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