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