arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/msr.h>
  42 #include <asm/desc.h>
  43 #include <asm/mtrr.h>
  44
  45 #define MAX_IO_MSRS 256
  46 #define CR0_RESERVED_BITS                                               \
  47         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  48                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  49                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  50 #define CR4_RESERVED_BITS                                               \
  51         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  52                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  53                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  54                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  55
  56 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  57 /* EFER defaults:
  58  * - enable syscall per default because its emulated by KVM
  59  * - enable LME and LMA per default on 64 bit KVM
  60  */
  61 #ifdef CONFIG_X86_64
  62 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  63 #else
  64 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  65 #endif
  66
  67 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  68 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  69
  70 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  71                                     struct kvm_cpuid_entry2 __user *entries);
  72
  73 struct kvm_x86_ops *kvm_x86_ops;
  74 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  75
  76 struct kvm_stats_debugfs_item debugfs_entries[] = {
  77         { "pf_fixed", VCPU_STAT(pf_fixed) },
  78         { "pf_guest", VCPU_STAT(pf_guest) },
  79         { "tlb_flush", VCPU_STAT(tlb_flush) },
  80         { "invlpg", VCPU_STAT(invlpg) },
  81         { "exits", VCPU_STAT(exits) },
  82         { "io_exits", VCPU_STAT(io_exits) },
  83         { "mmio_exits", VCPU_STAT(mmio_exits) },
  84         { "signal_exits", VCPU_STAT(signal_exits) },
  85         { "irq_window", VCPU_STAT(irq_window_exits) },
  86         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  87         { "halt_exits", VCPU_STAT(halt_exits) },
  88         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  89         { "hypercalls", VCPU_STAT(hypercalls) },
  90         { "request_irq", VCPU_STAT(request_irq_exits) },
  91         { "request_nmi", VCPU_STAT(request_nmi_exits) },
  92         { "irq_exits", VCPU_STAT(irq_exits) },
  93         { "host_state_reload", VCPU_STAT(host_state_reload) },
  94         { "efer_reload", VCPU_STAT(efer_reload) },
  95         { "fpu_reload", VCPU_STAT(fpu_reload) },
  96         { "insn_emulation", VCPU_STAT(insn_emulation) },
  97         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  98         { "irq_injections", VCPU_STAT(irq_injections) },
  99         { "nmi_injections", VCPU_STAT(nmi_injections) },
 100         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 101         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 102         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 103         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 104         { "mmu_flooded", VM_STAT(mmu_flooded) },
 105         { "mmu_recycled", VM_STAT(mmu_recycled) },
 106         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 107         { "mmu_unsync", VM_STAT(mmu_unsync) },
 108         { "mmu_unsync_global", VM_STAT(mmu_unsync_global) },
 109         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 110         { "largepages", VM_STAT(lpages) },
 111         { NULL }
 112 };
 113
 114 unsigned long segment_base(u16 selector)
 115 {
 116         struct descriptor_table gdt;
 117         struct desc_struct *d;
 118         unsigned long table_base;
 119         unsigned long v;
 120
 121         if (selector == 0)
 122                 return 0;
 123
 124         asm("sgdt %0" : "=m"(gdt));
 125         table_base = gdt.base;
 126
 127         if (selector & 4) {           /* from ldt */
 128                 u16 ldt_selector;
 129
 130                 asm("sldt %0" : "=g"(ldt_selector));
 131                 table_base = segment_base(ldt_selector);
 132         }
 133         d = (struct desc_struct *)(table_base + (selector & ~7));
 134         v = d->base0 | ((unsigned long)d->base1 << 16) |
 135                 ((unsigned long)d->base2 << 24);
 136 #ifdef CONFIG_X86_64
 137         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 138                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 139 #endif
 140         return v;
 141 }
 142 EXPORT_SYMBOL_GPL(segment_base);
 143
 144 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 145 {
 146         if (irqchip_in_kernel(vcpu->kvm))
 147                 return vcpu->arch.apic_base;
 148         else
 149                 return vcpu->arch.apic_base;
 150 }
 151 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 152
 153 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 154 {
 155         /* TODO: reserve bits check */
 156         if (irqchip_in_kernel(vcpu->kvm))
 157                 kvm_lapic_set_base(vcpu, data);
 158         else
 159                 vcpu->arch.apic_base = data;
 160 }
 161 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 162
 163 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 164 {
 165         WARN_ON(vcpu->arch.exception.pending);
 166         vcpu->arch.exception.pending = true;
 167         vcpu->arch.exception.has_error_code = false;
 168         vcpu->arch.exception.nr = nr;
 169 }
 170 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 171
 172 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 173                            u32 error_code)
 174 {
 175         ++vcpu->stat.pf_guest;
 176         if (vcpu->arch.exception.pending) {
 177                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 178                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 179                                         " double fault 0x%lx\n", addr);
 180                         vcpu->arch.exception.nr = DF_VECTOR;
 181                         vcpu->arch.exception.error_code = 0;
 182                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 183                         /* triple fault -> shutdown */
 184                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 185                 }
 186                 return;
 187         }
 188         vcpu->arch.cr2 = addr;
 189         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 190 }
 191
 192 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 193 {
 194         vcpu->arch.nmi_pending = 1;
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 197
 198 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 199 {
 200         WARN_ON(vcpu->arch.exception.pending);
 201         vcpu->arch.exception.pending = true;
 202         vcpu->arch.exception.has_error_code = true;
 203         vcpu->arch.exception.nr = nr;
 204         vcpu->arch.exception.error_code = error_code;
 205 }
 206 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 207
 208 static void __queue_exception(struct kvm_vcpu *vcpu)
 209 {
 210         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 211                                      vcpu->arch.exception.has_error_code,
 212                                      vcpu->arch.exception.error_code);
 213 }
 214
 215 /*
 216  * Load the pae pdptrs.  Return true is they are all valid.
 217  */
 218 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 219 {
 220         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 221         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 222         int i;
 223         int ret;
 224         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 225
 226         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 227                                   offset * sizeof(u64), sizeof(pdpte));
 228         if (ret < 0) {
 229                 ret = 0;
 230                 goto out;
 231         }
 232         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 233                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 234                         ret = 0;
 235                         goto out;
 236                 }
 237         }
 238         ret = 1;
 239
 240         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 241 out:
 242
 243         return ret;
 244 }
 245 EXPORT_SYMBOL_GPL(load_pdptrs);
 246
 247 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 248 {
 249         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 250         bool changed = true;
 251         int r;
 252
 253         if (is_long_mode(vcpu) || !is_pae(vcpu))
 254                 return false;
 255
 256         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 257         if (r < 0)
 258                 goto out;
 259         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 260 out:
 261
 262         return changed;
 263 }
 264
 265 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 266 {
 267         if (cr0 & CR0_RESERVED_BITS) {
 268                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 269                        cr0, vcpu->arch.cr0);
 270                 kvm_inject_gp(vcpu, 0);
 271                 return;
 272         }
 273
 274         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 275                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 276                 kvm_inject_gp(vcpu, 0);
 277                 return;
 278         }
 279
 280         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 281                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 282                        "and a clear PE flag\n");
 283                 kvm_inject_gp(vcpu, 0);
 284                 return;
 285         }
 286
 287         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 288 #ifdef CONFIG_X86_64
 289                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 290                         int cs_db, cs_l;
 291
 292                         if (!is_pae(vcpu)) {
 293                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 294                                        "in long mode while PAE is disabled\n");
 295                                 kvm_inject_gp(vcpu, 0);
 296                                 return;
 297                         }
 298                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 299                         if (cs_l) {
 300                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 301                                        "in long mode while CS.L == 1\n");
 302                                 kvm_inject_gp(vcpu, 0);
 303                                 return;
 304
 305                         }
 306                 } else
 307 #endif
 308                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 309                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 310                                "reserved bits\n");
 311                         kvm_inject_gp(vcpu, 0);
 312                         return;
 313                 }
 314
 315         }
 316
 317         kvm_x86_ops->set_cr0(vcpu, cr0);
 318         vcpu->arch.cr0 = cr0;
 319
 320         kvm_mmu_sync_global(vcpu);
 321         kvm_mmu_reset_context(vcpu);
 322         return;
 323 }
 324 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 325
 326 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 327 {
 328         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 329         KVMTRACE_1D(LMSW, vcpu,
 330                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 331                     handler);
 332 }
 333 EXPORT_SYMBOL_GPL(kvm_lmsw);
 334
 335 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 336 {
 337         if (cr4 & CR4_RESERVED_BITS) {
 338                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 339                 kvm_inject_gp(vcpu, 0);
 340                 return;
 341         }
 342
 343         if (is_long_mode(vcpu)) {
 344                 if (!(cr4 & X86_CR4_PAE)) {
 345                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 346                                "in long mode\n");
 347                         kvm_inject_gp(vcpu, 0);
 348                         return;
 349                 }
 350         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 351                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 352                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 353                 kvm_inject_gp(vcpu, 0);
 354                 return;
 355         }
 356
 357         if (cr4 & X86_CR4_VMXE) {
 358                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 359                 kvm_inject_gp(vcpu, 0);
 360                 return;
 361         }
 362         kvm_x86_ops->set_cr4(vcpu, cr4);
 363         vcpu->arch.cr4 = cr4;
 364         kvm_mmu_sync_global(vcpu);
 365         kvm_mmu_reset_context(vcpu);
 366 }
 367 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 368
 369 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 370 {
 371         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 372                 kvm_mmu_sync_roots(vcpu);
 373                 kvm_mmu_flush_tlb(vcpu);
 374                 return;
 375         }
 376
 377         if (is_long_mode(vcpu)) {
 378                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 379                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 380                         kvm_inject_gp(vcpu, 0);
 381                         return;
 382                 }
 383         } else {
 384                 if (is_pae(vcpu)) {
 385                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 386                                 printk(KERN_DEBUG
 387                                        "set_cr3: #GP, reserved bits\n");
 388                                 kvm_inject_gp(vcpu, 0);
 389                                 return;
 390                         }
 391                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 392                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 393                                        "reserved bits\n");
 394                                 kvm_inject_gp(vcpu, 0);
 395                                 return;
 396                         }
 397                 }
 398                 /*
 399                  * We don't check reserved bits in nonpae mode, because
 400                  * this isn't enforced, and VMware depends on this.
 401                  */
 402         }
 403
 404         /*
 405          * Does the new cr3 value map to physical memory? (Note, we
 406          * catch an invalid cr3 even in real-mode, because it would
 407          * cause trouble later on when we turn on paging anyway.)
 408          *
 409          * A real CPU would silently accept an invalid cr3 and would
 410          * attempt to use it - with largely undefined (and often hard
 411          * to debug) behavior on the guest side.
 412          */
 413         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 414                 kvm_inject_gp(vcpu, 0);
 415         else {
 416                 vcpu->arch.cr3 = cr3;
 417                 vcpu->arch.mmu.new_cr3(vcpu);
 418         }
 419 }
 420 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 421
 422 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 423 {
 424         if (cr8 & CR8_RESERVED_BITS) {
 425                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 426                 kvm_inject_gp(vcpu, 0);
 427                 return;
 428         }
 429         if (irqchip_in_kernel(vcpu->kvm))
 430                 kvm_lapic_set_tpr(vcpu, cr8);
 431         else
 432                 vcpu->arch.cr8 = cr8;
 433 }
 434 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 435
 436 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 437 {
 438         if (irqchip_in_kernel(vcpu->kvm))
 439                 return kvm_lapic_get_cr8(vcpu);
 440         else
 441                 return vcpu->arch.cr8;
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 444
 445 /*
 446  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 447  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 448  *
 449  * This list is modified at module load time to reflect the
 450  * capabilities of the host cpu.
 451  */
 452 static u32 msrs_to_save[] = {
 453         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 454         MSR_K6_STAR,
 455 #ifdef CONFIG_X86_64
 456         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 457 #endif
 458         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 459         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 460 };
 461
 462 static unsigned num_msrs_to_save;
 463
 464 static u32 emulated_msrs[] = {
 465         MSR_IA32_MISC_ENABLE,
 466 };
 467
 468 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 469 {
 470         if (efer & efer_reserved_bits) {
 471                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 472                        efer);
 473                 kvm_inject_gp(vcpu, 0);
 474                 return;
 475         }
 476
 477         if (is_paging(vcpu)
 478             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 479                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 480                 kvm_inject_gp(vcpu, 0);
 481                 return;
 482         }
 483
 484         kvm_x86_ops->set_efer(vcpu, efer);
 485
 486         efer &= ~EFER_LMA;
 487         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 488
 489         vcpu->arch.shadow_efer = efer;
 490 }
 491
 492 void kvm_enable_efer_bits(u64 mask)
 493 {
 494        efer_reserved_bits &= ~mask;
 495 }
 496 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 497
 498
 499 /*
 500  * Writes msr value into into the appropriate "register".
 501  * Returns 0 on success, non-0 otherwise.
 502  * Assumes vcpu_load() was already called.
 503  */
 504 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 505 {
 506         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 507 }
 508
 509 /*
 510  * Adapt set_msr() to msr_io()'s calling convention
 511  */
 512 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 513 {
 514         return kvm_set_msr(vcpu, index, *data);
 515 }
 516
 517 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 518 {
 519         static int version;
 520         struct pvclock_wall_clock wc;
 521         struct timespec now, sys, boot;
 522
 523         if (!wall_clock)
 524                 return;
 525
 526         version++;
 527
 528         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 529
 530         /*
 531          * The guest calculates current wall clock time by adding
 532          * system time (updated by kvm_write_guest_time below) to the
 533          * wall clock specified here.  guest system time equals host
 534          * system time for us, thus we must fill in host boot time here.
 535          */
 536         now = current_kernel_time();
 537         ktime_get_ts(&sys);
 538         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 539
 540         wc.sec = boot.tv_sec;
 541         wc.nsec = boot.tv_nsec;
 542         wc.version = version;
 543
 544         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 545
 546         version++;
 547         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 548 }
 549
 550 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 551 {
 552         uint32_t quotient, remainder;
 553
 554         /* Don't try to replace with do_div(), this one calculates
 555          * "(dividend << 32) / divisor" */
 556         __asm__ ( "divl %4"
 557                   : "=a" (quotient), "=d" (remainder)
 558                   : "0" (0), "1" (dividend), "r" (divisor) );
 559         return quotient;
 560 }
 561
 562 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 563 {
 564         uint64_t nsecs = 1000000000LL;
 565         int32_t  shift = 0;
 566         uint64_t tps64;
 567         uint32_t tps32;
 568
 569         tps64 = tsc_khz * 1000LL;
 570         while (tps64 > nsecs*2) {
 571                 tps64 >>= 1;
 572                 shift--;
 573         }
 574
 575         tps32 = (uint32_t)tps64;
 576         while (tps32 <= (uint32_t)nsecs) {
 577                 tps32 <<= 1;
 578                 shift++;
 579         }
 580
 581         hv_clock->tsc_shift = shift;
 582         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 583
 584         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 585                  __func__, tsc_khz, hv_clock->tsc_shift,
 586                  hv_clock->tsc_to_system_mul);
 587 }
 588
 589 static void kvm_write_guest_time(struct kvm_vcpu *v)
 590 {
 591         struct timespec ts;
 592         unsigned long flags;
 593         struct kvm_vcpu_arch *vcpu = &v->arch;
 594         void *shared_kaddr;
 595
 596         if ((!vcpu->time_page))
 597                 return;
 598
 599         if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
 600                 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
 601                 vcpu->hv_clock_tsc_khz = tsc_khz;
 602         }
 603
 604         /* Keep irq disabled to prevent changes to the clock */
 605         local_irq_save(flags);
 606         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 607                           &vcpu->hv_clock.tsc_timestamp);
 608         ktime_get_ts(&ts);
 609         local_irq_restore(flags);
 610
 611         /* With all the info we got, fill in the values */
 612
 613         vcpu->hv_clock.system_time = ts.tv_nsec +
 614                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 615         /*
 616          * The interface expects us to write an even number signaling that the
 617          * update is finished. Since the guest won't see the intermediate
 618          * state, we just increase by 2 at the end.
 619          */
 620         vcpu->hv_clock.version += 2;
 621
 622         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 623
 624         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 625                sizeof(vcpu->hv_clock));
 626
 627         kunmap_atomic(shared_kaddr, KM_USER0);
 628
 629         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 630 }
 631
 632 static bool msr_mtrr_valid(unsigned msr)
 633 {
 634         switch (msr) {
 635         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 636         case MSR_MTRRfix64K_00000:
 637         case MSR_MTRRfix16K_80000:
 638         case MSR_MTRRfix16K_A0000:
 639         case MSR_MTRRfix4K_C0000:
 640         case MSR_MTRRfix4K_C8000:
 641         case MSR_MTRRfix4K_D0000:
 642         case MSR_MTRRfix4K_D8000:
 643         case MSR_MTRRfix4K_E0000:
 644         case MSR_MTRRfix4K_E8000:
 645         case MSR_MTRRfix4K_F0000:
 646         case MSR_MTRRfix4K_F8000:
 647         case MSR_MTRRdefType:
 648         case MSR_IA32_CR_PAT:
 649                 return true;
 650         case 0x2f8:
 651                 return true;
 652         }
 653         return false;
 654 }
 655
 656 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 657 {
 658         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 659
 660         if (!msr_mtrr_valid(msr))
 661                 return 1;
 662
 663         if (msr == MSR_MTRRdefType) {
 664                 vcpu->arch.mtrr_state.def_type = data;
 665                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 666         } else if (msr == MSR_MTRRfix64K_00000)
 667                 p[0] = data;
 668         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 669                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 670         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 671                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 672         else if (msr == MSR_IA32_CR_PAT)
 673                 vcpu->arch.pat = data;
 674         else {  /* Variable MTRRs */
 675                 int idx, is_mtrr_mask;
 676                 u64 *pt;
 677
 678                 idx = (msr - 0x200) / 2;
 679                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 680                 if (!is_mtrr_mask)
 681                         pt =
 682                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 683                 else
 684                         pt =
 685                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 686                 *pt = data;
 687         }
 688
 689         kvm_mmu_reset_context(vcpu);
 690         return 0;
 691 }
 692
 693 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 694 {
 695         switch (msr) {
 696         case MSR_EFER:
 697                 set_efer(vcpu, data);
 698                 break;
 699         case MSR_IA32_MC0_STATUS:
 700                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 701                        __func__, data);
 702                 break;
 703         case MSR_IA32_MCG_STATUS:
 704                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 705                         __func__, data);
 706                 break;
 707         case MSR_IA32_MCG_CTL:
 708                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 709                         __func__, data);
 710                 break;
 711         case MSR_IA32_DEBUGCTLMSR:
 712                 if (!data) {
 713                         /* We support the non-activated case already */
 714                         break;
 715                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 716                         /* Values other than LBR and BTF are vendor-specific,
 717                            thus reserved and should throw a #GP */
 718                         return 1;
 719                 }
 720                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 721                         __func__, data);
 722                 break;
 723         case MSR_IA32_UCODE_REV:
 724         case MSR_IA32_UCODE_WRITE:
 725                 break;
 726         case 0x200 ... 0x2ff:
 727                 return set_msr_mtrr(vcpu, msr, data);
 728         case MSR_IA32_APICBASE:
 729                 kvm_set_apic_base(vcpu, data);
 730                 break;
 731         case MSR_IA32_MISC_ENABLE:
 732                 vcpu->arch.ia32_misc_enable_msr = data;
 733                 break;
 734         case MSR_KVM_WALL_CLOCK:
 735                 vcpu->kvm->arch.wall_clock = data;
 736                 kvm_write_wall_clock(vcpu->kvm, data);
 737                 break;
 738         case MSR_KVM_SYSTEM_TIME: {
 739                 if (vcpu->arch.time_page) {
 740                         kvm_release_page_dirty(vcpu->arch.time_page);
 741                         vcpu->arch.time_page = NULL;
 742                 }
 743
 744                 vcpu->arch.time = data;
 745
 746                 /* we verify if the enable bit is set... */
 747                 if (!(data & 1))
 748                         break;
 749
 750                 /* ...but clean it before doing the actual write */
 751                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 752
 753                 vcpu->arch.time_page =
 754                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 755
 756                 if (is_error_page(vcpu->arch.time_page)) {
 757                         kvm_release_page_clean(vcpu->arch.time_page);
 758                         vcpu->arch.time_page = NULL;
 759                 }
 760
 761                 kvm_write_guest_time(vcpu);
 762                 break;
 763         }
 764         default:
 765                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 766                 return 1;
 767         }
 768         return 0;
 769 }
 770 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 771
 772
 773 /*
 774  * Reads an msr value (of 'msr_index') into 'pdata'.
 775  * Returns 0 on success, non-0 otherwise.
 776  * Assumes vcpu_load() was already called.
 777  */
 778 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 779 {
 780         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 781 }
 782
 783 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 784 {
 785         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 786
 787         if (!msr_mtrr_valid(msr))
 788                 return 1;
 789
 790         if (msr == MSR_MTRRdefType)
 791                 *pdata = vcpu->arch.mtrr_state.def_type +
 792                          (vcpu->arch.mtrr_state.enabled << 10);
 793         else if (msr == MSR_MTRRfix64K_00000)
 794                 *pdata = p[0];
 795         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 796                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 797         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 798                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 799         else if (msr == MSR_IA32_CR_PAT)
 800                 *pdata = vcpu->arch.pat;
 801         else {  /* Variable MTRRs */
 802                 int idx, is_mtrr_mask;
 803                 u64 *pt;
 804
 805                 idx = (msr - 0x200) / 2;
 806                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 807                 if (!is_mtrr_mask)
 808                         pt =
 809                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 810                 else
 811                         pt =
 812                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 813                 *pdata = *pt;
 814         }
 815
 816         return 0;
 817 }
 818
 819 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 820 {
 821         u64 data;
 822
 823         switch (msr) {
 824         case 0xc0010010: /* SYSCFG */
 825         case 0xc0010015: /* HWCR */
 826         case MSR_IA32_PLATFORM_ID:
 827         case MSR_IA32_P5_MC_ADDR:
 828         case MSR_IA32_P5_MC_TYPE:
 829         case MSR_IA32_MC0_CTL:
 830         case MSR_IA32_MCG_STATUS:
 831         case MSR_IA32_MCG_CAP:
 832         case MSR_IA32_MCG_CTL:
 833         case MSR_IA32_MC0_MISC:
 834         case MSR_IA32_MC0_MISC+4:
 835         case MSR_IA32_MC0_MISC+8:
 836         case MSR_IA32_MC0_MISC+12:
 837         case MSR_IA32_MC0_MISC+16:
 838         case MSR_IA32_MC0_MISC+20:
 839         case MSR_IA32_UCODE_REV:
 840         case MSR_IA32_EBL_CR_POWERON:
 841         case MSR_IA32_DEBUGCTLMSR:
 842         case MSR_IA32_LASTBRANCHFROMIP:
 843         case MSR_IA32_LASTBRANCHTOIP:
 844         case MSR_IA32_LASTINTFROMIP:
 845         case MSR_IA32_LASTINTTOIP:
 846                 data = 0;
 847                 break;
 848         case MSR_MTRRcap:
 849                 data = 0x500 | KVM_NR_VAR_MTRR;
 850                 break;
 851         case 0x200 ... 0x2ff:
 852                 return get_msr_mtrr(vcpu, msr, pdata);
 853         case 0xcd: /* fsb frequency */
 854                 data = 3;
 855                 break;
 856         case MSR_IA32_APICBASE:
 857                 data = kvm_get_apic_base(vcpu);
 858                 break;
 859         case MSR_IA32_MISC_ENABLE:
 860                 data = vcpu->arch.ia32_misc_enable_msr;
 861                 break;
 862         case MSR_IA32_PERF_STATUS:
 863                 /* TSC increment by tick */
 864                 data = 1000ULL;
 865                 /* CPU multiplier */
 866                 data |= (((uint64_t)4ULL) << 40);
 867                 break;
 868         case MSR_EFER:
 869                 data = vcpu->arch.shadow_efer;
 870                 break;
 871         case MSR_KVM_WALL_CLOCK:
 872                 data = vcpu->kvm->arch.wall_clock;
 873                 break;
 874         case MSR_KVM_SYSTEM_TIME:
 875                 data = vcpu->arch.time;
 876                 break;
 877         default:
 878                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 879                 return 1;
 880         }
 881         *pdata = data;
 882         return 0;
 883 }
 884 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 885
 886 /*
 887  * Read or write a bunch of msrs. All parameters are kernel addresses.
 888  *
 889  * @return number of msrs set successfully.
 890  */
 891 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 892                     struct kvm_msr_entry *entries,
 893                     int (*do_msr)(struct kvm_vcpu *vcpu,
 894                                   unsigned index, u64 *data))
 895 {
 896         int i;
 897
 898         vcpu_load(vcpu);
 899
 900         down_read(&vcpu->kvm->slots_lock);
 901         for (i = 0; i < msrs->nmsrs; ++i)
 902                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 903                         break;
 904         up_read(&vcpu->kvm->slots_lock);
 905
 906         vcpu_put(vcpu);
 907
 908         return i;
 909 }
 910
 911 /*
 912  * Read or write a bunch of msrs. Parameters are user addresses.
 913  *
 914  * @return number of msrs set successfully.
 915  */
 916 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 917                   int (*do_msr)(struct kvm_vcpu *vcpu,
 918                                 unsigned index, u64 *data),
 919                   int writeback)
 920 {
 921         struct kvm_msrs msrs;
 922         struct kvm_msr_entry *entries;
 923         int r, n;
 924         unsigned size;
 925
 926         r = -EFAULT;
 927         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 928                 goto out;
 929
 930         r = -E2BIG;
 931         if (msrs.nmsrs >= MAX_IO_MSRS)
 932                 goto out;
 933
 934         r = -ENOMEM;
 935         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 936         entries = vmalloc(size);
 937         if (!entries)
 938                 goto out;
 939
 940         r = -EFAULT;
 941         if (copy_from_user(entries, user_msrs->entries, size))
 942                 goto out_free;
 943
 944         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 945         if (r < 0)
 946                 goto out_free;
 947
 948         r = -EFAULT;
 949         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 950                 goto out_free;
 951
 952         r = n;
 953
 954 out_free:
 955         vfree(entries);
 956 out:
 957         return r;
 958 }
 959
 960 int kvm_dev_ioctl_check_extension(long ext)
 961 {
 962         int r;
 963
 964         switch (ext) {
 965         case KVM_CAP_IRQCHIP:
 966         case KVM_CAP_HLT:
 967         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 968         case KVM_CAP_SET_TSS_ADDR:
 969         case KVM_CAP_EXT_CPUID:
 970         case KVM_CAP_PIT:
 971         case KVM_CAP_NOP_IO_DELAY:
 972         case KVM_CAP_MP_STATE:
 973         case KVM_CAP_SYNC_MMU:
 974                 r = 1;
 975                 break;
 976         case KVM_CAP_COALESCED_MMIO:
 977                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 978                 break;
 979         case KVM_CAP_VAPIC:
 980                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 981                 break;
 982         case KVM_CAP_NR_VCPUS:
 983                 r = KVM_MAX_VCPUS;
 984                 break;
 985         case KVM_CAP_NR_MEMSLOTS:
 986                 r = KVM_MEMORY_SLOTS;
 987                 break;
 988         case KVM_CAP_PV_MMU:
 989                 r = !tdp_enabled;
 990                 break;
 991         case KVM_CAP_IOMMU:
 992                 r = iommu_found();
 993                 break;
 994         case KVM_CAP_CLOCKSOURCE:
 995                 r = boot_cpu_has(X86_FEATURE_CONSTANT_TSC);
 996                 break;
 997         default:
 998                 r = 0;
 999                 break;
1000         }
1001         return r;
1002
1003 }
1004
1005 long kvm_arch_dev_ioctl(struct file *filp,
1006                         unsigned int ioctl, unsigned long arg)
1007 {
1008         void __user *argp = (void __user *)arg;
1009         long r;
1010
1011         switch (ioctl) {
1012         case KVM_GET_MSR_INDEX_LIST: {
1013                 struct kvm_msr_list __user *user_msr_list = argp;
1014                 struct kvm_msr_list msr_list;
1015                 unsigned n;
1016
1017                 r = -EFAULT;
1018                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1019                         goto out;
1020                 n = msr_list.nmsrs;
1021                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1022                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1023                         goto out;
1024                 r = -E2BIG;
1025                 if (n < num_msrs_to_save)
1026                         goto out;
1027                 r = -EFAULT;
1028                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1029                                  num_msrs_to_save * sizeof(u32)))
1030                         goto out;
1031                 if (copy_to_user(user_msr_list->indices
1032                                  + num_msrs_to_save * sizeof(u32),
1033                                  &emulated_msrs,
1034                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1035                         goto out;
1036                 r = 0;
1037                 break;
1038         }
1039         case KVM_GET_SUPPORTED_CPUID: {
1040                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1041                 struct kvm_cpuid2 cpuid;
1042
1043                 r = -EFAULT;
1044                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1045                         goto out;
1046                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1047                         cpuid_arg->entries);
1048                 if (r)
1049                         goto out;
1050
1051                 r = -EFAULT;
1052                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1053                         goto out;
1054                 r = 0;
1055                 break;
1056         }
1057         default:
1058                 r = -EINVAL;
1059         }
1060 out:
1061         return r;
1062 }
1063
1064 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1065 {
1066         kvm_x86_ops->vcpu_load(vcpu, cpu);
1067         kvm_write_guest_time(vcpu);
1068 }
1069
1070 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1071 {
1072         kvm_x86_ops->vcpu_put(vcpu);
1073         kvm_put_guest_fpu(vcpu);
1074 }
1075
1076 static int is_efer_nx(void)
1077 {
1078         u64 efer;
1079
1080         rdmsrl(MSR_EFER, efer);
1081         return efer & EFER_NX;
1082 }
1083
1084 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1085 {
1086         int i;
1087         struct kvm_cpuid_entry2 *e, *entry;
1088
1089         entry = NULL;
1090         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1091                 e = &vcpu->arch.cpuid_entries[i];
1092                 if (e->function == 0x80000001) {
1093                         entry = e;
1094                         break;
1095                 }
1096         }
1097         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1098                 entry->edx &= ~(1 << 20);
1099                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1100         }
1101 }
1102
1103 /* when an old userspace process fills a new kernel module */
1104 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1105                                     struct kvm_cpuid *cpuid,
1106                                     struct kvm_cpuid_entry __user *entries)
1107 {
1108         int r, i;
1109         struct kvm_cpuid_entry *cpuid_entries;
1110
1111         r = -E2BIG;
1112         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1113                 goto out;
1114         r = -ENOMEM;
1115         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1116         if (!cpuid_entries)
1117                 goto out;
1118         r = -EFAULT;
1119         if (copy_from_user(cpuid_entries, entries,
1120                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1121                 goto out_free;
1122         for (i = 0; i < cpuid->nent; i++) {
1123                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1124                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1125                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1126                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1127                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1128                 vcpu->arch.cpuid_entries[i].index = 0;
1129                 vcpu->arch.cpuid_entries[i].flags = 0;
1130                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1131                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1132                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1133         }
1134         vcpu->arch.cpuid_nent = cpuid->nent;
1135         cpuid_fix_nx_cap(vcpu);
1136         r = 0;
1137
1138 out_free:
1139         vfree(cpuid_entries);
1140 out:
1141         return r;
1142 }
1143
1144 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1145                                     struct kvm_cpuid2 *cpuid,
1146                                     struct kvm_cpuid_entry2 __user *entries)
1147 {
1148         int r;
1149
1150         r = -E2BIG;
1151         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1152                 goto out;
1153         r = -EFAULT;
1154         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1155                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1156                 goto out;
1157         vcpu->arch.cpuid_nent = cpuid->nent;
1158         return 0;
1159
1160 out:
1161         return r;
1162 }
1163
1164 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1165                                     struct kvm_cpuid2 *cpuid,
1166                                     struct kvm_cpuid_entry2 __user *entries)
1167 {
1168         int r;
1169
1170         r = -E2BIG;
1171         if (cpuid->nent < vcpu->arch.cpuid_nent)
1172                 goto out;
1173         r = -EFAULT;
1174         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1175                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1176                 goto out;
1177         return 0;
1178
1179 out:
1180         cpuid->nent = vcpu->arch.cpuid_nent;
1181         return r;
1182 }
1183
1184 static inline u32 bit(int bitno)
1185 {
1186         return 1 << (bitno & 31);
1187 }
1188
1189 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1190                           u32 index)
1191 {
1192         entry->function = function;
1193         entry->index = index;
1194         cpuid_count(entry->function, entry->index,
1195                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1196         entry->flags = 0;
1197 }
1198
1199 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1200                          u32 index, int *nent, int maxnent)
1201 {
1202         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1203                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1204                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1205                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1206                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1207                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1208                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1209                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1210                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1211                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1212         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1213                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1214                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1215                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1216                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1217                 bit(X86_FEATURE_PGE) |
1218                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1219                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1220                 bit(X86_FEATURE_SYSCALL) |
1221                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1222 #ifdef CONFIG_X86_64
1223                 bit(X86_FEATURE_LM) |
1224 #endif
1225                 bit(X86_FEATURE_MMXEXT) |
1226                 bit(X86_FEATURE_3DNOWEXT) |
1227                 bit(X86_FEATURE_3DNOW);
1228         const u32 kvm_supported_word3_x86_features =
1229                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1230         const u32 kvm_supported_word6_x86_features =
1231                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1232
1233         /* all func 2 cpuid_count() should be called on the same cpu */
1234         get_cpu();
1235         do_cpuid_1_ent(entry, function, index);
1236         ++*nent;
1237
1238         switch (function) {
1239         case 0:
1240                 entry->eax = min(entry->eax, (u32)0xb);
1241                 break;
1242         case 1:
1243                 entry->edx &= kvm_supported_word0_x86_features;
1244                 entry->ecx &= kvm_supported_word3_x86_features;
1245                 break;
1246         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1247          * may return different values. This forces us to get_cpu() before
1248          * issuing the first command, and also to emulate this annoying behavior
1249          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1250         case 2: {
1251                 int t, times = entry->eax & 0xff;
1252
1253                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1254                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1255                 for (t = 1; t < times && *nent < maxnent; ++t) {
1256                         do_cpuid_1_ent(&entry[t], function, 0);
1257                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1258                         ++*nent;
1259                 }
1260                 break;
1261         }
1262         /* function 4 and 0xb have additional index. */
1263         case 4: {
1264                 int i, cache_type;
1265
1266                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1267                 /* read more entries until cache_type is zero */
1268                 for (i = 1; *nent < maxnent; ++i) {
1269                         cache_type = entry[i - 1].eax & 0x1f;
1270                         if (!cache_type)
1271                                 break;
1272                         do_cpuid_1_ent(&entry[i], function, i);
1273                         entry[i].flags |=
1274                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1275                         ++*nent;
1276                 }
1277                 break;
1278         }
1279         case 0xb: {
1280                 int i, level_type;
1281
1282                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1283                 /* read more entries until level_type is zero */
1284                 for (i = 1; *nent < maxnent; ++i) {
1285                         level_type = entry[i - 1].ecx & 0xff00;
1286                         if (!level_type)
1287                                 break;
1288                         do_cpuid_1_ent(&entry[i], function, i);
1289                         entry[i].flags |=
1290                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1291                         ++*nent;
1292                 }
1293                 break;
1294         }
1295         case 0x80000000:
1296                 entry->eax = min(entry->eax, 0x8000001a);
1297                 break;
1298         case 0x80000001:
1299                 entry->edx &= kvm_supported_word1_x86_features;
1300                 entry->ecx &= kvm_supported_word6_x86_features;
1301                 break;
1302         }
1303         put_cpu();
1304 }
1305
1306 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1307                                     struct kvm_cpuid_entry2 __user *entries)
1308 {
1309         struct kvm_cpuid_entry2 *cpuid_entries;
1310         int limit, nent = 0, r = -E2BIG;
1311         u32 func;
1312
1313         if (cpuid->nent < 1)
1314                 goto out;
1315         r = -ENOMEM;
1316         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1317         if (!cpuid_entries)
1318                 goto out;
1319
1320         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1321         limit = cpuid_entries[0].eax;
1322         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1323                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1324                                 &nent, cpuid->nent);
1325         r = -E2BIG;
1326         if (nent >= cpuid->nent)
1327                 goto out_free;
1328
1329         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1330         limit = cpuid_entries[nent - 1].eax;
1331         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1332                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1333                                &nent, cpuid->nent);
1334         r = -EFAULT;
1335         if (copy_to_user(entries, cpuid_entries,
1336                         nent * sizeof(struct kvm_cpuid_entry2)))
1337                 goto out_free;
1338         cpuid->nent = nent;
1339         r = 0;
1340
1341 out_free:
1342         vfree(cpuid_entries);
1343 out:
1344         return r;
1345 }
1346
1347 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1348                                     struct kvm_lapic_state *s)
1349 {
1350         vcpu_load(vcpu);
1351         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1352         vcpu_put(vcpu);
1353
1354         return 0;
1355 }
1356
1357 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1358                                     struct kvm_lapic_state *s)
1359 {
1360         vcpu_load(vcpu);
1361         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1362         kvm_apic_post_state_restore(vcpu);
1363         vcpu_put(vcpu);
1364
1365         return 0;
1366 }
1367
1368 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1369                                     struct kvm_interrupt *irq)
1370 {
1371         if (irq->irq < 0 || irq->irq >= 256)
1372                 return -EINVAL;
1373         if (irqchip_in_kernel(vcpu->kvm))
1374                 return -ENXIO;
1375         vcpu_load(vcpu);
1376
1377         set_bit(irq->irq, vcpu->arch.irq_pending);
1378         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1379
1380         vcpu_put(vcpu);
1381
1382         return 0;
1383 }
1384
1385 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1386 {
1387         vcpu_load(vcpu);
1388         kvm_inject_nmi(vcpu);
1389         vcpu_put(vcpu);
1390
1391         return 0;
1392 }
1393
1394 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1395                                            struct kvm_tpr_access_ctl *tac)
1396 {
1397         if (tac->flags)
1398                 return -EINVAL;
1399         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1400         return 0;
1401 }
1402
1403 long kvm_arch_vcpu_ioctl(struct file *filp,
1404                          unsigned int ioctl, unsigned long arg)
1405 {
1406         struct kvm_vcpu *vcpu = filp->private_data;
1407         void __user *argp = (void __user *)arg;
1408         int r;
1409         struct kvm_lapic_state *lapic = NULL;
1410
1411         switch (ioctl) {
1412         case KVM_GET_LAPIC: {
1413                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1414
1415                 r = -ENOMEM;
1416                 if (!lapic)
1417                         goto out;
1418                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1419                 if (r)
1420                         goto out;
1421                 r = -EFAULT;
1422                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1423                         goto out;
1424                 r = 0;
1425                 break;
1426         }
1427         case KVM_SET_LAPIC: {
1428                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1429                 r = -ENOMEM;
1430                 if (!lapic)
1431                         goto out;
1432                 r = -EFAULT;
1433                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1434                         goto out;
1435                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1436                 if (r)
1437                         goto out;
1438                 r = 0;
1439                 break;
1440         }
1441         case KVM_INTERRUPT: {
1442                 struct kvm_interrupt irq;
1443
1444                 r = -EFAULT;
1445                 if (copy_from_user(&irq, argp, sizeof irq))
1446                         goto out;
1447                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1448                 if (r)
1449                         goto out;
1450                 r = 0;
1451                 break;
1452         }
1453         case KVM_NMI: {
1454                 r = kvm_vcpu_ioctl_nmi(vcpu);
1455                 if (r)
1456                         goto out;
1457                 r = 0;
1458                 break;
1459         }
1460         case KVM_SET_CPUID: {
1461                 struct kvm_cpuid __user *cpuid_arg = argp;
1462                 struct kvm_cpuid cpuid;
1463
1464                 r = -EFAULT;
1465                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1466                         goto out;
1467                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1468                 if (r)
1469                         goto out;
1470                 break;
1471         }
1472         case KVM_SET_CPUID2: {
1473                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1474                 struct kvm_cpuid2 cpuid;
1475
1476                 r = -EFAULT;
1477                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1478                         goto out;
1479                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1480                                 cpuid_arg->entries);
1481                 if (r)
1482                         goto out;
1483                 break;
1484         }
1485         case KVM_GET_CPUID2: {
1486                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1487                 struct kvm_cpuid2 cpuid;
1488
1489                 r = -EFAULT;
1490                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1491                         goto out;
1492                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1493                                 cpuid_arg->entries);
1494                 if (r)
1495                         goto out;
1496                 r = -EFAULT;
1497                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1498                         goto out;
1499                 r = 0;
1500                 break;
1501         }
1502         case KVM_GET_MSRS:
1503                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1504                 break;
1505         case KVM_SET_MSRS:
1506                 r = msr_io(vcpu, argp, do_set_msr, 0);
1507                 break;
1508         case KVM_TPR_ACCESS_REPORTING: {
1509                 struct kvm_tpr_access_ctl tac;
1510
1511                 r = -EFAULT;
1512                 if (copy_from_user(&tac, argp, sizeof tac))
1513                         goto out;
1514                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1515                 if (r)
1516                         goto out;
1517                 r = -EFAULT;
1518                 if (copy_to_user(argp, &tac, sizeof tac))
1519                         goto out;
1520                 r = 0;
1521                 break;
1522         };
1523         case KVM_SET_VAPIC_ADDR: {
1524                 struct kvm_vapic_addr va;
1525
1526                 r = -EINVAL;
1527                 if (!irqchip_in_kernel(vcpu->kvm))
1528                         goto out;
1529                 r = -EFAULT;
1530                 if (copy_from_user(&va, argp, sizeof va))
1531                         goto out;
1532                 r = 0;
1533                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1534                 break;
1535         }
1536         default:
1537                 r = -EINVAL;
1538         }
1539 out:
1540         if (lapic)
1541                 kfree(lapic);
1542         return r;
1543 }
1544
1545 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1546 {
1547         int ret;
1548
1549         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1550                 return -1;
1551         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1552         return ret;
1553 }
1554
1555 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1556                                           u32 kvm_nr_mmu_pages)
1557 {
1558         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1559                 return -EINVAL;
1560
1561         down_write(&kvm->slots_lock);
1562
1563         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1564         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1565
1566         up_write(&kvm->slots_lock);
1567         return 0;
1568 }
1569
1570 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1571 {
1572         return kvm->arch.n_alloc_mmu_pages;
1573 }
1574
1575 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1576 {
1577         int i;
1578         struct kvm_mem_alias *alias;
1579
1580         for (i = 0; i < kvm->arch.naliases; ++i) {
1581                 alias = &kvm->arch.aliases[i];
1582                 if (gfn >= alias->base_gfn
1583                     && gfn < alias->base_gfn + alias->npages)
1584                         return alias->target_gfn + gfn - alias->base_gfn;
1585         }
1586         return gfn;
1587 }
1588
1589 /*
1590  * Set a new alias region.  Aliases map a portion of physical memory into
1591  * another portion.  This is useful for memory windows, for example the PC
1592  * VGA region.
1593  */
1594 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1595                                          struct kvm_memory_alias *alias)
1596 {
1597         int r, n;
1598         struct kvm_mem_alias *p;
1599
1600         r = -EINVAL;
1601         /* General sanity checks */
1602         if (alias->memory_size & (PAGE_SIZE - 1))
1603                 goto out;
1604         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1605                 goto out;
1606         if (alias->slot >= KVM_ALIAS_SLOTS)
1607                 goto out;
1608         if (alias->guest_phys_addr + alias->memory_size
1609             < alias->guest_phys_addr)
1610                 goto out;
1611         if (alias->target_phys_addr + alias->memory_size
1612             < alias->target_phys_addr)
1613                 goto out;
1614
1615         down_write(&kvm->slots_lock);
1616         spin_lock(&kvm->mmu_lock);
1617
1618         p = &kvm->arch.aliases[alias->slot];
1619         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1620         p->npages = alias->memory_size >> PAGE_SHIFT;
1621         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1622
1623         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1624                 if (kvm->arch.aliases[n - 1].npages)
1625                         break;
1626         kvm->arch.naliases = n;
1627
1628         spin_unlock(&kvm->mmu_lock);
1629         kvm_mmu_zap_all(kvm);
1630
1631         up_write(&kvm->slots_lock);
1632
1633         return 0;
1634
1635 out:
1636         return r;
1637 }
1638
1639 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1640 {
1641         int r;
1642
1643         r = 0;
1644         switch (chip->chip_id) {
1645         case KVM_IRQCHIP_PIC_MASTER:
1646                 memcpy(&chip->chip.pic,
1647                         &pic_irqchip(kvm)->pics[0],
1648                         sizeof(struct kvm_pic_state));
1649                 break;
1650         case KVM_IRQCHIP_PIC_SLAVE:
1651                 memcpy(&chip->chip.pic,
1652                         &pic_irqchip(kvm)->pics[1],
1653                         sizeof(struct kvm_pic_state));
1654                 break;
1655         case KVM_IRQCHIP_IOAPIC:
1656                 memcpy(&chip->chip.ioapic,
1657                         ioapic_irqchip(kvm),
1658                         sizeof(struct kvm_ioapic_state));
1659                 break;
1660         default:
1661                 r = -EINVAL;
1662                 break;
1663         }
1664         return r;
1665 }
1666
1667 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1668 {
1669         int r;
1670
1671         r = 0;
1672         switch (chip->chip_id) {
1673         case KVM_IRQCHIP_PIC_MASTER:
1674                 memcpy(&pic_irqchip(kvm)->pics[0],
1675                         &chip->chip.pic,
1676                         sizeof(struct kvm_pic_state));
1677                 break;
1678         case KVM_IRQCHIP_PIC_SLAVE:
1679                 memcpy(&pic_irqchip(kvm)->pics[1],
1680                         &chip->chip.pic,
1681                         sizeof(struct kvm_pic_state));
1682                 break;
1683         case KVM_IRQCHIP_IOAPIC:
1684                 memcpy(ioapic_irqchip(kvm),
1685                         &chip->chip.ioapic,
1686                         sizeof(struct kvm_ioapic_state));
1687                 break;
1688         default:
1689                 r = -EINVAL;
1690                 break;
1691         }
1692         kvm_pic_update_irq(pic_irqchip(kvm));
1693         return r;
1694 }
1695
1696 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1697 {
1698         int r = 0;
1699
1700         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1701         return r;
1702 }
1703
1704 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1705 {
1706         int r = 0;
1707
1708         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1709         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1710         return r;
1711 }
1712
1713 /*
1714  * Get (and clear) the dirty memory log for a memory slot.
1715  */
1716 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1717                                       struct kvm_dirty_log *log)
1718 {
1719         int r;
1720         int n;
1721         struct kvm_memory_slot *memslot;
1722         int is_dirty = 0;
1723
1724         down_write(&kvm->slots_lock);
1725
1726         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1727         if (r)
1728                 goto out;
1729
1730         /* If nothing is dirty, don't bother messing with page tables. */
1731         if (is_dirty) {
1732                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1733                 kvm_flush_remote_tlbs(kvm);
1734                 memslot = &kvm->memslots[log->slot];
1735                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1736                 memset(memslot->dirty_bitmap, 0, n);
1737         }
1738         r = 0;
1739 out:
1740         up_write(&kvm->slots_lock);
1741         return r;
1742 }
1743
1744 long kvm_arch_vm_ioctl(struct file *filp,
1745                        unsigned int ioctl, unsigned long arg)
1746 {
1747         struct kvm *kvm = filp->private_data;
1748         void __user *argp = (void __user *)arg;
1749         int r = -EINVAL;
1750         /*
1751          * This union makes it completely explicit to gcc-3.x
1752          * that these two variables' stack usage should be
1753          * combined, not added together.
1754          */
1755         union {
1756                 struct kvm_pit_state ps;
1757                 struct kvm_memory_alias alias;
1758         } u;
1759
1760         switch (ioctl) {
1761         case KVM_SET_TSS_ADDR:
1762                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1763                 if (r < 0)
1764                         goto out;
1765                 break;
1766         case KVM_SET_MEMORY_REGION: {
1767                 struct kvm_memory_region kvm_mem;
1768                 struct kvm_userspace_memory_region kvm_userspace_mem;
1769
1770                 r = -EFAULT;
1771                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1772                         goto out;
1773                 kvm_userspace_mem.slot = kvm_mem.slot;
1774                 kvm_userspace_mem.flags = kvm_mem.flags;
1775                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1776                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1777                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1778                 if (r)
1779                         goto out;
1780                 break;
1781         }
1782         case KVM_SET_NR_MMU_PAGES:
1783                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1784                 if (r)
1785                         goto out;
1786                 break;
1787         case KVM_GET_NR_MMU_PAGES:
1788                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1789                 break;
1790         case KVM_SET_MEMORY_ALIAS:
1791                 r = -EFAULT;
1792                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1793                         goto out;
1794                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1795                 if (r)
1796                         goto out;
1797                 break;
1798         case KVM_CREATE_IRQCHIP:
1799                 r = -ENOMEM;
1800                 kvm->arch.vpic = kvm_create_pic(kvm);
1801                 if (kvm->arch.vpic) {
1802                         r = kvm_ioapic_init(kvm);
1803                         if (r) {
1804                                 kfree(kvm->arch.vpic);
1805                                 kvm->arch.vpic = NULL;
1806                                 goto out;
1807                         }
1808                 } else
1809                         goto out;
1810                 break;
1811         case KVM_CREATE_PIT:
1812                 r = -ENOMEM;
1813                 kvm->arch.vpit = kvm_create_pit(kvm);
1814                 if (kvm->arch.vpit)
1815                         r = 0;
1816                 break;
1817         case KVM_IRQ_LINE: {
1818                 struct kvm_irq_level irq_event;
1819
1820                 r = -EFAULT;
1821                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1822                         goto out;
1823                 if (irqchip_in_kernel(kvm)) {
1824                         mutex_lock(&kvm->lock);
1825                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1826                                     irq_event.irq, irq_event.level);
1827                         mutex_unlock(&kvm->lock);
1828                         r = 0;
1829                 }
1830                 break;
1831         }
1832         case KVM_GET_IRQCHIP: {
1833                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1834                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1835
1836                 r = -ENOMEM;
1837                 if (!chip)
1838                         goto out;
1839                 r = -EFAULT;
1840                 if (copy_from_user(chip, argp, sizeof *chip))
1841                         goto get_irqchip_out;
1842                 r = -ENXIO;
1843                 if (!irqchip_in_kernel(kvm))
1844                         goto get_irqchip_out;
1845                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1846                 if (r)
1847                         goto get_irqchip_out;
1848                 r = -EFAULT;
1849                 if (copy_to_user(argp, chip, sizeof *chip))
1850                         goto get_irqchip_out;
1851                 r = 0;
1852         get_irqchip_out:
1853                 kfree(chip);
1854                 if (r)
1855                         goto out;
1856                 break;
1857         }
1858         case KVM_SET_IRQCHIP: {
1859                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1860                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1861
1862                 r = -ENOMEM;
1863                 if (!chip)
1864                         goto out;
1865                 r = -EFAULT;
1866                 if (copy_from_user(chip, argp, sizeof *chip))
1867                         goto set_irqchip_out;
1868                 r = -ENXIO;
1869                 if (!irqchip_in_kernel(kvm))
1870                         goto set_irqchip_out;
1871                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1872                 if (r)
1873                         goto set_irqchip_out;
1874                 r = 0;
1875         set_irqchip_out:
1876                 kfree(chip);
1877                 if (r)
1878                         goto out;
1879                 break;
1880         }
1881         case KVM_GET_PIT: {
1882                 r = -EFAULT;
1883                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1884                         goto out;
1885                 r = -ENXIO;
1886                 if (!kvm->arch.vpit)
1887                         goto out;
1888                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1889                 if (r)
1890                         goto out;
1891                 r = -EFAULT;
1892                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1893                         goto out;
1894                 r = 0;
1895                 break;
1896         }
1897         case KVM_SET_PIT: {
1898                 r = -EFAULT;
1899                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1900                         goto out;
1901                 r = -ENXIO;
1902                 if (!kvm->arch.vpit)
1903                         goto out;
1904                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1905                 if (r)
1906                         goto out;
1907                 r = 0;
1908                 break;
1909         }
1910         default:
1911                 ;
1912         }
1913 out:
1914         return r;
1915 }
1916
1917 static void kvm_init_msr_list(void)
1918 {
1919         u32 dummy[2];
1920         unsigned i, j;
1921
1922         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1923                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1924                         continue;
1925                 if (j < i)
1926                         msrs_to_save[j] = msrs_to_save[i];
1927                 j++;
1928         }
1929         num_msrs_to_save = j;
1930 }
1931
1932 /*
1933  * Only apic need an MMIO device hook, so shortcut now..
1934  */
1935 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1936                                                 gpa_t addr, int len,
1937                                                 int is_write)
1938 {
1939         struct kvm_io_device *dev;
1940
1941         if (vcpu->arch.apic) {
1942                 dev = &vcpu->arch.apic->dev;
1943                 if (dev->in_range(dev, addr, len, is_write))
1944                         return dev;
1945         }
1946         return NULL;
1947 }
1948
1949
1950 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1951                                                 gpa_t addr, int len,
1952                                                 int is_write)
1953 {
1954         struct kvm_io_device *dev;
1955
1956         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1957         if (dev == NULL)
1958                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1959                                           is_write);
1960         return dev;
1961 }
1962
1963 int emulator_read_std(unsigned long addr,
1964                              void *val,
1965                              unsigned int bytes,
1966                              struct kvm_vcpu *vcpu)
1967 {
1968         void *data = val;
1969         int r = X86EMUL_CONTINUE;
1970
1971         while (bytes) {
1972                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1973                 unsigned offset = addr & (PAGE_SIZE-1);
1974                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1975                 int ret;
1976
1977                 if (gpa == UNMAPPED_GVA) {
1978                         r = X86EMUL_PROPAGATE_FAULT;
1979                         goto out;
1980                 }
1981                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1982                 if (ret < 0) {
1983                         r = X86EMUL_UNHANDLEABLE;
1984                         goto out;
1985                 }
1986
1987                 bytes -= tocopy;
1988                 data += tocopy;
1989                 addr += tocopy;
1990         }
1991 out:
1992         return r;
1993 }
1994 EXPORT_SYMBOL_GPL(emulator_read_std);
1995
1996 static int emulator_read_emulated(unsigned long addr,
1997                                   void *val,
1998                                   unsigned int bytes,
1999                                   struct kvm_vcpu *vcpu)
2000 {
2001         struct kvm_io_device *mmio_dev;
2002         gpa_t                 gpa;
2003
2004         if (vcpu->mmio_read_completed) {
2005                 memcpy(val, vcpu->mmio_data, bytes);
2006                 vcpu->mmio_read_completed = 0;
2007                 return X86EMUL_CONTINUE;
2008         }
2009
2010         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2011
2012         /* For APIC access vmexit */
2013         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2014                 goto mmio;
2015
2016         if (emulator_read_std(addr, val, bytes, vcpu)
2017                         == X86EMUL_CONTINUE)
2018                 return X86EMUL_CONTINUE;
2019         if (gpa == UNMAPPED_GVA)
2020                 return X86EMUL_PROPAGATE_FAULT;
2021
2022 mmio:
2023         /*
2024          * Is this MMIO handled locally?
2025          */
2026         mutex_lock(&vcpu->kvm->lock);
2027         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2028         if (mmio_dev) {
2029                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2030                 mutex_unlock(&vcpu->kvm->lock);
2031                 return X86EMUL_CONTINUE;
2032         }
2033         mutex_unlock(&vcpu->kvm->lock);
2034
2035         vcpu->mmio_needed = 1;
2036         vcpu->mmio_phys_addr = gpa;
2037         vcpu->mmio_size = bytes;
2038         vcpu->mmio_is_write = 0;
2039
2040         return X86EMUL_UNHANDLEABLE;
2041 }
2042
2043 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2044                           const void *val, int bytes)
2045 {
2046         int ret;
2047
2048         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2049         if (ret < 0)
2050                 return 0;
2051         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2052         return 1;
2053 }
2054
2055 static int emulator_write_emulated_onepage(unsigned long addr,
2056                                            const void *val,
2057                                            unsigned int bytes,
2058                                            struct kvm_vcpu *vcpu)
2059 {
2060         struct kvm_io_device *mmio_dev;
2061         gpa_t                 gpa;
2062
2063         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2064
2065         if (gpa == UNMAPPED_GVA) {
2066                 kvm_inject_page_fault(vcpu, addr, 2);
2067                 return X86EMUL_PROPAGATE_FAULT;
2068         }
2069
2070         /* For APIC access vmexit */
2071         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2072                 goto mmio;
2073
2074         if (emulator_write_phys(vcpu, gpa, val, bytes))
2075                 return X86EMUL_CONTINUE;
2076
2077 mmio:
2078         /*
2079          * Is this MMIO handled locally?
2080          */
2081         mutex_lock(&vcpu->kvm->lock);
2082         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2083         if (mmio_dev) {
2084                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2085                 mutex_unlock(&vcpu->kvm->lock);
2086                 return X86EMUL_CONTINUE;
2087         }
2088         mutex_unlock(&vcpu->kvm->lock);
2089
2090         vcpu->mmio_needed = 1;
2091         vcpu->mmio_phys_addr = gpa;
2092         vcpu->mmio_size = bytes;
2093         vcpu->mmio_is_write = 1;
2094         memcpy(vcpu->mmio_data, val, bytes);
2095
2096         return X86EMUL_CONTINUE;
2097 }
2098
2099 int emulator_write_emulated(unsigned long addr,
2100                                    const void *val,
2101                                    unsigned int bytes,
2102                                    struct kvm_vcpu *vcpu)
2103 {
2104         /* Crossing a page boundary? */
2105         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2106                 int rc, now;
2107
2108                 now = -addr & ~PAGE_MASK;
2109                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2110                 if (rc != X86EMUL_CONTINUE)
2111                         return rc;
2112                 addr += now;
2113                 val += now;
2114                 bytes -= now;
2115         }
2116         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2117 }
2118 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2119
2120 static int emulator_cmpxchg_emulated(unsigned long addr,
2121                                      const void *old,
2122                                      const void *new,
2123                                      unsigned int bytes,
2124                                      struct kvm_vcpu *vcpu)
2125 {
2126         static int reported;
2127
2128         if (!reported) {
2129                 reported = 1;
2130                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2131         }
2132 #ifndef CONFIG_X86_64
2133         /* guests cmpxchg8b have to be emulated atomically */
2134         if (bytes == 8) {
2135                 gpa_t gpa;
2136                 struct page *page;
2137                 char *kaddr;
2138                 u64 val;
2139
2140                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2141
2142                 if (gpa == UNMAPPED_GVA ||
2143                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2144                         goto emul_write;
2145
2146                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2147                         goto emul_write;
2148
2149                 val = *(u64 *)new;
2150
2151                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2152
2153                 kaddr = kmap_atomic(page, KM_USER0);
2154                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2155                 kunmap_atomic(kaddr, KM_USER0);
2156                 kvm_release_page_dirty(page);
2157         }
2158 emul_write:
2159 #endif
2160
2161         return emulator_write_emulated(addr, new, bytes, vcpu);
2162 }
2163
2164 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2165 {
2166         return kvm_x86_ops->get_segment_base(vcpu, seg);
2167 }
2168
2169 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2170 {
2171         kvm_mmu_invlpg(vcpu, address);
2172         return X86EMUL_CONTINUE;
2173 }
2174
2175 int emulate_clts(struct kvm_vcpu *vcpu)
2176 {
2177         KVMTRACE_0D(CLTS, vcpu, handler);
2178         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2179         return X86EMUL_CONTINUE;
2180 }
2181
2182 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2183 {
2184         struct kvm_vcpu *vcpu = ctxt->vcpu;
2185
2186         switch (dr) {
2187         case 0 ... 3:
2188                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2189                 return X86EMUL_CONTINUE;
2190         default:
2191                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2192                 return X86EMUL_UNHANDLEABLE;
2193         }
2194 }
2195
2196 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2197 {
2198         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2199         int exception;
2200
2201         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2202         if (exception) {
2203                 /* FIXME: better handling */
2204                 return X86EMUL_UNHANDLEABLE;
2205         }
2206         return X86EMUL_CONTINUE;
2207 }
2208
2209 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2210 {
2211         u8 opcodes[4];
2212         unsigned long rip = kvm_rip_read(vcpu);
2213         unsigned long rip_linear;
2214
2215         if (!printk_ratelimit())
2216                 return;
2217
2218         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2219
2220         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2221
2222         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2223                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2224 }
2225 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2226
2227 static struct x86_emulate_ops emulate_ops = {
2228         .read_std            = emulator_read_std,
2229         .read_emulated       = emulator_read_emulated,
2230         .write_emulated      = emulator_write_emulated,
2231         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2232 };
2233
2234 static void cache_all_regs(struct kvm_vcpu *vcpu)
2235 {
2236         kvm_register_read(vcpu, VCPU_REGS_RAX);
2237         kvm_register_read(vcpu, VCPU_REGS_RSP);
2238         kvm_register_read(vcpu, VCPU_REGS_RIP);
2239         vcpu->arch.regs_dirty = ~0;
2240 }
2241
2242 int emulate_instruction(struct kvm_vcpu *vcpu,
2243                         struct kvm_run *run,
2244                         unsigned long cr2,
2245                         u16 error_code,
2246                         int emulation_type)
2247 {
2248         int r;
2249         struct decode_cache *c;
2250
2251         kvm_clear_exception_queue(vcpu);
2252         vcpu->arch.mmio_fault_cr2 = cr2;
2253         /*
2254          * TODO: fix x86_emulate.c to use guest_read/write_register
2255          * instead of direct ->regs accesses, can save hundred cycles
2256          * on Intel for instructions that don't read/change RSP, for
2257          * for example.
2258          */
2259         cache_all_regs(vcpu);
2260
2261         vcpu->mmio_is_write = 0;
2262         vcpu->arch.pio.string = 0;
2263
2264         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2265                 int cs_db, cs_l;
2266                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2267
2268                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2269                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2270                 vcpu->arch.emulate_ctxt.mode =
2271                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2272                         ? X86EMUL_MODE_REAL : cs_l
2273                         ? X86EMUL_MODE_PROT64 : cs_db
2274                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2275
2276                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2277
2278                 /* Reject the instructions other than VMCALL/VMMCALL when
2279                  * try to emulate invalid opcode */
2280                 c = &vcpu->arch.emulate_ctxt.decode;
2281                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2282                     (!(c->twobyte && c->b == 0x01 &&
2283                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2284                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2285                         return EMULATE_FAIL;
2286
2287                 ++vcpu->stat.insn_emulation;
2288                 if (r)  {
2289                         ++vcpu->stat.insn_emulation_fail;
2290                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2291                                 return EMULATE_DONE;
2292                         return EMULATE_FAIL;
2293                 }
2294         }
2295
2296         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2297
2298         if (vcpu->arch.pio.string)
2299                 return EMULATE_DO_MMIO;
2300
2301         if ((r || vcpu->mmio_is_write) && run) {
2302                 run->exit_reason = KVM_EXIT_MMIO;
2303                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2304                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2305                 run->mmio.len = vcpu->mmio_size;
2306                 run->mmio.is_write = vcpu->mmio_is_write;
2307         }
2308
2309         if (r) {
2310                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2311                         return EMULATE_DONE;
2312                 if (!vcpu->mmio_needed) {
2313                         kvm_report_emulation_failure(vcpu, "mmio");
2314                         return EMULATE_FAIL;
2315                 }
2316                 return EMULATE_DO_MMIO;
2317         }
2318
2319         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2320
2321         if (vcpu->mmio_is_write) {
2322                 vcpu->mmio_needed = 0;
2323                 return EMULATE_DO_MMIO;
2324         }
2325
2326         return EMULATE_DONE;
2327 }
2328 EXPORT_SYMBOL_GPL(emulate_instruction);
2329
2330 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2331 {
2332         int i;
2333
2334         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2335                 if (vcpu->arch.pio.guest_pages[i]) {
2336                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2337                         vcpu->arch.pio.guest_pages[i] = NULL;
2338                 }
2339 }
2340
2341 static int pio_copy_data(struct kvm_vcpu *vcpu)
2342 {
2343         void *p = vcpu->arch.pio_data;
2344         void *q;
2345         unsigned bytes;
2346         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2347
2348         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2349                  PAGE_KERNEL);
2350         if (!q) {
2351                 free_pio_guest_pages(vcpu);
2352                 return -ENOMEM;
2353         }
2354         q += vcpu->arch.pio.guest_page_offset;
2355         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2356         if (vcpu->arch.pio.in)
2357                 memcpy(q, p, bytes);
2358         else
2359                 memcpy(p, q, bytes);
2360         q -= vcpu->arch.pio.guest_page_offset;
2361         vunmap(q);
2362         free_pio_guest_pages(vcpu);
2363         return 0;
2364 }
2365
2366 int complete_pio(struct kvm_vcpu *vcpu)
2367 {
2368         struct kvm_pio_request *io = &vcpu->arch.pio;
2369         long delta;
2370         int r;
2371         unsigned long val;
2372
2373         if (!io->string) {
2374                 if (io->in) {
2375                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2376                         memcpy(&val, vcpu->arch.pio_data, io->size);
2377                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2378                 }
2379         } else {
2380                 if (io->in) {
2381                         r = pio_copy_data(vcpu);
2382                         if (r)
2383                                 return r;
2384                 }
2385
2386                 delta = 1;
2387                 if (io->rep) {
2388                         delta *= io->cur_count;
2389                         /*
2390                          * The size of the register should really depend on
2391                          * current address size.
2392                          */
2393                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2394                         val -= delta;
2395                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2396                 }
2397                 if (io->down)
2398                         delta = -delta;
2399                 delta *= io->size;
2400                 if (io->in) {
2401                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2402                         val += delta;
2403                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2404                 } else {
2405                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2406                         val += delta;
2407                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2408                 }
2409         }
2410
2411         io->count -= io->cur_count;
2412         io->cur_count = 0;
2413
2414         return 0;
2415 }
2416
2417 static void kernel_pio(struct kvm_io_device *pio_dev,
2418                        struct kvm_vcpu *vcpu,
2419                        void *pd)
2420 {
2421         /* TODO: String I/O for in kernel device */
2422
2423         mutex_lock(&vcpu->kvm->lock);
2424         if (vcpu->arch.pio.in)
2425                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2426                                   vcpu->arch.pio.size,
2427                                   pd);
2428         else
2429                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2430                                    vcpu->arch.pio.size,
2431                                    pd);
2432         mutex_unlock(&vcpu->kvm->lock);
2433 }
2434
2435 static void pio_string_write(struct kvm_io_device *pio_dev,
2436                              struct kvm_vcpu *vcpu)
2437 {
2438         struct kvm_pio_request *io = &vcpu->arch.pio;
2439         void *pd = vcpu->arch.pio_data;
2440         int i;
2441
2442         mutex_lock(&vcpu->kvm->lock);
2443         for (i = 0; i < io->cur_count; i++) {
2444                 kvm_iodevice_write(pio_dev, io->port,
2445                                    io->size,
2446                                    pd);
2447                 pd += io->size;
2448         }
2449         mutex_unlock(&vcpu->kvm->lock);
2450 }
2451
2452 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2453                                                gpa_t addr, int len,
2454                                                int is_write)
2455 {
2456         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2457 }
2458
2459 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2460                   int size, unsigned port)
2461 {
2462         struct kvm_io_device *pio_dev;
2463         unsigned long val;
2464
2465         vcpu->run->exit_reason = KVM_EXIT_IO;
2466         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2467         vcpu->run->io.size = vcpu->arch.pio.size = size;
2468         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2469         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2470         vcpu->run->io.port = vcpu->arch.pio.port = port;
2471         vcpu->arch.pio.in = in;
2472         vcpu->arch.pio.string = 0;
2473         vcpu->arch.pio.down = 0;
2474         vcpu->arch.pio.guest_page_offset = 0;
2475         vcpu->arch.pio.rep = 0;
2476
2477         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2478                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2479                             handler);
2480         else
2481                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2482                             handler);
2483
2484         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2485         memcpy(vcpu->arch.pio_data, &val, 4);
2486
2487         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2488         if (pio_dev) {
2489                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2490                 complete_pio(vcpu);
2491                 return 1;
2492         }
2493         return 0;
2494 }
2495 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2496
2497 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2498                   int size, unsigned long count, int down,
2499                   gva_t address, int rep, unsigned port)
2500 {
2501         unsigned now, in_page;
2502         int i, ret = 0;
2503         int nr_pages = 1;
2504         struct page *page;
2505         struct kvm_io_device *pio_dev;
2506
2507         vcpu->run->exit_reason = KVM_EXIT_IO;
2508         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2509         vcpu->run->io.size = vcpu->arch.pio.size = size;
2510         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2511         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2512         vcpu->run->io.port = vcpu->arch.pio.port = port;
2513         vcpu->arch.pio.in = in;
2514         vcpu->arch.pio.string = 1;
2515         vcpu->arch.pio.down = down;
2516         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2517         vcpu->arch.pio.rep = rep;
2518
2519         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2520                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2521                             handler);
2522         else
2523                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2524                             handler);
2525
2526         if (!count) {
2527                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2528                 return 1;
2529         }
2530
2531         if (!down)
2532                 in_page = PAGE_SIZE - offset_in_page(address);
2533         else
2534                 in_page = offset_in_page(address) + size;
2535         now = min(count, (unsigned long)in_page / size);
2536         if (!now) {
2537                 /*
2538                  * String I/O straddles page boundary.  Pin two guest pages
2539                  * so that we satisfy atomicity constraints.  Do just one
2540                  * transaction to avoid complexity.
2541                  */
2542                 nr_pages = 2;
2543                 now = 1;
2544         }
2545         if (down) {
2546                 /*
2547                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2548                  */
2549                 pr_unimpl(vcpu, "guest string pio down\n");
2550                 kvm_inject_gp(vcpu, 0);
2551                 return 1;
2552         }
2553         vcpu->run->io.count = now;
2554         vcpu->arch.pio.cur_count = now;
2555
2556         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2557                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2558
2559         for (i = 0; i < nr_pages; ++i) {
2560                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2561                 vcpu->arch.pio.guest_pages[i] = page;
2562                 if (!page) {
2563                         kvm_inject_gp(vcpu, 0);
2564                         free_pio_guest_pages(vcpu);
2565                         return 1;
2566                 }
2567         }
2568
2569         pio_dev = vcpu_find_pio_dev(vcpu, port,
2570                                     vcpu->arch.pio.cur_count,
2571                                     !vcpu->arch.pio.in);
2572         if (!vcpu->arch.pio.in) {
2573                 /* string PIO write */
2574                 ret = pio_copy_data(vcpu);
2575                 if (ret >= 0 && pio_dev) {
2576                         pio_string_write(pio_dev, vcpu);
2577                         complete_pio(vcpu);
2578                         if (vcpu->arch.pio.count == 0)
2579                                 ret = 1;
2580                 }
2581         } else if (pio_dev)
2582                 pr_unimpl(vcpu, "no string pio read support yet, "
2583                        "port %x size %d count %ld\n",
2584                         port, size, count);
2585
2586         return ret;
2587 }
2588 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2589
2590 int kvm_arch_init(void *opaque)
2591 {
2592         int r;
2593         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2594
2595         if (kvm_x86_ops) {
2596                 printk(KERN_ERR "kvm: already loaded the other module\n");
2597                 r = -EEXIST;
2598                 goto out;
2599         }
2600
2601         if (!ops->cpu_has_kvm_support()) {
2602                 printk(KERN_ERR "kvm: no hardware support\n");
2603                 r = -EOPNOTSUPP;
2604                 goto out;
2605         }
2606         if (ops->disabled_by_bios()) {
2607                 printk(KERN_ERR "kvm: disabled by bios\n");
2608                 r = -EOPNOTSUPP;
2609                 goto out;
2610         }
2611
2612         r = kvm_mmu_module_init();
2613         if (r)
2614                 goto out;
2615
2616         kvm_init_msr_list();
2617
2618         kvm_x86_ops = ops;
2619         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2620         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2621         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2622                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2623         return 0;
2624
2625 out:
2626         return r;
2627 }
2628
2629 void kvm_arch_exit(void)
2630 {
2631         kvm_x86_ops = NULL;
2632         kvm_mmu_module_exit();
2633 }
2634
2635 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2636 {
2637         ++vcpu->stat.halt_exits;
2638         KVMTRACE_0D(HLT, vcpu, handler);
2639         if (irqchip_in_kernel(vcpu->kvm)) {
2640                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2641                 return 1;
2642         } else {
2643                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2644                 return 0;
2645         }
2646 }
2647 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2648
2649 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2650                            unsigned long a1)
2651 {
2652         if (is_long_mode(vcpu))
2653                 return a0;
2654         else
2655                 return a0 | ((gpa_t)a1 << 32);
2656 }
2657
2658 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2659 {
2660         unsigned long nr, a0, a1, a2, a3, ret;
2661         int r = 1;
2662
2663         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2664         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2665         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2666         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2667         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2668
2669         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2670
2671         if (!is_long_mode(vcpu)) {
2672                 nr &= 0xFFFFFFFF;
2673                 a0 &= 0xFFFFFFFF;
2674                 a1 &= 0xFFFFFFFF;
2675                 a2 &= 0xFFFFFFFF;
2676                 a3 &= 0xFFFFFFFF;
2677         }
2678
2679         switch (nr) {
2680         case KVM_HC_VAPIC_POLL_IRQ:
2681                 ret = 0;
2682                 break;
2683         case KVM_HC_MMU_OP:
2684                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2685                 break;
2686         default:
2687                 ret = -KVM_ENOSYS;
2688                 break;
2689         }
2690         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2691         ++vcpu->stat.hypercalls;
2692         return r;
2693 }
2694 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2695
2696 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2697 {
2698         char instruction[3];
2699         int ret = 0;
2700         unsigned long rip = kvm_rip_read(vcpu);
2701
2702
2703         /*
2704          * Blow out the MMU to ensure that no other VCPU has an active mapping
2705          * to ensure that the updated hypercall appears atomically across all
2706          * VCPUs.
2707          */
2708         kvm_mmu_zap_all(vcpu->kvm);
2709
2710         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2711         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2712             != X86EMUL_CONTINUE)
2713                 ret = -EFAULT;
2714
2715         return ret;
2716 }
2717
2718 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2719 {
2720         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2721 }
2722
2723 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2724 {
2725         struct descriptor_table dt = { limit, base };
2726
2727         kvm_x86_ops->set_gdt(vcpu, &dt);
2728 }
2729
2730 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2731 {
2732         struct descriptor_table dt = { limit, base };
2733
2734         kvm_x86_ops->set_idt(vcpu, &dt);
2735 }
2736
2737 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2738                    unsigned long *rflags)
2739 {
2740         kvm_lmsw(vcpu, msw);
2741         *rflags = kvm_x86_ops->get_rflags(vcpu);
2742 }
2743
2744 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2745 {
2746         unsigned long value;
2747
2748         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2749         switch (cr) {
2750         case 0:
2751                 value = vcpu->arch.cr0;
2752                 break;
2753         case 2:
2754                 value = vcpu->arch.cr2;
2755                 break;
2756         case 3:
2757                 value = vcpu->arch.cr3;
2758                 break;
2759         case 4:
2760                 value = vcpu->arch.cr4;
2761                 break;
2762         case 8:
2763                 value = kvm_get_cr8(vcpu);
2764                 break;
2765         default:
2766                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2767                 return 0;
2768         }
2769         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2770                     (u32)((u64)value >> 32), handler);
2771
2772         return value;
2773 }
2774
2775 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2776                      unsigned long *rflags)
2777 {
2778         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2779                     (u32)((u64)val >> 32), handler);
2780
2781         switch (cr) {
2782         case 0:
2783                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2784                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2785                 break;
2786         case 2:
2787                 vcpu->arch.cr2 = val;
2788                 break;
2789         case 3:
2790                 kvm_set_cr3(vcpu, val);
2791                 break;
2792         case 4:
2793                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2794                 break;
2795         case 8:
2796                 kvm_set_cr8(vcpu, val & 0xfUL);
2797                 break;
2798         default:
2799                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2800         }
2801 }
2802
2803 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2804 {
2805         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2806         int j, nent = vcpu->arch.cpuid_nent;
2807
2808         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2809         /* when no next entry is found, the current entry[i] is reselected */
2810         for (j = i + 1; ; j = (j + 1) % nent) {
2811                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2812                 if (ej->function == e->function) {
2813                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2814                         return j;
2815                 }
2816         }
2817         return 0; /* silence gcc, even though control never reaches here */
2818 }
2819
2820 /* find an entry with matching function, matching index (if needed), and that
2821  * should be read next (if it's stateful) */
2822 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2823         u32 function, u32 index)
2824 {
2825         if (e->function != function)
2826                 return 0;
2827         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2828                 return 0;
2829         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2830                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2831                 return 0;
2832         return 1;
2833 }
2834
2835 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2836 {
2837         int i;
2838         u32 function, index;
2839         struct kvm_cpuid_entry2 *e, *best;
2840
2841         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2842         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2843         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2844         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2845         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2846         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2847         best = NULL;
2848         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2849                 e = &vcpu->arch.cpuid_entries[i];
2850                 if (is_matching_cpuid_entry(e, function, index)) {
2851                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2852                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2853                         best = e;
2854                         break;
2855                 }
2856                 /*
2857                  * Both basic or both extended?
2858                  */
2859                 if (((e->function ^ function) & 0x80000000) == 0)
2860                         if (!best || e->function > best->function)
2861                                 best = e;
2862         }
2863         if (best) {
2864                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2865                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2866                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2867                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2868         }
2869         kvm_x86_ops->skip_emulated_instruction(vcpu);
2870         KVMTRACE_5D(CPUID, vcpu, function,
2871                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2872                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2873                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2874                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2875 }
2876 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2877
2878 /*
2879  * Check if userspace requested an interrupt window, and that the
2880  * interrupt window is open.
2881  *
2882  * No need to exit to userspace if we already have an interrupt queued.
2883  */
2884 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2885                                           struct kvm_run *kvm_run)
2886 {
2887         return (!vcpu->arch.irq_summary &&
2888                 kvm_run->request_interrupt_window &&
2889                 vcpu->arch.interrupt_window_open &&
2890                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2891 }
2892
2893 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2894                               struct kvm_run *kvm_run)
2895 {
2896         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2897         kvm_run->cr8 = kvm_get_cr8(vcpu);
2898         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2899         if (irqchip_in_kernel(vcpu->kvm))
2900                 kvm_run->ready_for_interrupt_injection = 1;
2901         else
2902                 kvm_run->ready_for_interrupt_injection =
2903                                         (vcpu->arch.interrupt_window_open &&
2904                                          vcpu->arch.irq_summary == 0);
2905 }
2906
2907 static void vapic_enter(struct kvm_vcpu *vcpu)
2908 {
2909         struct kvm_lapic *apic = vcpu->arch.apic;
2910         struct page *page;
2911
2912         if (!apic || !apic->vapic_addr)
2913                 return;
2914
2915         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2916
2917         vcpu->arch.apic->vapic_page = page;
2918 }
2919
2920 static void vapic_exit(struct kvm_vcpu *vcpu)
2921 {
2922         struct kvm_lapic *apic = vcpu->arch.apic;
2923
2924         if (!apic || !apic->vapic_addr)
2925                 return;
2926
2927         down_read(&vcpu->kvm->slots_lock);
2928         kvm_release_page_dirty(apic->vapic_page);
2929         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2930         up_read(&vcpu->kvm->slots_lock);
2931 }
2932
2933 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2934 {
2935         int r;
2936
2937         if (vcpu->requests)
2938                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2939                         kvm_mmu_unload(vcpu);
2940
2941         r = kvm_mmu_reload(vcpu);
2942         if (unlikely(r))
2943                 goto out;
2944
2945         if (vcpu->requests) {
2946                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2947                         __kvm_migrate_timers(vcpu);
2948                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2949                         kvm_mmu_sync_roots(vcpu);
2950                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2951                         kvm_x86_ops->tlb_flush(vcpu);
2952                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2953                                        &vcpu->requests)) {
2954                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2955                         r = 0;
2956                         goto out;
2957                 }
2958                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2959                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2960                         r = 0;
2961                         goto out;
2962                 }
2963         }
2964
2965         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2966         kvm_inject_pending_timer_irqs(vcpu);
2967
2968         preempt_disable();
2969
2970         kvm_x86_ops->prepare_guest_switch(vcpu);
2971         kvm_load_guest_fpu(vcpu);
2972
2973         local_irq_disable();
2974
2975         if (vcpu->requests || need_resched() || signal_pending(current)) {
2976                 local_irq_enable();
2977                 preempt_enable();
2978                 r = 1;
2979                 goto out;
2980         }
2981
2982         if (vcpu->guest_debug.enabled)
2983                 kvm_x86_ops->guest_debug_pre(vcpu);
2984
2985         vcpu->guest_mode = 1;
2986         /*
2987          * Make sure that guest_mode assignment won't happen after
2988          * testing the pending IRQ vector bitmap.
2989          */
2990         smp_wmb();
2991
2992         if (vcpu->arch.exception.pending)
2993                 __queue_exception(vcpu);
2994         else if (irqchip_in_kernel(vcpu->kvm))
2995                 kvm_x86_ops->inject_pending_irq(vcpu);
2996         else
2997                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2998
2999         kvm_lapic_sync_to_vapic(vcpu);
3000
3001         up_read(&vcpu->kvm->slots_lock);
3002
3003         kvm_guest_enter();
3004
3005
3006         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3007         kvm_x86_ops->run(vcpu, kvm_run);
3008
3009         vcpu->guest_mode = 0;
3010         local_irq_enable();
3011
3012         ++vcpu->stat.exits;
3013
3014         /*
3015          * We must have an instruction between local_irq_enable() and
3016          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3017          * the interrupt shadow.  The stat.exits increment will do nicely.
3018          * But we need to prevent reordering, hence this barrier():
3019          */
3020         barrier();
3021
3022         kvm_guest_exit();
3023
3024         preempt_enable();
3025
3026         down_read(&vcpu->kvm->slots_lock);
3027
3028         /*
3029          * Profile KVM exit RIPs:
3030          */
3031         if (unlikely(prof_on == KVM_PROFILING)) {
3032                 unsigned long rip = kvm_rip_read(vcpu);
3033                 profile_hit(KVM_PROFILING, (void *)rip);
3034         }
3035
3036         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3037                 vcpu->arch.exception.pending = false;
3038
3039         kvm_lapic_sync_from_vapic(vcpu);
3040
3041         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3042 out:
3043         return r;
3044 }
3045
3046 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3047 {
3048         int r;
3049
3050         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3051                 pr_debug("vcpu %d received sipi with vector # %x\n",
3052                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3053                 kvm_lapic_reset(vcpu);
3054                 r = kvm_arch_vcpu_reset(vcpu);
3055                 if (r)
3056                         return r;
3057                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3058         }
3059
3060         down_read(&vcpu->kvm->slots_lock);
3061         vapic_enter(vcpu);
3062
3063         r = 1;
3064         while (r > 0) {
3065                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3066                         r = vcpu_enter_guest(vcpu, kvm_run);
3067                 else {
3068                         up_read(&vcpu->kvm->slots_lock);
3069                         kvm_vcpu_block(vcpu);
3070                         down_read(&vcpu->kvm->slots_lock);
3071                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3072                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3073                                         vcpu->arch.mp_state =
3074                                                         KVM_MP_STATE_RUNNABLE;
3075                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3076                                 r = -EINTR;
3077                 }
3078
3079                 if (r > 0) {
3080                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3081                                 r = -EINTR;
3082                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3083                                 ++vcpu->stat.request_irq_exits;
3084                         }
3085                         if (signal_pending(current)) {
3086                                 r = -EINTR;
3087                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3088                                 ++vcpu->stat.signal_exits;
3089                         }
3090                         if (need_resched()) {
3091                                 up_read(&vcpu->kvm->slots_lock);
3092                                 kvm_resched(vcpu);
3093                                 down_read(&vcpu->kvm->slots_lock);
3094                         }
3095                 }
3096         }
3097
3098         up_read(&vcpu->kvm->slots_lock);
3099         post_kvm_run_save(vcpu, kvm_run);
3100
3101         vapic_exit(vcpu);
3102
3103         return r;
3104 }
3105
3106 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3107 {
3108         int r;
3109         sigset_t sigsaved;
3110
3111         vcpu_load(vcpu);
3112
3113         if (vcpu->sigset_active)
3114                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3115
3116         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3117                 kvm_vcpu_block(vcpu);
3118                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3119                 r = -EAGAIN;
3120                 goto out;
3121         }
3122
3123         /* re-sync apic's tpr */
3124         if (!irqchip_in_kernel(vcpu->kvm))
3125                 kvm_set_cr8(vcpu, kvm_run->cr8);
3126
3127         if (vcpu->arch.pio.cur_count) {
3128                 r = complete_pio(vcpu);
3129                 if (r)
3130                         goto out;
3131         }
3132 #if CONFIG_HAS_IOMEM
3133         if (vcpu->mmio_needed) {
3134                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3135                 vcpu->mmio_read_completed = 1;
3136                 vcpu->mmio_needed = 0;
3137
3138                 down_read(&vcpu->kvm->slots_lock);
3139                 r = emulate_instruction(vcpu, kvm_run,
3140                                         vcpu->arch.mmio_fault_cr2, 0,
3141                                         EMULTYPE_NO_DECODE);
3142                 up_read(&vcpu->kvm->slots_lock);
3143                 if (r == EMULATE_DO_MMIO) {
3144                         /*
3145                          * Read-modify-write.  Back to userspace.
3146                          */
3147                         r = 0;
3148                         goto out;
3149                 }
3150         }
3151 #endif
3152         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3153                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3154                                      kvm_run->hypercall.ret);
3155
3156         r = __vcpu_run(vcpu, kvm_run);
3157
3158 out:
3159         if (vcpu->sigset_active)
3160                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3161
3162         vcpu_put(vcpu);
3163         return r;
3164 }
3165
3166 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3167 {
3168         vcpu_load(vcpu);
3169
3170         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3171         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3172         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3173         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3174         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3175         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3176         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3177         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3178 #ifdef CONFIG_X86_64
3179         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3180         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3181         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3182         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3183         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3184         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3185         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3186         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3187 #endif
3188
3189         regs->rip = kvm_rip_read(vcpu);
3190         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3191
3192         /*
3193          * Don't leak debug flags in case they were set for guest debugging
3194          */
3195         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3196                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3197
3198         vcpu_put(vcpu);
3199
3200         return 0;
3201 }
3202
3203 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3204 {
3205         vcpu_load(vcpu);
3206
3207         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3208         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3209         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3210         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3211         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3212         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3213         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3214         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3215 #ifdef CONFIG_X86_64
3216         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3217         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3218         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3219         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3220         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3221         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3222         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3223         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3224
3225 #endif
3226
3227         kvm_rip_write(vcpu, regs->rip);
3228         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3229
3230
3231         vcpu->arch.exception.pending = false;
3232
3233         vcpu_put(vcpu);
3234
3235         return 0;
3236 }
3237
3238 void kvm_get_segment(struct kvm_vcpu *vcpu,
3239                      struct kvm_segment *var, int seg)
3240 {
3241         kvm_x86_ops->get_segment(vcpu, var, seg);
3242 }
3243
3244 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3245 {
3246         struct kvm_segment cs;
3247
3248         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3249         *db = cs.db;
3250         *l = cs.l;
3251 }
3252 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3253
3254 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3255                                   struct kvm_sregs *sregs)
3256 {
3257         struct descriptor_table dt;
3258         int pending_vec;
3259
3260         vcpu_load(vcpu);
3261
3262         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3263         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3264         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3265         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3266         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3267         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3268
3269         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3270         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3271
3272         kvm_x86_ops->get_idt(vcpu, &dt);
3273         sregs->idt.limit = dt.limit;
3274         sregs->idt.base = dt.base;
3275         kvm_x86_ops->get_gdt(vcpu, &dt);
3276         sregs->gdt.limit = dt.limit;
3277         sregs->gdt.base = dt.base;
3278
3279         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3280         sregs->cr0 = vcpu->arch.cr0;
3281         sregs->cr2 = vcpu->arch.cr2;
3282         sregs->cr3 = vcpu->arch.cr3;
3283         sregs->cr4 = vcpu->arch.cr4;
3284         sregs->cr8 = kvm_get_cr8(vcpu);
3285         sregs->efer = vcpu->arch.shadow_efer;
3286         sregs->apic_base = kvm_get_apic_base(vcpu);
3287
3288         if (irqchip_in_kernel(vcpu->kvm)) {
3289                 memset(sregs->interrupt_bitmap, 0,
3290                        sizeof sregs->interrupt_bitmap);
3291                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3292                 if (pending_vec >= 0)
3293                         set_bit(pending_vec,
3294                                 (unsigned long *)sregs->interrupt_bitmap);
3295         } else
3296                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3297                        sizeof sregs->interrupt_bitmap);
3298
3299         vcpu_put(vcpu);
3300
3301         return 0;
3302 }
3303
3304 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3305                                     struct kvm_mp_state *mp_state)
3306 {
3307         vcpu_load(vcpu);
3308         mp_state->mp_state = vcpu->arch.mp_state;
3309         vcpu_put(vcpu);
3310         return 0;
3311 }
3312
3313 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3314                                     struct kvm_mp_state *mp_state)
3315 {
3316         vcpu_load(vcpu);
3317         vcpu->arch.mp_state = mp_state->mp_state;
3318         vcpu_put(vcpu);
3319         return 0;
3320 }
3321
3322 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3323                         struct kvm_segment *var, int seg)
3324 {
3325         kvm_x86_ops->set_segment(vcpu, var, seg);
3326 }
3327
3328 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3329                                    struct kvm_segment *kvm_desct)
3330 {
3331         kvm_desct->base = seg_desc->base0;
3332         kvm_desct->base |= seg_desc->base1 << 16;
3333         kvm_desct->base |= seg_desc->base2 << 24;
3334         kvm_desct->limit = seg_desc->limit0;
3335         kvm_desct->limit |= seg_desc->limit << 16;
3336         if (seg_desc->g) {
3337                 kvm_desct->limit <<= 12;
3338                 kvm_desct->limit |= 0xfff;
3339         }
3340         kvm_desct->selector = selector;
3341         kvm_desct->type = seg_desc->type;
3342         kvm_desct->present = seg_desc->p;
3343         kvm_desct->dpl = seg_desc->dpl;
3344         kvm_desct->db = seg_desc->d;
3345         kvm_desct->s = seg_desc->s;
3346         kvm_desct->l = seg_desc->l;
3347         kvm_desct->g = seg_desc->g;
3348         kvm_desct->avl = seg_desc->avl;
3349         if (!selector)
3350                 kvm_desct->unusable = 1;
3351         else
3352                 kvm_desct->unusable = 0;
3353         kvm_desct->padding = 0;
3354 }
3355
3356 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3357                                           u16 selector,
3358                                           struct descriptor_table *dtable)
3359 {
3360         if (selector & 1 << 2) {
3361                 struct kvm_segment kvm_seg;
3362
3363                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3364
3365                 if (kvm_seg.unusable)
3366                         dtable->limit = 0;
3367                 else
3368                         dtable->limit = kvm_seg.limit;
3369                 dtable->base = kvm_seg.base;
3370         }
3371         else
3372                 kvm_x86_ops->get_gdt(vcpu, dtable);
3373 }
3374
3375 /* allowed just for 8 bytes segments */
3376 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3377                                          struct desc_struct *seg_desc)
3378 {
3379         gpa_t gpa;
3380         struct descriptor_table dtable;
3381         u16 index = selector >> 3;
3382
3383         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3384
3385         if (dtable.limit < index * 8 + 7) {
3386                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3387                 return 1;
3388         }
3389         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3390         gpa += index * 8;
3391         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3392 }
3393
3394 /* allowed just for 8 bytes segments */
3395 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3396                                          struct desc_struct *seg_desc)
3397 {
3398         gpa_t gpa;
3399         struct descriptor_table dtable;
3400         u16 index = selector >> 3;
3401
3402         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3403
3404         if (dtable.limit < index * 8 + 7)
3405                 return 1;
3406         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3407         gpa += index * 8;
3408         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3409 }
3410
3411 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3412                              struct desc_struct *seg_desc)
3413 {
3414         u32 base_addr;
3415
3416         base_addr = seg_desc->base0;
3417         base_addr |= (seg_desc->base1 << 16);
3418         base_addr |= (seg_desc->base2 << 24);
3419
3420         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3421 }
3422
3423 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3424 {
3425         struct kvm_segment kvm_seg;
3426
3427         kvm_get_segment(vcpu, &kvm_seg, seg);
3428         return kvm_seg.selector;
3429 }
3430
3431 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3432                                                 u16 selector,
3433                                                 struct kvm_segment *kvm_seg)
3434 {
3435         struct desc_struct seg_desc;
3436
3437         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3438                 return 1;
3439         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3440         return 0;
3441 }
3442
3443 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3444 {
3445         struct kvm_segment segvar = {
3446                 .base = selector << 4,
3447                 .limit = 0xffff,
3448                 .selector = selector,
3449                 .type = 3,
3450                 .present = 1,
3451                 .dpl = 3,
3452                 .db = 0,
3453                 .s = 1,
3454                 .l = 0,
3455                 .g = 0,
3456                 .avl = 0,
3457                 .unusable = 0,
3458         };
3459         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3460         return 0;
3461 }
3462
3463 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3464                                 int type_bits, int seg)
3465 {
3466         struct kvm_segment kvm_seg;
3467
3468         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3469                 return kvm_load_realmode_segment(vcpu, selector, seg);
3470         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3471                 return 1;
3472         kvm_seg.type |= type_bits;
3473
3474         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3475             seg != VCPU_SREG_LDTR)
3476                 if (!kvm_seg.s)
3477                         kvm_seg.unusable = 1;
3478
3479         kvm_set_segment(vcpu, &kvm_seg, seg);
3480         return 0;
3481 }
3482
3483 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3484                                 struct tss_segment_32 *tss)
3485 {
3486         tss->cr3 = vcpu->arch.cr3;
3487         tss->eip = kvm_rip_read(vcpu);
3488         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3489         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3490         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3491         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3492         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3493         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3494         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3495         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3496         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3497         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3498         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3499         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3500         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3501         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3502         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3503         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3504         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3505 }
3506
3507 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3508                                   struct tss_segment_32 *tss)
3509 {
3510         kvm_set_cr3(vcpu, tss->cr3);
3511
3512         kvm_rip_write(vcpu, tss->eip);
3513         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3514
3515         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3516         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3517         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3518         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3519         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3520         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3521         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3522         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3523
3524         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3525                 return 1;
3526
3527         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3528                 return 1;
3529
3530         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3531                 return 1;
3532
3533         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3534                 return 1;
3535
3536         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3537                 return 1;
3538
3539         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3540                 return 1;
3541
3542         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3543                 return 1;
3544         return 0;
3545 }
3546
3547 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3548                                 struct tss_segment_16 *tss)
3549 {
3550         tss->ip = kvm_rip_read(vcpu);
3551         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3552         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3553         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3554         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3555         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3556         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3557         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3558         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3559         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3560
3561         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3562         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3563         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3564         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3565         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3566         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3567 }
3568
3569 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3570                                  struct tss_segment_16 *tss)
3571 {
3572         kvm_rip_write(vcpu, tss->ip);
3573         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3574         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3575         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3576         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3577         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3578         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3579         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3580         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3581         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3582
3583         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3584                 return 1;
3585
3586         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3587                 return 1;
3588
3589         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3590                 return 1;
3591
3592         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3593                 return 1;
3594
3595         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3596                 return 1;
3597         return 0;
3598 }
3599
3600 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3601                        u32 old_tss_base,
3602                        struct desc_struct *nseg_desc)
3603 {
3604         struct tss_segment_16 tss_segment_16;
3605         int ret = 0;
3606
3607         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3608                            sizeof tss_segment_16))
3609                 goto out;
3610
3611         save_state_to_tss16(vcpu, &tss_segment_16);
3612
3613         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3614                             sizeof tss_segment_16))
3615                 goto out;
3616
3617         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3618                            &tss_segment_16, sizeof tss_segment_16))
3619                 goto out;
3620
3621         if (load_state_from_tss16(vcpu, &tss_segment_16))
3622                 goto out;
3623
3624         ret = 1;
3625 out:
3626         return ret;
3627 }
3628
3629 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3630                        u32 old_tss_base,
3631                        struct desc_struct *nseg_desc)
3632 {
3633         struct tss_segment_32 tss_segment_32;
3634         int ret = 0;
3635
3636         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3637                            sizeof tss_segment_32))
3638                 goto out;
3639
3640         save_state_to_tss32(vcpu, &tss_segment_32);
3641
3642         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3643                             sizeof tss_segment_32))
3644                 goto out;
3645
3646         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3647                            &tss_segment_32, sizeof tss_segment_32))
3648                 goto out;
3649
3650         if (load_state_from_tss32(vcpu, &tss_segment_32))
3651                 goto out;
3652
3653         ret = 1;
3654 out:
3655         return ret;
3656 }
3657
3658 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3659 {
3660         struct kvm_segment tr_seg;
3661         struct desc_struct cseg_desc;
3662         struct desc_struct nseg_desc;
3663         int ret = 0;
3664         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3665         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3666
3667         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3668
3669         /* FIXME: Handle errors. Failure to read either TSS or their
3670          * descriptors should generate a pagefault.
3671          */
3672         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3673                 goto out;
3674
3675         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3676                 goto out;
3677
3678         if (reason != TASK_SWITCH_IRET) {
3679                 int cpl;
3680
3681                 cpl = kvm_x86_ops->get_cpl(vcpu);
3682                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3683                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3684                         return 1;
3685                 }
3686         }
3687
3688         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3689                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3690                 return 1;
3691         }
3692
3693         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3694                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3695                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3696         }
3697
3698         if (reason == TASK_SWITCH_IRET) {
3699                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3700                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3701         }
3702
3703         kvm_x86_ops->skip_emulated_instruction(vcpu);
3704
3705         if (nseg_desc.type & 8)
3706                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3707                                          &nseg_desc);
3708         else
3709                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3710                                          &nseg_desc);
3711
3712         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3713                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3714                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3715         }
3716
3717         if (reason != TASK_SWITCH_IRET) {
3718                 nseg_desc.type |= (1 << 1);
3719                 save_guest_segment_descriptor(vcpu, tss_selector,
3720                                               &nseg_desc);
3721         }
3722
3723         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3724         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3725         tr_seg.type = 11;
3726         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3727 out:
3728         return ret;
3729 }
3730 EXPORT_SYMBOL_GPL(kvm_task_switch);
3731
3732 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3733                                   struct kvm_sregs *sregs)
3734 {
3735         int mmu_reset_needed = 0;
3736         int i, pending_vec, max_bits;
3737         struct descriptor_table dt;
3738
3739         vcpu_load(vcpu);
3740
3741         dt.limit = sregs->idt.limit;
3742         dt.base = sregs->idt.base;
3743         kvm_x86_ops->set_idt(vcpu, &dt);
3744         dt.limit = sregs->gdt.limit;
3745         dt.base = sregs->gdt.base;
3746         kvm_x86_ops->set_gdt(vcpu, &dt);
3747
3748         vcpu->arch.cr2 = sregs->cr2;
3749         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3750         vcpu->arch.cr3 = sregs->cr3;
3751
3752         kvm_set_cr8(vcpu, sregs->cr8);
3753
3754         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3755         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3756         kvm_set_apic_base(vcpu, sregs->apic_base);
3757
3758         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3759
3760         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3761         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3762         vcpu->arch.cr0 = sregs->cr0;
3763
3764         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3765         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3766         if (!is_long_mode(vcpu) && is_pae(vcpu))
3767                 load_pdptrs(vcpu, vcpu->arch.cr3);
3768
3769         if (mmu_reset_needed)
3770                 kvm_mmu_reset_context(vcpu);
3771
3772         if (!irqchip_in_kernel(vcpu->kvm)) {
3773                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3774                        sizeof vcpu->arch.irq_pending);
3775                 vcpu->arch.irq_summary = 0;
3776                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3777                         if (vcpu->arch.irq_pending[i])
3778                                 __set_bit(i, &vcpu->arch.irq_summary);
3779         } else {
3780                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3781                 pending_vec = find_first_bit(
3782                         (const unsigned long *)sregs->interrupt_bitmap,
3783                         max_bits);
3784                 /* Only pending external irq is handled here */
3785                 if (pending_vec < max_bits) {
3786                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3787                         pr_debug("Set back pending irq %d\n",
3788                                  pending_vec);
3789                 }
3790                 kvm_pic_clear_isr_ack(vcpu->kvm);
3791         }
3792
3793         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3794         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3795         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3796         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3797         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3798         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3799
3800         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3801         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3802
3803         /* Older userspace won't unhalt the vcpu on reset. */
3804         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3805             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3806             !(vcpu->arch.cr0 & X86_CR0_PE))
3807                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3808
3809         vcpu_put(vcpu);
3810
3811         return 0;
3812 }
3813
3814 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3815                                     struct kvm_debug_guest *dbg)
3816 {
3817         int r;
3818
3819         vcpu_load(vcpu);
3820
3821         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3822
3823         vcpu_put(vcpu);
3824
3825         return r;
3826 }
3827
3828 /*
3829  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3830  * we have asm/x86/processor.h
3831  */
3832 struct fxsave {
3833         u16     cwd;
3834         u16     swd;
3835         u16     twd;
3836         u16     fop;
3837         u64     rip;
3838         u64     rdp;
3839         u32     mxcsr;
3840         u32     mxcsr_mask;
3841         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3842 #ifdef CONFIG_X86_64
3843         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3844 #else
3845         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3846 #endif
3847 };
3848
3849 /*
3850  * Translate a guest virtual address to a guest physical address.
3851  */
3852 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3853                                     struct kvm_translation *tr)
3854 {
3855         unsigned long vaddr = tr->linear_address;
3856         gpa_t gpa;
3857
3858         vcpu_load(vcpu);
3859         down_read(&vcpu->kvm->slots_lock);
3860         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3861         up_read(&vcpu->kvm->slots_lock);
3862         tr->physical_address = gpa;
3863         tr->valid = gpa != UNMAPPED_GVA;
3864         tr->writeable = 1;
3865         tr->usermode = 0;
3866         vcpu_put(vcpu);
3867
3868         return 0;
3869 }
3870
3871 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3872 {
3873         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3874
3875         vcpu_load(vcpu);
3876
3877         memcpy(fpu->fpr, fxsave->st_space, 128);
3878         fpu->fcw = fxsave->cwd;
3879         fpu->fsw = fxsave->swd;
3880         fpu->ftwx = fxsave->twd;
3881         fpu->last_opcode = fxsave->fop;
3882         fpu->last_ip = fxsave->rip;
3883         fpu->last_dp = fxsave->rdp;
3884         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3885
3886         vcpu_put(vcpu);
3887
3888         return 0;
3889 }
3890
3891 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3892 {
3893         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3894
3895         vcpu_load(vcpu);
3896
3897         memcpy(fxsave->st_space, fpu->fpr, 128);
3898         fxsave->cwd = fpu->fcw;
3899         fxsave->swd = fpu->fsw;
3900         fxsave->twd = fpu->ftwx;
3901         fxsave->fop = fpu->last_opcode;
3902         fxsave->rip = fpu->last_ip;
3903         fxsave->rdp = fpu->last_dp;
3904         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3905
3906         vcpu_put(vcpu);
3907
3908         return 0;
3909 }
3910
3911 void fx_init(struct kvm_vcpu *vcpu)
3912 {
3913         unsigned after_mxcsr_mask;
3914
3915         /*
3916          * Touch the fpu the first time in non atomic context as if
3917          * this is the first fpu instruction the exception handler
3918          * will fire before the instruction returns and it'll have to
3919          * allocate ram with GFP_KERNEL.
3920          */
3921         if (!used_math())
3922                 kvm_fx_save(&vcpu->arch.host_fx_image);
3923
3924         /* Initialize guest FPU by resetting ours and saving into guest's */
3925         preempt_disable();
3926         kvm_fx_save(&vcpu->arch.host_fx_image);
3927         kvm_fx_finit();
3928         kvm_fx_save(&vcpu->arch.guest_fx_image);
3929         kvm_fx_restore(&vcpu->arch.host_fx_image);
3930         preempt_enable();
3931
3932         vcpu->arch.cr0 |= X86_CR0_ET;
3933         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3934         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3935         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3936                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3937 }
3938 EXPORT_SYMBOL_GPL(fx_init);
3939
3940 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3941 {
3942         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3943                 return;
3944
3945         vcpu->guest_fpu_loaded = 1;
3946         kvm_fx_save(&vcpu->arch.host_fx_image);
3947         kvm_fx_restore(&vcpu->arch.guest_fx_image);
3948 }
3949 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3950
3951 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3952 {
3953         if (!vcpu->guest_fpu_loaded)
3954                 return;
3955
3956         vcpu->guest_fpu_loaded = 0;
3957         kvm_fx_save(&vcpu->arch.guest_fx_image);
3958         kvm_fx_restore(&vcpu->arch.host_fx_image);
3959         ++vcpu->stat.fpu_reload;
3960 }
3961 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3962
3963 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3964 {
3965         kvm_x86_ops->vcpu_free(vcpu);
3966 }
3967
3968 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3969                                                 unsigned int id)
3970 {
3971         return kvm_x86_ops->vcpu_create(kvm, id);
3972 }
3973
3974 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3975 {
3976         int r;
3977
3978         /* We do fxsave: this must be aligned. */
3979         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3980
3981         vcpu->arch.mtrr_state.have_fixed = 1;
3982         vcpu_load(vcpu);
3983         r = kvm_arch_vcpu_reset(vcpu);
3984         if (r == 0)
3985                 r = kvm_mmu_setup(vcpu);
3986         vcpu_put(vcpu);
3987         if (r < 0)
3988                 goto free_vcpu;
3989
3990         return 0;
3991 free_vcpu:
3992         kvm_x86_ops->vcpu_free(vcpu);
3993         return r;
3994 }
3995
3996 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3997 {
3998         vcpu_load(vcpu);
3999         kvm_mmu_unload(vcpu);
4000         vcpu_put(vcpu);
4001
4002         kvm_x86_ops->vcpu_free(vcpu);
4003 }
4004
4005 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4006 {
4007         vcpu->arch.nmi_pending = false;
4008         vcpu->arch.nmi_injected = false;
4009
4010         return kvm_x86_ops->vcpu_reset(vcpu);
4011 }
4012
4013 void kvm_arch_hardware_enable(void *garbage)
4014 {
4015         kvm_x86_ops->hardware_enable(garbage);
4016 }
4017
4018 void kvm_arch_hardware_disable(void *garbage)
4019 {
4020         kvm_x86_ops->hardware_disable(garbage);
4021 }
4022
4023 int kvm_arch_hardware_setup(void)
4024 {
4025         return kvm_x86_ops->hardware_setup();
4026 }
4027
4028 void kvm_arch_hardware_unsetup(void)
4029 {
4030         kvm_x86_ops->hardware_unsetup();
4031 }
4032
4033 void kvm_arch_check_processor_compat(void *rtn)
4034 {
4035         kvm_x86_ops->check_processor_compatibility(rtn);
4036 }
4037
4038 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4039 {
4040         struct page *page;
4041         struct kvm *kvm;
4042         int r;
4043
4044         BUG_ON(vcpu->kvm == NULL);
4045         kvm = vcpu->kvm;
4046
4047         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4048         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4049                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4050         else
4051                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4052
4053         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4054         if (!page) {
4055                 r = -ENOMEM;
4056                 goto fail;
4057         }
4058         vcpu->arch.pio_data = page_address(page);
4059
4060         r = kvm_mmu_create(vcpu);
4061         if (r < 0)
4062                 goto fail_free_pio_data;
4063
4064         if (irqchip_in_kernel(kvm)) {
4065                 r = kvm_create_lapic(vcpu);
4066                 if (r < 0)
4067                         goto fail_mmu_destroy;
4068         }
4069
4070         return 0;
4071
4072 fail_mmu_destroy:
4073         kvm_mmu_destroy(vcpu);
4074 fail_free_pio_data:
4075         free_page((unsigned long)vcpu->arch.pio_data);
4076 fail:
4077         return r;
4078 }
4079
4080 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4081 {
4082         kvm_free_lapic(vcpu);
4083         down_read(&vcpu->kvm->slots_lock);
4084         kvm_mmu_destroy(vcpu);
4085         up_read(&vcpu->kvm->slots_lock);
4086         free_page((unsigned long)vcpu->arch.pio_data);
4087 }
4088
4089 struct  kvm *kvm_arch_create_vm(void)
4090 {
4091         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4092
4093         if (!kvm)
4094                 return ERR_PTR(-ENOMEM);
4095
4096         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4097         INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4098         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4099
4100         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4101         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4102
4103         return kvm;
4104 }
4105
4106 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4107 {
4108         vcpu_load(vcpu);
4109         kvm_mmu_unload(vcpu);
4110         vcpu_put(vcpu);
4111 }
4112
4113 static void kvm_free_vcpus(struct kvm *kvm)
4114 {
4115         unsigned int i;
4116
4117         /*
4118          * Unpin any mmu pages first.
4119          */
4120         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4121                 if (kvm->vcpus[i])
4122                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4123         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4124                 if (kvm->vcpus[i]) {
4125                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4126                         kvm->vcpus[i] = NULL;
4127                 }
4128         }
4129
4130 }
4131
4132 void kvm_arch_sync_events(struct kvm *kvm)
4133 {
4134         kvm_free_all_assigned_devices(kvm);
4135 }
4136
4137 void kvm_arch_destroy_vm(struct kvm *kvm)
4138 {
4139         kvm_iommu_unmap_guest(kvm);
4140         kvm_free_pit(kvm);
4141         kfree(kvm->arch.vpic);
4142         kfree(kvm->arch.vioapic);
4143         kvm_free_vcpus(kvm);
4144         kvm_free_physmem(kvm);
4145         if (kvm->arch.apic_access_page)
4146                 put_page(kvm->arch.apic_access_page);
4147         if (kvm->arch.ept_identity_pagetable)
4148                 put_page(kvm->arch.ept_identity_pagetable);
4149         kfree(kvm);
4150 }
4151
4152 int kvm_arch_set_memory_region(struct kvm *kvm,
4153                                 struct kvm_userspace_memory_region *mem,
4154                                 struct kvm_memory_slot old,
4155                                 int user_alloc)
4156 {
4157         int npages = mem->memory_size >> PAGE_SHIFT;
4158         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4159
4160         /*To keep backward compatibility with older userspace,
4161          *x86 needs to hanlde !user_alloc case.
4162          */
4163         if (!user_alloc) {
4164                 if (npages && !old.rmap) {
4165                         unsigned long userspace_addr;
4166
4167                         down_write(&current->mm->mmap_sem);
4168                         userspace_addr = do_mmap(NULL, 0,
4169                                                  npages * PAGE_SIZE,
4170                                                  PROT_READ | PROT_WRITE,
4171                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4172                                                  0);
4173                         up_write(&current->mm->mmap_sem);
4174
4175                         if (IS_ERR((void *)userspace_addr))
4176                                 return PTR_ERR((void *)userspace_addr);
4177
4178                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4179                         spin_lock(&kvm->mmu_lock);
4180                         memslot->userspace_addr = userspace_addr;
4181                         spin_unlock(&kvm->mmu_lock);
4182                 } else {
4183                         if (!old.user_alloc && old.rmap) {
4184                                 int ret;
4185
4186                                 down_write(&current->mm->mmap_sem);
4187                                 ret = do_munmap(current->mm, old.userspace_addr,
4188                                                 old.npages * PAGE_SIZE);
4189                                 up_write(&current->mm->mmap_sem);
4190                                 if (ret < 0)
4191                                         printk(KERN_WARNING
4192                                        "kvm_vm_ioctl_set_memory_region: "
4193                                        "failed to munmap memory\n");
4194                         }
4195                 }
4196         }
4197
4198         if (!kvm->arch.n_requested_mmu_pages) {
4199                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4200                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4201         }
4202
4203         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4204         kvm_flush_remote_tlbs(kvm);
4205
4206         return 0;
4207 }
4208
4209 void kvm_arch_flush_shadow(struct kvm *kvm)
4210 {
4211         kvm_mmu_zap_all(kvm);
4212 }
4213
4214 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4215 {
4216         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4217                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4218                || vcpu->arch.nmi_pending;
4219 }
4220
4221 static void vcpu_kick_intr(void *info)
4222 {
4223 #ifdef DEBUG
4224         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4225         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4226 #endif
4227 }
4228
4229 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4230 {
4231         int ipi_pcpu = vcpu->cpu;
4232         int cpu = get_cpu();
4233
4234         if (waitqueue_active(&vcpu->wq)) {
4235                 wake_up_interruptible(&vcpu->wq);
4236                 ++vcpu->stat.halt_wakeup;
4237         }
4238         /*
4239          * We may be called synchronously with irqs disabled in guest mode,
4240          * So need not to call smp_call_function_single() in that case.
4241          */
4242         if (vcpu->guest_mode && vcpu->cpu != cpu)
4243                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4244         put_cpu();
4245 }