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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * 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                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1250                 for (t = 1; t < times && *nent < maxnent; ++t) {
1251                         do_cpuid_1_ent(&entry[t], function, 0);
1252                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1253                         ++*nent;
1254                 }
1255                 break;
1256         }
1257         /* function 4 and 0xb have additional index. */
1258         case 4: {
1259                 int i, cache_type;
1260
1261                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1262                 /* read more entries until cache_type is zero */
1263                 for (i = 1; *nent < maxnent; ++i) {
1264                         cache_type = entry[i - 1].eax & 0x1f;
1265                         if (!cache_type)
1266                                 break;
1267                         do_cpuid_1_ent(&entry[i], function, i);
1268                         entry[i].flags |=
1269                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1270                         ++*nent;
1271                 }
1272                 break;
1273         }
1274         case 0xb: {
1275                 int i, level_type;
1276
1277                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1278                 /* read more entries until level_type is zero */
1279                 for (i = 1; *nent < maxnent; ++i) {
1280                         level_type = entry[i - 1].ecx & 0xff00;
1281                         if (!level_type)
1282                                 break;
1283                         do_cpuid_1_ent(&entry[i], function, i);
1284                         entry[i].flags |=
1285                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1286                         ++*nent;
1287                 }
1288                 break;
1289         }
1290         case 0x80000000:
1291                 entry->eax = min(entry->eax, 0x8000001a);
1292                 break;
1293         case 0x80000001:
1294                 entry->edx &= kvm_supported_word1_x86_features;
1295                 entry->ecx &= kvm_supported_word6_x86_features;
1296                 break;
1297         }
1298         put_cpu();
1299 }
1300
1301 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1302                                     struct kvm_cpuid_entry2 __user *entries)
1303 {
1304         struct kvm_cpuid_entry2 *cpuid_entries;
1305         int limit, nent = 0, r = -E2BIG;
1306         u32 func;
1307
1308         if (cpuid->nent < 1)
1309                 goto out;
1310         r = -ENOMEM;
1311         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1312         if (!cpuid_entries)
1313                 goto out;
1314
1315         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1316         limit = cpuid_entries[0].eax;
1317         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1318                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1319                                 &nent, cpuid->nent);
1320         r = -E2BIG;
1321         if (nent >= cpuid->nent)
1322                 goto out_free;
1323
1324         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1325         limit = cpuid_entries[nent - 1].eax;
1326         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1327                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1328                                &nent, cpuid->nent);
1329         r = -EFAULT;
1330         if (copy_to_user(entries, cpuid_entries,
1331                         nent * sizeof(struct kvm_cpuid_entry2)))
1332                 goto out_free;
1333         cpuid->nent = nent;
1334         r = 0;
1335
1336 out_free:
1337         vfree(cpuid_entries);
1338 out:
1339         return r;
1340 }
1341
1342 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1343                                     struct kvm_lapic_state *s)
1344 {
1345         vcpu_load(vcpu);
1346         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1347         vcpu_put(vcpu);
1348
1349         return 0;
1350 }
1351
1352 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1353                                     struct kvm_lapic_state *s)
1354 {
1355         vcpu_load(vcpu);
1356         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1357         kvm_apic_post_state_restore(vcpu);
1358         vcpu_put(vcpu);
1359
1360         return 0;
1361 }
1362
1363 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1364                                     struct kvm_interrupt *irq)
1365 {
1366         if (irq->irq < 0 || irq->irq >= 256)
1367                 return -EINVAL;
1368         if (irqchip_in_kernel(vcpu->kvm))
1369                 return -ENXIO;
1370         vcpu_load(vcpu);
1371
1372         set_bit(irq->irq, vcpu->arch.irq_pending);
1373         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1374
1375         vcpu_put(vcpu);
1376
1377         return 0;
1378 }
1379
1380 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1381 {
1382         vcpu_load(vcpu);
1383         kvm_inject_nmi(vcpu);
1384         vcpu_put(vcpu);
1385
1386         return 0;
1387 }
1388
1389 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1390                                            struct kvm_tpr_access_ctl *tac)
1391 {
1392         if (tac->flags)
1393                 return -EINVAL;
1394         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1395         return 0;
1396 }
1397
1398 long kvm_arch_vcpu_ioctl(struct file *filp,
1399                          unsigned int ioctl, unsigned long arg)
1400 {
1401         struct kvm_vcpu *vcpu = filp->private_data;
1402         void __user *argp = (void __user *)arg;
1403         int r;
1404         struct kvm_lapic_state *lapic = NULL;
1405
1406         switch (ioctl) {
1407         case KVM_GET_LAPIC: {
1408                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1409
1410                 r = -ENOMEM;
1411                 if (!lapic)
1412                         goto out;
1413                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1414                 if (r)
1415                         goto out;
1416                 r = -EFAULT;
1417                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1418                         goto out;
1419                 r = 0;
1420                 break;
1421         }
1422         case KVM_SET_LAPIC: {
1423                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1424                 r = -ENOMEM;
1425                 if (!lapic)
1426                         goto out;
1427                 r = -EFAULT;
1428                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1429                         goto out;
1430                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1431                 if (r)
1432                         goto out;
1433                 r = 0;
1434                 break;
1435         }
1436         case KVM_INTERRUPT: {
1437                 struct kvm_interrupt irq;
1438
1439                 r = -EFAULT;
1440                 if (copy_from_user(&irq, argp, sizeof irq))
1441                         goto out;
1442                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1443                 if (r)
1444                         goto out;
1445                 r = 0;
1446                 break;
1447         }
1448         case KVM_NMI: {
1449                 r = kvm_vcpu_ioctl_nmi(vcpu);
1450                 if (r)
1451                         goto out;
1452                 r = 0;
1453                 break;
1454         }
1455         case KVM_SET_CPUID: {
1456                 struct kvm_cpuid __user *cpuid_arg = argp;
1457                 struct kvm_cpuid cpuid;
1458
1459                 r = -EFAULT;
1460                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1461                         goto out;
1462                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1463                 if (r)
1464                         goto out;
1465                 break;
1466         }
1467         case KVM_SET_CPUID2: {
1468                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1469                 struct kvm_cpuid2 cpuid;
1470
1471                 r = -EFAULT;
1472                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1473                         goto out;
1474                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1475                                 cpuid_arg->entries);
1476                 if (r)
1477                         goto out;
1478                 break;
1479         }
1480         case KVM_GET_CPUID2: {
1481                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1482                 struct kvm_cpuid2 cpuid;
1483
1484                 r = -EFAULT;
1485                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1486                         goto out;
1487                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1488                                 cpuid_arg->entries);
1489                 if (r)
1490                         goto out;
1491                 r = -EFAULT;
1492                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1493                         goto out;
1494                 r = 0;
1495                 break;
1496         }
1497         case KVM_GET_MSRS:
1498                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1499                 break;
1500         case KVM_SET_MSRS:
1501                 r = msr_io(vcpu, argp, do_set_msr, 0);
1502                 break;
1503         case KVM_TPR_ACCESS_REPORTING: {
1504                 struct kvm_tpr_access_ctl tac;
1505
1506                 r = -EFAULT;
1507                 if (copy_from_user(&tac, argp, sizeof tac))
1508                         goto out;
1509                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1510                 if (r)
1511                         goto out;
1512                 r = -EFAULT;
1513                 if (copy_to_user(argp, &tac, sizeof tac))
1514                         goto out;
1515                 r = 0;
1516                 break;
1517         };
1518         case KVM_SET_VAPIC_ADDR: {
1519                 struct kvm_vapic_addr va;
1520
1521                 r = -EINVAL;
1522                 if (!irqchip_in_kernel(vcpu->kvm))
1523                         goto out;
1524                 r = -EFAULT;
1525                 if (copy_from_user(&va, argp, sizeof va))
1526                         goto out;
1527                 r = 0;
1528                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1529                 break;
1530         }
1531         default:
1532                 r = -EINVAL;
1533         }
1534 out:
1535         if (lapic)
1536                 kfree(lapic);
1537         return r;
1538 }
1539
1540 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1541 {
1542         int ret;
1543
1544         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1545                 return -1;
1546         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1547         return ret;
1548 }
1549
1550 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1551                                           u32 kvm_nr_mmu_pages)
1552 {
1553         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1554                 return -EINVAL;
1555
1556         down_write(&kvm->slots_lock);
1557
1558         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1559         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1560
1561         up_write(&kvm->slots_lock);
1562         return 0;
1563 }
1564
1565 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1566 {
1567         return kvm->arch.n_alloc_mmu_pages;
1568 }
1569
1570 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1571 {
1572         int i;
1573         struct kvm_mem_alias *alias;
1574
1575         for (i = 0; i < kvm->arch.naliases; ++i) {
1576                 alias = &kvm->arch.aliases[i];
1577                 if (gfn >= alias->base_gfn
1578                     && gfn < alias->base_gfn + alias->npages)
1579                         return alias->target_gfn + gfn - alias->base_gfn;
1580         }
1581         return gfn;
1582 }
1583
1584 /*
1585  * Set a new alias region.  Aliases map a portion of physical memory into
1586  * another portion.  This is useful for memory windows, for example the PC
1587  * VGA region.
1588  */
1589 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1590                                          struct kvm_memory_alias *alias)
1591 {
1592         int r, n;
1593         struct kvm_mem_alias *p;
1594
1595         r = -EINVAL;
1596         /* General sanity checks */
1597         if (alias->memory_size & (PAGE_SIZE - 1))
1598                 goto out;
1599         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1600                 goto out;
1601         if (alias->slot >= KVM_ALIAS_SLOTS)
1602                 goto out;
1603         if (alias->guest_phys_addr + alias->memory_size
1604             < alias->guest_phys_addr)
1605                 goto out;
1606         if (alias->target_phys_addr + alias->memory_size
1607             < alias->target_phys_addr)
1608                 goto out;
1609
1610         down_write(&kvm->slots_lock);
1611         spin_lock(&kvm->mmu_lock);
1612
1613         p = &kvm->arch.aliases[alias->slot];
1614         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1615         p->npages = alias->memory_size >> PAGE_SHIFT;
1616         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1617
1618         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1619                 if (kvm->arch.aliases[n - 1].npages)
1620                         break;
1621         kvm->arch.naliases = n;
1622
1623         spin_unlock(&kvm->mmu_lock);
1624         kvm_mmu_zap_all(kvm);
1625
1626         up_write(&kvm->slots_lock);
1627
1628         return 0;
1629
1630 out:
1631         return r;
1632 }
1633
1634 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1635 {
1636         int r;
1637
1638         r = 0;
1639         switch (chip->chip_id) {
1640         case KVM_IRQCHIP_PIC_MASTER:
1641                 memcpy(&chip->chip.pic,
1642                         &pic_irqchip(kvm)->pics[0],
1643                         sizeof(struct kvm_pic_state));
1644                 break;
1645         case KVM_IRQCHIP_PIC_SLAVE:
1646                 memcpy(&chip->chip.pic,
1647                         &pic_irqchip(kvm)->pics[1],
1648                         sizeof(struct kvm_pic_state));
1649                 break;
1650         case KVM_IRQCHIP_IOAPIC:
1651                 memcpy(&chip->chip.ioapic,
1652                         ioapic_irqchip(kvm),
1653                         sizeof(struct kvm_ioapic_state));
1654                 break;
1655         default:
1656                 r = -EINVAL;
1657                 break;
1658         }
1659         return r;
1660 }
1661
1662 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1663 {
1664         int r;
1665
1666         r = 0;
1667         switch (chip->chip_id) {
1668         case KVM_IRQCHIP_PIC_MASTER:
1669                 memcpy(&pic_irqchip(kvm)->pics[0],
1670                         &chip->chip.pic,
1671                         sizeof(struct kvm_pic_state));
1672                 break;
1673         case KVM_IRQCHIP_PIC_SLAVE:
1674                 memcpy(&pic_irqchip(kvm)->pics[1],
1675                         &chip->chip.pic,
1676                         sizeof(struct kvm_pic_state));
1677                 break;
1678         case KVM_IRQCHIP_IOAPIC:
1679                 memcpy(ioapic_irqchip(kvm),
1680                         &chip->chip.ioapic,
1681                         sizeof(struct kvm_ioapic_state));
1682                 break;
1683         default:
1684                 r = -EINVAL;
1685                 break;
1686         }
1687         kvm_pic_update_irq(pic_irqchip(kvm));
1688         return r;
1689 }
1690
1691 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1692 {
1693         int r = 0;
1694
1695         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1696         return r;
1697 }
1698
1699 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1700 {
1701         int r = 0;
1702
1703         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1704         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1705         return r;
1706 }
1707
1708 /*
1709  * Get (and clear) the dirty memory log for a memory slot.
1710  */
1711 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1712                                       struct kvm_dirty_log *log)
1713 {
1714         int r;
1715         int n;
1716         struct kvm_memory_slot *memslot;
1717         int is_dirty = 0;
1718
1719         down_write(&kvm->slots_lock);
1720
1721         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1722         if (r)
1723                 goto out;
1724
1725         /* If nothing is dirty, don't bother messing with page tables. */
1726         if (is_dirty) {
1727                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1728                 kvm_flush_remote_tlbs(kvm);
1729                 memslot = &kvm->memslots[log->slot];
1730                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1731                 memset(memslot->dirty_bitmap, 0, n);
1732         }
1733         r = 0;
1734 out:
1735         up_write(&kvm->slots_lock);
1736         return r;
1737 }
1738
1739 long kvm_arch_vm_ioctl(struct file *filp,
1740                        unsigned int ioctl, unsigned long arg)
1741 {
1742         struct kvm *kvm = filp->private_data;
1743         void __user *argp = (void __user *)arg;
1744         int r = -EINVAL;
1745         /*
1746          * This union makes it completely explicit to gcc-3.x
1747          * that these two variables' stack usage should be
1748          * combined, not added together.
1749          */
1750         union {
1751                 struct kvm_pit_state ps;
1752                 struct kvm_memory_alias alias;
1753         } u;
1754
1755         switch (ioctl) {
1756         case KVM_SET_TSS_ADDR:
1757                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1758                 if (r < 0)
1759                         goto out;
1760                 break;
1761         case KVM_SET_MEMORY_REGION: {
1762                 struct kvm_memory_region kvm_mem;
1763                 struct kvm_userspace_memory_region kvm_userspace_mem;
1764
1765                 r = -EFAULT;
1766                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1767                         goto out;
1768                 kvm_userspace_mem.slot = kvm_mem.slot;
1769                 kvm_userspace_mem.flags = kvm_mem.flags;
1770                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1771                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1772                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1773                 if (r)
1774                         goto out;
1775                 break;
1776         }
1777         case KVM_SET_NR_MMU_PAGES:
1778                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1779                 if (r)
1780                         goto out;
1781                 break;
1782         case KVM_GET_NR_MMU_PAGES:
1783                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1784                 break;
1785         case KVM_SET_MEMORY_ALIAS:
1786                 r = -EFAULT;
1787                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1788                         goto out;
1789                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1790                 if (r)
1791                         goto out;
1792                 break;
1793         case KVM_CREATE_IRQCHIP:
1794                 r = -ENOMEM;
1795                 kvm->arch.vpic = kvm_create_pic(kvm);
1796                 if (kvm->arch.vpic) {
1797                         r = kvm_ioapic_init(kvm);
1798                         if (r) {
1799                                 kfree(kvm->arch.vpic);
1800                                 kvm->arch.vpic = NULL;
1801                                 goto out;
1802                         }
1803                 } else
1804                         goto out;
1805                 break;
1806         case KVM_CREATE_PIT:
1807                 r = -ENOMEM;
1808                 kvm->arch.vpit = kvm_create_pit(kvm);
1809                 if (kvm->arch.vpit)
1810                         r = 0;
1811                 break;
1812         case KVM_IRQ_LINE: {
1813                 struct kvm_irq_level irq_event;
1814
1815                 r = -EFAULT;
1816                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1817                         goto out;
1818                 if (irqchip_in_kernel(kvm)) {
1819                         mutex_lock(&kvm->lock);
1820                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1821                                     irq_event.irq, irq_event.level);
1822                         mutex_unlock(&kvm->lock);
1823                         r = 0;
1824                 }
1825                 break;
1826         }
1827         case KVM_GET_IRQCHIP: {
1828                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1829                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1830
1831                 r = -ENOMEM;
1832                 if (!chip)
1833                         goto out;
1834                 r = -EFAULT;
1835                 if (copy_from_user(chip, argp, sizeof *chip))
1836                         goto get_irqchip_out;
1837                 r = -ENXIO;
1838                 if (!irqchip_in_kernel(kvm))
1839                         goto get_irqchip_out;
1840                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1841                 if (r)
1842                         goto get_irqchip_out;
1843                 r = -EFAULT;
1844                 if (copy_to_user(argp, chip, sizeof *chip))
1845                         goto get_irqchip_out;
1846                 r = 0;
1847         get_irqchip_out:
1848                 kfree(chip);
1849                 if (r)
1850                         goto out;
1851                 break;
1852         }
1853         case KVM_SET_IRQCHIP: {
1854                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1855                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1856
1857                 r = -ENOMEM;
1858                 if (!chip)
1859                         goto out;
1860                 r = -EFAULT;
1861                 if (copy_from_user(chip, argp, sizeof *chip))
1862                         goto set_irqchip_out;
1863                 r = -ENXIO;
1864                 if (!irqchip_in_kernel(kvm))
1865                         goto set_irqchip_out;
1866                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1867                 if (r)
1868                         goto set_irqchip_out;
1869                 r = 0;
1870         set_irqchip_out:
1871                 kfree(chip);
1872                 if (r)
1873                         goto out;
1874                 break;
1875         }
1876         case KVM_GET_PIT: {
1877                 r = -EFAULT;
1878                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1879                         goto out;
1880                 r = -ENXIO;
1881                 if (!kvm->arch.vpit)
1882                         goto out;
1883                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1884                 if (r)
1885                         goto out;
1886                 r = -EFAULT;
1887                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1888                         goto out;
1889                 r = 0;
1890                 break;
1891         }
1892         case KVM_SET_PIT: {
1893                 r = -EFAULT;
1894                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1895                         goto out;
1896                 r = -ENXIO;
1897                 if (!kvm->arch.vpit)
1898                         goto out;
1899                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1900                 if (r)
1901                         goto out;
1902                 r = 0;
1903                 break;
1904         }
1905         default:
1906                 ;
1907         }
1908 out:
1909         return r;
1910 }
1911
1912 static void kvm_init_msr_list(void)
1913 {
1914         u32 dummy[2];
1915         unsigned i, j;
1916
1917         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1918                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1919                         continue;
1920                 if (j < i)
1921                         msrs_to_save[j] = msrs_to_save[i];
1922                 j++;
1923         }
1924         num_msrs_to_save = j;
1925 }
1926
1927 /*
1928  * Only apic need an MMIO device hook, so shortcut now..
1929  */
1930 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1931                                                 gpa_t addr, int len,
1932                                                 int is_write)
1933 {
1934         struct kvm_io_device *dev;
1935
1936         if (vcpu->arch.apic) {
1937                 dev = &vcpu->arch.apic->dev;
1938                 if (dev->in_range(dev, addr, len, is_write))
1939                         return dev;
1940         }
1941         return NULL;
1942 }
1943
1944
1945 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1946                                                 gpa_t addr, int len,
1947                                                 int is_write)
1948 {
1949         struct kvm_io_device *dev;
1950
1951         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1952         if (dev == NULL)
1953                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1954                                           is_write);
1955         return dev;
1956 }
1957
1958 int emulator_read_std(unsigned long addr,
1959                              void *val,
1960                              unsigned int bytes,
1961                              struct kvm_vcpu *vcpu)
1962 {
1963         void *data = val;
1964         int r = X86EMUL_CONTINUE;
1965
1966         while (bytes) {
1967                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1968                 unsigned offset = addr & (PAGE_SIZE-1);
1969                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1970                 int ret;
1971
1972                 if (gpa == UNMAPPED_GVA) {
1973                         r = X86EMUL_PROPAGATE_FAULT;
1974                         goto out;
1975                 }
1976                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1977                 if (ret < 0) {
1978                         r = X86EMUL_UNHANDLEABLE;
1979                         goto out;
1980                 }
1981
1982                 bytes -= tocopy;
1983                 data += tocopy;
1984                 addr += tocopy;
1985         }
1986 out:
1987         return r;
1988 }
1989 EXPORT_SYMBOL_GPL(emulator_read_std);
1990
1991 static int emulator_read_emulated(unsigned long addr,
1992                                   void *val,
1993                                   unsigned int bytes,
1994                                   struct kvm_vcpu *vcpu)
1995 {
1996         struct kvm_io_device *mmio_dev;
1997         gpa_t                 gpa;
1998
1999         if (vcpu->mmio_read_completed) {
2000                 memcpy(val, vcpu->mmio_data, bytes);
2001                 vcpu->mmio_read_completed = 0;
2002                 return X86EMUL_CONTINUE;
2003         }
2004
2005         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2006
2007         /* For APIC access vmexit */
2008         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2009                 goto mmio;
2010
2011         if (emulator_read_std(addr, val, bytes, vcpu)
2012                         == X86EMUL_CONTINUE)
2013                 return X86EMUL_CONTINUE;
2014         if (gpa == UNMAPPED_GVA)
2015                 return X86EMUL_PROPAGATE_FAULT;
2016
2017 mmio:
2018         /*
2019          * Is this MMIO handled locally?
2020          */
2021         mutex_lock(&vcpu->kvm->lock);
2022         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2023         if (mmio_dev) {
2024                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2025                 mutex_unlock(&vcpu->kvm->lock);
2026                 return X86EMUL_CONTINUE;
2027         }
2028         mutex_unlock(&vcpu->kvm->lock);
2029
2030         vcpu->mmio_needed = 1;
2031         vcpu->mmio_phys_addr = gpa;
2032         vcpu->mmio_size = bytes;
2033         vcpu->mmio_is_write = 0;
2034
2035         return X86EMUL_UNHANDLEABLE;
2036 }
2037
2038 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2039                           const void *val, int bytes)
2040 {
2041         int ret;
2042
2043         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2044         if (ret < 0)
2045                 return 0;
2046         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2047         return 1;
2048 }
2049
2050 static int emulator_write_emulated_onepage(unsigned long addr,
2051                                            const void *val,
2052                                            unsigned int bytes,
2053                                            struct kvm_vcpu *vcpu)
2054 {
2055         struct kvm_io_device *mmio_dev;
2056         gpa_t                 gpa;
2057
2058         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2059
2060         if (gpa == UNMAPPED_GVA) {
2061                 kvm_inject_page_fault(vcpu, addr, 2);
2062                 return X86EMUL_PROPAGATE_FAULT;
2063         }
2064
2065         /* For APIC access vmexit */
2066         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2067                 goto mmio;
2068
2069         if (emulator_write_phys(vcpu, gpa, val, bytes))
2070                 return X86EMUL_CONTINUE;
2071
2072 mmio:
2073         /*
2074          * Is this MMIO handled locally?
2075          */
2076         mutex_lock(&vcpu->kvm->lock);
2077         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2078         if (mmio_dev) {
2079                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2080                 mutex_unlock(&vcpu->kvm->lock);
2081                 return X86EMUL_CONTINUE;
2082         }
2083         mutex_unlock(&vcpu->kvm->lock);
2084
2085         vcpu->mmio_needed = 1;
2086         vcpu->mmio_phys_addr = gpa;
2087         vcpu->mmio_size = bytes;
2088         vcpu->mmio_is_write = 1;
2089         memcpy(vcpu->mmio_data, val, bytes);
2090
2091         return X86EMUL_CONTINUE;
2092 }
2093
2094 int emulator_write_emulated(unsigned long addr,
2095                                    const void *val,
2096                                    unsigned int bytes,
2097                                    struct kvm_vcpu *vcpu)
2098 {
2099         /* Crossing a page boundary? */
2100         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2101                 int rc, now;
2102
2103                 now = -addr & ~PAGE_MASK;
2104                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2105                 if (rc != X86EMUL_CONTINUE)
2106                         return rc;
2107                 addr += now;
2108                 val += now;
2109                 bytes -= now;
2110         }
2111         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2112 }
2113 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2114
2115 static int emulator_cmpxchg_emulated(unsigned long addr,
2116                                      const void *old,
2117                                      const void *new,
2118                                      unsigned int bytes,
2119                                      struct kvm_vcpu *vcpu)
2120 {
2121         static int reported;
2122
2123         if (!reported) {
2124                 reported = 1;
2125                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2126         }
2127 #ifndef CONFIG_X86_64
2128         /* guests cmpxchg8b have to be emulated atomically */
2129         if (bytes == 8) {
2130                 gpa_t gpa;
2131                 struct page *page;
2132                 char *kaddr;
2133                 u64 val;
2134
2135                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2136
2137                 if (gpa == UNMAPPED_GVA ||
2138                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2139                         goto emul_write;
2140
2141                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2142                         goto emul_write;
2143
2144                 val = *(u64 *)new;
2145
2146                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2147
2148                 kaddr = kmap_atomic(page, KM_USER0);
2149                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2150                 kunmap_atomic(kaddr, KM_USER0);
2151                 kvm_release_page_dirty(page);
2152         }
2153 emul_write:
2154 #endif
2155
2156         return emulator_write_emulated(addr, new, bytes, vcpu);
2157 }
2158
2159 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2160 {
2161         return kvm_x86_ops->get_segment_base(vcpu, seg);
2162 }
2163
2164 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2165 {
2166         kvm_mmu_invlpg(vcpu, address);
2167         return X86EMUL_CONTINUE;
2168 }
2169
2170 int emulate_clts(struct kvm_vcpu *vcpu)
2171 {
2172         KVMTRACE_0D(CLTS, vcpu, handler);
2173         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2174         return X86EMUL_CONTINUE;
2175 }
2176
2177 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2178 {
2179         struct kvm_vcpu *vcpu = ctxt->vcpu;
2180
2181         switch (dr) {
2182         case 0 ... 3:
2183                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2184                 return X86EMUL_CONTINUE;
2185         default:
2186                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2187                 return X86EMUL_UNHANDLEABLE;
2188         }
2189 }
2190
2191 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2192 {
2193         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2194         int exception;
2195
2196         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2197         if (exception) {
2198                 /* FIXME: better handling */
2199                 return X86EMUL_UNHANDLEABLE;
2200         }
2201         return X86EMUL_CONTINUE;
2202 }
2203
2204 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2205 {
2206         u8 opcodes[4];
2207         unsigned long rip = kvm_rip_read(vcpu);
2208         unsigned long rip_linear;
2209
2210         if (!printk_ratelimit())
2211                 return;
2212
2213         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2214
2215         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2216
2217         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2218                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2219 }
2220 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2221
2222 static struct x86_emulate_ops emulate_ops = {
2223         .read_std            = emulator_read_std,
2224         .read_emulated       = emulator_read_emulated,
2225         .write_emulated      = emulator_write_emulated,
2226         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2227 };
2228
2229 static void cache_all_regs(struct kvm_vcpu *vcpu)
2230 {
2231         kvm_register_read(vcpu, VCPU_REGS_RAX);
2232         kvm_register_read(vcpu, VCPU_REGS_RSP);
2233         kvm_register_read(vcpu, VCPU_REGS_RIP);
2234         vcpu->arch.regs_dirty = ~0;
2235 }
2236
2237 int emulate_instruction(struct kvm_vcpu *vcpu,
2238                         struct kvm_run *run,
2239                         unsigned long cr2,
2240                         u16 error_code,
2241                         int emulation_type)
2242 {
2243         int r;
2244         struct decode_cache *c;
2245
2246         kvm_clear_exception_queue(vcpu);
2247         vcpu->arch.mmio_fault_cr2 = cr2;
2248         /*
2249          * TODO: fix x86_emulate.c to use guest_read/write_register
2250          * instead of direct ->regs accesses, can save hundred cycles
2251          * on Intel for instructions that don't read/change RSP, for
2252          * for example.
2253          */
2254         cache_all_regs(vcpu);
2255
2256         vcpu->mmio_is_write = 0;
2257         vcpu->arch.pio.string = 0;
2258
2259         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2260                 int cs_db, cs_l;
2261                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2262
2263                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2264                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2265                 vcpu->arch.emulate_ctxt.mode =
2266                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2267                         ? X86EMUL_MODE_REAL : cs_l
2268                         ? X86EMUL_MODE_PROT64 : cs_db
2269                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2270
2271                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2272
2273                 /* Reject the instructions other than VMCALL/VMMCALL when
2274                  * try to emulate invalid opcode */
2275                 c = &vcpu->arch.emulate_ctxt.decode;
2276                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2277                     (!(c->twobyte && c->b == 0x01 &&
2278                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2279                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2280                         return EMULATE_FAIL;
2281
2282                 ++vcpu->stat.insn_emulation;
2283                 if (r)  {
2284                         ++vcpu->stat.insn_emulation_fail;
2285                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2286                                 return EMULATE_DONE;
2287                         return EMULATE_FAIL;
2288                 }
2289         }
2290
2291         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2292
2293         if (vcpu->arch.pio.string)
2294                 return EMULATE_DO_MMIO;
2295
2296         if ((r || vcpu->mmio_is_write) && run) {
2297                 run->exit_reason = KVM_EXIT_MMIO;
2298                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2299                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2300                 run->mmio.len = vcpu->mmio_size;
2301                 run->mmio.is_write = vcpu->mmio_is_write;
2302         }
2303
2304         if (r) {
2305                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2306                         return EMULATE_DONE;
2307                 if (!vcpu->mmio_needed) {
2308                         kvm_report_emulation_failure(vcpu, "mmio");
2309                         return EMULATE_FAIL;
2310                 }
2311                 return EMULATE_DO_MMIO;
2312         }
2313
2314         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2315
2316         if (vcpu->mmio_is_write) {
2317                 vcpu->mmio_needed = 0;
2318                 return EMULATE_DO_MMIO;
2319         }
2320
2321         return EMULATE_DONE;
2322 }
2323 EXPORT_SYMBOL_GPL(emulate_instruction);
2324
2325 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2326 {
2327         int i;
2328
2329         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2330                 if (vcpu->arch.pio.guest_pages[i]) {
2331                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2332                         vcpu->arch.pio.guest_pages[i] = NULL;
2333                 }
2334 }
2335
2336 static int pio_copy_data(struct kvm_vcpu *vcpu)
2337 {
2338         void *p = vcpu->arch.pio_data;
2339         void *q;
2340         unsigned bytes;
2341         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2342
2343         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2344                  PAGE_KERNEL);
2345         if (!q) {
2346                 free_pio_guest_pages(vcpu);
2347                 return -ENOMEM;
2348         }
2349         q += vcpu->arch.pio.guest_page_offset;
2350         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2351         if (vcpu->arch.pio.in)
2352                 memcpy(q, p, bytes);
2353         else
2354                 memcpy(p, q, bytes);
2355         q -= vcpu->arch.pio.guest_page_offset;
2356         vunmap(q);
2357         free_pio_guest_pages(vcpu);
2358         return 0;
2359 }
2360
2361 int complete_pio(struct kvm_vcpu *vcpu)
2362 {
2363         struct kvm_pio_request *io = &vcpu->arch.pio;
2364         long delta;
2365         int r;
2366         unsigned long val;
2367
2368         if (!io->string) {
2369                 if (io->in) {
2370                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2371                         memcpy(&val, vcpu->arch.pio_data, io->size);
2372                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2373                 }
2374         } else {
2375                 if (io->in) {
2376                         r = pio_copy_data(vcpu);
2377                         if (r)
2378                                 return r;
2379                 }
2380
2381                 delta = 1;
2382                 if (io->rep) {
2383                         delta *= io->cur_count;
2384                         /*
2385                          * The size of the register should really depend on
2386                          * current address size.
2387                          */
2388                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2389                         val -= delta;
2390                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2391                 }
2392                 if (io->down)
2393                         delta = -delta;
2394                 delta *= io->size;
2395                 if (io->in) {
2396                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2397                         val += delta;
2398                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2399                 } else {
2400                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2401                         val += delta;
2402                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2403                 }
2404         }
2405
2406         io->count -= io->cur_count;
2407         io->cur_count = 0;
2408
2409         return 0;
2410 }
2411
2412 static void kernel_pio(struct kvm_io_device *pio_dev,
2413                        struct kvm_vcpu *vcpu,
2414                        void *pd)
2415 {
2416         /* TODO: String I/O for in kernel device */
2417
2418         mutex_lock(&vcpu->kvm->lock);
2419         if (vcpu->arch.pio.in)
2420                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2421                                   vcpu->arch.pio.size,
2422                                   pd);
2423         else
2424                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2425                                    vcpu->arch.pio.size,
2426                                    pd);
2427         mutex_unlock(&vcpu->kvm->lock);
2428 }
2429
2430 static void pio_string_write(struct kvm_io_device *pio_dev,
2431                              struct kvm_vcpu *vcpu)
2432 {
2433         struct kvm_pio_request *io = &vcpu->arch.pio;
2434         void *pd = vcpu->arch.pio_data;
2435         int i;
2436
2437         mutex_lock(&vcpu->kvm->lock);
2438         for (i = 0; i < io->cur_count; i++) {
2439                 kvm_iodevice_write(pio_dev, io->port,
2440                                    io->size,
2441                                    pd);
2442                 pd += io->size;
2443         }
2444         mutex_unlock(&vcpu->kvm->lock);
2445 }
2446
2447 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2448                                                gpa_t addr, int len,
2449                                                int is_write)
2450 {
2451         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2452 }
2453
2454 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2455                   int size, unsigned port)
2456 {
2457         struct kvm_io_device *pio_dev;
2458         unsigned long val;
2459
2460         vcpu->run->exit_reason = KVM_EXIT_IO;
2461         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2462         vcpu->run->io.size = vcpu->arch.pio.size = size;
2463         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2464         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2465         vcpu->run->io.port = vcpu->arch.pio.port = port;
2466         vcpu->arch.pio.in = in;
2467         vcpu->arch.pio.string = 0;
2468         vcpu->arch.pio.down = 0;
2469         vcpu->arch.pio.guest_page_offset = 0;
2470         vcpu->arch.pio.rep = 0;
2471
2472         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2473                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2474                             handler);
2475         else
2476                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2477                             handler);
2478
2479         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2480         memcpy(vcpu->arch.pio_data, &val, 4);
2481
2482         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2483         if (pio_dev) {
2484                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2485                 complete_pio(vcpu);
2486                 return 1;
2487         }
2488         return 0;
2489 }
2490 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2491
2492 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2493                   int size, unsigned long count, int down,
2494                   gva_t address, int rep, unsigned port)
2495 {
2496         unsigned now, in_page;
2497         int i, ret = 0;
2498         int nr_pages = 1;
2499         struct page *page;
2500         struct kvm_io_device *pio_dev;
2501
2502         vcpu->run->exit_reason = KVM_EXIT_IO;
2503         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2504         vcpu->run->io.size = vcpu->arch.pio.size = size;
2505         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2506         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2507         vcpu->run->io.port = vcpu->arch.pio.port = port;
2508         vcpu->arch.pio.in = in;
2509         vcpu->arch.pio.string = 1;
2510         vcpu->arch.pio.down = down;
2511         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2512         vcpu->arch.pio.rep = rep;
2513
2514         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2515                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2516                             handler);
2517         else
2518                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2519                             handler);
2520
2521         if (!count) {
2522                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2523                 return 1;
2524         }
2525
2526         if (!down)
2527                 in_page = PAGE_SIZE - offset_in_page(address);
2528         else
2529                 in_page = offset_in_page(address) + size;
2530         now = min(count, (unsigned long)in_page / size);
2531         if (!now) {
2532                 /*
2533                  * String I/O straddles page boundary.  Pin two guest pages
2534                  * so that we satisfy atomicity constraints.  Do just one
2535                  * transaction to avoid complexity.
2536                  */
2537                 nr_pages = 2;
2538                 now = 1;
2539         }
2540         if (down) {
2541                 /*
2542                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2543                  */
2544                 pr_unimpl(vcpu, "guest string pio down\n");
2545                 kvm_inject_gp(vcpu, 0);
2546                 return 1;
2547         }
2548         vcpu->run->io.count = now;
2549         vcpu->arch.pio.cur_count = now;
2550
2551         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2552                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2553
2554         for (i = 0; i < nr_pages; ++i) {
2555                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2556                 vcpu->arch.pio.guest_pages[i] = page;
2557                 if (!page) {
2558                         kvm_inject_gp(vcpu, 0);
2559                         free_pio_guest_pages(vcpu);
2560                         return 1;
2561                 }
2562         }
2563
2564         pio_dev = vcpu_find_pio_dev(vcpu, port,
2565                                     vcpu->arch.pio.cur_count,
2566                                     !vcpu->arch.pio.in);
2567         if (!vcpu->arch.pio.in) {
2568                 /* string PIO write */
2569                 ret = pio_copy_data(vcpu);
2570                 if (ret >= 0 && pio_dev) {
2571                         pio_string_write(pio_dev, vcpu);
2572                         complete_pio(vcpu);
2573                         if (vcpu->arch.pio.count == 0)
2574                                 ret = 1;
2575                 }
2576         } else if (pio_dev)
2577                 pr_unimpl(vcpu, "no string pio read support yet, "
2578                        "port %x size %d count %ld\n",
2579                         port, size, count);
2580
2581         return ret;
2582 }
2583 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2584
2585 int kvm_arch_init(void *opaque)
2586 {
2587         int r;
2588         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2589
2590         if (kvm_x86_ops) {
2591                 printk(KERN_ERR "kvm: already loaded the other module\n");
2592                 r = -EEXIST;
2593                 goto out;
2594         }
2595
2596         if (!ops->cpu_has_kvm_support()) {
2597                 printk(KERN_ERR "kvm: no hardware support\n");
2598                 r = -EOPNOTSUPP;
2599                 goto out;
2600         }
2601         if (ops->disabled_by_bios()) {
2602                 printk(KERN_ERR "kvm: disabled by bios\n");
2603                 r = -EOPNOTSUPP;
2604                 goto out;
2605         }
2606
2607         r = kvm_mmu_module_init();
2608         if (r)
2609                 goto out;
2610
2611         kvm_init_msr_list();
2612
2613         kvm_x86_ops = ops;
2614         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2615         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2616         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2617                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2618         return 0;
2619
2620 out:
2621         return r;
2622 }
2623
2624 void kvm_arch_exit(void)
2625 {
2626         kvm_x86_ops = NULL;
2627         kvm_mmu_module_exit();
2628 }
2629
2630 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2631 {
2632         ++vcpu->stat.halt_exits;
2633         KVMTRACE_0D(HLT, vcpu, handler);
2634         if (irqchip_in_kernel(vcpu->kvm)) {
2635                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2636                 return 1;
2637         } else {
2638                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2639                 return 0;
2640         }
2641 }
2642 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2643
2644 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2645                            unsigned long a1)
2646 {
2647         if (is_long_mode(vcpu))
2648                 return a0;
2649         else
2650                 return a0 | ((gpa_t)a1 << 32);
2651 }
2652
2653 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2654 {
2655         unsigned long nr, a0, a1, a2, a3, ret;
2656         int r = 1;
2657
2658         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2659         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2660         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2661         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2662         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2663
2664         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2665
2666         if (!is_long_mode(vcpu)) {
2667                 nr &= 0xFFFFFFFF;
2668                 a0 &= 0xFFFFFFFF;
2669                 a1 &= 0xFFFFFFFF;
2670                 a2 &= 0xFFFFFFFF;
2671                 a3 &= 0xFFFFFFFF;
2672         }
2673
2674         switch (nr) {
2675         case KVM_HC_VAPIC_POLL_IRQ:
2676                 ret = 0;
2677                 break;
2678         case KVM_HC_MMU_OP:
2679                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2680                 break;
2681         default:
2682                 ret = -KVM_ENOSYS;
2683                 break;
2684         }
2685         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2686         ++vcpu->stat.hypercalls;
2687         return r;
2688 }
2689 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2690
2691 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2692 {
2693         char instruction[3];
2694         int ret = 0;
2695         unsigned long rip = kvm_rip_read(vcpu);
2696
2697
2698         /*
2699          * Blow out the MMU to ensure that no other VCPU has an active mapping
2700          * to ensure that the updated hypercall appears atomically across all
2701          * VCPUs.
2702          */
2703         kvm_mmu_zap_all(vcpu->kvm);
2704
2705         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2706         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2707             != X86EMUL_CONTINUE)
2708                 ret = -EFAULT;
2709
2710         return ret;
2711 }
2712
2713 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2714 {
2715         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2716 }
2717
2718 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2719 {
2720         struct descriptor_table dt = { limit, base };
2721
2722         kvm_x86_ops->set_gdt(vcpu, &dt);
2723 }
2724
2725 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2726 {
2727         struct descriptor_table dt = { limit, base };
2728
2729         kvm_x86_ops->set_idt(vcpu, &dt);
2730 }
2731
2732 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2733                    unsigned long *rflags)
2734 {
2735         kvm_lmsw(vcpu, msw);
2736         *rflags = kvm_x86_ops->get_rflags(vcpu);
2737 }
2738
2739 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2740 {
2741         unsigned long value;
2742
2743         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2744         switch (cr) {
2745         case 0:
2746                 value = vcpu->arch.cr0;
2747                 break;
2748         case 2:
2749                 value = vcpu->arch.cr2;
2750                 break;
2751         case 3:
2752                 value = vcpu->arch.cr3;
2753                 break;
2754         case 4:
2755                 value = vcpu->arch.cr4;
2756                 break;
2757         case 8:
2758                 value = kvm_get_cr8(vcpu);
2759                 break;
2760         default:
2761                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2762                 return 0;
2763         }
2764         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2765                     (u32)((u64)value >> 32), handler);
2766
2767         return value;
2768 }
2769
2770 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2771                      unsigned long *rflags)
2772 {
2773         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2774                     (u32)((u64)val >> 32), handler);
2775
2776         switch (cr) {
2777         case 0:
2778                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2779                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2780                 break;
2781         case 2:
2782                 vcpu->arch.cr2 = val;
2783                 break;
2784         case 3:
2785                 kvm_set_cr3(vcpu, val);
2786                 break;
2787         case 4:
2788                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2789                 break;
2790         case 8:
2791                 kvm_set_cr8(vcpu, val & 0xfUL);
2792                 break;
2793         default:
2794                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2795         }
2796 }
2797
2798 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2799 {
2800         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2801         int j, nent = vcpu->arch.cpuid_nent;
2802
2803         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2804         /* when no next entry is found, the current entry[i] is reselected */
2805         for (j = i + 1; ; j = (j + 1) % nent) {
2806                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2807                 if (ej->function == e->function) {
2808                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2809                         return j;
2810                 }
2811         }
2812         return 0; /* silence gcc, even though control never reaches here */
2813 }
2814
2815 /* find an entry with matching function, matching index (if needed), and that
2816  * should be read next (if it's stateful) */
2817 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2818         u32 function, u32 index)
2819 {
2820         if (e->function != function)
2821                 return 0;
2822         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2823                 return 0;
2824         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2825                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2826                 return 0;
2827         return 1;
2828 }
2829
2830 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2831 {
2832         int i;
2833         u32 function, index;
2834         struct kvm_cpuid_entry2 *e, *best;
2835
2836         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2837         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2838         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2839         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2840         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2841         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2842         best = NULL;
2843         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2844                 e = &vcpu->arch.cpuid_entries[i];
2845                 if (is_matching_cpuid_entry(e, function, index)) {
2846                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2847                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2848                         best = e;
2849                         break;
2850                 }
2851                 /*
2852                  * Both basic or both extended?
2853                  */
2854                 if (((e->function ^ function) & 0x80000000) == 0)
2855                         if (!best || e->function > best->function)
2856                                 best = e;
2857         }
2858         if (best) {
2859                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2860                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2861                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2862                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2863         }
2864         kvm_x86_ops->skip_emulated_instruction(vcpu);
2865         KVMTRACE_5D(CPUID, vcpu, function,
2866                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2867                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2868                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2869                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2870 }
2871 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2872
2873 /*
2874  * Check if userspace requested an interrupt window, and that the
2875  * interrupt window is open.
2876  *
2877  * No need to exit to userspace if we already have an interrupt queued.
2878  */
2879 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2880                                           struct kvm_run *kvm_run)
2881 {
2882         return (!vcpu->arch.irq_summary &&
2883                 kvm_run->request_interrupt_window &&
2884                 vcpu->arch.interrupt_window_open &&
2885                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2886 }
2887
2888 /*
2889  * Check if userspace requested a NMI window, and that the NMI window
2890  * is open.
2891  *
2892  * No need to exit to userspace if we already have a NMI queued.
2893  */
2894 static int dm_request_for_nmi_injection(struct kvm_vcpu *vcpu,
2895                                         struct kvm_run *kvm_run)
2896 {
2897         return (!vcpu->arch.nmi_pending &&
2898                 kvm_run->request_nmi_window &&
2899                 vcpu->arch.nmi_window_open);
2900 }
2901
2902 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2903                               struct kvm_run *kvm_run)
2904 {
2905         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2906         kvm_run->cr8 = kvm_get_cr8(vcpu);
2907         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2908         if (irqchip_in_kernel(vcpu->kvm)) {
2909                 kvm_run->ready_for_interrupt_injection = 1;
2910                 kvm_run->ready_for_nmi_injection = 1;
2911         } else {
2912                 kvm_run->ready_for_interrupt_injection =
2913                                         (vcpu->arch.interrupt_window_open &&
2914                                          vcpu->arch.irq_summary == 0);
2915                 kvm_run->ready_for_nmi_injection =
2916                                         (vcpu->arch.nmi_window_open &&
2917                                          vcpu->arch.nmi_pending == 0);
2918         }
2919 }
2920
2921 static void vapic_enter(struct kvm_vcpu *vcpu)
2922 {
2923         struct kvm_lapic *apic = vcpu->arch.apic;
2924         struct page *page;
2925
2926         if (!apic || !apic->vapic_addr)
2927                 return;
2928
2929         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2930
2931         vcpu->arch.apic->vapic_page = page;
2932 }
2933
2934 static void vapic_exit(struct kvm_vcpu *vcpu)
2935 {
2936         struct kvm_lapic *apic = vcpu->arch.apic;
2937
2938         if (!apic || !apic->vapic_addr)
2939                 return;
2940
2941         down_read(&vcpu->kvm->slots_lock);
2942         kvm_release_page_dirty(apic->vapic_page);
2943         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2944         up_read(&vcpu->kvm->slots_lock);
2945 }
2946
2947 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2948 {
2949         int r;
2950
2951         if (vcpu->requests)
2952                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2953                         kvm_mmu_unload(vcpu);
2954
2955         r = kvm_mmu_reload(vcpu);
2956         if (unlikely(r))
2957                 goto out;
2958
2959         if (vcpu->requests) {
2960                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2961                         __kvm_migrate_timers(vcpu);
2962                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2963                         kvm_mmu_sync_roots(vcpu);
2964                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2965                         kvm_x86_ops->tlb_flush(vcpu);
2966                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2967                                        &vcpu->requests)) {
2968                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2969                         r = 0;
2970                         goto out;
2971                 }
2972                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2973                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2974                         r = 0;
2975                         goto out;
2976                 }
2977         }
2978
2979         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2980         kvm_inject_pending_timer_irqs(vcpu);
2981
2982         preempt_disable();
2983
2984         kvm_x86_ops->prepare_guest_switch(vcpu);
2985         kvm_load_guest_fpu(vcpu);
2986
2987         local_irq_disable();
2988
2989         if (vcpu->requests || need_resched() || signal_pending(current)) {
2990                 local_irq_enable();
2991                 preempt_enable();
2992                 r = 1;
2993                 goto out;
2994         }
2995
2996         if (vcpu->guest_debug.enabled)
2997                 kvm_x86_ops->guest_debug_pre(vcpu);
2998
2999         vcpu->guest_mode = 1;
3000         /*
3001          * Make sure that guest_mode assignment won't happen after
3002          * testing the pending IRQ vector bitmap.
3003          */
3004         smp_wmb();
3005
3006         if (vcpu->arch.exception.pending)
3007                 __queue_exception(vcpu);
3008         else if (irqchip_in_kernel(vcpu->kvm))
3009                 kvm_x86_ops->inject_pending_irq(vcpu);
3010         else
3011                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3012
3013         kvm_lapic_sync_to_vapic(vcpu);
3014
3015         up_read(&vcpu->kvm->slots_lock);
3016
3017         kvm_guest_enter();
3018
3019
3020         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3021         kvm_x86_ops->run(vcpu, kvm_run);
3022
3023         vcpu->guest_mode = 0;
3024         local_irq_enable();
3025
3026         ++vcpu->stat.exits;
3027
3028         /*
3029          * We must have an instruction between local_irq_enable() and
3030          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3031          * the interrupt shadow.  The stat.exits increment will do nicely.
3032          * But we need to prevent reordering, hence this barrier():
3033          */
3034         barrier();
3035
3036         kvm_guest_exit();
3037
3038         preempt_enable();
3039
3040         down_read(&vcpu->kvm->slots_lock);
3041
3042         /*
3043          * Profile KVM exit RIPs:
3044          */
3045         if (unlikely(prof_on == KVM_PROFILING)) {
3046                 unsigned long rip = kvm_rip_read(vcpu);
3047                 profile_hit(KVM_PROFILING, (void *)rip);
3048         }
3049
3050         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3051                 vcpu->arch.exception.pending = false;
3052
3053         kvm_lapic_sync_from_vapic(vcpu);
3054
3055         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3056 out:
3057         return r;
3058 }
3059
3060 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3061 {
3062         int r;
3063
3064         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3065                 pr_debug("vcpu %d received sipi with vector # %x\n",
3066                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3067                 kvm_lapic_reset(vcpu);
3068                 r = kvm_arch_vcpu_reset(vcpu);
3069                 if (r)
3070                         return r;
3071                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3072         }
3073
3074         down_read(&vcpu->kvm->slots_lock);
3075         vapic_enter(vcpu);
3076
3077         r = 1;
3078         while (r > 0) {
3079                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3080                         r = vcpu_enter_guest(vcpu, kvm_run);
3081                 else {
3082                         up_read(&vcpu->kvm->slots_lock);
3083                         kvm_vcpu_block(vcpu);
3084                         down_read(&vcpu->kvm->slots_lock);
3085                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3086                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3087                                         vcpu->arch.mp_state =
3088                                                         KVM_MP_STATE_RUNNABLE;
3089                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3090                                 r = -EINTR;
3091                 }
3092
3093                 if (r > 0) {
3094                         if (dm_request_for_nmi_injection(vcpu, kvm_run)) {
3095                                 r = -EINTR;
3096                                 kvm_run->exit_reason = KVM_EXIT_NMI;
3097                                 ++vcpu->stat.request_nmi_exits;
3098                         }
3099                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3100                                 r = -EINTR;
3101                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3102                                 ++vcpu->stat.request_irq_exits;
3103                         }
3104                         if (signal_pending(current)) {
3105                                 r = -EINTR;
3106                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3107                                 ++vcpu->stat.signal_exits;
3108                         }
3109                         if (need_resched()) {
3110                                 up_read(&vcpu->kvm->slots_lock);
3111                                 kvm_resched(vcpu);
3112                                 down_read(&vcpu->kvm->slots_lock);
3113                         }
3114                 }
3115         }
3116
3117         up_read(&vcpu->kvm->slots_lock);
3118         post_kvm_run_save(vcpu, kvm_run);
3119
3120         vapic_exit(vcpu);
3121
3122         return r;
3123 }
3124
3125 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3126 {
3127         int r;
3128         sigset_t sigsaved;
3129
3130         vcpu_load(vcpu);
3131
3132         if (vcpu->sigset_active)
3133                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3134
3135         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3136                 kvm_vcpu_block(vcpu);
3137                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3138                 r = -EAGAIN;
3139                 goto out;
3140         }
3141
3142         /* re-sync apic's tpr */
3143         if (!irqchip_in_kernel(vcpu->kvm))
3144                 kvm_set_cr8(vcpu, kvm_run->cr8);
3145
3146         if (vcpu->arch.pio.cur_count) {
3147                 r = complete_pio(vcpu);
3148                 if (r)
3149                         goto out;
3150         }
3151 #if CONFIG_HAS_IOMEM
3152         if (vcpu->mmio_needed) {
3153                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3154                 vcpu->mmio_read_completed = 1;
3155                 vcpu->mmio_needed = 0;
3156
3157                 down_read(&vcpu->kvm->slots_lock);
3158                 r = emulate_instruction(vcpu, kvm_run,
3159                                         vcpu->arch.mmio_fault_cr2, 0,
3160                                         EMULTYPE_NO_DECODE);
3161                 up_read(&vcpu->kvm->slots_lock);
3162                 if (r == EMULATE_DO_MMIO) {
3163                         /*
3164                          * Read-modify-write.  Back to userspace.
3165                          */
3166                         r = 0;
3167                         goto out;
3168                 }
3169         }
3170 #endif
3171         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3172                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3173                                      kvm_run->hypercall.ret);
3174
3175         r = __vcpu_run(vcpu, kvm_run);
3176
3177 out:
3178         if (vcpu->sigset_active)
3179                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3180
3181         vcpu_put(vcpu);
3182         return r;
3183 }
3184
3185 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3186 {
3187         vcpu_load(vcpu);
3188
3189         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3190         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3191         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3192         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3193         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3194         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3195         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3196         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3197 #ifdef CONFIG_X86_64
3198         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3199         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3200         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3201         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3202         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3203         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3204         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3205         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3206 #endif
3207
3208         regs->rip = kvm_rip_read(vcpu);
3209         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3210
3211         /*
3212          * Don't leak debug flags in case they were set for guest debugging
3213          */
3214         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3215                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3216
3217         vcpu_put(vcpu);
3218
3219         return 0;
3220 }
3221
3222 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3223 {
3224         vcpu_load(vcpu);
3225
3226         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3227         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3228         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3229         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3230         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3231         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3232         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3233         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3234 #ifdef CONFIG_X86_64
3235         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3236         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3237         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3238         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3239         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3240         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3241         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3242         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3243
3244 #endif
3245
3246         kvm_rip_write(vcpu, regs->rip);
3247         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3248
3249
3250         vcpu->arch.exception.pending = false;
3251
3252         vcpu_put(vcpu);
3253
3254         return 0;
3255 }
3256
3257 void kvm_get_segment(struct kvm_vcpu *vcpu,
3258                      struct kvm_segment *var, int seg)
3259 {
3260         kvm_x86_ops->get_segment(vcpu, var, seg);
3261 }
3262
3263 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3264 {
3265         struct kvm_segment cs;
3266
3267         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3268         *db = cs.db;
3269         *l = cs.l;
3270 }
3271 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3272
3273 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3274                                   struct kvm_sregs *sregs)
3275 {
3276         struct descriptor_table dt;
3277         int pending_vec;
3278
3279         vcpu_load(vcpu);
3280
3281         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3282         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3283         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3284         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3285         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3286         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3287
3288         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3289         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3290
3291         kvm_x86_ops->get_idt(vcpu, &dt);
3292         sregs->idt.limit = dt.limit;
3293         sregs->idt.base = dt.base;
3294         kvm_x86_ops->get_gdt(vcpu, &dt);
3295         sregs->gdt.limit = dt.limit;
3296         sregs->gdt.base = dt.base;
3297
3298         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3299         sregs->cr0 = vcpu->arch.cr0;
3300         sregs->cr2 = vcpu->arch.cr2;
3301         sregs->cr3 = vcpu->arch.cr3;
3302         sregs->cr4 = vcpu->arch.cr4;
3303         sregs->cr8 = kvm_get_cr8(vcpu);
3304         sregs->efer = vcpu->arch.shadow_efer;
3305         sregs->apic_base = kvm_get_apic_base(vcpu);
3306
3307         if (irqchip_in_kernel(vcpu->kvm)) {
3308                 memset(sregs->interrupt_bitmap, 0,
3309                        sizeof sregs->interrupt_bitmap);
3310                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3311                 if (pending_vec >= 0)
3312                         set_bit(pending_vec,
3313                                 (unsigned long *)sregs->interrupt_bitmap);
3314         } else
3315                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3316                        sizeof sregs->interrupt_bitmap);
3317
3318         vcpu_put(vcpu);
3319
3320         return 0;
3321 }
3322
3323 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3324                                     struct kvm_mp_state *mp_state)
3325 {
3326         vcpu_load(vcpu);
3327         mp_state->mp_state = vcpu->arch.mp_state;
3328         vcpu_put(vcpu);
3329         return 0;
3330 }
3331
3332 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3333                                     struct kvm_mp_state *mp_state)
3334 {
3335         vcpu_load(vcpu);
3336         vcpu->arch.mp_state = mp_state->mp_state;
3337         vcpu_put(vcpu);
3338         return 0;
3339 }
3340
3341 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3342                         struct kvm_segment *var, int seg)
3343 {
3344         kvm_x86_ops->set_segment(vcpu, var, seg);
3345 }
3346
3347 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3348                                    struct kvm_segment *kvm_desct)
3349 {
3350         kvm_desct->base = seg_desc->base0;
3351         kvm_desct->base |= seg_desc->base1 << 16;
3352         kvm_desct->base |= seg_desc->base2 << 24;
3353         kvm_desct->limit = seg_desc->limit0;
3354         kvm_desct->limit |= seg_desc->limit << 16;
3355         if (seg_desc->g) {
3356                 kvm_desct->limit <<= 12;
3357                 kvm_desct->limit |= 0xfff;
3358         }
3359         kvm_desct->selector = selector;
3360         kvm_desct->type = seg_desc->type;
3361         kvm_desct->present = seg_desc->p;
3362         kvm_desct->dpl = seg_desc->dpl;
3363         kvm_desct->db = seg_desc->d;
3364         kvm_desct->s = seg_desc->s;
3365         kvm_desct->l = seg_desc->l;
3366         kvm_desct->g = seg_desc->g;
3367         kvm_desct->avl = seg_desc->avl;
3368         if (!selector)
3369                 kvm_desct->unusable = 1;
3370         else
3371                 kvm_desct->unusable = 0;
3372         kvm_desct->padding = 0;
3373 }
3374
3375 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3376                                           u16 selector,
3377                                           struct descriptor_table *dtable)
3378 {
3379         if (selector & 1 << 2) {
3380                 struct kvm_segment kvm_seg;
3381
3382                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3383
3384                 if (kvm_seg.unusable)
3385                         dtable->limit = 0;
3386                 else
3387                         dtable->limit = kvm_seg.limit;
3388                 dtable->base = kvm_seg.base;
3389         }
3390         else
3391                 kvm_x86_ops->get_gdt(vcpu, dtable);
3392 }
3393
3394 /* allowed just for 8 bytes segments */
3395 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3396                                          struct desc_struct *seg_desc)
3397 {
3398         gpa_t gpa;
3399         struct descriptor_table dtable;
3400         u16 index = selector >> 3;
3401
3402         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3403
3404         if (dtable.limit < index * 8 + 7) {
3405                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3406                 return 1;
3407         }
3408         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3409         gpa += index * 8;
3410         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3411 }
3412
3413 /* allowed just for 8 bytes segments */
3414 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3415                                          struct desc_struct *seg_desc)
3416 {
3417         gpa_t gpa;
3418         struct descriptor_table dtable;
3419         u16 index = selector >> 3;
3420
3421         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3422
3423         if (dtable.limit < index * 8 + 7)
3424                 return 1;
3425         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3426         gpa += index * 8;
3427         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3428 }
3429
3430 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3431                              struct desc_struct *seg_desc)
3432 {
3433         u32 base_addr;
3434
3435         base_addr = seg_desc->base0;
3436         base_addr |= (seg_desc->base1 << 16);
3437         base_addr |= (seg_desc->base2 << 24);
3438
3439         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3440 }
3441
3442 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3443 {
3444         struct kvm_segment kvm_seg;
3445
3446         kvm_get_segment(vcpu, &kvm_seg, seg);
3447         return kvm_seg.selector;
3448 }
3449
3450 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3451                                                 u16 selector,
3452                                                 struct kvm_segment *kvm_seg)
3453 {
3454         struct desc_struct seg_desc;
3455
3456         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3457                 return 1;
3458         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3459         return 0;
3460 }
3461
3462 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3463 {
3464         struct kvm_segment segvar = {
3465                 .base = selector << 4,
3466                 .limit = 0xffff,
3467                 .selector = selector,
3468                 .type = 3,
3469                 .present = 1,
3470                 .dpl = 3,
3471                 .db = 0,
3472                 .s = 1,
3473                 .l = 0,
3474                 .g = 0,
3475                 .avl = 0,
3476                 .unusable = 0,
3477         };
3478         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3479         return 0;
3480 }
3481
3482 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3483                                 int type_bits, int seg)
3484 {
3485         struct kvm_segment kvm_seg;
3486
3487         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3488                 return kvm_load_realmode_segment(vcpu, selector, seg);
3489         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3490                 return 1;
3491         kvm_seg.type |= type_bits;
3492
3493         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3494             seg != VCPU_SREG_LDTR)
3495                 if (!kvm_seg.s)
3496                         kvm_seg.unusable = 1;
3497
3498         kvm_set_segment(vcpu, &kvm_seg, seg);
3499         return 0;
3500 }
3501
3502 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3503                                 struct tss_segment_32 *tss)
3504 {
3505         tss->cr3 = vcpu->arch.cr3;
3506         tss->eip = kvm_rip_read(vcpu);
3507         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3508         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3509         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3510         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3511         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3512         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3513         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3514         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3515         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3516         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3517         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3518         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3519         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3520         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3521         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3522         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3523         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3524 }
3525
3526 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3527                                   struct tss_segment_32 *tss)
3528 {
3529         kvm_set_cr3(vcpu, tss->cr3);
3530
3531         kvm_rip_write(vcpu, tss->eip);
3532         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3533
3534         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3535         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3536         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3537         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3538         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3539         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3540         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3541         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3542
3543         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3544                 return 1;
3545
3546         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3547                 return 1;
3548
3549         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3550                 return 1;
3551
3552         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3553                 return 1;
3554
3555         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3556                 return 1;
3557
3558         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3559                 return 1;
3560
3561         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3562                 return 1;
3563         return 0;
3564 }
3565
3566 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3567                                 struct tss_segment_16 *tss)
3568 {
3569         tss->ip = kvm_rip_read(vcpu);
3570         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3571         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3572         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3573         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3574         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3575         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3576         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3577         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3578         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3579
3580         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3581         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3582         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3583         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3584         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3585         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3586 }
3587
3588 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3589                                  struct tss_segment_16 *tss)
3590 {
3591         kvm_rip_write(vcpu, tss->ip);
3592         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3593         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3594         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3595         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3596         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3597         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3598         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3599         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3600         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3601
3602         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3603                 return 1;
3604
3605         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3606                 return 1;
3607
3608         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3609                 return 1;
3610
3611         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3612                 return 1;
3613
3614         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3615                 return 1;
3616         return 0;
3617 }
3618
3619 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3620                        u32 old_tss_base,
3621                        struct desc_struct *nseg_desc)
3622 {
3623         struct tss_segment_16 tss_segment_16;
3624         int ret = 0;
3625
3626         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3627                            sizeof tss_segment_16))
3628                 goto out;
3629
3630         save_state_to_tss16(vcpu, &tss_segment_16);
3631
3632         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3633                             sizeof tss_segment_16))
3634                 goto out;
3635
3636         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3637                            &tss_segment_16, sizeof tss_segment_16))
3638                 goto out;
3639
3640         if (load_state_from_tss16(vcpu, &tss_segment_16))
3641                 goto out;
3642
3643         ret = 1;
3644 out:
3645         return ret;
3646 }
3647
3648 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3649                        u32 old_tss_base,
3650                        struct desc_struct *nseg_desc)
3651 {
3652         struct tss_segment_32 tss_segment_32;
3653         int ret = 0;
3654
3655         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3656                            sizeof tss_segment_32))
3657                 goto out;
3658
3659         save_state_to_tss32(vcpu, &tss_segment_32);
3660
3661         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3662                             sizeof tss_segment_32))
3663                 goto out;
3664
3665         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3666                            &tss_segment_32, sizeof tss_segment_32))
3667                 goto out;
3668
3669         if (load_state_from_tss32(vcpu, &tss_segment_32))
3670                 goto out;
3671
3672         ret = 1;
3673 out:
3674         return ret;
3675 }
3676
3677 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3678 {
3679         struct kvm_segment tr_seg;
3680         struct desc_struct cseg_desc;
3681         struct desc_struct nseg_desc;
3682         int ret = 0;
3683         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3684         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3685
3686         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3687
3688         /* FIXME: Handle errors. Failure to read either TSS or their
3689          * descriptors should generate a pagefault.
3690          */
3691         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3692                 goto out;
3693
3694         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3695                 goto out;
3696
3697         if (reason != TASK_SWITCH_IRET) {
3698                 int cpl;
3699
3700                 cpl = kvm_x86_ops->get_cpl(vcpu);
3701                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3702                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3703                         return 1;
3704                 }
3705         }
3706
3707         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3708                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3709                 return 1;
3710         }
3711
3712         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3713                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3714                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3715         }
3716
3717         if (reason == TASK_SWITCH_IRET) {
3718                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3719                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3720         }
3721
3722         kvm_x86_ops->skip_emulated_instruction(vcpu);
3723
3724         if (nseg_desc.type & 8)
3725                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3726                                          &nseg_desc);
3727         else
3728                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3729                                          &nseg_desc);
3730
3731         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3732                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3733                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3734         }
3735
3736         if (reason != TASK_SWITCH_IRET) {
3737                 nseg_desc.type |= (1 << 1);
3738                 save_guest_segment_descriptor(vcpu, tss_selector,
3739                                               &nseg_desc);
3740         }
3741
3742         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3743         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3744         tr_seg.type = 11;
3745         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3746 out:
3747         return ret;
3748 }
3749 EXPORT_SYMBOL_GPL(kvm_task_switch);
3750
3751 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3752                                   struct kvm_sregs *sregs)
3753 {
3754         int mmu_reset_needed = 0;
3755         int i, pending_vec, max_bits;
3756         struct descriptor_table dt;
3757
3758         vcpu_load(vcpu);
3759
3760         dt.limit = sregs->idt.limit;
3761         dt.base = sregs->idt.base;
3762         kvm_x86_ops->set_idt(vcpu, &dt);
3763         dt.limit = sregs->gdt.limit;
3764         dt.base = sregs->gdt.base;
3765         kvm_x86_ops->set_gdt(vcpu, &dt);
3766
3767         vcpu->arch.cr2 = sregs->cr2;
3768         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3769         vcpu->arch.cr3 = sregs->cr3;
3770
3771         kvm_set_cr8(vcpu, sregs->cr8);
3772
3773         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3774         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3775         kvm_set_apic_base(vcpu, sregs->apic_base);
3776
3777         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3778
3779         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3780         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3781         vcpu->arch.cr0 = sregs->cr0;
3782
3783         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3784         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3785         if (!is_long_mode(vcpu) && is_pae(vcpu))
3786                 load_pdptrs(vcpu, vcpu->arch.cr3);
3787
3788         if (mmu_reset_needed)
3789                 kvm_mmu_reset_context(vcpu);
3790
3791         if (!irqchip_in_kernel(vcpu->kvm)) {
3792                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3793                        sizeof vcpu->arch.irq_pending);
3794                 vcpu->arch.irq_summary = 0;
3795                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3796                         if (vcpu->arch.irq_pending[i])
3797                                 __set_bit(i, &vcpu->arch.irq_summary);
3798         } else {
3799                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3800                 pending_vec = find_first_bit(
3801                         (const unsigned long *)sregs->interrupt_bitmap,
3802                         max_bits);
3803                 /* Only pending external irq is handled here */
3804                 if (pending_vec < max_bits) {
3805                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3806                         pr_debug("Set back pending irq %d\n",
3807                                  pending_vec);
3808                 }
3809                 kvm_pic_clear_isr_ack(vcpu->kvm);
3810         }
3811
3812         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3813         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3814         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3815         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3816         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3817         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3818
3819         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3820         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3821
3822         /* Older userspace won't unhalt the vcpu on reset. */
3823         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3824             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3825             !(vcpu->arch.cr0 & X86_CR0_PE))
3826                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3827
3828         vcpu_put(vcpu);
3829
3830         return 0;
3831 }
3832
3833 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3834                                     struct kvm_debug_guest *dbg)
3835 {
3836         int r;
3837
3838         vcpu_load(vcpu);
3839
3840         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3841
3842         vcpu_put(vcpu);
3843
3844         return r;
3845 }
3846
3847 /*
3848  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3849  * we have asm/x86/processor.h
3850  */
3851 struct fxsave {
3852         u16     cwd;
3853         u16     swd;
3854         u16     twd;
3855         u16     fop;
3856         u64     rip;
3857         u64     rdp;
3858         u32     mxcsr;
3859         u32     mxcsr_mask;
3860         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3861 #ifdef CONFIG_X86_64
3862         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3863 #else
3864         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3865 #endif
3866 };
3867
3868 /*
3869  * Translate a guest virtual address to a guest physical address.
3870  */
3871 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3872                                     struct kvm_translation *tr)
3873 {
3874         unsigned long vaddr = tr->linear_address;
3875         gpa_t gpa;
3876
3877         vcpu_load(vcpu);
3878         down_read(&vcpu->kvm->slots_lock);
3879         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3880         up_read(&vcpu->kvm->slots_lock);
3881         tr->physical_address = gpa;
3882         tr->valid = gpa != UNMAPPED_GVA;
3883         tr->writeable = 1;
3884         tr->usermode = 0;
3885         vcpu_put(vcpu);
3886
3887         return 0;
3888 }
3889
3890 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3891 {
3892         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3893
3894         vcpu_load(vcpu);
3895
3896         memcpy(fpu->fpr, fxsave->st_space, 128);
3897         fpu->fcw = fxsave->cwd;
3898         fpu->fsw = fxsave->swd;
3899         fpu->ftwx = fxsave->twd;
3900         fpu->last_opcode = fxsave->fop;
3901         fpu->last_ip = fxsave->rip;
3902         fpu->last_dp = fxsave->rdp;
3903         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3904
3905         vcpu_put(vcpu);
3906
3907         return 0;
3908 }
3909
3910 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3911 {
3912         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3913
3914         vcpu_load(vcpu);
3915
3916         memcpy(fxsave->st_space, fpu->fpr, 128);
3917         fxsave->cwd = fpu->fcw;
3918         fxsave->swd = fpu->fsw;
3919         fxsave->twd = fpu->ftwx;
3920         fxsave->fop = fpu->last_opcode;
3921         fxsave->rip = fpu->last_ip;
3922         fxsave->rdp = fpu->last_dp;
3923         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3924
3925         vcpu_put(vcpu);
3926
3927         return 0;
3928 }
3929
3930 void fx_init(struct kvm_vcpu *vcpu)
3931 {
3932         unsigned after_mxcsr_mask;
3933
3934         /*
3935          * Touch the fpu the first time in non atomic context as if
3936          * this is the first fpu instruction the exception handler
3937          * will fire before the instruction returns and it'll have to
3938          * allocate ram with GFP_KERNEL.
3939          */
3940         if (!used_math())
3941                 kvm_fx_save(&vcpu->arch.host_fx_image);
3942
3943         /* Initialize guest FPU by resetting ours and saving into guest's */
3944         preempt_disable();
3945         kvm_fx_save(&vcpu->arch.host_fx_image);
3946         kvm_fx_finit();
3947         kvm_fx_save(&vcpu->arch.guest_fx_image);
3948         kvm_fx_restore(&vcpu->arch.host_fx_image);
3949         preempt_enable();
3950
3951         vcpu->arch.cr0 |= X86_CR0_ET;
3952         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3953         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3954         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3955                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3956 }
3957 EXPORT_SYMBOL_GPL(fx_init);
3958
3959 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3960 {
3961         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3962                 return;
3963
3964         vcpu->guest_fpu_loaded = 1;
3965         kvm_fx_save(&vcpu->arch.host_fx_image);
3966         kvm_fx_restore(&vcpu->arch.guest_fx_image);
3967 }
3968 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3969
3970 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3971 {
3972         if (!vcpu->guest_fpu_loaded)
3973                 return;
3974
3975         vcpu->guest_fpu_loaded = 0;
3976         kvm_fx_save(&vcpu->arch.guest_fx_image);
3977         kvm_fx_restore(&vcpu->arch.host_fx_image);
3978         ++vcpu->stat.fpu_reload;
3979 }
3980 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3981
3982 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3983 {
3984         kvm_x86_ops->vcpu_free(vcpu);
3985 }
3986
3987 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3988                                                 unsigned int id)
3989 {
3990         return kvm_x86_ops->vcpu_create(kvm, id);
3991 }
3992
3993 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3994 {
3995         int r;
3996
3997         /* We do fxsave: this must be aligned. */
3998         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3999
4000         vcpu->arch.mtrr_state.have_fixed = 1;
4001         vcpu_load(vcpu);
4002         r = kvm_arch_vcpu_reset(vcpu);
4003         if (r == 0)
4004                 r = kvm_mmu_setup(vcpu);
4005         vcpu_put(vcpu);
4006         if (r < 0)
4007                 goto free_vcpu;
4008
4009         return 0;
4010 free_vcpu:
4011         kvm_x86_ops->vcpu_free(vcpu);
4012         return r;
4013 }
4014
4015 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4016 {
4017         vcpu_load(vcpu);
4018         kvm_mmu_unload(vcpu);
4019         vcpu_put(vcpu);
4020
4021         kvm_x86_ops->vcpu_free(vcpu);
4022 }
4023
4024 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4025 {
4026         vcpu->arch.nmi_pending = false;
4027         vcpu->arch.nmi_injected = false;
4028
4029         return kvm_x86_ops->vcpu_reset(vcpu);
4030 }
4031
4032 void kvm_arch_hardware_enable(void *garbage)
4033 {
4034         kvm_x86_ops->hardware_enable(garbage);
4035 }
4036
4037 void kvm_arch_hardware_disable(void *garbage)
4038 {
4039         kvm_x86_ops->hardware_disable(garbage);
4040 }
4041
4042 int kvm_arch_hardware_setup(void)
4043 {
4044         return kvm_x86_ops->hardware_setup();
4045 }
4046
4047 void kvm_arch_hardware_unsetup(void)
4048 {
4049         kvm_x86_ops->hardware_unsetup();
4050 }
4051
4052 void kvm_arch_check_processor_compat(void *rtn)
4053 {
4054         kvm_x86_ops->check_processor_compatibility(rtn);
4055 }
4056
4057 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4058 {
4059         struct page *page;
4060         struct kvm *kvm;
4061         int r;
4062
4063         BUG_ON(vcpu->kvm == NULL);
4064         kvm = vcpu->kvm;
4065
4066         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4067         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4068                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4069         else
4070                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4071
4072         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4073         if (!page) {
4074                 r = -ENOMEM;
4075                 goto fail;
4076         }
4077         vcpu->arch.pio_data = page_address(page);
4078
4079         r = kvm_mmu_create(vcpu);
4080         if (r < 0)
4081                 goto fail_free_pio_data;
4082
4083         if (irqchip_in_kernel(kvm)) {
4084                 r = kvm_create_lapic(vcpu);
4085                 if (r < 0)
4086                         goto fail_mmu_destroy;
4087         }
4088
4089         return 0;
4090
4091 fail_mmu_destroy:
4092         kvm_mmu_destroy(vcpu);
4093 fail_free_pio_data:
4094         free_page((unsigned long)vcpu->arch.pio_data);
4095 fail:
4096         return r;
4097 }
4098
4099 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4100 {
4101         kvm_free_lapic(vcpu);
4102         down_read(&vcpu->kvm->slots_lock);
4103         kvm_mmu_destroy(vcpu);
4104         up_read(&vcpu->kvm->slots_lock);
4105         free_page((unsigned long)vcpu->arch.pio_data);
4106 }
4107
4108 struct  kvm *kvm_arch_create_vm(void)
4109 {
4110         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4111
4112         if (!kvm)
4113                 return ERR_PTR(-ENOMEM);
4114
4115         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4116         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4117
4118         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4119         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4120
4121         return kvm;
4122 }
4123
4124 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4125 {
4126         vcpu_load(vcpu);
4127         kvm_mmu_unload(vcpu);
4128         vcpu_put(vcpu);
4129 }
4130
4131 static void kvm_free_vcpus(struct kvm *kvm)
4132 {
4133         unsigned int i;
4134
4135         /*
4136          * Unpin any mmu pages first.
4137          */
4138         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4139                 if (kvm->vcpus[i])
4140                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4141         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4142                 if (kvm->vcpus[i]) {
4143                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4144                         kvm->vcpus[i] = NULL;
4145                 }
4146         }
4147
4148 }
4149
4150 void kvm_arch_destroy_vm(struct kvm *kvm)
4151 {
4152         kvm_free_all_assigned_devices(kvm);
4153         kvm_iommu_unmap_guest(kvm);
4154         kvm_free_pit(kvm);
4155         kfree(kvm->arch.vpic);
4156         kfree(kvm->arch.vioapic);
4157         kvm_free_vcpus(kvm);
4158         kvm_free_physmem(kvm);
4159         if (kvm->arch.apic_access_page)
4160                 put_page(kvm->arch.apic_access_page);
4161         if (kvm->arch.ept_identity_pagetable)
4162                 put_page(kvm->arch.ept_identity_pagetable);
4163         kfree(kvm);
4164 }
4165
4166 int kvm_arch_set_memory_region(struct kvm *kvm,
4167                                 struct kvm_userspace_memory_region *mem,
4168                                 struct kvm_memory_slot old,
4169                                 int user_alloc)
4170 {
4171         int npages = mem->memory_size >> PAGE_SHIFT;
4172         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4173
4174         /*To keep backward compatibility with older userspace,
4175          *x86 needs to hanlde !user_alloc case.
4176          */
4177         if (!user_alloc) {
4178                 if (npages && !old.rmap) {
4179                         unsigned long userspace_addr;
4180
4181                         down_write(&current->mm->mmap_sem);
4182                         userspace_addr = do_mmap(NULL, 0,
4183                                                  npages * PAGE_SIZE,
4184                                                  PROT_READ | PROT_WRITE,
4185                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4186                                                  0);
4187                         up_write(&current->mm->mmap_sem);
4188
4189                         if (IS_ERR((void *)userspace_addr))
4190                                 return PTR_ERR((void *)userspace_addr);
4191
4192                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4193                         spin_lock(&kvm->mmu_lock);
4194                         memslot->userspace_addr = userspace_addr;
4195                         spin_unlock(&kvm->mmu_lock);
4196                 } else {
4197                         if (!old.user_alloc && old.rmap) {
4198                                 int ret;
4199
4200                                 down_write(&current->mm->mmap_sem);
4201                                 ret = do_munmap(current->mm, old.userspace_addr,
4202                                                 old.npages * PAGE_SIZE);
4203                                 up_write(&current->mm->mmap_sem);
4204                                 if (ret < 0)
4205                                         printk(KERN_WARNING
4206                                        "kvm_vm_ioctl_set_memory_region: "
4207                                        "failed to munmap memory\n");
4208                         }
4209                 }
4210         }
4211
4212         if (!kvm->arch.n_requested_mmu_pages) {
4213                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4214                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4215         }
4216
4217         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4218         kvm_flush_remote_tlbs(kvm);
4219
4220         return 0;
4221 }
4222
4223 void kvm_arch_flush_shadow(struct kvm *kvm)
4224 {
4225         kvm_mmu_zap_all(kvm);
4226 }
4227
4228 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4229 {
4230         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4231                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4232                || vcpu->arch.nmi_pending;
4233 }
4234
4235 static void vcpu_kick_intr(void *info)
4236 {
4237 #ifdef DEBUG
4238         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4239         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4240 #endif
4241 }
4242
4243 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4244 {
4245         int ipi_pcpu = vcpu->cpu;
4246         int cpu = get_cpu();
4247
4248         if (waitqueue_active(&vcpu->wq)) {
4249                 wake_up_interruptible(&vcpu->wq);
4250                 ++vcpu->stat.halt_wakeup;
4251         }
4252         /*
4253          * We may be called synchronously with irqs disabled in guest mode,
4254          * So need not to call smp_call_function_single() in that case.
4255          */
4256         if (vcpu->guest_mode && vcpu->cpu != cpu)
4257                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4258         put_cpu();
4259 }