626         kvm_mmu_module_exit();
2627 }
2628
2629 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2630 {
2631         ++vcpu->stat.halt_exits;
2632         KVMTRACE_0D(HLT, vcpu, handler);
2633         if (irqchip_in_kernel(vcpu->kvm)) {
2634                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2635                 return 1;
2636         } else {
2637                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2638                 return 0;
2639         }
2640 }
2641 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2642
2643 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2644                            unsigned long a1)
2645 {
2646         if (is_long_mode(vcpu))
2647                 return a0;
2648         else
2649                 return a0 | ((gpa_t)a1 << 32);
2650 }
2651
2652 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2653 {
2654         unsigned long nr, a0, a1, a2, a3, ret;
2655         int r = 1;
2656
2657         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2658         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2659         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2660         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2661         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2662
2663         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2664
2665         if (!is_long_mode(vcpu)) {
2666                 nr &= 0xFFFFFFFF;
2667                 a0 &= 0xFFFFFFFF;
2668                 a1 &= 0xFFFFFFFF;
2669                 a2 &= 0xFFFFFFFF;
2670                 a3 &= 0xFFFFFFFF;
2671         }
2672
2673         switch (nr) {
2674         case KVM_HC_VAPIC_POLL_IRQ:
2675                 ret = 0;
2676                 break;
2677         case KVM_HC_MMU_OP:
2678                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2679                 break;
2680         default:
2681                 ret = -KVM_ENOSYS;
2682                 break;
2683         }
2684         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2685         ++vcpu->stat.hypercalls;
2686         return r;
2687 }
2688 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2689
2690 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2691 {
2692         char instruction[3];
2693         int ret = 0;
2694         unsigned long rip = kvm_rip_read(vcpu);
2695
2696
2697         /*
2698          * Blow out the MMU to ensure that no other VCPU has an active mapping
2699          * to ensure that the updated hypercall appears atomically across all
2700          * VCPUs.
2701          */
2702         kvm_mmu_zap_all(vcpu->kvm);
2703
2704         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2705         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2706             != X86EMUL_CONTINUE)
2707                 ret = -EFAULT;
2708
2709         return ret;
2710 }
2711
2712 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2713 {
2714         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2715 }
2716
2717 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2718 {
2719         struct descriptor_table dt = { limit, base };
2720
2721         kvm_x86_ops->set_gdt(vcpu, &dt);
2722 }
2723
2724 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2725 {
2726         struct descriptor_table dt = { limit, base };
2727
2728         kvm_x86_ops->set_idt(vcpu, &dt);
2729 }
2730
2731 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2732                    unsigned long *rflags)
2733 {
2734         kvm_lmsw(vcpu, msw);
2735         *rflags = kvm_x86_ops->get_rflags(vcpu);
2736 }
2737
2738 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2739 {
2740         unsigned long value;
2741
2742         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2743         switch (cr) {
2744         case 0:
2745                 value = vcpu->arch.cr0;
2746                 break;
2747         case 2:
2748                 value = vcpu->arch.cr2;
2749                 break;
2750         case 3:
2751                 value = vcpu->arch.cr3;
2752                 break;
2753         case 4:
2754                 value = vcpu->arch.cr4;
2755                 break;
2756         case 8:
2757                 value = kvm_get_cr8(vcpu);
2758                 break;
2759         default:
2760                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2761                 return 0;
2762         }
2763         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2764                     (u32)((u64)value >> 32), handler);
2765
2766         return value;
2767 }
2768
2769 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2770                      unsigned long *rflags)
2771 {
2772         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2773                     (u32)((u64)val >> 32), handler);
2774
2775         switch (cr) {
2776         case 0:
2777                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2778                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2779                 break;
2780         case 2:
2781                 vcpu->arch.cr2 = val;
2782                 break;
2783         case 3:
2784                 kvm_set_cr3(vcpu, val);
2785                 break;
2786         case 4:
2787                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2788                 break;
2789         case 8:
2790                 kvm_set_cr8(vcpu, val & 0xfUL);
2791                 break;
2792         default:
2793                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2794         }
2795 }
2796
2797 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2798 {
2799         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2800         int j, nent = vcpu->arch.cpuid_nent;
2801
2802         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2803         /* when no next entry is found, the current entry[i] is reselected */
2804         for (j = i + 1; j == i; j = (j + 1) % nent) {
2805                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2806                 if (ej->function == e->function) {
2807                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2808                         return j;
2809                 }
2810         }
2811         return 0; /* silence gcc, even though control never reaches here */
2812 }
2813
2814 /* find an entry with matching function, matching index (if needed), and that
2815  * should be read next (if it's stateful) */
2816 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2817         u32 function, u32 index)
2818 {
2819         if (e->function != function)
2820                 return 0;
2821         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2822                 return 0;
2823         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2824                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2825                 return 0;
2826         return 1;
2827 }
2828
2829 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2830 {
2831         int i;
2832         u32 function, index;
2833         struct kvm_cpuid_entry2 *e, *best;
2834
2835         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2836         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2837         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2838         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2839         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2840         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2841         best = NULL;
2842         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2843                 e = &vcpu->arch.cpuid_entries[i];
2844                 if (is_matching_cpuid_entry(e, function, index)) {
2845                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2846                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2847                         best = e;
2848                         break;
2849                 }
2850                 /*
2851                  * Both basic or both extended?
2852                  */
2853                 if (((e->function ^ function) & 0x80000000) == 0)
2854                         if (!best || e->function > best->function)
2855                                 best = e;
2856         }
2857         if (best) {
2858                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2859                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2860                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2861                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2862         }
2863         kvm_x86_ops->skip_emulated_instruction(vcpu);
2864         KVMTRACE_5D(CPUID, vcpu, function,
2865                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2866                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2867                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2868                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2869 }
2870 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2871
2872 /*
2873  * Check if userspace requested an interrupt window, and that the
2874  * interrupt window is open.
2875  *
2876  * No need to exit to userspace if we already have an interrupt queued.
2877  */
2878 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2879                                           struct kvm_run *kvm_run)
2880 {
2881         return (!vcpu->arch.irq_summary &&
2882                 kvm_run->request_interrupt_window &&
2883                 vcpu->arch.interrupt_window_open &&
2884                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2885 }
2886
2887 /*
2888  * Check if userspace requested a NMI window, and that the NMI window
2889  * is open.
2890  *
2891  * No need to exit to userspace if we already have a NMI queued.
2892  */
2893 static int dm_request_for_nmi_injection(struct kvm_vcpu *vcpu,
2894                                         struct kvm_run *kvm_run)
2895 {
2896         return (!vcpu->arch.nmi_pending &&
2897                 kvm_run->request_nmi_window &&
2898                 vcpu->arch.nmi_window_open);
2899 }
2900
2901 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2902                               struct kvm_run *kvm_run)
2903 {
2904         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2905         kvm_run->cr8 = kvm_get_cr8(vcpu);
2906         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2907         if (irqchip_in_kernel(vcpu->kvm)) {
2908                 kvm_run->ready_for_interrupt_injection = 1;
2909                 kvm_run->ready_for_nmi_injection = 1;
2910         } else {
2911                 kvm_run->ready_for_interrupt_injection =
2912                                         (vcpu->arch.interrupt_window_open &&
2913                                          vcpu->arch.irq_summary == 0);
2914                 kvm_run->ready_for_nmi_injection =
2915                                         (vcpu->arch.nmi_window_open &&
2916                                          vcpu->arch.nmi_pending == 0);
2917         }
2918 }
2919
2920 static void vapic_enter(struct kvm_vcpu *vcpu)
2921 {
2922         struct kvm_lapic *apic = vcpu->arch.apic;
2923         struct page *page;
2924
2925         if (!apic || !apic->vapic_addr)
2926                 return;
2927
2928         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2929
2930         vcpu->arch.apic->vapic_page = page;
2931 }
2932
2933 static void vapic_exit(struct kvm_vcpu *vcpu)
2934 {
2935         struct kvm_lapic *apic = vcpu->arch.apic;
2936
2937         if (!apic || !apic->vapic_addr)
2938                 return;
2939
2940         down_read(&vcpu->kvm->slots_lock);
2941         kvm_release_page_dirty(apic->vapic_page);
2942         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2943         up_read(&vcpu->kvm->slots_lock);
2944 }
2945
2946 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2947 {
2948         int r;
2949
2950         if (vcpu->requests)
2951                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2952                         kvm_mmu_unload(vcpu);
2953
2954         r = kvm_mmu_reload(vcpu);
2955         if (unlikely(r))
2956                 goto out;
2957
2958         if (vcpu->requests) {
2959                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2960                         __kvm_migrate_timers(vcpu);
2961                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2962                         kvm_mmu_sync_roots(vcpu);
2963                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2964                         kvm_x86_ops->tlb_flush(vcpu);
2965                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2966                                        &vcpu->requests)) {
2967                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2968                         r = 0;
2969                         goto out;
2970                 }
2971                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2972                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2973                         r = 0;
2974                         goto out;
2975                 }
2976         }
2977
2978         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2979         kvm_inject_pending_timer_irqs(vcpu);
2980
2981         preempt_disable();
2982
2983         kvm_x86_ops->prepare_guest_switch(vcpu);
2984         kvm_load_guest_fpu(vcpu);
2985
2986         local_irq_disable();
2987
2988         if (vcpu->requests || need_resched() || signal_pending(current)) {
2989                 local_irq_enable();
2990                 preempt_enable();
2991                 r = 1;
2992                 goto out;
2993         }
2994
2995         if (vcpu->guest_debug.enabled)
2996                 kvm_x86_ops->guest_debug_pre(vcpu);
2997
2998         vcpu->guest_mode = 1;
2999         /*
3000          * Make sure that guest_mode assignment won't happen after
3001          * testing the pending IRQ vector bitmap.
3002          */
3003         smp_wmb();
3004
3005         if (vcpu->arch.exception.pending)
3006                 __queue_exception(vcpu);
3007         else if (irqchip_in_kernel(vcpu->kvm))
3008                 kvm_x86_ops->inject_pending_irq(vcpu);
3009         else
3010                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3011
3012         kvm_lapic_sync_to_vapic(vcpu);
3013
3014         up_read(&vcpu->kvm->slots_lock);
3015
3016         kvm_guest_enter();
3017
3018
3019         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3020         kvm_x86_ops->run(vcpu, kvm_run);
3021
3022         vcpu->guest_mode = 0;
3023         local_irq_enable();
3024
3025         ++vcpu->stat.exits;
3026
3027         /*
3028          * We must have an instruction between local_irq_enable() and
3029          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3030          * the interrupt shadow.  The stat.exits increment will do nicely.
3031          * But we need to prevent reordering, hence this barrier():
3032          */
3033         barrier();
3034
3035         kvm_guest_exit();
3036
3037         preempt_enable();
3038
3039         down_read(&vcpu->kvm->slots_lock);
3040
3041         /*
3042          * Profile KVM exit RIPs:
3043          */
3044         if (unlikely(prof_on == KVM_PROFILING)) {
3045                 unsigned long rip = kvm_rip_read(vcpu);
3046                 profile_hit(KVM_PROFILING, (void *)rip);
3047         }
3048
3049         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3050                 vcpu->arch.exception.pending = false;
3051
3052         kvm_lapic_sync_from_vapic(vcpu);
3053
3054         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3055 out:
3056         return r;
3057 }
3058
3059 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3060 {
3061         int r;
3062
3063         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3064                 pr_debug("vcpu %d received sipi with vector # %x\n",
3065                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3066                 kvm_lapic_reset(vcpu);
3067                 r = kvm_arch_vcpu_reset(vcpu);
3068                 if (r)
3069                         return r;
3070                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3071         }
3072
3073         down_read(&vcpu->kvm->slots_lock);
3074         vapic_enter(vcpu);
3075
3076         r = 1;
3077         while (r > 0) {
3078                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3079                         r = vcpu_enter_guest(vcpu, kvm_run);
3080                 else {
3081                         up_read(&vcpu->kvm->slots_lock);
3082                         kvm_vcpu_block(vcpu);
3083                         down_read(&vcpu->kvm->slots_lock);
3084                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3085                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3086                                         vcpu->arch.mp_state =
3087                                                         KVM_MP_STATE_RUNNABLE;
3088                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3089                                 r = -EINTR;
3090                 }
3091
3092                 if (r > 0) {
3093                         if (dm_request_for_nmi_injection(vcpu, kvm_run)) {
3094                                 r = -EINTR;
3095                                 kvm_run->exit_reason = KVM_EXIT_NMI;
3096                                 ++vcpu->stat.request_nmi_exits;
3097                         }
3098                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3099                                 r = -EINTR;
3100                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3101                                 ++vcpu->stat.request_irq_exits;
3102                         }
3103                         if (signal_pending(current)) {
3104                                 r = -EINTR;
3105                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3106                                 ++vcpu->stat.signal_exits;
3107                         }
3108                         if (need_resched()) {
3109                                 up_read(&vcpu->kvm->slots_lock);
3110                                 kvm_resched(vcpu);
3111                                 down_read(&vcpu->kvm->slots_lock);
3112                         }
3113                 }
3114         }
3115
3116         up_read(&vcpu->kvm->slots_lock);
3117         post_kvm_run_save(vcpu, kvm_run);
3118
3119         vapic_exit(vcpu);
3120
3121         return r;
3122 }
3123
3124 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3125 {
3126         int r;
3127         sigset_t sigsaved;
3128
3129         vcpu_load(vcpu);
3130
3131         if (vcpu->sigset_active)
3132                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3133
3134         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3135                 kvm_vcpu_block(vcpu);
3136                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3137                 r = -EAGAIN;
3138                 goto out;
3139         }
3140
3141         /* re-sync apic's tpr */
3142         if (!irqchip_in_kernel(vcpu->kvm))
3143                 kvm_set_cr8(vcpu, kvm_run->cr8);
3144
3145         if (vcpu->arch.pio.cur_count) {
3146                 r = complete_pio(vcpu);
3147                 if (r)
3148                         goto out;
3149         }
3150 #if CONFIG_HAS_IOMEM
3151         if (vcpu->mmio_needed) {
3152                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3153                 vcpu->mmio_read_completed = 1;
3154                 vcpu->mmio_needed = 0;
3155
3156                 down_read(&vcpu->kvm->slots_lock);
3157                 r = emulate_instruction(vcpu, kvm_run,
3158                                         vcpu->arch.mmio_fault_cr2, 0,
3159                                         EMULTYPE_NO_DECODE);
3160                 up_read(&vcpu->kvm->slots_lock);
3161                 if (r == EMULATE_DO_MMIO) {
3162                         /*
3163                          * Read-modify-write.  Back to userspace.
3164                          */
3165                         r = 0;
3166                         goto out;
3167                 }
3168         }
3169 #endif
3170         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3171                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3172                                      kvm_run->hypercall.ret);
3173
3174         r = __vcpu_run(vcpu, kvm_run);
3175
3176 out:
3177         if (vcpu->sigset_active)
3178                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3179
3180         vcpu_put(vcpu);
3181         return r;
3182 }
3183
3184 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3185 {
3186         vcpu_load(vcpu);
3187
3188         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3189         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3190         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3191         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3192         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3193         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3194         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3195         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3196 #ifdef CONFIG_X86_64
3197         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3198         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3199         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3200         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3201         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3202         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3203         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3204         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3205 #endif
3206
3207         regs->rip = kvm_rip_read(vcpu);
3208         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3209
3210         /*
3211          * Don't leak debug flags in case they were set for guest debugging
3212          */
3213         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3214                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3215
3216         vcpu_put(vcpu);
3217
3218         return 0;
3219 }
3220
3221 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3222 {
3223         vcpu_load(vcpu);
3224
3225         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3226         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3227         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3228         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3229         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3230         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3231         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3232         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3233 #ifdef CONFIG_X86_64
3234         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3235         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3236         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3237         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3238         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3239         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3240         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3241         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3242
3243 #endif
3244
3245         kvm_rip_write(vcpu, regs->rip);
3246         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3247
3248
3249         vcpu->arch.exception.pending = false;
3250
3251         vcpu_put(vcpu);
3252
3253         return 0;
3254 }
3255
3256 void kvm_get_segment(struct kvm_vcpu *vcpu,
3257                      struct kvm_segment *var, int seg)
3258 {
3259         kvm_x86_ops->get_segment(vcpu, var, seg);
3260 }
3261
3262 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3263 {
3264         struct kvm_segment cs;
3265
3266         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3267         *db = cs.db;
3268         *l = cs.l;
3269 }
3270 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3271
3272 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3273                                   struct kvm_sregs *sregs)
3274 {
3275         struct descriptor_table dt;
3276         int pending_vec;
3277
3278         vcpu_load(vcpu);
3279
3280         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3281         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3282         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3283         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3284         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3285         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3286
3287         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3288         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3289
3290         kvm_x86_ops->get_idt(vcpu, &dt);
3291         sregs->idt.limit = dt.limit;
3292         sregs->idt.base = dt.base;
3293         kvm_x86_ops->get_gdt(vcpu, &dt);
3294         sregs->gdt.limit = dt.limit;
3295         sregs->gdt.base = dt.base;
3296
3297         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3298         sregs->cr0 = vcpu->arch.cr0;
3299         sregs->cr2 = vcpu->arch.cr2;
3300         sregs->cr3 = vcpu->arch.cr3;
3301         sregs->cr4 = vcpu->arch.cr4;
3302         sregs->cr8 = kvm_get_cr8(vcpu);
3303         sregs->efer = vcpu->arch.shadow_efer;
3304         sregs->apic_base = kvm_get_apic_base(vcpu);
3305
3306         if (irqchip_in_kernel(vcpu->kvm)) {
3307                 memset(sregs->interrupt_bitmap, 0,
3308                        sizeof sregs->interrupt_bitmap);
3309                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3310                 if (pending_vec >= 0)
3311                         set_bit(pending_vec,
3312                                 (unsigned long *)sregs->interrupt_bitmap);
3313         } else
3314                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3315                        sizeof sregs->interrupt_bitmap);
3316
3317         vcpu_put(vcpu);
3318
3319         return 0;
3320 }
3321
3322 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3323                                     struct kvm_mp_state *mp_state)
3324 {
3325         vcpu_load(vcpu);
3326         mp_state->mp_state = vcpu->arch.mp_state;
3327         vcpu_put(vcpu);
3328         return 0;
3329 }
3330
3331 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3332                                     struct kvm_mp_state *mp_state)
3333 {
3334         vcpu_load(vcpu);
3335         vcpu->arch.mp_state = mp_state->mp_state;
3336         vcpu_put(vcpu);
3337         return 0;
3338 }
3339
3340 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3341                         struct kvm_segment *var, int seg)
3342 {
3343         kvm_x86_ops->set_segment(vcpu, var, seg);
3344 }
3345
3346 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3347                                    struct kvm_segment *kvm_desct)
3348 {
3349         kvm_desct->base = seg_desc->base0;
3350         kvm_desct->base |= seg_desc->base1 << 16;
3351         kvm_desct->base |= seg_desc->base2 << 24;
3352         kvm_desct->limit = seg_desc->limit0;
3353         kvm_desct->limit |= seg_desc->limit << 16;
3354         if (seg_desc->g) {
3355                 kvm_desct->limit <<= 12;
3356                 kvm_desct->limit |= 0xfff;
3357         }
3358         kvm_desct->selector = selector;
3359         kvm_desct->type = seg_desc->type;
3360         kvm_desct->present = seg_desc->p;
3361         kvm_desct->dpl = seg_desc->dpl;
3362         kvm_desct->db = seg_desc->d;
3363         kvm_desct->s = seg_desc->s;
3364         kvm_desct->l = seg_desc->l;
3365         kvm_desct->g = seg_desc->g;
3366         kvm_desct->avl = seg_desc->avl;
3367         if (!selector)
3368                 kvm_desct->unusable = 1;
3369         else
3370                 kvm_desct->unusable = 0;
3371         kvm_desct->padding = 0;
3372 }
3373
3374 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3375                                           u16 selector,
3376                                           struct descriptor_table *dtable)
3377 {
3378         if (selector & 1 << 2) {
3379                 struct kvm_segment kvm_seg;
3380
3381                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3382
3383                 if (kvm_seg.unusable)
3384                         dtable->limit = 0;
3385                 else
3386                         dtable->limit = kvm_seg.limit;
3387                 dtable->base = kvm_seg.base;
3388         }
3389         else
3390                 kvm_x86_ops->get_gdt(vcpu, dtable);
3391 }
3392
3393 /* allowed just for 8 bytes segments */
3394 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3395                                          struct desc_struct *seg_desc)
3396 {
3397         gpa_t gpa;
3398         struct descriptor_table dtable;
3399         u16 index = selector >> 3;
3400
3401         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3402
3403         if (dtable.limit < index * 8 + 7) {
3404                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3405                 return 1;
3406         }
3407         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3408         gpa += index * 8;
3409         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3410 }
3411
3412 /* allowed just for 8 bytes segments */
3413 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3414                                          struct desc_struct *seg_desc)
3415 {
3416         gpa_t gpa;
3417         struct descriptor_table dtable;
3418         u16 index = selector >> 3;
3419
3420         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3421
3422         if (dtable.limit < index * 8 + 7)
3423                 return 1;
3424         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3425         gpa += index * 8;
3426         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3427 }
3428
3429 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3430                              struct desc_struct *seg_desc)
3431 {
3432         u32 base_addr;
3433
3434         base_addr = seg_desc->base0;
3435         base_addr |= (seg_desc->base1 << 16);
3436         base_addr |= (seg_desc->base2 << 24);
3437
3438         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3439 }
3440
3441 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3442 {
3443         struct kvm_segment kvm_seg;
3444
3445         kvm_get_segment(vcpu, &kvm_seg, seg);
3446         return kvm_seg.selector;
3447 }
3448
3449 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3450                                                 u16 selector,
3451                                                 struct kvm_segment *kvm_seg)
3452 {
3453         struct desc_struct seg_desc;
3454
3455         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3456                 return 1;
3457         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3458         return 0;
3459 }
3460
3461 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3462 {
3463         struct kvm_segment segvar = {
3464                 .base = selector << 4,
3465                 .limit = 0xffff,
3466                 .selector = selector,
3467                 .type = 3,
3468                 .present = 1,
3469                 .dpl = 3,
3470                 .db = 0,
3471                 .s = 1,
3472                 .l = 0,
3473                 .g = 0,
3474                 .avl = 0,
3475                 .unusable = 0,
3476         };
3477         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3478         return 0;
3479 }
3480
3481 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3482                                 int type_bits, int seg)
3483 {
3484         struct kvm_segment kvm_seg;
3485
3486         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3487                 return kvm_load_realmode_segment(vcpu, selector, seg);
3488         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3489                 return 1;
3490         kvm_seg.type |= type_bits;
3491
3492         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3493             seg != VCPU_SREG_LDTR)
3494                 if (!kvm_seg.s)
3495                         kvm_seg.unusable = 1;
3496
3497         kvm_set_segment(vcpu, &kvm_seg, seg);
3498         return 0;
3499 }
3500
3501 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3502                                 struct tss_segment_32 *tss)
3503 {
3504         tss->cr3 = vcpu->arch.cr3;
3505         tss->eip = kvm_rip_read(vcpu);
3506         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3507         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3508         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3509         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3510         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3511         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3512         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3513         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3514         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3515         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3516         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3517         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3518         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3519         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3520         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3521         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3522         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3523 }
3524
3525 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3526                                   struct tss_segment_32 *tss)
3527 {
3528         kvm_set_cr3(vcpu, tss->cr3);
3529
3530         kvm_rip_write(vcpu, tss->eip);
3531         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3532
3533         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3534         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3535         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3536         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3537         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3538         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3539         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3540         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3541
3542         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3543                 return 1;
3544
3545         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3546                 return 1;
3547
3548         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3549                 return 1;
3550
3551         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3552                 return 1;
3553
3554         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3555                 return 1;
3556
3557         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3558                 return 1;
3559
3560         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3561                 return 1;
3562         return 0;
3563 }
3564
3565 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3566                                 struct tss_segment_16 *tss)
3567 {
3568         tss->ip = kvm_rip_read(vcpu);
3569         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3570         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3571         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3572         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3573         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3574         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3575         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3576         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3577         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3578
3579         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3580         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3581         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3582         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3583         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3584         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3585 }
3586
3587 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3588                                  struct tss_segment_16 *tss)
3589 {
3590         kvm_rip_write(vcpu, tss->ip);
3591         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3592         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3593         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3594         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3595         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3596         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3597         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3598         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3599         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3600
3601         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3602                 return 1;
3603
3604         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3605                 return 1;
3606
3607         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3608                 return 1;
3609
3610         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3611                 return 1;
3612
3613         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3614                 return 1;
3615         return 0;
3616 }
3617
3618 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3619                        u32 old_tss_base,
3620                        struct desc_struct *nseg_desc)
3621 {
3622         struct tss_segment_16 tss_segment_16;
3623         int ret = 0;
3624
3625         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3626                            sizeof tss_segment_16))
3627                 goto out;
3628
3629         save_state_to_tss16(vcpu, &tss_segment_16);
3630
3631         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3632                             sizeof tss_segment_16))
3633                 goto out;
3634
3635         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3636                            &tss_segment_16, sizeof tss_segment_16))
3637                 goto out;
3638
3639         if (load_state_from_tss16(vcpu, &tss_segment_16))
3640                 goto out;
3641
3642         ret = 1;
3643 out:
3644         return ret;
3645 }
3646
3647 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3648                        u32 old_tss_base,
3649                        struct desc_struct *nseg_desc)
3650 {
3651         struct tss_segment_32 tss_segment_32;
3652         int ret = 0;
3653
3654         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3655                            sizeof tss_segment_32))
3656                 goto out;
3657
3658         save_state_to_tss32(vcpu, &tss_segment_32);
3659
3660         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3661                             sizeof tss_segment_32))
3662                 goto out;
3663
3664         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3665                            &tss_segment_32, sizeof tss_segment_32))
3666                 goto out;
3667
3668         if (load_state_from_tss32(vcpu, &tss_segment_32))
3669                 goto out;
3670
3671         ret = 1;
3672 out:
3673         return ret;
3674 }
3675
3676 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3677 {
3678         struct kvm_segment tr_seg;
3679         struct desc_struct cseg_desc;
3680         struct desc_struct nseg_desc;
3681         int ret = 0;
3682         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3683         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3684
3685         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3686
3687         /* FIXME: Handle errors. Failure to read either TSS or their
3688          * descriptors should generate a pagefault.
3689          */
3690         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3691                 goto out;
3692
3693         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3694                 goto out;
3695
3696         if (reason != TASK_SWITCH_IRET) {
3697                 int cpl;
3698
3699                 cpl = kvm_x86_ops->get_cpl(vcpu);
3700                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3701                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3702                         return 1;
3703                 }
3704         }
3705
3706         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3707                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3708                 return 1;
3709         }
3710
3711         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3712                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3713                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3714         }
3715
3716         if (reason == TASK_SWITCH_IRET) {
3717                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3718                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3719         }
3720
3721         kvm_x86_ops->skip_emulated_instruction(vcpu);
3722
3723         if (nseg_desc.type & 8)
3724                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3725                                          &nseg_desc);
3726         else
3727                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3728                                          &nseg_desc);
3729
3730         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3731                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3732                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3733         }
3734
3735         if (reason != TASK_SWITCH_IRET) {
3736                 nseg_desc.type |= (1 << 1);
3737                 save_guest_segment_descriptor(vcpu, tss_selector,
3738                                               &nseg_desc);
3739         }
3740
3741         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3742         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3743         tr_seg.type = 11;
3744         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3745 out:
3746         return ret;
3747 }
3748 EXPORT_SYMBOL_GPL(kvm_task_switch);
3749
3750 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3751                                   struct kvm_sregs *sregs)
3752 {
3753         int mmu_reset_needed = 0;
3754         int i, pending_vec, max_bits;
3755         struct descriptor_table dt;
3756
3757         vcpu_load(vcpu);
3758
3759         dt.limit = sregs->idt.limit;
3760         dt.base = sregs->idt.base;
3761         kvm_x86_ops->set_idt(vcpu, &dt);
3762         dt.limit = sregs->gdt.limit;
3763         dt.base = sregs->gdt.base;
3764         kvm_x86_ops->set_gdt(vcpu, &dt);
3765
3766         vcpu->arch.cr2 = sregs->cr2;
3767         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3768         vcpu->arch.cr3 = sregs->cr3;
3769
3770         kvm_set_cr8(vcpu, sregs->cr8);
3771
3772         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3773         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3774         kvm_set_apic_base(vcpu, sregs->apic_base);
3775
3776         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3777
3778         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3779         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3780         vcpu->arch.cr0 = sregs->cr0;
3781
3782         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3783         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3784         if (!is_long_mode(vcpu) && is_pae(vcpu))
3785                 load_pdptrs(vcpu, vcpu->arch.cr3);
3786
3787         if (mmu_reset_needed)
3788                 kvm_mmu_reset_context(vcpu);
3789
3790         if (!irqchip_in_kernel(vcpu->kvm)) {
3791                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3792                        sizeof vcpu->arch.irq_pending);
3793                 vcpu->arch.irq_summary = 0;
3794                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3795                         if (vcpu->arch.irq_pending[i])
3796                                 __set_bit(i, &vcpu->arch.irq_summary);
3797         } else {
3798                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3799                 pending_vec = find_first_bit(
3800                         (const unsigned long *)sregs->interrupt_bitmap,
3801                         max_bits);
3802                 /* Only pending external irq is handled here */
3803                 if (pending_vec < max_bits) {
3804                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3805                         pr_debug("Set back pending irq %d\n",
3806                                  pending_vec);
3807                 }
3808                 kvm_pic_clear_isr_ack(vcpu->kvm);
3809         }
3810
3811         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3812         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3813         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3814         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3815         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3816         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3817
3818         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3819         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3820
3821         /* Older userspace won't unhalt the vcpu on reset. */
3822         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3823             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3824             !(vcpu->arch.cr0 & X86_CR0_PE))
3825                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3826
3827         vcpu_put(vcpu);
3828
3829         return 0;
3830 }
3831
3832 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3833                                     struct kvm_debug_guest *dbg)
3834 {
3835         int r;
3836
3837         vcpu_load(vcpu);
3838
3839         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3840
3841         vcpu_put(vcpu);
3842
3843         return r;
3844 }
3845
3846 /*
3847  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3848  * we have asm/x86/processor.h
3849  */
3850 struct fxsave {
3851         u16     cwd;
3852         u16     swd;
3853         u16     twd;
3854         u16     fop;
3855         u64     rip;
3856         u64     rdp;
3857         u32     mxcsr;
3858         u32     mxcsr_mask;
3859         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3860 #ifdef CONFIG_X86_64
3861         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3862 #else
3863         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3864 #endif
3865 };
3866
3867 /*
3868  * Translate a guest virtual address to a guest physical address.
3869  */
3870 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3871                                     struct kvm_translation *tr)
3872 {
3873         unsigned long vaddr = tr->linear_address;
3874         gpa_t gpa;
3875
3876         vcpu_load(vcpu);
3877         down_read(&vcpu->kvm->slots_lock);
3878         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3879         up_read(&vcpu->kvm->slots_lock);
3880         tr->physical_address = gpa;
3881         tr->valid = gpa != UNMAPPED_GVA;
3882         tr->writeable = 1;
3883         tr->usermode = 0;
3884         vcpu_put(vcpu);
3885
3886         return 0;
3887 }
3888
3889 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3890 {
3891         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3892
3893         vcpu_load(vcpu);
3894
3895         memcpy(fpu->fpr, fxsave->st_space, 128);
3896         fpu->fcw = fxsave->cwd;
3897         fpu->fsw = fxsave->swd;
3898         fpu->ftwx = fxsave->twd;
3899         fpu->last_opcode = fxsave->fop;
3900         fpu->last_ip = fxsave->rip;
3901         fpu->last_dp = fxsave->rdp;
3902         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3903
3904         vcpu_put(vcpu);
3905
3906         return 0;
3907 }
3908
3909 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3910 {
3911         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3912
3913         vcpu_load(vcpu);
3914
3915         memcpy(fxsave->st_space, fpu->fpr, 128);
3916         fxsave->cwd = fpu->fcw;
3917         fxsave->swd = fpu->fsw;
3918         fxsave->twd = fpu->ftwx;
3919         fxsave->fop = fpu->last_opcode;
3920         fxsave->rip = fpu->last_ip;
3921         fxsave->rdp = fpu->last_dp;
3922         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3923
3924         vcpu_put(vcpu);
3925
3926         return 0;
3927 }
3928
3929 void fx_init(struct kvm_vcpu *vcpu)
3930 {
3931         unsigned after_mxcsr_mask;
3932
3933         /*
3934          * Touch the fpu the first time in non atomic context as if
3935          * this is the first fpu instruction the exception handler
3936          * will fire before the instruction returns and it'll have to
3937          * allocate ram with GFP_KERNEL.
3938          */
3939         if (!used_math())
3940                 kvm_fx_save(&vcpu->arch.host_fx_image);
3941
3942         /* Initialize guest FPU by resetting ours and saving into guest's */
3943         preempt_disable();
3944         kvm_fx_save(&vcpu->arch.host_fx_image);
3945         kvm_fx_finit();
3946         kvm_fx_save(&vcpu->arch.guest_fx_image);
3947         kvm_fx_restore(&vcpu->arch.host_fx_image);
3948         preempt_enable();
3949
3950         vcpu->arch.cr0 |= X86_CR0_ET;
3951         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3952         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3953         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3954                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3955 }
3956 EXPORT_SYMBOL_GPL(fx_init);
3957
3958 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3959 {
3960         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3961                 return;
3962
3963         vcpu->guest_fpu_loaded = 1;
3964         kvm_fx_save(&vcpu->arch.host_fx_image);
3965         kvm_fx_restore(&vcpu->arch.guest_fx_image);
3966 }
3967 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3968
3969 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3970 {
3971         if (!vcpu->guest_fpu_loaded)
3972                 return;
3973
3974         vcpu->guest_fpu_loaded = 0;
3975         kvm_fx_save(&vcpu->arch.guest_fx_image);
3976         kvm_fx_restore(&vcpu->arch.host_fx_image);
3977         ++vcpu->stat.fpu_reload;
3978 }
3979 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3980
3981 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3982 {
3983         kvm_x86_ops->vcpu_free(vcpu);
3984 }
3985
3986 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3987                                                 unsigned int id)
3988 {
3989         return kvm_x86_ops->vcpu_create(kvm, id);
3990 }
3991
3992 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3993 {
3994         int r;
3995
3996         /* We do fxsave: this must be aligned. */
3997         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3998
3999         vcpu->arch.mtrr_state.have_fixed = 1;
4000         vcpu_load(vcpu);
4001         r = kvm_arch_vcpu_reset(vcpu);
4002         if (r == 0)
4003                 r = kvm_mmu_setup(vcpu);
4004         vcpu_put(vcpu);
4005         if (r < 0)
4006                 goto free_vcpu;
4007
4008         return 0;
4009 free_vcpu:
4010         kvm_x86_ops->vcpu_free(vcpu);
4011         return r;
4012 }
4013
4014 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4015 {
4016         vcpu_load(vcpu);
4017         kvm_mmu_unload(vcpu);
4018         vcpu_put(vcpu);
4019
4020         kvm_x86_ops->vcpu_free(vcpu);
4021 }
4022
4023 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4024 {
4025         vcpu->arch.nmi_pending = false;
4026         vcpu->arch.nmi_injected = false;
4027
4028         return kvm_x86_ops->vcpu_reset(vcpu);
4029 }
4030
4031 void kvm_arch_hardware_enable(void *garbage)
4032 {
4033         kvm_x86_ops->hardware_enable(garbage);
4034 }
4035
4036 void kvm_arch_hardware_disable(void *garbage)
4037 {
4038         kvm_x86_ops->hardware_disable(garbage);
4039 }
4040
4041 int kvm_arch_hardware_setup(void)
4042 {
4043         return kvm_x86_ops->hardware_setup();
4044 }
4045
4046 void kvm_arch_hardware_unsetup(void)
4047 {
4048         kvm_x86_ops->hardware_unsetup();
4049 }
4050
4051 void kvm_arch_check_processor_compat(void *rtn)
4052 {
4053         kvm_x86_ops->check_processor_compatibility(rtn);
4054 }
4055
4056 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4057 {
4058         struct page *page;
4059         struct kvm *kvm;
4060         int r;
4061
4062         BUG_ON(vcpu->kvm == NULL);
4063         kvm = vcpu->kvm;
4064
4065         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4066         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4067                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4068         else
4069                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4070
4071         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4072         if (!page) {
4073                 r = -ENOMEM;
4074                 goto fail;
4075         }
4076         vcpu->arch.pio_data = page_address(page);
4077
4078         r = kvm_mmu_create(vcpu);
4079         if (r < 0)
4080                 goto fail_free_pio_data;
4081
4082         if (irqchip_in_kernel(kvm)) {
4083                 r = kvm_create_lapic(vcpu);
4084                 if (r < 0)
4085                         goto fail_mmu_destroy;
4086         }
4087
4088         return 0;
4089
4090 fail_mmu_destroy:
4091         kvm_mmu_destroy(vcpu);
4092 fail_free_pio_data:
4093         free_page((unsigned long)vcpu->arch.pio_data);
4094 fail:
4095         return r;
4096 }
4097
4098 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4099 {
4100         kvm_free_lapic(vcpu);
4101         down_read(&vcpu->kvm->slots_lock);
4102         kvm_mmu_destroy(vcpu);
4103         up_read(&vcpu->kvm->slots_lock);
4104         free_page((unsigned long)vcpu->arch.pio_data);
4105 }
4106
4107 struct  kvm *kvm_arch_create_vm(void)
4108 {
4109         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4110
4111         if (!kvm)
4112                 return ERR_PTR(-ENOMEM);
4113
4114         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4115         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4116
4117         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4118         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4119
4120         return kvm;
4121 }
4122
4123 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4124 {
4125         vcpu_load(vcpu);
4126         kvm_mmu_unload(vcpu);
4127         vcpu_put(vcpu);
4128 }
4129
4130 static void kvm_free_vcpus(struct kvm *kvm)
4131 {
4132         unsigned int i;
4133
4134         /*
4135          * Unpin any mmu pages first.
4136          */
4137         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4138                 if (kvm->vcpus[i])
4139                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4140         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4141                 if (kvm->vcpus[i]) {
4142                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4143                         kvm->vcpus[i] = NULL;
4144                 }
4145         }
4146
4147 }
4148
4149 void kvm_arch_destroy_vm(struct kvm *kvm)
4150 {
4151         kvm_free_all_assigned_devices(kvm);
4152         kvm_iommu_unmap_guest(kvm);
4153         kvm_free_pit(kvm);
4154         kfree(kvm->arch.vpic);
4155         kfree(kvm->arch.vioapic);
4156         kvm_free_vcpus(kvm);
4157         kvm_free_physmem(kvm);
4158         if (kvm->arch.apic_access_page)
4159                 put_page(kvm->arch.apic_access_page);
4160         if (kvm->arch.ept_identity_pagetable)
4161                 put_page(kvm->arch.ept_identity_pagetable);
4162         kfree(kvm);
4163 }
4164
4165 int kvm_arch_set_memory_region(struct kvm *kvm,
4166                                 struct kvm_userspace_memory_region *mem,
4167                                 struct kvm_memory_slot old,
4168                                 int user_alloc)
4169 {
4170         int npages = mem->memory_size >> PAGE_SHIFT;
4171         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4172
4173         /*To keep backward compatibility with older userspace,
4174          *x86 needs to hanlde !user_alloc case.
4175          */
4176         if (!user_alloc) {
4177                 if (npages && !old.rmap) {
4178                         unsigned long userspace_addr;
4179
4180                         down_write(&current->mm->mmap_sem);
4181                         userspace_addr = do_mmap(NULL, 0,
4182                                                  npages * PAGE_SIZE,
4183                                                  PROT_READ | PROT_WRITE,
4184                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4185                                                  0);
4186                         up_write(&current->mm->mmap_sem);
4187
4188                         if (IS_ERR((void *)userspace_addr))
4189                                 return PTR_ERR((void *)userspace_addr);
4190
4191                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4192                         spin_lock(&kvm->mmu_lock);
4193                         memslot->userspace_addr = userspace_addr;
4194                         spin_unlock(&kvm->mmu_lock);
4195                 } else {
4196                         if (!old.user_alloc && old.rmap) {
4197                                 int ret;
4198
4199                                 down_write(&current->mm->mmap_sem);
4200                                 ret = do_munmap(current->mm, old.userspace_addr,
4201                                                 old.npages * PAGE_SIZE);
4202                                 up_write(&current->mm->mmap_sem);
4203                                 if (ret < 0)
4204                                         printk(KERN_WARNING
4205                                        "kvm_vm_ioctl_set_memory_region: "
4206                                        "failed to munmap memory\n");
4207                         }
4208                 }
4209         }
4210
4211         if (!kvm->arch.n_requested_mmu_pages) {
4212                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4213                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4214         }
4215
4216         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4217         kvm_flush_remote_tlbs(kvm);
4218
4219         return 0;
4220 }
4221
4222 void kvm_arch_flush_shadow(struct kvm *kvm)
4223 {
4224         kvm_mmu_zap_all(kvm);
4225 }
4226
4227 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4228 {
4229         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4230                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4231                || vcpu->arch.nmi_pending;
4232 }
4233
4234 static void vcpu_kick_intr(void *info)
4235 {
4236 #ifdef DEBUG
4237         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4238         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4239 #endif
4240 }
4241
4242 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4243 {
4244         int ipi_pcpu = vcpu->cpu;
4245         int cpu = get_cpu();
4246
4247         if (waitqueue_active(&vcpu->wq)) {
4248                 wake_up_interruptible(&vcpu->wq);
4249                 ++vcpu->stat.halt_wakeup;
4250         }
4251         /*
4252          * We may be called synchronously with irqs disabled in guest mode,
4253          * So need not to call smp_call_function_single() in that case.
4254          */
4255         if (vcpu->guest_mode && vcpu->cpu != cpu)
4256                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4257         put_cpu();
4258 }