drivers/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  *
   8  * Authors:
   9  *   Avi Kivity   <avi@qumranet.com>
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include "kvm.h"
  18 #include "x86.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/fs.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/module.h>
  27 #include <linux/mman.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/msr.h>
  31
  32 #define MAX_IO_MSRS 256
  33 #define CR0_RESERVED_BITS                                               \
  34         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  35                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  36                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  37 #define CR4_RESERVED_BITS                                               \
  38         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  39                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  40                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  41                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  42
  43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  44 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  45
  46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  48
  49 struct kvm_x86_ops *kvm_x86_ops;
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "pf_fixed", VCPU_STAT(pf_fixed) },
  53         { "pf_guest", VCPU_STAT(pf_guest) },
  54         { "tlb_flush", VCPU_STAT(tlb_flush) },
  55         { "invlpg", VCPU_STAT(invlpg) },
  56         { "exits", VCPU_STAT(exits) },
  57         { "io_exits", VCPU_STAT(io_exits) },
  58         { "mmio_exits", VCPU_STAT(mmio_exits) },
  59         { "signal_exits", VCPU_STAT(signal_exits) },
  60         { "irq_window", VCPU_STAT(irq_window_exits) },
  61         { "halt_exits", VCPU_STAT(halt_exits) },
  62         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  63         { "request_irq", VCPU_STAT(request_irq_exits) },
  64         { "irq_exits", VCPU_STAT(irq_exits) },
  65         { "host_state_reload", VCPU_STAT(host_state_reload) },
  66         { "efer_reload", VCPU_STAT(efer_reload) },
  67         { "fpu_reload", VCPU_STAT(fpu_reload) },
  68         { "insn_emulation", VCPU_STAT(insn_emulation) },
  69         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  70         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  71         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  72         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  73         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  74         { "mmu_flooded", VM_STAT(mmu_flooded) },
  75         { "mmu_recycled", VM_STAT(mmu_recycled) },
  76         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  77         { NULL }
  78 };
  79
  80
  81 unsigned long segment_base(u16 selector)
  82 {
  83         struct descriptor_table gdt;
  84         struct segment_descriptor *d;
  85         unsigned long table_base;
  86         unsigned long v;
  87
  88         if (selector == 0)
  89                 return 0;
  90
  91         asm("sgdt %0" : "=m"(gdt));
  92         table_base = gdt.base;
  93
  94         if (selector & 4) {           /* from ldt */
  95                 u16 ldt_selector;
  96
  97                 asm("sldt %0" : "=g"(ldt_selector));
  98                 table_base = segment_base(ldt_selector);
  99         }
 100         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 101         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 102                 ((unsigned long)d->base_high << 24);
 103 #ifdef CONFIG_X86_64
 104         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 105                 v |= ((unsigned long) \
 106                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 107 #endif
 108         return v;
 109 }
 110 EXPORT_SYMBOL_GPL(segment_base);
 111
 112 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 113 {
 114         if (irqchip_in_kernel(vcpu->kvm))
 115                 return vcpu->apic_base;
 116         else
 117                 return vcpu->apic_base;
 118 }
 119 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 120
 121 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 122 {
 123         /* TODO: reserve bits check */
 124         if (irqchip_in_kernel(vcpu->kvm))
 125                 kvm_lapic_set_base(vcpu, data);
 126         else
 127                 vcpu->apic_base = data;
 128 }
 129 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 130
 131 static void inject_gp(struct kvm_vcpu *vcpu)
 132 {
 133         kvm_x86_ops->inject_gp(vcpu, 0);
 134 }
 135
 136 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 137 {
 138         WARN_ON(vcpu->exception.pending);
 139         vcpu->exception.pending = true;
 140         vcpu->exception.has_error_code = false;
 141         vcpu->exception.nr = nr;
 142 }
 143 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 144
 145 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 146 {
 147         WARN_ON(vcpu->exception.pending);
 148         vcpu->exception.pending = true;
 149         vcpu->exception.has_error_code = true;
 150         vcpu->exception.nr = nr;
 151         vcpu->exception.error_code = error_code;
 152 }
 153 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 154
 155 static void __queue_exception(struct kvm_vcpu *vcpu)
 156 {
 157         kvm_x86_ops->queue_exception(vcpu, vcpu->exception.nr,
 158                                      vcpu->exception.has_error_code,
 159                                      vcpu->exception.error_code);
 160 }
 161
 162 /*
 163  * Load the pae pdptrs.  Return true is they are all valid.
 164  */
 165 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 166 {
 167         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 168         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 169         int i;
 170         int ret;
 171         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 172
 173         mutex_lock(&vcpu->kvm->lock);
 174         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 175                                   offset * sizeof(u64), sizeof(pdpte));
 176         if (ret < 0) {
 177                 ret = 0;
 178                 goto out;
 179         }
 180         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 181                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 182                         ret = 0;
 183                         goto out;
 184                 }
 185         }
 186         ret = 1;
 187
 188         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 189 out:
 190         mutex_unlock(&vcpu->kvm->lock);
 191
 192         return ret;
 193 }
 194
 195 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 196 {
 197         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 198         bool changed = true;
 199         int r;
 200
 201         if (is_long_mode(vcpu) || !is_pae(vcpu))
 202                 return false;
 203
 204         mutex_lock(&vcpu->kvm->lock);
 205         r = kvm_read_guest(vcpu->kvm, vcpu->cr3 & ~31u, pdpte, sizeof(pdpte));
 206         if (r < 0)
 207                 goto out;
 208         changed = memcmp(pdpte, vcpu->pdptrs, sizeof(pdpte)) != 0;
 209 out:
 210         mutex_unlock(&vcpu->kvm->lock);
 211
 212         return changed;
 213 }
 214
 215 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 216 {
 217         if (cr0 & CR0_RESERVED_BITS) {
 218                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 219                        cr0, vcpu->cr0);
 220                 inject_gp(vcpu);
 221                 return;
 222         }
 223
 224         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 225                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 226                 inject_gp(vcpu);
 227                 return;
 228         }
 229
 230         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 231                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 232                        "and a clear PE flag\n");
 233                 inject_gp(vcpu);
 234                 return;
 235         }
 236
 237         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 238 #ifdef CONFIG_X86_64
 239                 if ((vcpu->shadow_efer & EFER_LME)) {
 240                         int cs_db, cs_l;
 241
 242                         if (!is_pae(vcpu)) {
 243                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 244                                        "in long mode while PAE is disabled\n");
 245                                 inject_gp(vcpu);
 246                                 return;
 247                         }
 248                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 249                         if (cs_l) {
 250                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 251                                        "in long mode while CS.L == 1\n");
 252                                 inject_gp(vcpu);
 253                                 return;
 254
 255                         }
 256                 } else
 257 #endif
 258                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 259                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 260                                "reserved bits\n");
 261                         inject_gp(vcpu);
 262                         return;
 263                 }
 264
 265         }
 266
 267         kvm_x86_ops->set_cr0(vcpu, cr0);
 268         vcpu->cr0 = cr0;
 269
 270         mutex_lock(&vcpu->kvm->lock);
 271         kvm_mmu_reset_context(vcpu);
 272         mutex_unlock(&vcpu->kvm->lock);
 273         return;
 274 }
 275 EXPORT_SYMBOL_GPL(set_cr0);
 276
 277 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 278 {
 279         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 280 }
 281 EXPORT_SYMBOL_GPL(lmsw);
 282
 283 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 284 {
 285         if (cr4 & CR4_RESERVED_BITS) {
 286                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 287                 inject_gp(vcpu);
 288                 return;
 289         }
 290
 291         if (is_long_mode(vcpu)) {
 292                 if (!(cr4 & X86_CR4_PAE)) {
 293                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 294                                "in long mode\n");
 295                         inject_gp(vcpu);
 296                         return;
 297                 }
 298         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 299                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 300                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 301                 inject_gp(vcpu);
 302                 return;
 303         }
 304
 305         if (cr4 & X86_CR4_VMXE) {
 306                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 307                 inject_gp(vcpu);
 308                 return;
 309         }
 310         kvm_x86_ops->set_cr4(vcpu, cr4);
 311         vcpu->cr4 = cr4;
 312         mutex_lock(&vcpu->kvm->lock);
 313         kvm_mmu_reset_context(vcpu);
 314         mutex_unlock(&vcpu->kvm->lock);
 315 }
 316 EXPORT_SYMBOL_GPL(set_cr4);
 317
 318 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 319 {
 320         if (cr3 == vcpu->cr3 && !pdptrs_changed(vcpu)) {
 321                 kvm_mmu_flush_tlb(vcpu);
 322                 return;
 323         }
 324
 325         if (is_long_mode(vcpu)) {
 326                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 327                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 328                         inject_gp(vcpu);
 329                         return;
 330                 }
 331         } else {
 332                 if (is_pae(vcpu)) {
 333                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 334                                 printk(KERN_DEBUG
 335                                        "set_cr3: #GP, reserved bits\n");
 336                                 inject_gp(vcpu);
 337                                 return;
 338                         }
 339                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 340                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 341                                        "reserved bits\n");
 342                                 inject_gp(vcpu);
 343                                 return;
 344                         }
 345                 }
 346                 /*
 347                  * We don't check reserved bits in nonpae mode, because
 348                  * this isn't enforced, and VMware depends on this.
 349                  */
 350         }
 351
 352         mutex_lock(&vcpu->kvm->lock);
 353         /*
 354          * Does the new cr3 value map to physical memory? (Note, we
 355          * catch an invalid cr3 even in real-mode, because it would
 356          * cause trouble later on when we turn on paging anyway.)
 357          *
 358          * A real CPU would silently accept an invalid cr3 and would
 359          * attempt to use it - with largely undefined (and often hard
 360          * to debug) behavior on the guest side.
 361          */
 362         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 363                 inject_gp(vcpu);
 364         else {
 365                 vcpu->cr3 = cr3;
 366                 vcpu->mmu.new_cr3(vcpu);
 367         }
 368         mutex_unlock(&vcpu->kvm->lock);
 369 }
 370 EXPORT_SYMBOL_GPL(set_cr3);
 371
 372 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 373 {
 374         if (cr8 & CR8_RESERVED_BITS) {
 375                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 376                 inject_gp(vcpu);
 377                 return;
 378         }
 379         if (irqchip_in_kernel(vcpu->kvm))
 380                 kvm_lapic_set_tpr(vcpu, cr8);
 381         else
 382                 vcpu->cr8 = cr8;
 383 }
 384 EXPORT_SYMBOL_GPL(set_cr8);
 385
 386 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 387 {
 388         if (irqchip_in_kernel(vcpu->kvm))
 389                 return kvm_lapic_get_cr8(vcpu);
 390         else
 391                 return vcpu->cr8;
 392 }
 393 EXPORT_SYMBOL_GPL(get_cr8);
 394
 395 /*
 396  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 397  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 398  *
 399  * This list is modified at module load time to reflect the
 400  * capabilities of the host cpu.
 401  */
 402 static u32 msrs_to_save[] = {
 403         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 404         MSR_K6_STAR,
 405 #ifdef CONFIG_X86_64
 406         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 407 #endif
 408         MSR_IA32_TIME_STAMP_COUNTER,
 409 };
 410
 411 static unsigned num_msrs_to_save;
 412
 413 static u32 emulated_msrs[] = {
 414         MSR_IA32_MISC_ENABLE,
 415 };
 416
 417 #ifdef CONFIG_X86_64
 418
 419 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 420 {
 421         if (efer & EFER_RESERVED_BITS) {
 422                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 423                        efer);
 424                 inject_gp(vcpu);
 425                 return;
 426         }
 427
 428         if (is_paging(vcpu)
 429             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 430                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 431                 inject_gp(vcpu);
 432                 return;
 433         }
 434
 435         kvm_x86_ops->set_efer(vcpu, efer);
 436
 437         efer &= ~EFER_LMA;
 438         efer |= vcpu->shadow_efer & EFER_LMA;
 439
 440         vcpu->shadow_efer = efer;
 441 }
 442
 443 #endif
 444
 445 /*
 446  * Writes msr value into into the appropriate "register".
 447  * Returns 0 on success, non-0 otherwise.
 448  * Assumes vcpu_load() was already called.
 449  */
 450 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 451 {
 452         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 453 }
 454
 455 /*
 456  * Adapt set_msr() to msr_io()'s calling convention
 457  */
 458 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 459 {
 460         return kvm_set_msr(vcpu, index, *data);
 461 }
 462
 463
 464 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 465 {
 466         switch (msr) {
 467 #ifdef CONFIG_X86_64
 468         case MSR_EFER:
 469                 set_efer(vcpu, data);
 470                 break;
 471 #endif
 472         case MSR_IA32_MC0_STATUS:
 473                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 474                        __FUNCTION__, data);
 475                 break;
 476         case MSR_IA32_MCG_STATUS:
 477                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 478                         __FUNCTION__, data);
 479                 break;
 480         case MSR_IA32_UCODE_REV:
 481         case MSR_IA32_UCODE_WRITE:
 482         case 0x200 ... 0x2ff: /* MTRRs */
 483                 break;
 484         case MSR_IA32_APICBASE:
 485                 kvm_set_apic_base(vcpu, data);
 486                 break;
 487         case MSR_IA32_MISC_ENABLE:
 488                 vcpu->ia32_misc_enable_msr = data;
 489                 break;
 490         default:
 491                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
 492                 return 1;
 493         }
 494         return 0;
 495 }
 496 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 497
 498
 499 /*
 500  * Reads an msr value (of 'msr_index') into 'pdata'.
 501  * Returns 0 on success, non-0 otherwise.
 502  * Assumes vcpu_load() was already called.
 503  */
 504 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 505 {
 506         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 507 }
 508
 509 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 510 {
 511         u64 data;
 512
 513         switch (msr) {
 514         case 0xc0010010: /* SYSCFG */
 515         case 0xc0010015: /* HWCR */
 516         case MSR_IA32_PLATFORM_ID:
 517         case MSR_IA32_P5_MC_ADDR:
 518         case MSR_IA32_P5_MC_TYPE:
 519         case MSR_IA32_MC0_CTL:
 520         case MSR_IA32_MCG_STATUS:
 521         case MSR_IA32_MCG_CAP:
 522         case MSR_IA32_MC0_MISC:
 523         case MSR_IA32_MC0_MISC+4:
 524         case MSR_IA32_MC0_MISC+8:
 525         case MSR_IA32_MC0_MISC+12:
 526         case MSR_IA32_MC0_MISC+16:
 527         case MSR_IA32_UCODE_REV:
 528         case MSR_IA32_PERF_STATUS:
 529         case MSR_IA32_EBL_CR_POWERON:
 530                 /* MTRR registers */
 531         case 0xfe:
 532         case 0x200 ... 0x2ff:
 533                 data = 0;
 534                 break;
 535         case 0xcd: /* fsb frequency */
 536                 data = 3;
 537                 break;
 538         case MSR_IA32_APICBASE:
 539                 data = kvm_get_apic_base(vcpu);
 540                 break;
 541         case MSR_IA32_MISC_ENABLE:
 542                 data = vcpu->ia32_misc_enable_msr;
 543                 break;
 544 #ifdef CONFIG_X86_64
 545         case MSR_EFER:
 546                 data = vcpu->shadow_efer;
 547                 break;
 548 #endif
 549         default:
 550                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 551                 return 1;
 552         }
 553         *pdata = data;
 554         return 0;
 555 }
 556 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 557
 558 /*
 559  * Read or write a bunch of msrs. All parameters are kernel addresses.
 560  *
 561  * @return number of msrs set successfully.
 562  */
 563 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 564                     struct kvm_msr_entry *entries,
 565                     int (*do_msr)(struct kvm_vcpu *vcpu,
 566                                   unsigned index, u64 *data))
 567 {
 568         int i;
 569
 570         vcpu_load(vcpu);
 571
 572         for (i = 0; i < msrs->nmsrs; ++i)
 573                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 574                         break;
 575
 576         vcpu_put(vcpu);
 577
 578         return i;
 579 }
 580
 581 /*
 582  * Read or write a bunch of msrs. Parameters are user addresses.
 583  *
 584  * @return number of msrs set successfully.
 585  */
 586 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 587                   int (*do_msr)(struct kvm_vcpu *vcpu,
 588                                 unsigned index, u64 *data),
 589                   int writeback)
 590 {
 591         struct kvm_msrs msrs;
 592         struct kvm_msr_entry *entries;
 593         int r, n;
 594         unsigned size;
 595
 596         r = -EFAULT;
 597         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 598                 goto out;
 599
 600         r = -E2BIG;
 601         if (msrs.nmsrs >= MAX_IO_MSRS)
 602                 goto out;
 603
 604         r = -ENOMEM;
 605         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 606         entries = vmalloc(size);
 607         if (!entries)
 608                 goto out;
 609
 610         r = -EFAULT;
 611         if (copy_from_user(entries, user_msrs->entries, size))
 612                 goto out_free;
 613
 614         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 615         if (r < 0)
 616                 goto out_free;
 617
 618         r = -EFAULT;
 619         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 620                 goto out_free;
 621
 622         r = n;
 623
 624 out_free:
 625         vfree(entries);
 626 out:
 627         return r;
 628 }
 629
 630 /*
 631  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 632  * cached on it.
 633  */
 634 void decache_vcpus_on_cpu(int cpu)
 635 {
 636         struct kvm *vm;
 637         struct kvm_vcpu *vcpu;
 638         int i;
 639
 640         spin_lock(&kvm_lock);
 641         list_for_each_entry(vm, &vm_list, vm_list)
 642                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 643                         vcpu = vm->vcpus[i];
 644                         if (!vcpu)
 645                                 continue;
 646                         /*
 647                          * If the vcpu is locked, then it is running on some
 648                          * other cpu and therefore it is not cached on the
 649                          * cpu in question.
 650                          *
 651                          * If it's not locked, check the last cpu it executed
 652                          * on.
 653                          */
 654                         if (mutex_trylock(&vcpu->mutex)) {
 655                                 if (vcpu->cpu == cpu) {
 656                                         kvm_x86_ops->vcpu_decache(vcpu);
 657                                         vcpu->cpu = -1;
 658                                 }
 659                                 mutex_unlock(&vcpu->mutex);
 660                         }
 661                 }
 662         spin_unlock(&kvm_lock);
 663 }
 664
 665 int kvm_dev_ioctl_check_extension(long ext)
 666 {
 667         int r;
 668
 669         switch (ext) {
 670         case KVM_CAP_IRQCHIP:
 671         case KVM_CAP_HLT:
 672         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 673         case KVM_CAP_USER_MEMORY:
 674         case KVM_CAP_SET_TSS_ADDR:
 675         case KVM_CAP_EXT_CPUID:
 676                 r = 1;
 677                 break;
 678         default:
 679                 r = 0;
 680                 break;
 681         }
 682         return r;
 683
 684 }
 685
 686 long kvm_arch_dev_ioctl(struct file *filp,
 687                         unsigned int ioctl, unsigned long arg)
 688 {
 689         void __user *argp = (void __user *)arg;
 690         long r;
 691
 692         switch (ioctl) {
 693         case KVM_GET_MSR_INDEX_LIST: {
 694                 struct kvm_msr_list __user *user_msr_list = argp;
 695                 struct kvm_msr_list msr_list;
 696                 unsigned n;
 697
 698                 r = -EFAULT;
 699                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 700                         goto out;
 701                 n = msr_list.nmsrs;
 702                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 703                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 704                         goto out;
 705                 r = -E2BIG;
 706                 if (n < num_msrs_to_save)
 707                         goto out;
 708                 r = -EFAULT;
 709                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 710                                  num_msrs_to_save * sizeof(u32)))
 711                         goto out;
 712                 if (copy_to_user(user_msr_list->indices
 713                                  + num_msrs_to_save * sizeof(u32),
 714                                  &emulated_msrs,
 715                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 716                         goto out;
 717                 r = 0;
 718                 break;
 719         }
 720         default:
 721                 r = -EINVAL;
 722         }
 723 out:
 724         return r;
 725 }
 726
 727 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 728 {
 729         kvm_x86_ops->vcpu_load(vcpu, cpu);
 730 }
 731
 732 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 733 {
 734         kvm_x86_ops->vcpu_put(vcpu);
 735         kvm_put_guest_fpu(vcpu);
 736 }
 737
 738 static int is_efer_nx(void)
 739 {
 740         u64 efer;
 741
 742         rdmsrl(MSR_EFER, efer);
 743         return efer & EFER_NX;
 744 }
 745
 746 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 747 {
 748         int i;
 749         struct kvm_cpuid_entry2 *e, *entry;
 750
 751         entry = NULL;
 752         for (i = 0; i < vcpu->cpuid_nent; ++i) {
 753                 e = &vcpu->cpuid_entries[i];
 754                 if (e->function == 0x80000001) {
 755                         entry = e;
 756                         break;
 757                 }
 758         }
 759         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 760                 entry->edx &= ~(1 << 20);
 761                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 762         }
 763 }
 764
 765 /* when an old userspace process fills a new kernel module */
 766 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 767                                     struct kvm_cpuid *cpuid,
 768                                     struct kvm_cpuid_entry __user *entries)
 769 {
 770         int r, i;
 771         struct kvm_cpuid_entry *cpuid_entries;
 772
 773         r = -E2BIG;
 774         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 775                 goto out;
 776         r = -ENOMEM;
 777         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 778         if (!cpuid_entries)
 779                 goto out;
 780         r = -EFAULT;
 781         if (copy_from_user(cpuid_entries, entries,
 782                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 783                 goto out_free;
 784         for (i = 0; i < cpuid->nent; i++) {
 785                 vcpu->cpuid_entries[i].function = cpuid_entries[i].function;
 786                 vcpu->cpuid_entries[i].eax = cpuid_entries[i].eax;
 787                 vcpu->cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 788                 vcpu->cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 789                 vcpu->cpuid_entries[i].edx = cpuid_entries[i].edx;
 790                 vcpu->cpuid_entries[i].index = 0;
 791                 vcpu->cpuid_entries[i].flags = 0;
 792                 vcpu->cpuid_entries[i].padding[0] = 0;
 793                 vcpu->cpuid_entries[i].padding[1] = 0;
 794                 vcpu->cpuid_entries[i].padding[2] = 0;
 795         }
 796         vcpu->cpuid_nent = cpuid->nent;
 797         cpuid_fix_nx_cap(vcpu);
 798         r = 0;
 799
 800 out_free:
 801         vfree(cpuid_entries);
 802 out:
 803         return r;
 804 }
 805
 806 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 807                                     struct kvm_cpuid2 *cpuid,
 808                                     struct kvm_cpuid_entry2 __user *entries)
 809 {
 810         int r;
 811
 812         r = -E2BIG;
 813         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 814                 goto out;
 815         r = -EFAULT;
 816         if (copy_from_user(&vcpu->cpuid_entries, entries,
 817                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 818                 goto out;
 819         vcpu->cpuid_nent = cpuid->nent;
 820         return 0;
 821
 822 out:
 823         return r;
 824 }
 825
 826 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 827                                     struct kvm_cpuid2 *cpuid,
 828                                     struct kvm_cpuid_entry2 __user *entries)
 829 {
 830         int r;
 831
 832         r = -E2BIG;
 833         if (cpuid->nent < vcpu->cpuid_nent)
 834                 goto out;
 835         r = -EFAULT;
 836         if (copy_to_user(entries, &vcpu->cpuid_entries,
 837                            vcpu->cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 838                 goto out;
 839         return 0;
 840
 841 out:
 842         cpuid->nent = vcpu->cpuid_nent;
 843         return r;
 844 }
 845
 846 static inline u32 bit(int bitno)
 847 {
 848         return 1 << (bitno & 31);
 849 }
 850
 851 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 852                           u32 index)
 853 {
 854         entry->function = function;
 855         entry->index = index;
 856         cpuid_count(entry->function, entry->index,
 857                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 858         entry->flags = 0;
 859 }
 860
 861 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 862                          u32 index, int *nent, int maxnent)
 863 {
 864         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
 865                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 866                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 867                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 868                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 869                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
 870                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 871                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
 872                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
 873                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
 874         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
 875                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 876                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 877                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 878                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 879                 bit(X86_FEATURE_PGE) |
 880                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 881                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
 882                 bit(X86_FEATURE_SYSCALL) |
 883                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
 884 #ifdef CONFIG_X86_64
 885                 bit(X86_FEATURE_LM) |
 886 #endif
 887                 bit(X86_FEATURE_MMXEXT) |
 888                 bit(X86_FEATURE_3DNOWEXT) |
 889                 bit(X86_FEATURE_3DNOW);
 890         const u32 kvm_supported_word3_x86_features =
 891                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
 892         const u32 kvm_supported_word6_x86_features =
 893                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
 894
 895         /* all func 2 cpuid_count() should be called on the same cpu */
 896         get_cpu();
 897         do_cpuid_1_ent(entry, function, index);
 898         ++*nent;
 899
 900         switch (function) {
 901         case 0:
 902                 entry->eax = min(entry->eax, (u32)0xb);
 903                 break;
 904         case 1:
 905                 entry->edx &= kvm_supported_word0_x86_features;
 906                 entry->ecx &= kvm_supported_word3_x86_features;
 907                 break;
 908         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 909          * may return different values. This forces us to get_cpu() before
 910          * issuing the first command, and also to emulate this annoying behavior
 911          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 912         case 2: {
 913                 int t, times = entry->eax & 0xff;
 914
 915                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 916                 for (t = 1; t < times && *nent < maxnent; ++t) {
 917                         do_cpuid_1_ent(&entry[t], function, 0);
 918                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 919                         ++*nent;
 920                 }
 921                 break;
 922         }
 923         /* function 4 and 0xb have additional index. */
 924         case 4: {
 925                 int index, cache_type;
 926
 927                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 928                 /* read more entries until cache_type is zero */
 929                 for (index = 1; *nent < maxnent; ++index) {
 930                         cache_type = entry[index - 1].eax & 0x1f;
 931                         if (!cache_type)
 932                                 break;
 933                         do_cpuid_1_ent(&entry[index], function, index);
 934                         entry[index].flags |=
 935                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 936                         ++*nent;
 937                 }
 938                 break;
 939         }
 940         case 0xb: {
 941                 int index, level_type;
 942
 943                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 944                 /* read more entries until level_type is zero */
 945                 for (index = 1; *nent < maxnent; ++index) {
 946                         level_type = entry[index - 1].ecx & 0xff;
 947                         if (!level_type)
 948                                 break;
 949                         do_cpuid_1_ent(&entry[index], function, index);
 950                         entry[index].flags |=
 951                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 952                         ++*nent;
 953                 }
 954                 break;
 955         }
 956         case 0x80000000:
 957                 entry->eax = min(entry->eax, 0x8000001a);
 958                 break;
 959         case 0x80000001:
 960                 entry->edx &= kvm_supported_word1_x86_features;
 961                 entry->ecx &= kvm_supported_word6_x86_features;
 962                 break;
 963         }
 964         put_cpu();
 965 }
 966
 967 static int kvm_vm_ioctl_get_supported_cpuid(struct kvm *kvm,
 968                                     struct kvm_cpuid2 *cpuid,
 969                                     struct kvm_cpuid_entry2 __user *entries)
 970 {
 971         struct kvm_cpuid_entry2 *cpuid_entries;
 972         int limit, nent = 0, r = -E2BIG;
 973         u32 func;
 974
 975         if (cpuid->nent < 1)
 976                 goto out;
 977         r = -ENOMEM;
 978         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 979         if (!cpuid_entries)
 980                 goto out;
 981
 982         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
 983         limit = cpuid_entries[0].eax;
 984         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
 985                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
 986                                 &nent, cpuid->nent);
 987         r = -E2BIG;
 988         if (nent >= cpuid->nent)
 989                 goto out_free;
 990
 991         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
 992         limit = cpuid_entries[nent - 1].eax;
 993         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
 994                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
 995                                &nent, cpuid->nent);
 996         r = -EFAULT;
 997         if (copy_to_user(entries, cpuid_entries,
 998                         nent * sizeof(struct kvm_cpuid_entry2)))
 999                 goto out_free;
1000         cpuid->nent = nent;
1001         r = 0;
1002
1003 out_free:
1004         vfree(cpuid_entries);
1005 out:
1006         return r;
1007 }
1008
1009 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1010                                     struct kvm_lapic_state *s)
1011 {
1012         vcpu_load(vcpu);
1013         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
1014         vcpu_put(vcpu);
1015
1016         return 0;
1017 }
1018
1019 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1020                                     struct kvm_lapic_state *s)
1021 {
1022         vcpu_load(vcpu);
1023         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
1024         kvm_apic_post_state_restore(vcpu);
1025         vcpu_put(vcpu);
1026
1027         return 0;
1028 }
1029
1030 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1031                                     struct kvm_interrupt *irq)
1032 {
1033         if (irq->irq < 0 || irq->irq >= 256)
1034                 return -EINVAL;
1035         if (irqchip_in_kernel(vcpu->kvm))
1036                 return -ENXIO;
1037         vcpu_load(vcpu);
1038
1039         set_bit(irq->irq, vcpu->irq_pending);
1040         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1041
1042         vcpu_put(vcpu);
1043
1044         return 0;
1045 }
1046
1047 long kvm_arch_vcpu_ioctl(struct file *filp,
1048                          unsigned int ioctl, unsigned long arg)
1049 {
1050         struct kvm_vcpu *vcpu = filp->private_data;
1051         void __user *argp = (void __user *)arg;
1052         int r;
1053
1054         switch (ioctl) {
1055         case KVM_GET_LAPIC: {
1056                 struct kvm_lapic_state lapic;
1057
1058                 memset(&lapic, 0, sizeof lapic);
1059                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1060                 if (r)
1061                         goto out;
1062                 r = -EFAULT;
1063                 if (copy_to_user(argp, &lapic, sizeof lapic))
1064                         goto out;
1065                 r = 0;
1066                 break;
1067         }
1068         case KVM_SET_LAPIC: {
1069                 struct kvm_lapic_state lapic;
1070
1071                 r = -EFAULT;
1072                 if (copy_from_user(&lapic, argp, sizeof lapic))
1073                         goto out;
1074                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1075                 if (r)
1076                         goto out;
1077                 r = 0;
1078                 break;
1079         }
1080         case KVM_INTERRUPT: {
1081                 struct kvm_interrupt irq;
1082
1083                 r = -EFAULT;
1084                 if (copy_from_user(&irq, argp, sizeof irq))
1085                         goto out;
1086                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1087                 if (r)
1088                         goto out;
1089                 r = 0;
1090                 break;
1091         }
1092         case KVM_SET_CPUID: {
1093                 struct kvm_cpuid __user *cpuid_arg = argp;
1094                 struct kvm_cpuid cpuid;
1095
1096                 r = -EFAULT;
1097                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1098                         goto out;
1099                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1100                 if (r)
1101                         goto out;
1102                 break;
1103         }
1104         case KVM_SET_CPUID2: {
1105                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1106                 struct kvm_cpuid2 cpuid;
1107
1108                 r = -EFAULT;
1109                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1110                         goto out;
1111                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1112                                 cpuid_arg->entries);
1113                 if (r)
1114                         goto out;
1115                 break;
1116         }
1117         case KVM_GET_CPUID2: {
1118                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1119                 struct kvm_cpuid2 cpuid;
1120
1121                 r = -EFAULT;
1122                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1123                         goto out;
1124                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1125                                 cpuid_arg->entries);
1126                 if (r)
1127                         goto out;
1128                 r = -EFAULT;
1129                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1130                         goto out;
1131                 r = 0;
1132                 break;
1133         }
1134         case KVM_GET_MSRS:
1135                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1136                 break;
1137         case KVM_SET_MSRS:
1138                 r = msr_io(vcpu, argp, do_set_msr, 0);
1139                 break;
1140         default:
1141                 r = -EINVAL;
1142         }
1143 out:
1144         return r;
1145 }
1146
1147 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1148 {
1149         int ret;
1150
1151         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1152                 return -1;
1153         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1154         return ret;
1155 }
1156
1157 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1158                                           u32 kvm_nr_mmu_pages)
1159 {
1160         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1161                 return -EINVAL;
1162
1163         mutex_lock(&kvm->lock);
1164
1165         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1166         kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
1167
1168         mutex_unlock(&kvm->lock);
1169         return 0;
1170 }
1171
1172 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1173 {
1174         return kvm->n_alloc_mmu_pages;
1175 }
1176
1177 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1178 {
1179         int i;
1180         struct kvm_mem_alias *alias;
1181
1182         for (i = 0; i < kvm->naliases; ++i) {
1183                 alias = &kvm->aliases[i];
1184                 if (gfn >= alias->base_gfn
1185                     && gfn < alias->base_gfn + alias->npages)
1186                         return alias->target_gfn + gfn - alias->base_gfn;
1187         }
1188         return gfn;
1189 }
1190
1191 /*
1192  * Set a new alias region.  Aliases map a portion of physical memory into
1193  * another portion.  This is useful for memory windows, for example the PC
1194  * VGA region.
1195  */
1196 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1197                                          struct kvm_memory_alias *alias)
1198 {
1199         int r, n;
1200         struct kvm_mem_alias *p;
1201
1202         r = -EINVAL;
1203         /* General sanity checks */
1204         if (alias->memory_size & (PAGE_SIZE - 1))
1205                 goto out;
1206         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1207                 goto out;
1208         if (alias->slot >= KVM_ALIAS_SLOTS)
1209                 goto out;
1210         if (alias->guest_phys_addr + alias->memory_size
1211             < alias->guest_phys_addr)
1212                 goto out;
1213         if (alias->target_phys_addr + alias->memory_size
1214             < alias->target_phys_addr)
1215                 goto out;
1216
1217         mutex_lock(&kvm->lock);
1218
1219         p = &kvm->aliases[alias->slot];
1220         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1221         p->npages = alias->memory_size >> PAGE_SHIFT;
1222         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1223
1224         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1225                 if (kvm->aliases[n - 1].npages)
1226                         break;
1227         kvm->naliases = n;
1228
1229         kvm_mmu_zap_all(kvm);
1230
1231         mutex_unlock(&kvm->lock);
1232
1233         return 0;
1234
1235 out:
1236         return r;
1237 }
1238
1239 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1240 {
1241         int r;
1242
1243         r = 0;
1244         switch (chip->chip_id) {
1245         case KVM_IRQCHIP_PIC_MASTER:
1246                 memcpy(&chip->chip.pic,
1247                         &pic_irqchip(kvm)->pics[0],
1248                         sizeof(struct kvm_pic_state));
1249                 break;
1250         case KVM_IRQCHIP_PIC_SLAVE:
1251                 memcpy(&chip->chip.pic,
1252                         &pic_irqchip(kvm)->pics[1],
1253                         sizeof(struct kvm_pic_state));
1254                 break;
1255         case KVM_IRQCHIP_IOAPIC:
1256                 memcpy(&chip->chip.ioapic,
1257                         ioapic_irqchip(kvm),
1258                         sizeof(struct kvm_ioapic_state));
1259                 break;
1260         default:
1261                 r = -EINVAL;
1262                 break;
1263         }
1264         return r;
1265 }
1266
1267 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1268 {
1269         int r;
1270
1271         r = 0;
1272         switch (chip->chip_id) {
1273         case KVM_IRQCHIP_PIC_MASTER:
1274                 memcpy(&pic_irqchip(kvm)->pics[0],
1275                         &chip->chip.pic,
1276                         sizeof(struct kvm_pic_state));
1277                 break;
1278         case KVM_IRQCHIP_PIC_SLAVE:
1279                 memcpy(&pic_irqchip(kvm)->pics[1],
1280                         &chip->chip.pic,
1281                         sizeof(struct kvm_pic_state));
1282                 break;
1283         case KVM_IRQCHIP_IOAPIC:
1284                 memcpy(ioapic_irqchip(kvm),
1285                         &chip->chip.ioapic,
1286                         sizeof(struct kvm_ioapic_state));
1287                 break;
1288         default:
1289                 r = -EINVAL;
1290                 break;
1291         }
1292         kvm_pic_update_irq(pic_irqchip(kvm));
1293         return r;
1294 }
1295
1296 /*
1297  * Get (and clear) the dirty memory log for a memory slot.
1298  */
1299 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1300                                       struct kvm_dirty_log *log)
1301 {
1302         int r;
1303         int n;
1304         struct kvm_memory_slot *memslot;
1305         int is_dirty = 0;
1306
1307         mutex_lock(&kvm->lock);
1308
1309         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1310         if (r)
1311                 goto out;
1312
1313         /* If nothing is dirty, don't bother messing with page tables. */
1314         if (is_dirty) {
1315                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1316                 kvm_flush_remote_tlbs(kvm);
1317                 memslot = &kvm->memslots[log->slot];
1318                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1319                 memset(memslot->dirty_bitmap, 0, n);
1320         }
1321         r = 0;
1322 out:
1323         mutex_unlock(&kvm->lock);
1324         return r;
1325 }
1326
1327 long kvm_arch_vm_ioctl(struct file *filp,
1328                        unsigned int ioctl, unsigned long arg)
1329 {
1330         struct kvm *kvm = filp->private_data;
1331         void __user *argp = (void __user *)arg;
1332         int r = -EINVAL;
1333
1334         switch (ioctl) {
1335         case KVM_SET_TSS_ADDR:
1336                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1337                 if (r < 0)
1338                         goto out;
1339                 break;
1340         case KVM_SET_MEMORY_REGION: {
1341                 struct kvm_memory_region kvm_mem;
1342                 struct kvm_userspace_memory_region kvm_userspace_mem;
1343
1344                 r = -EFAULT;
1345                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1346                         goto out;
1347                 kvm_userspace_mem.slot = kvm_mem.slot;
1348                 kvm_userspace_mem.flags = kvm_mem.flags;
1349                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1350                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1351                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1352                 if (r)
1353                         goto out;
1354                 break;
1355         }
1356         case KVM_SET_NR_MMU_PAGES:
1357                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1358                 if (r)
1359                         goto out;
1360                 break;
1361         case KVM_GET_NR_MMU_PAGES:
1362                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1363                 break;
1364         case KVM_SET_MEMORY_ALIAS: {
1365                 struct kvm_memory_alias alias;
1366
1367                 r = -EFAULT;
1368                 if (copy_from_user(&alias, argp, sizeof alias))
1369                         goto out;
1370                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1371                 if (r)
1372                         goto out;
1373                 break;
1374         }
1375         case KVM_CREATE_IRQCHIP:
1376                 r = -ENOMEM;
1377                 kvm->vpic = kvm_create_pic(kvm);
1378                 if (kvm->vpic) {
1379                         r = kvm_ioapic_init(kvm);
1380                         if (r) {
1381                                 kfree(kvm->vpic);
1382                                 kvm->vpic = NULL;
1383                                 goto out;
1384                         }
1385                 } else
1386                         goto out;
1387                 break;
1388         case KVM_IRQ_LINE: {
1389                 struct kvm_irq_level irq_event;
1390
1391                 r = -EFAULT;
1392                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1393                         goto out;
1394                 if (irqchip_in_kernel(kvm)) {
1395                         mutex_lock(&kvm->lock);
1396                         if (irq_event.irq < 16)
1397                                 kvm_pic_set_irq(pic_irqchip(kvm),
1398                                         irq_event.irq,
1399                                         irq_event.level);
1400                         kvm_ioapic_set_irq(kvm->vioapic,
1401                                         irq_event.irq,
1402                                         irq_event.level);
1403                         mutex_unlock(&kvm->lock);
1404                         r = 0;
1405                 }
1406                 break;
1407         }
1408         case KVM_GET_IRQCHIP: {
1409                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1410                 struct kvm_irqchip chip;
1411
1412                 r = -EFAULT;
1413                 if (copy_from_user(&chip, argp, sizeof chip))
1414                         goto out;
1415                 r = -ENXIO;
1416                 if (!irqchip_in_kernel(kvm))
1417                         goto out;
1418                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1419                 if (r)
1420                         goto out;
1421                 r = -EFAULT;
1422                 if (copy_to_user(argp, &chip, sizeof chip))
1423                         goto out;
1424                 r = 0;
1425                 break;
1426         }
1427         case KVM_SET_IRQCHIP: {
1428                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1429                 struct kvm_irqchip chip;
1430
1431                 r = -EFAULT;
1432                 if (copy_from_user(&chip, argp, sizeof chip))
1433                         goto out;
1434                 r = -ENXIO;
1435                 if (!irqchip_in_kernel(kvm))
1436                         goto out;
1437                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1438                 if (r)
1439                         goto out;
1440                 r = 0;
1441                 break;
1442         }
1443         case KVM_GET_SUPPORTED_CPUID: {
1444                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1445                 struct kvm_cpuid2 cpuid;
1446
1447                 r = -EFAULT;
1448                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1449                         goto out;
1450                 r = kvm_vm_ioctl_get_supported_cpuid(kvm, &cpuid,
1451                         cpuid_arg->entries);
1452                 if (r)
1453                         goto out;
1454
1455                 r = -EFAULT;
1456                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1457                         goto out;
1458                 r = 0;
1459                 break;
1460         }
1461         default:
1462                 ;
1463         }
1464 out:
1465         return r;
1466 }
1467
1468 static void kvm_init_msr_list(void)
1469 {
1470         u32 dummy[2];
1471         unsigned i, j;
1472
1473         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1474                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1475                         continue;
1476                 if (j < i)
1477                         msrs_to_save[j] = msrs_to_save[i];
1478                 j++;
1479         }
1480         num_msrs_to_save = j;
1481 }
1482
1483 /*
1484  * Only apic need an MMIO device hook, so shortcut now..
1485  */
1486 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1487                                                 gpa_t addr)
1488 {
1489         struct kvm_io_device *dev;
1490
1491         if (vcpu->apic) {
1492                 dev = &vcpu->apic->dev;
1493                 if (dev->in_range(dev, addr))
1494                         return dev;
1495         }
1496         return NULL;
1497 }
1498
1499
1500 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1501                                                 gpa_t addr)
1502 {
1503         struct kvm_io_device *dev;
1504
1505         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1506         if (dev == NULL)
1507                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1508         return dev;
1509 }
1510
1511 int emulator_read_std(unsigned long addr,
1512                              void *val,
1513                              unsigned int bytes,
1514                              struct kvm_vcpu *vcpu)
1515 {
1516         void *data = val;
1517
1518         while (bytes) {
1519                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1520                 unsigned offset = addr & (PAGE_SIZE-1);
1521                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1522                 int ret;
1523
1524                 if (gpa == UNMAPPED_GVA)
1525                         return X86EMUL_PROPAGATE_FAULT;
1526                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1527                 if (ret < 0)
1528                         return X86EMUL_UNHANDLEABLE;
1529
1530                 bytes -= tocopy;
1531                 data += tocopy;
1532                 addr += tocopy;
1533         }
1534
1535         return X86EMUL_CONTINUE;
1536 }
1537 EXPORT_SYMBOL_GPL(emulator_read_std);
1538
1539 static int emulator_read_emulated(unsigned long addr,
1540                                   void *val,
1541                                   unsigned int bytes,
1542                                   struct kvm_vcpu *vcpu)
1543 {
1544         struct kvm_io_device *mmio_dev;
1545         gpa_t                 gpa;
1546
1547         if (vcpu->mmio_read_completed) {
1548                 memcpy(val, vcpu->mmio_data, bytes);
1549                 vcpu->mmio_read_completed = 0;
1550                 return X86EMUL_CONTINUE;
1551         }
1552
1553         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1554
1555         /* For APIC access vmexit */
1556         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1557                 goto mmio;
1558
1559         if (emulator_read_std(addr, val, bytes, vcpu)
1560                         == X86EMUL_CONTINUE)
1561                 return X86EMUL_CONTINUE;
1562         if (gpa == UNMAPPED_GVA)
1563                 return X86EMUL_PROPAGATE_FAULT;
1564
1565 mmio:
1566         /*
1567          * Is this MMIO handled locally?
1568          */
1569         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1570         if (mmio_dev) {
1571                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1572                 return X86EMUL_CONTINUE;
1573         }
1574
1575         vcpu->mmio_needed = 1;
1576         vcpu->mmio_phys_addr = gpa;
1577         vcpu->mmio_size = bytes;
1578         vcpu->mmio_is_write = 0;
1579
1580         return X86EMUL_UNHANDLEABLE;
1581 }
1582
1583 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1584                                const void *val, int bytes)
1585 {
1586         int ret;
1587
1588         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1589         if (ret < 0)
1590                 return 0;
1591         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1592         return 1;
1593 }
1594
1595 static int emulator_write_emulated_onepage(unsigned long addr,
1596                                            const void *val,
1597                                            unsigned int bytes,
1598                                            struct kvm_vcpu *vcpu)
1599 {
1600         struct kvm_io_device *mmio_dev;
1601         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1602
1603         if (gpa == UNMAPPED_GVA) {
1604                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1605                 return X86EMUL_PROPAGATE_FAULT;
1606         }
1607
1608         /* For APIC access vmexit */
1609         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1610                 goto mmio;
1611
1612         if (emulator_write_phys(vcpu, gpa, val, bytes))
1613                 return X86EMUL_CONTINUE;
1614
1615 mmio:
1616         /*
1617          * Is this MMIO handled locally?
1618          */
1619         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1620         if (mmio_dev) {
1621                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1622                 return X86EMUL_CONTINUE;
1623         }
1624
1625         vcpu->mmio_needed = 1;
1626         vcpu->mmio_phys_addr = gpa;
1627         vcpu->mmio_size = bytes;
1628         vcpu->mmio_is_write = 1;
1629         memcpy(vcpu->mmio_data, val, bytes);
1630
1631         return X86EMUL_CONTINUE;
1632 }
1633
1634 int emulator_write_emulated(unsigned long addr,
1635                                    const void *val,
1636                                    unsigned int bytes,
1637                                    struct kvm_vcpu *vcpu)
1638 {
1639         /* Crossing a page boundary? */
1640         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1641                 int rc, now;
1642
1643                 now = -addr & ~PAGE_MASK;
1644                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1645                 if (rc != X86EMUL_CONTINUE)
1646                         return rc;
1647                 addr += now;
1648                 val += now;
1649                 bytes -= now;
1650         }
1651         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1652 }
1653 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1654
1655 static int emulator_cmpxchg_emulated(unsigned long addr,
1656                                      const void *old,
1657                                      const void *new,
1658                                      unsigned int bytes,
1659                                      struct kvm_vcpu *vcpu)
1660 {
1661         static int reported;
1662
1663         if (!reported) {
1664                 reported = 1;
1665                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1666         }
1667         return emulator_write_emulated(addr, new, bytes, vcpu);
1668 }
1669
1670 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1671 {
1672         return kvm_x86_ops->get_segment_base(vcpu, seg);
1673 }
1674
1675 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1676 {
1677         return X86EMUL_CONTINUE;
1678 }
1679
1680 int emulate_clts(struct kvm_vcpu *vcpu)
1681 {
1682         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1683         return X86EMUL_CONTINUE;
1684 }
1685
1686 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1687 {
1688         struct kvm_vcpu *vcpu = ctxt->vcpu;
1689
1690         switch (dr) {
1691         case 0 ... 3:
1692                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1693                 return X86EMUL_CONTINUE;
1694         default:
1695                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1696                 return X86EMUL_UNHANDLEABLE;
1697         }
1698 }
1699
1700 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1701 {
1702         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1703         int exception;
1704
1705         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1706         if (exception) {
1707                 /* FIXME: better handling */
1708                 return X86EMUL_UNHANDLEABLE;
1709         }
1710         return X86EMUL_CONTINUE;
1711 }
1712
1713 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1714 {
1715         static int reported;
1716         u8 opcodes[4];
1717         unsigned long rip = vcpu->rip;
1718         unsigned long rip_linear;
1719
1720         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1721
1722         if (reported)
1723                 return;
1724
1725         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1726
1727         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1728                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1729         reported = 1;
1730 }
1731 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1732
1733 struct x86_emulate_ops emulate_ops = {
1734         .read_std            = emulator_read_std,
1735         .read_emulated       = emulator_read_emulated,
1736         .write_emulated      = emulator_write_emulated,
1737         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1738 };
1739
1740 int emulate_instruction(struct kvm_vcpu *vcpu,
1741                         struct kvm_run *run,
1742                         unsigned long cr2,
1743                         u16 error_code,
1744                         int no_decode)
1745 {
1746         int r;
1747
1748         vcpu->mmio_fault_cr2 = cr2;
1749         kvm_x86_ops->cache_regs(vcpu);
1750
1751         vcpu->mmio_is_write = 0;
1752         vcpu->pio.string = 0;
1753
1754         if (!no_decode) {
1755                 int cs_db, cs_l;
1756                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1757
1758                 vcpu->emulate_ctxt.vcpu = vcpu;
1759                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1760                 vcpu->emulate_ctxt.mode =
1761                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1762                         ? X86EMUL_MODE_REAL : cs_l
1763                         ? X86EMUL_MODE_PROT64 : cs_db
1764                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1765
1766                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1767                         vcpu->emulate_ctxt.cs_base = 0;
1768                         vcpu->emulate_ctxt.ds_base = 0;
1769                         vcpu->emulate_ctxt.es_base = 0;
1770                         vcpu->emulate_ctxt.ss_base = 0;
1771                 } else {
1772                         vcpu->emulate_ctxt.cs_base =
1773                                         get_segment_base(vcpu, VCPU_SREG_CS);
1774                         vcpu->emulate_ctxt.ds_base =
1775                                         get_segment_base(vcpu, VCPU_SREG_DS);
1776                         vcpu->emulate_ctxt.es_base =
1777                                         get_segment_base(vcpu, VCPU_SREG_ES);
1778                         vcpu->emulate_ctxt.ss_base =
1779                                         get_segment_base(vcpu, VCPU_SREG_SS);
1780                 }
1781
1782                 vcpu->emulate_ctxt.gs_base =
1783                                         get_segment_base(vcpu, VCPU_SREG_GS);
1784                 vcpu->emulate_ctxt.fs_base =
1785                                         get_segment_base(vcpu, VCPU_SREG_FS);
1786
1787                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1788                 ++vcpu->stat.insn_emulation;
1789                 if (r)  {
1790                         ++vcpu->stat.insn_emulation_fail;
1791                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1792                                 return EMULATE_DONE;
1793                         return EMULATE_FAIL;
1794                 }
1795         }
1796
1797         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1798
1799         if (vcpu->pio.string)
1800                 return EMULATE_DO_MMIO;
1801
1802         if ((r || vcpu->mmio_is_write) && run) {
1803                 run->exit_reason = KVM_EXIT_MMIO;
1804                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1805                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1806                 run->mmio.len = vcpu->mmio_size;
1807                 run->mmio.is_write = vcpu->mmio_is_write;
1808         }
1809
1810         if (r) {
1811                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1812                         return EMULATE_DONE;
1813                 if (!vcpu->mmio_needed) {
1814                         kvm_report_emulation_failure(vcpu, "mmio");
1815                         return EMULATE_FAIL;
1816                 }
1817                 return EMULATE_DO_MMIO;
1818         }
1819
1820         kvm_x86_ops->decache_regs(vcpu);
1821         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1822
1823         if (vcpu->mmio_is_write) {
1824                 vcpu->mmio_needed = 0;
1825                 return EMULATE_DO_MMIO;
1826         }
1827
1828         return EMULATE_DONE;
1829 }
1830 EXPORT_SYMBOL_GPL(emulate_instruction);
1831
1832 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1833 {
1834         int i;
1835
1836         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1837                 if (vcpu->pio.guest_pages[i]) {
1838                         kvm_release_page_dirty(vcpu->pio.guest_pages[i]);
1839                         vcpu->pio.guest_pages[i] = NULL;
1840                 }
1841 }
1842
1843 static int pio_copy_data(struct kvm_vcpu *vcpu)
1844 {
1845         void *p = vcpu->pio_data;
1846         void *q;
1847         unsigned bytes;
1848         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1849
1850         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1851                  PAGE_KERNEL);
1852         if (!q) {
1853                 free_pio_guest_pages(vcpu);
1854                 return -ENOMEM;
1855         }
1856         q += vcpu->pio.guest_page_offset;
1857         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1858         if (vcpu->pio.in)
1859                 memcpy(q, p, bytes);
1860         else
1861                 memcpy(p, q, bytes);
1862         q -= vcpu->pio.guest_page_offset;
1863         vunmap(q);
1864         free_pio_guest_pages(vcpu);
1865         return 0;
1866 }
1867
1868 int complete_pio(struct kvm_vcpu *vcpu)
1869 {
1870         struct kvm_pio_request *io = &vcpu->pio;
1871         long delta;
1872         int r;
1873
1874         kvm_x86_ops->cache_regs(vcpu);
1875
1876         if (!io->string) {
1877                 if (io->in)
1878                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1879                                io->size);
1880         } else {
1881                 if (io->in) {
1882                         r = pio_copy_data(vcpu);
1883                         if (r) {
1884                                 kvm_x86_ops->cache_regs(vcpu);
1885                                 return r;
1886                         }
1887                 }
1888
1889                 delta = 1;
1890                 if (io->rep) {
1891                         delta *= io->cur_count;
1892                         /*
1893                          * The size of the register should really depend on
1894                          * current address size.
1895                          */
1896                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1897                 }
1898                 if (io->down)
1899                         delta = -delta;
1900                 delta *= io->size;
1901                 if (io->in)
1902                         vcpu->regs[VCPU_REGS_RDI] += delta;
1903                 else
1904                         vcpu->regs[VCPU_REGS_RSI] += delta;
1905         }
1906
1907         kvm_x86_ops->decache_regs(vcpu);
1908
1909         io->count -= io->cur_count;
1910         io->cur_count = 0;
1911
1912         return 0;
1913 }
1914
1915 static void kernel_pio(struct kvm_io_device *pio_dev,
1916                        struct kvm_vcpu *vcpu,
1917                        void *pd)
1918 {
1919         /* TODO: String I/O for in kernel device */
1920
1921         mutex_lock(&vcpu->kvm->lock);
1922         if (vcpu->pio.in)
1923                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1924                                   vcpu->pio.size,
1925                                   pd);
1926         else
1927                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1928                                    vcpu->pio.size,
1929                                    pd);
1930         mutex_unlock(&vcpu->kvm->lock);
1931 }
1932
1933 static void pio_string_write(struct kvm_io_device *pio_dev,
1934                              struct kvm_vcpu *vcpu)
1935 {
1936         struct kvm_pio_request *io = &vcpu->pio;
1937         void *pd = vcpu->pio_data;
1938         int i;
1939
1940         mutex_lock(&vcpu->kvm->lock);
1941         for (i = 0; i < io->cur_count; i++) {
1942                 kvm_iodevice_write(pio_dev, io->port,
1943                                    io->size,
1944                                    pd);
1945                 pd += io->size;
1946         }
1947         mutex_unlock(&vcpu->kvm->lock);
1948 }
1949
1950 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1951                                                gpa_t addr)
1952 {
1953         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1954 }
1955
1956 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1957                   int size, unsigned port)
1958 {
1959         struct kvm_io_device *pio_dev;
1960
1961         vcpu->run->exit_reason = KVM_EXIT_IO;
1962         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1963         vcpu->run->io.size = vcpu->pio.size = size;
1964         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1965         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1966         vcpu->run->io.port = vcpu->pio.port = port;
1967         vcpu->pio.in = in;
1968         vcpu->pio.string = 0;
1969         vcpu->pio.down = 0;
1970         vcpu->pio.guest_page_offset = 0;
1971         vcpu->pio.rep = 0;
1972
1973         kvm_x86_ops->cache_regs(vcpu);
1974         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1975         kvm_x86_ops->decache_regs(vcpu);
1976
1977         kvm_x86_ops->skip_emulated_instruction(vcpu);
1978
1979         pio_dev = vcpu_find_pio_dev(vcpu, port);
1980         if (pio_dev) {
1981                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1982                 complete_pio(vcpu);
1983                 return 1;
1984         }
1985         return 0;
1986 }
1987 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1988
1989 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1990                   int size, unsigned long count, int down,
1991                   gva_t address, int rep, unsigned port)
1992 {
1993         unsigned now, in_page;
1994         int i, ret = 0;
1995         int nr_pages = 1;
1996         struct page *page;
1997         struct kvm_io_device *pio_dev;
1998
1999         vcpu->run->exit_reason = KVM_EXIT_IO;
2000         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2001         vcpu->run->io.size = vcpu->pio.size = size;
2002         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2003         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2004         vcpu->run->io.port = vcpu->pio.port = port;
2005         vcpu->pio.in = in;
2006         vcpu->pio.string = 1;
2007         vcpu->pio.down = down;
2008         vcpu->pio.guest_page_offset = offset_in_page(address);
2009         vcpu->pio.rep = rep;
2010
2011         if (!count) {
2012                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2013                 return 1;
2014         }
2015
2016         if (!down)
2017                 in_page = PAGE_SIZE - offset_in_page(address);
2018         else
2019                 in_page = offset_in_page(address) + size;
2020         now = min(count, (unsigned long)in_page / size);
2021         if (!now) {
2022                 /*
2023                  * String I/O straddles page boundary.  Pin two guest pages
2024                  * so that we satisfy atomicity constraints.  Do just one
2025                  * transaction to avoid complexity.
2026                  */
2027                 nr_pages = 2;
2028                 now = 1;
2029         }
2030         if (down) {
2031                 /*
2032                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2033                  */
2034                 pr_unimpl(vcpu, "guest string pio down\n");
2035                 inject_gp(vcpu);
2036                 return 1;
2037         }
2038         vcpu->run->io.count = now;
2039         vcpu->pio.cur_count = now;
2040
2041         if (vcpu->pio.cur_count == vcpu->pio.count)
2042                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2043
2044         for (i = 0; i < nr_pages; ++i) {
2045                 mutex_lock(&vcpu->kvm->lock);
2046                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2047                 vcpu->pio.guest_pages[i] = page;
2048                 mutex_unlock(&vcpu->kvm->lock);
2049                 if (!page) {
2050                         inject_gp(vcpu);
2051                         free_pio_guest_pages(vcpu);
2052                         return 1;
2053                 }
2054         }
2055
2056         pio_dev = vcpu_find_pio_dev(vcpu, port);
2057         if (!vcpu->pio.in) {
2058                 /* string PIO write */
2059                 ret = pio_copy_data(vcpu);
2060                 if (ret >= 0 && pio_dev) {
2061                         pio_string_write(pio_dev, vcpu);
2062                         complete_pio(vcpu);
2063                         if (vcpu->pio.count == 0)
2064                                 ret = 1;
2065                 }
2066         } else if (pio_dev)
2067                 pr_unimpl(vcpu, "no string pio read support yet, "
2068                        "port %x size %d count %ld\n",
2069                         port, size, count);
2070
2071         return ret;
2072 }
2073 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2074
2075 int kvm_arch_init(void *opaque)
2076 {
2077         int r;
2078         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2079
2080         r = kvm_mmu_module_init();
2081         if (r)
2082                 goto out_fail;
2083
2084         kvm_init_msr_list();
2085
2086         if (kvm_x86_ops) {
2087                 printk(KERN_ERR "kvm: already loaded the other module\n");
2088                 r = -EEXIST;
2089                 goto out;
2090         }
2091
2092         if (!ops->cpu_has_kvm_support()) {
2093                 printk(KERN_ERR "kvm: no hardware support\n");
2094                 r = -EOPNOTSUPP;
2095                 goto out;
2096         }
2097         if (ops->disabled_by_bios()) {
2098                 printk(KERN_ERR "kvm: disabled by bios\n");
2099                 r = -EOPNOTSUPP;
2100                 goto out;
2101         }
2102
2103         kvm_x86_ops = ops;
2104         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2105         return 0;
2106
2107 out:
2108         kvm_mmu_module_exit();
2109 out_fail:
2110         return r;
2111 }
2112
2113 void kvm_arch_exit(void)
2114 {
2115         kvm_x86_ops = NULL;
2116         kvm_mmu_module_exit();
2117 }
2118
2119 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2120 {
2121         ++vcpu->stat.halt_exits;
2122         if (irqchip_in_kernel(vcpu->kvm)) {
2123                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
2124                 kvm_vcpu_block(vcpu);
2125                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
2126                         return -EINTR;
2127                 return 1;
2128         } else {
2129                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2130                 return 0;
2131         }
2132 }
2133 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2134
2135 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2136 {
2137         unsigned long nr, a0, a1, a2, a3, ret;
2138
2139         kvm_x86_ops->cache_regs(vcpu);
2140
2141         nr = vcpu->regs[VCPU_REGS_RAX];
2142         a0 = vcpu->regs[VCPU_REGS_RBX];
2143         a1 = vcpu->regs[VCPU_REGS_RCX];
2144         a2 = vcpu->regs[VCPU_REGS_RDX];
2145         a3 = vcpu->regs[VCPU_REGS_RSI];
2146
2147         if (!is_long_mode(vcpu)) {
2148                 nr &= 0xFFFFFFFF;
2149                 a0 &= 0xFFFFFFFF;
2150                 a1 &= 0xFFFFFFFF;
2151                 a2 &= 0xFFFFFFFF;
2152                 a3 &= 0xFFFFFFFF;
2153         }
2154
2155         switch (nr) {
2156         default:
2157                 ret = -KVM_ENOSYS;
2158                 break;
2159         }
2160         vcpu->regs[VCPU_REGS_RAX] = ret;
2161         kvm_x86_ops->decache_regs(vcpu);
2162         return 0;
2163 }
2164 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2165
2166 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2167 {
2168         char instruction[3];
2169         int ret = 0;
2170
2171         mutex_lock(&vcpu->kvm->lock);
2172
2173         /*
2174          * Blow out the MMU to ensure that no other VCPU has an active mapping
2175          * to ensure that the updated hypercall appears atomically across all
2176          * VCPUs.
2177          */
2178         kvm_mmu_zap_all(vcpu->kvm);
2179
2180         kvm_x86_ops->cache_regs(vcpu);
2181         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2182         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
2183             != X86EMUL_CONTINUE)
2184                 ret = -EFAULT;
2185
2186         mutex_unlock(&vcpu->kvm->lock);
2187
2188         return ret;
2189 }
2190
2191 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2192 {
2193         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2194 }
2195
2196 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2197 {
2198         struct descriptor_table dt = { limit, base };
2199
2200         kvm_x86_ops->set_gdt(vcpu, &dt);
2201 }
2202
2203 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2204 {
2205         struct descriptor_table dt = { limit, base };
2206
2207         kvm_x86_ops->set_idt(vcpu, &dt);
2208 }
2209
2210 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2211                    unsigned long *rflags)
2212 {
2213         lmsw(vcpu, msw);
2214         *rflags = kvm_x86_ops->get_rflags(vcpu);
2215 }
2216
2217 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2218 {
2219         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2220         switch (cr) {
2221         case 0:
2222                 return vcpu->cr0;
2223         case 2:
2224                 return vcpu->cr2;
2225         case 3:
2226                 return vcpu->cr3;
2227         case 4:
2228                 return vcpu->cr4;
2229         default:
2230                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2231                 return 0;
2232         }
2233 }
2234
2235 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2236                      unsigned long *rflags)
2237 {
2238         switch (cr) {
2239         case 0:
2240                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
2241                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2242                 break;
2243         case 2:
2244                 vcpu->cr2 = val;
2245                 break;
2246         case 3:
2247                 set_cr3(vcpu, val);
2248                 break;
2249         case 4:
2250                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
2251                 break;
2252         default:
2253                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2254         }
2255 }
2256
2257 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2258 {
2259         struct kvm_cpuid_entry2 *e = &vcpu->cpuid_entries[i];
2260         int j, nent = vcpu->cpuid_nent;
2261
2262         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2263         /* when no next entry is found, the current entry[i] is reselected */
2264         for (j = i + 1; j == i; j = (j + 1) % nent) {
2265                 struct kvm_cpuid_entry2 *ej = &vcpu->cpuid_entries[j];
2266                 if (ej->function == e->function) {
2267                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2268                         return j;
2269                 }
2270         }
2271         return 0; /* silence gcc, even though control never reaches here */
2272 }
2273
2274 /* find an entry with matching function, matching index (if needed), and that
2275  * should be read next (if it's stateful) */
2276 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2277         u32 function, u32 index)
2278 {
2279         if (e->function != function)
2280                 return 0;
2281         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2282                 return 0;
2283         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2284                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2285                 return 0;
2286         return 1;
2287 }
2288
2289 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2290 {
2291         int i;
2292         u32 function, index;
2293         struct kvm_cpuid_entry2 *e, *best;
2294
2295         kvm_x86_ops->cache_regs(vcpu);
2296         function = vcpu->regs[VCPU_REGS_RAX];
2297         index = vcpu->regs[VCPU_REGS_RCX];
2298         vcpu->regs[VCPU_REGS_RAX] = 0;
2299         vcpu->regs[VCPU_REGS_RBX] = 0;
2300         vcpu->regs[VCPU_REGS_RCX] = 0;
2301         vcpu->regs[VCPU_REGS_RDX] = 0;
2302         best = NULL;
2303         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2304                 e = &vcpu->cpuid_entries[i];
2305                 if (is_matching_cpuid_entry(e, function, index)) {
2306                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2307                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2308                         best = e;
2309                         break;
2310                 }
2311                 /*
2312                  * Both basic or both extended?
2313                  */
2314                 if (((e->function ^ function) & 0x80000000) == 0)
2315                         if (!best || e->function > best->function)
2316                                 best = e;
2317         }
2318         if (best) {
2319                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
2320                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
2321                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
2322                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
2323         }
2324         kvm_x86_ops->decache_regs(vcpu);
2325         kvm_x86_ops->skip_emulated_instruction(vcpu);
2326 }
2327 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2328
2329 /*
2330  * Check if userspace requested an interrupt window, and that the
2331  * interrupt window is open.
2332  *
2333  * No need to exit to userspace if we already have an interrupt queued.
2334  */
2335 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2336                                           struct kvm_run *kvm_run)
2337 {
2338         return (!vcpu->irq_summary &&
2339                 kvm_run->request_interrupt_window &&
2340                 vcpu->interrupt_window_open &&
2341                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2342 }
2343
2344 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2345                               struct kvm_run *kvm_run)
2346 {
2347         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2348         kvm_run->cr8 = get_cr8(vcpu);
2349         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2350         if (irqchip_in_kernel(vcpu->kvm))
2351                 kvm_run->ready_for_interrupt_injection = 1;
2352         else
2353                 kvm_run->ready_for_interrupt_injection =
2354                                         (vcpu->interrupt_window_open &&
2355                                          vcpu->irq_summary == 0);
2356 }
2357
2358 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2359 {
2360         int r;
2361
2362         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2363                 pr_debug("vcpu %d received sipi with vector # %x\n",
2364                        vcpu->vcpu_id, vcpu->sipi_vector);
2365                 kvm_lapic_reset(vcpu);
2366                 r = kvm_x86_ops->vcpu_reset(vcpu);
2367                 if (r)
2368                         return r;
2369                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2370         }
2371
2372 preempted:
2373         if (vcpu->guest_debug.enabled)
2374                 kvm_x86_ops->guest_debug_pre(vcpu);
2375
2376 again:
2377         r = kvm_mmu_reload(vcpu);
2378         if (unlikely(r))
2379                 goto out;
2380
2381         kvm_inject_pending_timer_irqs(vcpu);
2382
2383         preempt_disable();
2384
2385         kvm_x86_ops->prepare_guest_switch(vcpu);
2386         kvm_load_guest_fpu(vcpu);
2387
2388         local_irq_disable();
2389
2390         if (signal_pending(current)) {
2391                 local_irq_enable();
2392                 preempt_enable();
2393                 r = -EINTR;
2394                 kvm_run->exit_reason = KVM_EXIT_INTR;
2395                 ++vcpu->stat.signal_exits;
2396                 goto out;
2397         }
2398
2399         if (vcpu->exception.pending)
2400                 __queue_exception(vcpu);
2401         else if (irqchip_in_kernel(vcpu->kvm))
2402                 kvm_x86_ops->inject_pending_irq(vcpu);
2403         else
2404                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2405
2406         vcpu->guest_mode = 1;
2407         kvm_guest_enter();
2408
2409         if (vcpu->requests)
2410                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2411                         kvm_x86_ops->tlb_flush(vcpu);
2412
2413         kvm_x86_ops->run(vcpu, kvm_run);
2414
2415         vcpu->guest_mode = 0;
2416         local_irq_enable();
2417
2418         ++vcpu->stat.exits;
2419
2420         /*
2421          * We must have an instruction between local_irq_enable() and
2422          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2423          * the interrupt shadow.  The stat.exits increment will do nicely.
2424          * But we need to prevent reordering, hence this barrier():
2425          */
2426         barrier();
2427
2428         kvm_guest_exit();
2429
2430         preempt_enable();
2431
2432         /*
2433          * Profile KVM exit RIPs:
2434          */
2435         if (unlikely(prof_on == KVM_PROFILING)) {
2436                 kvm_x86_ops->cache_regs(vcpu);
2437                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2438         }
2439
2440         if (vcpu->exception.pending && kvm_x86_ops->exception_injected(vcpu))
2441                 vcpu->exception.pending = false;
2442
2443         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2444
2445         if (r > 0) {
2446                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2447                         r = -EINTR;
2448                         kvm_run->exit_reason = KVM_EXIT_INTR;
2449                         ++vcpu->stat.request_irq_exits;
2450                         goto out;
2451                 }
2452                 if (!need_resched())
2453                         goto again;
2454         }
2455
2456 out:
2457         if (r > 0) {
2458                 kvm_resched(vcpu);
2459                 goto preempted;
2460         }
2461
2462         post_kvm_run_save(vcpu, kvm_run);
2463
2464         return r;
2465 }
2466
2467 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2468 {
2469         int r;
2470         sigset_t sigsaved;
2471
2472         vcpu_load(vcpu);
2473
2474         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2475                 kvm_vcpu_block(vcpu);
2476                 vcpu_put(vcpu);
2477                 return -EAGAIN;
2478         }
2479
2480         if (vcpu->sigset_active)
2481                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2482
2483         /* re-sync apic's tpr */
2484         if (!irqchip_in_kernel(vcpu->kvm))
2485                 set_cr8(vcpu, kvm_run->cr8);
2486
2487         if (vcpu->pio.cur_count) {
2488                 r = complete_pio(vcpu);
2489                 if (r)
2490                         goto out;
2491         }
2492 #if CONFIG_HAS_IOMEM
2493         if (vcpu->mmio_needed) {
2494                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2495                 vcpu->mmio_read_completed = 1;
2496                 vcpu->mmio_needed = 0;
2497                 r = emulate_instruction(vcpu, kvm_run,
2498                                         vcpu->mmio_fault_cr2, 0, 1);
2499                 if (r == EMULATE_DO_MMIO) {
2500                         /*
2501                          * Read-modify-write.  Back to userspace.
2502                          */
2503                         r = 0;
2504                         goto out;
2505                 }
2506         }
2507 #endif
2508         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2509                 kvm_x86_ops->cache_regs(vcpu);
2510                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2511                 kvm_x86_ops->decache_regs(vcpu);
2512         }
2513
2514         r = __vcpu_run(vcpu, kvm_run);
2515
2516 out:
2517         if (vcpu->sigset_active)
2518                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2519
2520         vcpu_put(vcpu);
2521         return r;
2522 }
2523
2524 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2525 {
2526         vcpu_load(vcpu);
2527
2528         kvm_x86_ops->cache_regs(vcpu);
2529
2530         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2531         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2532         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2533         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2534         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2535         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2536         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2537         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2538 #ifdef CONFIG_X86_64
2539         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2540         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2541         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2542         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2543         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2544         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2545         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2546         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2547 #endif
2548
2549         regs->rip = vcpu->rip;
2550         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2551
2552         /*
2553          * Don't leak debug flags in case they were set for guest debugging
2554          */
2555         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2556                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2557
2558         vcpu_put(vcpu);
2559
2560         return 0;
2561 }
2562
2563 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2564 {
2565         vcpu_load(vcpu);
2566
2567         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2568         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2569         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2570         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2571         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2572         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2573         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2574         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2575 #ifdef CONFIG_X86_64
2576         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2577         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2578         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2579         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2580         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2581         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2582         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2583         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2584 #endif
2585
2586         vcpu->rip = regs->rip;
2587         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2588
2589         kvm_x86_ops->decache_regs(vcpu);
2590
2591         vcpu_put(vcpu);
2592
2593         return 0;
2594 }
2595
2596 static void get_segment(struct kvm_vcpu *vcpu,
2597                         struct kvm_segment *var, int seg)
2598 {
2599         return kvm_x86_ops->get_segment(vcpu, var, seg);
2600 }
2601
2602 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2603 {
2604         struct kvm_segment cs;
2605
2606         get_segment(vcpu, &cs, VCPU_SREG_CS);
2607         *db = cs.db;
2608         *l = cs.l;
2609 }
2610 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2611
2612 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2613                                   struct kvm_sregs *sregs)
2614 {
2615         struct descriptor_table dt;
2616         int pending_vec;
2617
2618         vcpu_load(vcpu);
2619
2620         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2621         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2622         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2623         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2624         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2625         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2626
2627         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2628         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2629
2630         kvm_x86_ops->get_idt(vcpu, &dt);
2631         sregs->idt.limit = dt.limit;
2632         sregs->idt.base = dt.base;
2633         kvm_x86_ops->get_gdt(vcpu, &dt);
2634         sregs->gdt.limit = dt.limit;
2635         sregs->gdt.base = dt.base;
2636
2637         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2638         sregs->cr0 = vcpu->cr0;
2639         sregs->cr2 = vcpu->cr2;
2640         sregs->cr3 = vcpu->cr3;
2641         sregs->cr4 = vcpu->cr4;
2642         sregs->cr8 = get_cr8(vcpu);
2643         sregs->efer = vcpu->shadow_efer;
2644         sregs->apic_base = kvm_get_apic_base(vcpu);
2645
2646         if (irqchip_in_kernel(vcpu->kvm)) {
2647                 memset(sregs->interrupt_bitmap, 0,
2648                        sizeof sregs->interrupt_bitmap);
2649                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2650                 if (pending_vec >= 0)
2651                         set_bit(pending_vec,
2652                                 (unsigned long *)sregs->interrupt_bitmap);
2653         } else
2654                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2655                        sizeof sregs->interrupt_bitmap);
2656
2657         vcpu_put(vcpu);
2658
2659         return 0;
2660 }
2661
2662 static void set_segment(struct kvm_vcpu *vcpu,
2663                         struct kvm_segment *var, int seg)
2664 {
2665         return kvm_x86_ops->set_segment(vcpu, var, seg);
2666 }
2667
2668 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2669                                   struct kvm_sregs *sregs)
2670 {
2671         int mmu_reset_needed = 0;
2672         int i, pending_vec, max_bits;
2673         struct descriptor_table dt;
2674
2675         vcpu_load(vcpu);
2676
2677         dt.limit = sregs->idt.limit;
2678         dt.base = sregs->idt.base;
2679         kvm_x86_ops->set_idt(vcpu, &dt);
2680         dt.limit = sregs->gdt.limit;
2681         dt.base = sregs->gdt.base;
2682         kvm_x86_ops->set_gdt(vcpu, &dt);
2683
2684         vcpu->cr2 = sregs->cr2;
2685         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2686         vcpu->cr3 = sregs->cr3;
2687
2688         set_cr8(vcpu, sregs->cr8);
2689
2690         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2691 #ifdef CONFIG_X86_64
2692         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2693 #endif
2694         kvm_set_apic_base(vcpu, sregs->apic_base);
2695
2696         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2697
2698         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2699         vcpu->cr0 = sregs->cr0;
2700         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2701
2702         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2703         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2704         if (!is_long_mode(vcpu) && is_pae(vcpu))
2705                 load_pdptrs(vcpu, vcpu->cr3);
2706
2707         if (mmu_reset_needed)
2708                 kvm_mmu_reset_context(vcpu);
2709
2710         if (!irqchip_in_kernel(vcpu->kvm)) {
2711                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2712                        sizeof vcpu->irq_pending);
2713                 vcpu->irq_summary = 0;
2714                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2715                         if (vcpu->irq_pending[i])
2716                                 __set_bit(i, &vcpu->irq_summary);
2717         } else {
2718                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2719                 pending_vec = find_first_bit(
2720                         (const unsigned long *)sregs->interrupt_bitmap,
2721                         max_bits);
2722                 /* Only pending external irq is handled here */
2723                 if (pending_vec < max_bits) {
2724                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2725                         pr_debug("Set back pending irq %d\n",
2726                                  pending_vec);
2727                 }
2728         }
2729
2730         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2731         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2732         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2733         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2734         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2735         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2736
2737         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2738         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2739
2740         vcpu_put(vcpu);
2741
2742         return 0;
2743 }
2744
2745 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2746                                     struct kvm_debug_guest *dbg)
2747 {
2748         int r;
2749
2750         vcpu_load(vcpu);
2751
2752         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2753
2754         vcpu_put(vcpu);
2755
2756         return r;
2757 }
2758
2759 /*
2760  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2761  * we have asm/x86/processor.h
2762  */
2763 struct fxsave {
2764         u16     cwd;
2765         u16     swd;
2766         u16     twd;
2767         u16     fop;
2768         u64     rip;
2769         u64     rdp;
2770         u32     mxcsr;
2771         u32     mxcsr_mask;
2772         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2773 #ifdef CONFIG_X86_64
2774         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2775 #else
2776         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2777 #endif
2778 };
2779
2780 /*
2781  * Translate a guest virtual address to a guest physical address.
2782  */
2783 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2784                                     struct kvm_translation *tr)
2785 {
2786         unsigned long vaddr = tr->linear_address;
2787         gpa_t gpa;
2788
2789         vcpu_load(vcpu);
2790         mutex_lock(&vcpu->kvm->lock);
2791         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2792         tr->physical_address = gpa;
2793         tr->valid = gpa != UNMAPPED_GVA;
2794         tr->writeable = 1;
2795         tr->usermode = 0;
2796         mutex_unlock(&vcpu->kvm->lock);
2797         vcpu_put(vcpu);
2798
2799         return 0;
2800 }
2801
2802 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2803 {
2804         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2805
2806         vcpu_load(vcpu);
2807
2808         memcpy(fpu->fpr, fxsave->st_space, 128);
2809         fpu->fcw = fxsave->cwd;
2810         fpu->fsw = fxsave->swd;
2811         fpu->ftwx = fxsave->twd;
2812         fpu->last_opcode = fxsave->fop;
2813         fpu->last_ip = fxsave->rip;
2814         fpu->last_dp = fxsave->rdp;
2815         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2816
2817         vcpu_put(vcpu);
2818
2819         return 0;
2820 }
2821
2822 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2823 {
2824         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2825
2826         vcpu_load(vcpu);
2827
2828         memcpy(fxsave->st_space, fpu->fpr, 128);
2829         fxsave->cwd = fpu->fcw;
2830         fxsave->swd = fpu->fsw;
2831         fxsave->twd = fpu->ftwx;
2832         fxsave->fop = fpu->last_opcode;
2833         fxsave->rip = fpu->last_ip;
2834         fxsave->rdp = fpu->last_dp;
2835         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2836
2837         vcpu_put(vcpu);
2838
2839         return 0;
2840 }
2841
2842 void fx_init(struct kvm_vcpu *vcpu)
2843 {
2844         unsigned after_mxcsr_mask;
2845
2846         /* Initialize guest FPU by resetting ours and saving into guest's */
2847         preempt_disable();
2848         fx_save(&vcpu->host_fx_image);
2849         fpu_init();
2850         fx_save(&vcpu->guest_fx_image);
2851         fx_restore(&vcpu->host_fx_image);
2852         preempt_enable();
2853
2854         vcpu->cr0 |= X86_CR0_ET;
2855         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2856         vcpu->guest_fx_image.mxcsr = 0x1f80;
2857         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2858                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2859 }
2860 EXPORT_SYMBOL_GPL(fx_init);
2861
2862 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2863 {
2864         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2865                 return;
2866
2867         vcpu->guest_fpu_loaded = 1;
2868         fx_save(&vcpu->host_fx_image);
2869         fx_restore(&vcpu->guest_fx_image);
2870 }
2871 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2872
2873 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2874 {
2875         if (!vcpu->guest_fpu_loaded)
2876                 return;
2877
2878         vcpu->guest_fpu_loaded = 0;
2879         fx_save(&vcpu->guest_fx_image);
2880         fx_restore(&vcpu->host_fx_image);
2881         ++vcpu->stat.fpu_reload;
2882 }
2883 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
2884
2885 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2886 {
2887         kvm_x86_ops->vcpu_free(vcpu);
2888 }
2889
2890 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2891                                                 unsigned int id)
2892 {
2893         return kvm_x86_ops->vcpu_create(kvm, id);
2894 }
2895
2896 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2897 {
2898         int r;
2899
2900         /* We do fxsave: this must be aligned. */
2901         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2902
2903         vcpu_load(vcpu);
2904         r = kvm_arch_vcpu_reset(vcpu);
2905         if (r == 0)
2906                 r = kvm_mmu_setup(vcpu);
2907         vcpu_put(vcpu);
2908         if (r < 0)
2909                 goto free_vcpu;
2910
2911         return 0;
2912 free_vcpu:
2913         kvm_x86_ops->vcpu_free(vcpu);
2914         return r;
2915 }
2916
2917 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2918 {
2919         vcpu_load(vcpu);
2920         kvm_mmu_unload(vcpu);
2921         vcpu_put(vcpu);
2922
2923         kvm_x86_ops->vcpu_free(vcpu);
2924 }
2925
2926 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2927 {
2928         return kvm_x86_ops->vcpu_reset(vcpu);
2929 }
2930
2931 void kvm_arch_hardware_enable(void *garbage)
2932 {
2933         kvm_x86_ops->hardware_enable(garbage);
2934 }
2935
2936 void kvm_arch_hardware_disable(void *garbage)
2937 {
2938         kvm_x86_ops->hardware_disable(garbage);
2939 }
2940
2941 int kvm_arch_hardware_setup(void)
2942 {
2943         return kvm_x86_ops->hardware_setup();
2944 }
2945
2946 void kvm_arch_hardware_unsetup(void)
2947 {
2948         kvm_x86_ops->hardware_unsetup();
2949 }
2950
2951 void kvm_arch_check_processor_compat(void *rtn)
2952 {
2953         kvm_x86_ops->check_processor_compatibility(rtn);
2954 }
2955
2956 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2957 {
2958         struct page *page;
2959         struct kvm *kvm;
2960         int r;
2961
2962         BUG_ON(vcpu->kvm == NULL);
2963         kvm = vcpu->kvm;
2964
2965         vcpu->mmu.root_hpa = INVALID_PAGE;
2966         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2967                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2968         else
2969                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2970
2971         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2972         if (!page) {
2973                 r = -ENOMEM;
2974                 goto fail;
2975         }
2976         vcpu->pio_data = page_address(page);
2977
2978         r = kvm_mmu_create(vcpu);
2979         if (r < 0)
2980                 goto fail_free_pio_data;
2981
2982         if (irqchip_in_kernel(kvm)) {
2983                 r = kvm_create_lapic(vcpu);
2984                 if (r < 0)
2985                         goto fail_mmu_destroy;
2986         }
2987
2988         return 0;
2989
2990 fail_mmu_destroy:
2991         kvm_mmu_destroy(vcpu);
2992 fail_free_pio_data:
2993         free_page((unsigned long)vcpu->pio_data);
2994 fail:
2995         return r;
2996 }
2997
2998 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2999 {
3000         kvm_free_lapic(vcpu);
3001         kvm_mmu_destroy(vcpu);
3002         free_page((unsigned long)vcpu->pio_data);
3003 }
3004
3005 struct  kvm *kvm_arch_create_vm(void)
3006 {
3007         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3008
3009         if (!kvm)
3010                 return ERR_PTR(-ENOMEM);
3011
3012         INIT_LIST_HEAD(&kvm->active_mmu_pages);
3013
3014         return kvm;
3015 }
3016
3017 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3018 {
3019         vcpu_load(vcpu);
3020         kvm_mmu_unload(vcpu);
3021         vcpu_put(vcpu);
3022 }
3023
3024 static void kvm_free_vcpus(struct kvm *kvm)
3025 {
3026         unsigned int i;
3027
3028         /*
3029          * Unpin any mmu pages first.
3030          */
3031         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3032                 if (kvm->vcpus[i])
3033                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3034         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3035                 if (kvm->vcpus[i]) {
3036                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3037                         kvm->vcpus[i] = NULL;
3038                 }
3039         }
3040
3041 }
3042
3043 void kvm_arch_destroy_vm(struct kvm *kvm)
3044 {
3045         kfree(kvm->vpic);
3046         kfree(kvm->vioapic);
3047         kvm_free_vcpus(kvm);
3048         kvm_free_physmem(kvm);
3049         kfree(kvm);
3050 }
3051
3052 int kvm_arch_set_memory_region(struct kvm *kvm,
3053                                 struct kvm_userspace_memory_region *mem,
3054                                 struct kvm_memory_slot old,
3055                                 int user_alloc)
3056 {
3057         int npages = mem->memory_size >> PAGE_SHIFT;
3058         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3059
3060         /*To keep backward compatibility with older userspace,
3061          *x86 needs to hanlde !user_alloc case.
3062          */
3063         if (!user_alloc) {
3064                 if (npages && !old.rmap) {
3065                         down_write(&current->mm->mmap_sem);
3066                         memslot->userspace_addr = do_mmap(NULL, 0,
3067                                                      npages * PAGE_SIZE,
3068                                                      PROT_READ | PROT_WRITE,
3069                                                      MAP_SHARED | MAP_ANONYMOUS,
3070                                                      0);
3071                         up_write(&current->mm->mmap_sem);
3072
3073                         if (IS_ERR((void *)memslot->userspace_addr))
3074                                 return PTR_ERR((void *)memslot->userspace_addr);
3075                 } else {
3076                         if (!old.user_alloc && old.rmap) {
3077                                 int ret;
3078
3079                                 down_write(&current->mm->mmap_sem);
3080                                 ret = do_munmap(current->mm, old.userspace_addr,
3081                                                 old.npages * PAGE_SIZE);
3082                                 up_write(&current->mm->mmap_sem);
3083                                 if (ret < 0)
3084                                         printk(KERN_WARNING
3085                                        "kvm_vm_ioctl_set_memory_region: "
3086                                        "failed to munmap memory\n");
3087                         }
3088                 }
3089         }
3090
3091         if (!kvm->n_requested_mmu_pages) {
3092                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3093                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3094         }
3095
3096         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3097         kvm_flush_remote_tlbs(kvm);
3098
3099         return 0;
3100 }