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
  22 #include <linux/clocksource.h>
  23 #include <linux/kvm.h>
  24 #include <linux/fs.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/module.h>
  27 #include <linux/mman.h>
  28 #include <linux/highmem.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/msr.h>
  32 #include <asm/desc.h>
  33
  34 #define MAX_IO_MSRS 256
  35 #define CR0_RESERVED_BITS                                               \
  36         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  37                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  38                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  39 #define CR4_RESERVED_BITS                                               \
  40         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  41                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  42                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  43                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  44
  45 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  46 /* EFER defaults:
  47  * - enable syscall per default because its emulated by KVM
  48  * - enable LME and LMA per default on 64 bit KVM
  49  */
  50 #ifdef CONFIG_X86_64
  51 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  52 #else
  53 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  54 #endif
  55
  56 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  57 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  58
  59 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  60                                     struct kvm_cpuid_entry2 __user *entries);
  61
  62 struct kvm_x86_ops *kvm_x86_ops;
  63
  64 struct kvm_stats_debugfs_item debugfs_entries[] = {
  65         { "pf_fixed", VCPU_STAT(pf_fixed) },
  66         { "pf_guest", VCPU_STAT(pf_guest) },
  67         { "tlb_flush", VCPU_STAT(tlb_flush) },
  68         { "invlpg", VCPU_STAT(invlpg) },
  69         { "exits", VCPU_STAT(exits) },
  70         { "io_exits", VCPU_STAT(io_exits) },
  71         { "mmio_exits", VCPU_STAT(mmio_exits) },
  72         { "signal_exits", VCPU_STAT(signal_exits) },
  73         { "irq_window", VCPU_STAT(irq_window_exits) },
  74         { "halt_exits", VCPU_STAT(halt_exits) },
  75         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  76         { "hypercalls", VCPU_STAT(hypercalls) },
  77         { "request_irq", VCPU_STAT(request_irq_exits) },
  78         { "irq_exits", VCPU_STAT(irq_exits) },
  79         { "host_state_reload", VCPU_STAT(host_state_reload) },
  80         { "efer_reload", VCPU_STAT(efer_reload) },
  81         { "fpu_reload", VCPU_STAT(fpu_reload) },
  82         { "insn_emulation", VCPU_STAT(insn_emulation) },
  83         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  84         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  85         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  86         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  87         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  88         { "mmu_flooded", VM_STAT(mmu_flooded) },
  89         { "mmu_recycled", VM_STAT(mmu_recycled) },
  90         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
  91         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  92         { "largepages", VM_STAT(lpages) },
  93         { NULL }
  94 };
  95
  96
  97 unsigned long segment_base(u16 selector)
  98 {
  99         struct descriptor_table gdt;
 100         struct desc_struct *d;
 101         unsigned long table_base;
 102         unsigned long v;
 103
 104         if (selector == 0)
 105                 return 0;
 106
 107         asm("sgdt %0" : "=m"(gdt));
 108         table_base = gdt.base;
 109
 110         if (selector & 4) {           /* from ldt */
 111                 u16 ldt_selector;
 112
 113                 asm("sldt %0" : "=g"(ldt_selector));
 114                 table_base = segment_base(ldt_selector);
 115         }
 116         d = (struct desc_struct *)(table_base + (selector & ~7));
 117         v = d->base0 | ((unsigned long)d->base1 << 16) |
 118                 ((unsigned long)d->base2 << 24);
 119 #ifdef CONFIG_X86_64
 120         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 121                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 122 #endif
 123         return v;
 124 }
 125 EXPORT_SYMBOL_GPL(segment_base);
 126
 127 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 128 {
 129         if (irqchip_in_kernel(vcpu->kvm))
 130                 return vcpu->arch.apic_base;
 131         else
 132                 return vcpu->arch.apic_base;
 133 }
 134 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 135
 136 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 137 {
 138         /* TODO: reserve bits check */
 139         if (irqchip_in_kernel(vcpu->kvm))
 140                 kvm_lapic_set_base(vcpu, data);
 141         else
 142                 vcpu->arch.apic_base = data;
 143 }
 144 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 145
 146 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 147 {
 148         WARN_ON(vcpu->arch.exception.pending);
 149         vcpu->arch.exception.pending = true;
 150         vcpu->arch.exception.has_error_code = false;
 151         vcpu->arch.exception.nr = nr;
 152 }
 153 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 154
 155 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 156                            u32 error_code)
 157 {
 158         ++vcpu->stat.pf_guest;
 159         if (vcpu->arch.exception.pending) {
 160                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 161                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 162                                         " double fault 0x%lx\n", addr);
 163                         vcpu->arch.exception.nr = DF_VECTOR;
 164                         vcpu->arch.exception.error_code = 0;
 165                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 166                         /* triple fault -> shutdown */
 167                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 168                 }
 169                 return;
 170         }
 171         vcpu->arch.cr2 = addr;
 172         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 173 }
 174
 175 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 176 {
 177         WARN_ON(vcpu->arch.exception.pending);
 178         vcpu->arch.exception.pending = true;
 179         vcpu->arch.exception.has_error_code = true;
 180         vcpu->arch.exception.nr = nr;
 181         vcpu->arch.exception.error_code = error_code;
 182 }
 183 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 184
 185 static void __queue_exception(struct kvm_vcpu *vcpu)
 186 {
 187         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 188                                      vcpu->arch.exception.has_error_code,
 189                                      vcpu->arch.exception.error_code);
 190 }
 191
 192 /*
 193  * Load the pae pdptrs.  Return true is they are all valid.
 194  */
 195 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 196 {
 197         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 198         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 199         int i;
 200         int ret;
 201         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 202
 203         down_read(&vcpu->kvm->slots_lock);
 204         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 205                                   offset * sizeof(u64), sizeof(pdpte));
 206         if (ret < 0) {
 207                 ret = 0;
 208                 goto out;
 209         }
 210         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 211                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 212                         ret = 0;
 213                         goto out;
 214                 }
 215         }
 216         ret = 1;
 217
 218         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 219 out:
 220         up_read(&vcpu->kvm->slots_lock);
 221
 222         return ret;
 223 }
 224 EXPORT_SYMBOL_GPL(load_pdptrs);
 225
 226 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 227 {
 228         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 229         bool changed = true;
 230         int r;
 231
 232         if (is_long_mode(vcpu) || !is_pae(vcpu))
 233                 return false;
 234
 235         down_read(&vcpu->kvm->slots_lock);
 236         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 237         if (r < 0)
 238                 goto out;
 239         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 240 out:
 241         up_read(&vcpu->kvm->slots_lock);
 242
 243         return changed;
 244 }
 245
 246 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 247 {
 248         if (cr0 & CR0_RESERVED_BITS) {
 249                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 250                        cr0, vcpu->arch.cr0);
 251                 kvm_inject_gp(vcpu, 0);
 252                 return;
 253         }
 254
 255         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 256                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 257                 kvm_inject_gp(vcpu, 0);
 258                 return;
 259         }
 260
 261         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 262                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 263                        "and a clear PE flag\n");
 264                 kvm_inject_gp(vcpu, 0);
 265                 return;
 266         }
 267
 268         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 269 #ifdef CONFIG_X86_64
 270                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 271                         int cs_db, cs_l;
 272
 273                         if (!is_pae(vcpu)) {
 274                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 275                                        "in long mode while PAE is disabled\n");
 276                                 kvm_inject_gp(vcpu, 0);
 277                                 return;
 278                         }
 279                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 280                         if (cs_l) {
 281                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 282                                        "in long mode while CS.L == 1\n");
 283                                 kvm_inject_gp(vcpu, 0);
 284                                 return;
 285
 286                         }
 287                 } else
 288 #endif
 289                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 290                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 291                                "reserved bits\n");
 292                         kvm_inject_gp(vcpu, 0);
 293                         return;
 294                 }
 295
 296         }
 297
 298         kvm_x86_ops->set_cr0(vcpu, cr0);
 299         vcpu->arch.cr0 = cr0;
 300
 301         kvm_mmu_reset_context(vcpu);
 302         return;
 303 }
 304 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 305
 306 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 307 {
 308         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 309 }
 310 EXPORT_SYMBOL_GPL(kvm_lmsw);
 311
 312 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 313 {
 314         if (cr4 & CR4_RESERVED_BITS) {
 315                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 316                 kvm_inject_gp(vcpu, 0);
 317                 return;
 318         }
 319
 320         if (is_long_mode(vcpu)) {
 321                 if (!(cr4 & X86_CR4_PAE)) {
 322                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 323                                "in long mode\n");
 324                         kvm_inject_gp(vcpu, 0);
 325                         return;
 326                 }
 327         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 328                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 329                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 330                 kvm_inject_gp(vcpu, 0);
 331                 return;
 332         }
 333
 334         if (cr4 & X86_CR4_VMXE) {
 335                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 336                 kvm_inject_gp(vcpu, 0);
 337                 return;
 338         }
 339         kvm_x86_ops->set_cr4(vcpu, cr4);
 340         vcpu->arch.cr4 = cr4;
 341         kvm_mmu_reset_context(vcpu);
 342 }
 343 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 344
 345 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 346 {
 347         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 348                 kvm_mmu_flush_tlb(vcpu);
 349                 return;
 350         }
 351
 352         if (is_long_mode(vcpu)) {
 353                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 354                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 355                         kvm_inject_gp(vcpu, 0);
 356                         return;
 357                 }
 358         } else {
 359                 if (is_pae(vcpu)) {
 360                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 361                                 printk(KERN_DEBUG
 362                                        "set_cr3: #GP, reserved bits\n");
 363                                 kvm_inject_gp(vcpu, 0);
 364                                 return;
 365                         }
 366                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 367                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 368                                        "reserved bits\n");
 369                                 kvm_inject_gp(vcpu, 0);
 370                                 return;
 371                         }
 372                 }
 373                 /*
 374                  * We don't check reserved bits in nonpae mode, because
 375                  * this isn't enforced, and VMware depends on this.
 376                  */
 377         }
 378
 379         down_read(&vcpu->kvm->slots_lock);
 380         /*
 381          * Does the new cr3 value map to physical memory? (Note, we
 382          * catch an invalid cr3 even in real-mode, because it would
 383          * cause trouble later on when we turn on paging anyway.)
 384          *
 385          * A real CPU would silently accept an invalid cr3 and would
 386          * attempt to use it - with largely undefined (and often hard
 387          * to debug) behavior on the guest side.
 388          */
 389         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 390                 kvm_inject_gp(vcpu, 0);
 391         else {
 392                 vcpu->arch.cr3 = cr3;
 393                 vcpu->arch.mmu.new_cr3(vcpu);
 394         }
 395         up_read(&vcpu->kvm->slots_lock);
 396 }
 397 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 398
 399 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 400 {
 401         if (cr8 & CR8_RESERVED_BITS) {
 402                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 403                 kvm_inject_gp(vcpu, 0);
 404                 return;
 405         }
 406         if (irqchip_in_kernel(vcpu->kvm))
 407                 kvm_lapic_set_tpr(vcpu, cr8);
 408         else
 409                 vcpu->arch.cr8 = cr8;
 410 }
 411 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 412
 413 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 414 {
 415         if (irqchip_in_kernel(vcpu->kvm))
 416                 return kvm_lapic_get_cr8(vcpu);
 417         else
 418                 return vcpu->arch.cr8;
 419 }
 420 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 421
 422 /*
 423  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 424  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 425  *
 426  * This list is modified at module load time to reflect the
 427  * capabilities of the host cpu.
 428  */
 429 static u32 msrs_to_save[] = {
 430         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 431         MSR_K6_STAR,
 432 #ifdef CONFIG_X86_64
 433         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 434 #endif
 435         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 436         MSR_IA32_PERF_STATUS,
 437 };
 438
 439 static unsigned num_msrs_to_save;
 440
 441 static u32 emulated_msrs[] = {
 442         MSR_IA32_MISC_ENABLE,
 443 };
 444
 445 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 446 {
 447         if (efer & efer_reserved_bits) {
 448                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 449                        efer);
 450                 kvm_inject_gp(vcpu, 0);
 451                 return;
 452         }
 453
 454         if (is_paging(vcpu)
 455             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 456                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 457                 kvm_inject_gp(vcpu, 0);
 458                 return;
 459         }
 460
 461         kvm_x86_ops->set_efer(vcpu, efer);
 462
 463         efer &= ~EFER_LMA;
 464         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 465
 466         vcpu->arch.shadow_efer = efer;
 467 }
 468
 469 void kvm_enable_efer_bits(u64 mask)
 470 {
 471        efer_reserved_bits &= ~mask;
 472 }
 473 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 474
 475
 476 /*
 477  * Writes msr value into into the appropriate "register".
 478  * Returns 0 on success, non-0 otherwise.
 479  * Assumes vcpu_load() was already called.
 480  */
 481 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 482 {
 483         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 484 }
 485
 486 /*
 487  * Adapt set_msr() to msr_io()'s calling convention
 488  */
 489 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 490 {
 491         return kvm_set_msr(vcpu, index, *data);
 492 }
 493
 494 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 495 {
 496         static int version;
 497         struct kvm_wall_clock wc;
 498         struct timespec wc_ts;
 499
 500         if (!wall_clock)
 501                 return;
 502
 503         version++;
 504
 505         down_read(&kvm->slots_lock);
 506         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 507
 508         wc_ts = current_kernel_time();
 509         wc.wc_sec = wc_ts.tv_sec;
 510         wc.wc_nsec = wc_ts.tv_nsec;
 511         wc.wc_version = version;
 512
 513         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 514
 515         version++;
 516         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 517         up_read(&kvm->slots_lock);
 518 }
 519
 520 static void kvm_write_guest_time(struct kvm_vcpu *v)
 521 {
 522         struct timespec ts;
 523         unsigned long flags;
 524         struct kvm_vcpu_arch *vcpu = &v->arch;
 525         void *shared_kaddr;
 526
 527         if ((!vcpu->time_page))
 528                 return;
 529
 530         /* Keep irq disabled to prevent changes to the clock */
 531         local_irq_save(flags);
 532         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 533                           &vcpu->hv_clock.tsc_timestamp);
 534         ktime_get_ts(&ts);
 535         local_irq_restore(flags);
 536
 537         /* With all the info we got, fill in the values */
 538
 539         vcpu->hv_clock.system_time = ts.tv_nsec +
 540                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 541         /*
 542          * The interface expects us to write an even number signaling that the
 543          * update is finished. Since the guest won't see the intermediate
 544          * state, we just write "2" at the end
 545          */
 546         vcpu->hv_clock.version = 2;
 547
 548         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 549
 550         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 551                 sizeof(vcpu->hv_clock));
 552
 553         kunmap_atomic(shared_kaddr, KM_USER0);
 554
 555         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 556 }
 557
 558
 559 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 560 {
 561         switch (msr) {
 562         case MSR_EFER:
 563                 set_efer(vcpu, data);
 564                 break;
 565         case MSR_IA32_MC0_STATUS:
 566                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 567                        __func__, data);
 568                 break;
 569         case MSR_IA32_MCG_STATUS:
 570                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 571                         __func__, data);
 572                 break;
 573         case MSR_IA32_MCG_CTL:
 574                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 575                         __func__, data);
 576                 break;
 577         case MSR_IA32_UCODE_REV:
 578         case MSR_IA32_UCODE_WRITE:
 579         case 0x200 ... 0x2ff: /* MTRRs */
 580                 break;
 581         case MSR_IA32_APICBASE:
 582                 kvm_set_apic_base(vcpu, data);
 583                 break;
 584         case MSR_IA32_MISC_ENABLE:
 585                 vcpu->arch.ia32_misc_enable_msr = data;
 586                 break;
 587         case MSR_KVM_WALL_CLOCK:
 588                 vcpu->kvm->arch.wall_clock = data;
 589                 kvm_write_wall_clock(vcpu->kvm, data);
 590                 break;
 591         case MSR_KVM_SYSTEM_TIME: {
 592                 if (vcpu->arch.time_page) {
 593                         kvm_release_page_dirty(vcpu->arch.time_page);
 594                         vcpu->arch.time_page = NULL;
 595                 }
 596
 597                 vcpu->arch.time = data;
 598
 599                 /* we verify if the enable bit is set... */
 600                 if (!(data & 1))
 601                         break;
 602
 603                 /* ...but clean it before doing the actual write */
 604                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 605
 606                 vcpu->arch.hv_clock.tsc_to_system_mul =
 607                                         clocksource_khz2mult(tsc_khz, 22);
 608                 vcpu->arch.hv_clock.tsc_shift = 22;
 609
 610                 down_read(&current->mm->mmap_sem);
 611                 down_read(&vcpu->kvm->slots_lock);
 612                 vcpu->arch.time_page =
 613                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 614                 up_read(&vcpu->kvm->slots_lock);
 615                 up_read(&current->mm->mmap_sem);
 616
 617                 if (is_error_page(vcpu->arch.time_page)) {
 618                         kvm_release_page_clean(vcpu->arch.time_page);
 619                         vcpu->arch.time_page = NULL;
 620                 }
 621
 622                 kvm_write_guest_time(vcpu);
 623                 break;
 624         }
 625         default:
 626                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 627                 return 1;
 628         }
 629         return 0;
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 632
 633
 634 /*
 635  * Reads an msr value (of 'msr_index') into 'pdata'.
 636  * Returns 0 on success, non-0 otherwise.
 637  * Assumes vcpu_load() was already called.
 638  */
 639 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 640 {
 641         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 642 }
 643
 644 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 645 {
 646         u64 data;
 647
 648         switch (msr) {
 649         case 0xc0010010: /* SYSCFG */
 650         case 0xc0010015: /* HWCR */
 651         case MSR_IA32_PLATFORM_ID:
 652         case MSR_IA32_P5_MC_ADDR:
 653         case MSR_IA32_P5_MC_TYPE:
 654         case MSR_IA32_MC0_CTL:
 655         case MSR_IA32_MCG_STATUS:
 656         case MSR_IA32_MCG_CAP:
 657         case MSR_IA32_MCG_CTL:
 658         case MSR_IA32_MC0_MISC:
 659         case MSR_IA32_MC0_MISC+4:
 660         case MSR_IA32_MC0_MISC+8:
 661         case MSR_IA32_MC0_MISC+12:
 662         case MSR_IA32_MC0_MISC+16:
 663         case MSR_IA32_UCODE_REV:
 664         case MSR_IA32_EBL_CR_POWERON:
 665                 /* MTRR registers */
 666         case 0xfe:
 667         case 0x200 ... 0x2ff:
 668                 data = 0;
 669                 break;
 670         case 0xcd: /* fsb frequency */
 671                 data = 3;
 672                 break;
 673         case MSR_IA32_APICBASE:
 674                 data = kvm_get_apic_base(vcpu);
 675                 break;
 676         case MSR_IA32_MISC_ENABLE:
 677                 data = vcpu->arch.ia32_misc_enable_msr;
 678                 break;
 679         case MSR_IA32_PERF_STATUS:
 680                 /* TSC increment by tick */
 681                 data = 1000ULL;
 682                 /* CPU multiplier */
 683                 data |= (((uint64_t)4ULL) << 40);
 684                 break;
 685         case MSR_EFER:
 686                 data = vcpu->arch.shadow_efer;
 687                 break;
 688         case MSR_KVM_WALL_CLOCK:
 689                 data = vcpu->kvm->arch.wall_clock;
 690                 break;
 691         case MSR_KVM_SYSTEM_TIME:
 692                 data = vcpu->arch.time;
 693                 break;
 694         default:
 695                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 696                 return 1;
 697         }
 698         *pdata = data;
 699         return 0;
 700 }
 701 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 702
 703 /*
 704  * Read or write a bunch of msrs. All parameters are kernel addresses.
 705  *
 706  * @return number of msrs set successfully.
 707  */
 708 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 709                     struct kvm_msr_entry *entries,
 710                     int (*do_msr)(struct kvm_vcpu *vcpu,
 711                                   unsigned index, u64 *data))
 712 {
 713         int i;
 714
 715         vcpu_load(vcpu);
 716
 717         for (i = 0; i < msrs->nmsrs; ++i)
 718                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 719                         break;
 720
 721         vcpu_put(vcpu);
 722
 723         return i;
 724 }
 725
 726 /*
 727  * Read or write a bunch of msrs. Parameters are user addresses.
 728  *
 729  * @return number of msrs set successfully.
 730  */
 731 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 732                   int (*do_msr)(struct kvm_vcpu *vcpu,
 733                                 unsigned index, u64 *data),
 734                   int writeback)
 735 {
 736         struct kvm_msrs msrs;
 737         struct kvm_msr_entry *entries;
 738         int r, n;
 739         unsigned size;
 740
 741         r = -EFAULT;
 742         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 743                 goto out;
 744
 745         r = -E2BIG;
 746         if (msrs.nmsrs >= MAX_IO_MSRS)
 747                 goto out;
 748
 749         r = -ENOMEM;
 750         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 751         entries = vmalloc(size);
 752         if (!entries)
 753                 goto out;
 754
 755         r = -EFAULT;
 756         if (copy_from_user(entries, user_msrs->entries, size))
 757                 goto out_free;
 758
 759         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 760         if (r < 0)
 761                 goto out_free;
 762
 763         r = -EFAULT;
 764         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 765                 goto out_free;
 766
 767         r = n;
 768
 769 out_free:
 770         vfree(entries);
 771 out:
 772         return r;
 773 }
 774
 775 /*
 776  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 777  * cached on it.
 778  */
 779 void decache_vcpus_on_cpu(int cpu)
 780 {
 781         struct kvm *vm;
 782         struct kvm_vcpu *vcpu;
 783         int i;
 784
 785         spin_lock(&kvm_lock);
 786         list_for_each_entry(vm, &vm_list, vm_list)
 787                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 788                         vcpu = vm->vcpus[i];
 789                         if (!vcpu)
 790                                 continue;
 791                         /*
 792                          * If the vcpu is locked, then it is running on some
 793                          * other cpu and therefore it is not cached on the
 794                          * cpu in question.
 795                          *
 796                          * If it's not locked, check the last cpu it executed
 797                          * on.
 798                          */
 799                         if (mutex_trylock(&vcpu->mutex)) {
 800                                 if (vcpu->cpu == cpu) {
 801                                         kvm_x86_ops->vcpu_decache(vcpu);
 802                                         vcpu->cpu = -1;
 803                                 }
 804                                 mutex_unlock(&vcpu->mutex);
 805                         }
 806                 }
 807         spin_unlock(&kvm_lock);
 808 }
 809
 810 int kvm_dev_ioctl_check_extension(long ext)
 811 {
 812         int r;
 813
 814         switch (ext) {
 815         case KVM_CAP_IRQCHIP:
 816         case KVM_CAP_HLT:
 817         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 818         case KVM_CAP_USER_MEMORY:
 819         case KVM_CAP_SET_TSS_ADDR:
 820         case KVM_CAP_EXT_CPUID:
 821         case KVM_CAP_CLOCKSOURCE:
 822         case KVM_CAP_PIT:
 823         case KVM_CAP_NOP_IO_DELAY:
 824                 r = 1;
 825                 break;
 826         case KVM_CAP_VAPIC:
 827                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 828                 break;
 829         case KVM_CAP_NR_VCPUS:
 830                 r = KVM_MAX_VCPUS;
 831                 break;
 832         case KVM_CAP_NR_MEMSLOTS:
 833                 r = KVM_MEMORY_SLOTS;
 834                 break;
 835         default:
 836                 r = 0;
 837                 break;
 838         }
 839         return r;
 840
 841 }
 842
 843 long kvm_arch_dev_ioctl(struct file *filp,
 844                         unsigned int ioctl, unsigned long arg)
 845 {
 846         void __user *argp = (void __user *)arg;
 847         long r;
 848
 849         switch (ioctl) {
 850         case KVM_GET_MSR_INDEX_LIST: {
 851                 struct kvm_msr_list __user *user_msr_list = argp;
 852                 struct kvm_msr_list msr_list;
 853                 unsigned n;
 854
 855                 r = -EFAULT;
 856                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 857                         goto out;
 858                 n = msr_list.nmsrs;
 859                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 860                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 861                         goto out;
 862                 r = -E2BIG;
 863                 if (n < num_msrs_to_save)
 864                         goto out;
 865                 r = -EFAULT;
 866                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 867                                  num_msrs_to_save * sizeof(u32)))
 868                         goto out;
 869                 if (copy_to_user(user_msr_list->indices
 870                                  + num_msrs_to_save * sizeof(u32),
 871                                  &emulated_msrs,
 872                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 873                         goto out;
 874                 r = 0;
 875                 break;
 876         }
 877         case KVM_GET_SUPPORTED_CPUID: {
 878                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 879                 struct kvm_cpuid2 cpuid;
 880
 881                 r = -EFAULT;
 882                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 883                         goto out;
 884                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 885                         cpuid_arg->entries);
 886                 if (r)
 887                         goto out;
 888
 889                 r = -EFAULT;
 890                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 891                         goto out;
 892                 r = 0;
 893                 break;
 894         }
 895         default:
 896                 r = -EINVAL;
 897         }
 898 out:
 899         return r;
 900 }
 901
 902 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 903 {
 904         kvm_x86_ops->vcpu_load(vcpu, cpu);
 905         kvm_write_guest_time(vcpu);
 906 }
 907
 908 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 909 {
 910         kvm_x86_ops->vcpu_put(vcpu);
 911         kvm_put_guest_fpu(vcpu);
 912 }
 913
 914 static int is_efer_nx(void)
 915 {
 916         u64 efer;
 917
 918         rdmsrl(MSR_EFER, efer);
 919         return efer & EFER_NX;
 920 }
 921
 922 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 923 {
 924         int i;
 925         struct kvm_cpuid_entry2 *e, *entry;
 926
 927         entry = NULL;
 928         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 929                 e = &vcpu->arch.cpuid_entries[i];
 930                 if (e->function == 0x80000001) {
 931                         entry = e;
 932                         break;
 933                 }
 934         }
 935         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 936                 entry->edx &= ~(1 << 20);
 937                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 938         }
 939 }
 940
 941 /* when an old userspace process fills a new kernel module */
 942 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 943                                     struct kvm_cpuid *cpuid,
 944                                     struct kvm_cpuid_entry __user *entries)
 945 {
 946         int r, i;
 947         struct kvm_cpuid_entry *cpuid_entries;
 948
 949         r = -E2BIG;
 950         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 951                 goto out;
 952         r = -ENOMEM;
 953         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 954         if (!cpuid_entries)
 955                 goto out;
 956         r = -EFAULT;
 957         if (copy_from_user(cpuid_entries, entries,
 958                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 959                 goto out_free;
 960         for (i = 0; i < cpuid->nent; i++) {
 961                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 962                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 963                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 964                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 965                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 966                 vcpu->arch.cpuid_entries[i].index = 0;
 967                 vcpu->arch.cpuid_entries[i].flags = 0;
 968                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 969                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 970                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 971         }
 972         vcpu->arch.cpuid_nent = cpuid->nent;
 973         cpuid_fix_nx_cap(vcpu);
 974         r = 0;
 975
 976 out_free:
 977         vfree(cpuid_entries);
 978 out:
 979         return r;
 980 }
 981
 982 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 983                                     struct kvm_cpuid2 *cpuid,
 984                                     struct kvm_cpuid_entry2 __user *entries)
 985 {
 986         int r;
 987
 988         r = -E2BIG;
 989         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 990                 goto out;
 991         r = -EFAULT;
 992         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 993                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 994                 goto out;
 995         vcpu->arch.cpuid_nent = cpuid->nent;
 996         return 0;
 997
 998 out:
 999         return r;
1000 }
1001
1002 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1003                                     struct kvm_cpuid2 *cpuid,
1004                                     struct kvm_cpuid_entry2 __user *entries)
1005 {
1006         int r;
1007
1008         r = -E2BIG;
1009         if (cpuid->nent < vcpu->arch.cpuid_nent)
1010                 goto out;
1011         r = -EFAULT;
1012         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1013                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1014                 goto out;
1015         return 0;
1016
1017 out:
1018         cpuid->nent = vcpu->arch.cpuid_nent;
1019         return r;
1020 }
1021
1022 static inline u32 bit(int bitno)
1023 {
1024         return 1 << (bitno & 31);
1025 }
1026
1027 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1028                           u32 index)
1029 {
1030         entry->function = function;
1031         entry->index = index;
1032         cpuid_count(entry->function, entry->index,
1033                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1034         entry->flags = 0;
1035 }
1036
1037 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1038                          u32 index, int *nent, int maxnent)
1039 {
1040         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1041                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1042                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1043                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1044                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1045                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1046                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1047                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1048                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1049                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1050         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1051                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1052                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1053                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1054                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1055                 bit(X86_FEATURE_PGE) |
1056                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1057                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1058                 bit(X86_FEATURE_SYSCALL) |
1059                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1060 #ifdef CONFIG_X86_64
1061                 bit(X86_FEATURE_LM) |
1062 #endif
1063                 bit(X86_FEATURE_MMXEXT) |
1064                 bit(X86_FEATURE_3DNOWEXT) |
1065                 bit(X86_FEATURE_3DNOW);
1066         const u32 kvm_supported_word3_x86_features =
1067                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1068         const u32 kvm_supported_word6_x86_features =
1069                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1070
1071         /* all func 2 cpuid_count() should be called on the same cpu */
1072         get_cpu();
1073         do_cpuid_1_ent(entry, function, index);
1074         ++*nent;
1075
1076         switch (function) {
1077         case 0:
1078                 entry->eax = min(entry->eax, (u32)0xb);
1079                 break;
1080         case 1:
1081                 entry->edx &= kvm_supported_word0_x86_features;
1082                 entry->ecx &= kvm_supported_word3_x86_features;
1083                 break;
1084         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1085          * may return different values. This forces us to get_cpu() before
1086          * issuing the first command, and also to emulate this annoying behavior
1087          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1088         case 2: {
1089                 int t, times = entry->eax & 0xff;
1090
1091                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1092                 for (t = 1; t < times && *nent < maxnent; ++t) {
1093                         do_cpuid_1_ent(&entry[t], function, 0);
1094                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1095                         ++*nent;
1096                 }
1097                 break;
1098         }
1099         /* function 4 and 0xb have additional index. */
1100         case 4: {
1101                 int i, cache_type;
1102
1103                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1104                 /* read more entries until cache_type is zero */
1105                 for (i = 1; *nent < maxnent; ++i) {
1106                         cache_type = entry[i - 1].eax & 0x1f;
1107                         if (!cache_type)
1108                                 break;
1109                         do_cpuid_1_ent(&entry[i], function, i);
1110                         entry[i].flags |=
1111                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1112                         ++*nent;
1113                 }
1114                 break;
1115         }
1116         case 0xb: {
1117                 int i, level_type;
1118
1119                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1120                 /* read more entries until level_type is zero */
1121                 for (i = 1; *nent < maxnent; ++i) {
1122                         level_type = entry[i - 1].ecx & 0xff;
1123                         if (!level_type)
1124                                 break;
1125                         do_cpuid_1_ent(&entry[i], function, i);
1126                         entry[i].flags |=
1127                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1128                         ++*nent;
1129                 }
1130                 break;
1131         }
1132         case 0x80000000:
1133                 entry->eax = min(entry->eax, 0x8000001a);
1134                 break;
1135         case 0x80000001:
1136                 entry->edx &= kvm_supported_word1_x86_features;
1137                 entry->ecx &= kvm_supported_word6_x86_features;
1138                 break;
1139         }
1140         put_cpu();
1141 }
1142
1143 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1144                                     struct kvm_cpuid_entry2 __user *entries)
1145 {
1146         struct kvm_cpuid_entry2 *cpuid_entries;
1147         int limit, nent = 0, r = -E2BIG;
1148         u32 func;
1149
1150         if (cpuid->nent < 1)
1151                 goto out;
1152         r = -ENOMEM;
1153         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1154         if (!cpuid_entries)
1155                 goto out;
1156
1157         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1158         limit = cpuid_entries[0].eax;
1159         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1160                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1161                                 &nent, cpuid->nent);
1162         r = -E2BIG;
1163         if (nent >= cpuid->nent)
1164                 goto out_free;
1165
1166         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1167         limit = cpuid_entries[nent - 1].eax;
1168         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1169                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1170                                &nent, cpuid->nent);
1171         r = -EFAULT;
1172         if (copy_to_user(entries, cpuid_entries,
1173                         nent * sizeof(struct kvm_cpuid_entry2)))
1174                 goto out_free;
1175         cpuid->nent = nent;
1176         r = 0;
1177
1178 out_free:
1179         vfree(cpuid_entries);
1180 out:
1181         return r;
1182 }
1183
1184 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1185                                     struct kvm_lapic_state *s)
1186 {
1187         vcpu_load(vcpu);
1188         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1189         vcpu_put(vcpu);
1190
1191         return 0;
1192 }
1193
1194 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1195                                     struct kvm_lapic_state *s)
1196 {
1197         vcpu_load(vcpu);
1198         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1199         kvm_apic_post_state_restore(vcpu);
1200         vcpu_put(vcpu);
1201
1202         return 0;
1203 }
1204
1205 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1206                                     struct kvm_interrupt *irq)
1207 {
1208         if (irq->irq < 0 || irq->irq >= 256)
1209                 return -EINVAL;
1210         if (irqchip_in_kernel(vcpu->kvm))
1211                 return -ENXIO;
1212         vcpu_load(vcpu);
1213
1214         set_bit(irq->irq, vcpu->arch.irq_pending);
1215         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1216
1217         vcpu_put(vcpu);
1218
1219         return 0;
1220 }
1221
1222 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1223                                            struct kvm_tpr_access_ctl *tac)
1224 {
1225         if (tac->flags)
1226                 return -EINVAL;
1227         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1228         return 0;
1229 }
1230
1231 long kvm_arch_vcpu_ioctl(struct file *filp,
1232                          unsigned int ioctl, unsigned long arg)
1233 {
1234         struct kvm_vcpu *vcpu = filp->private_data;
1235         void __user *argp = (void __user *)arg;
1236         int r;
1237
1238         switch (ioctl) {
1239         case KVM_GET_LAPIC: {
1240                 struct kvm_lapic_state lapic;
1241
1242                 memset(&lapic, 0, sizeof lapic);
1243                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1244                 if (r)
1245                         goto out;
1246                 r = -EFAULT;
1247                 if (copy_to_user(argp, &lapic, sizeof lapic))
1248                         goto out;
1249                 r = 0;
1250                 break;
1251         }
1252         case KVM_SET_LAPIC: {
1253                 struct kvm_lapic_state lapic;
1254
1255                 r = -EFAULT;
1256                 if (copy_from_user(&lapic, argp, sizeof lapic))
1257                         goto out;
1258                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1259                 if (r)
1260                         goto out;
1261                 r = 0;
1262                 break;
1263         }
1264         case KVM_INTERRUPT: {
1265                 struct kvm_interrupt irq;
1266
1267                 r = -EFAULT;
1268                 if (copy_from_user(&irq, argp, sizeof irq))
1269                         goto out;
1270                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1271                 if (r)
1272                         goto out;
1273                 r = 0;
1274                 break;
1275         }
1276         case KVM_SET_CPUID: {
1277                 struct kvm_cpuid __user *cpuid_arg = argp;
1278                 struct kvm_cpuid cpuid;
1279
1280                 r = -EFAULT;
1281                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1282                         goto out;
1283                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1284                 if (r)
1285                         goto out;
1286                 break;
1287         }
1288         case KVM_SET_CPUID2: {
1289                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1290                 struct kvm_cpuid2 cpuid;
1291
1292                 r = -EFAULT;
1293                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1294                         goto out;
1295                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1296                                 cpuid_arg->entries);
1297                 if (r)
1298                         goto out;
1299                 break;
1300         }
1301         case KVM_GET_CPUID2: {
1302                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1303                 struct kvm_cpuid2 cpuid;
1304
1305                 r = -EFAULT;
1306                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1307                         goto out;
1308                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1309                                 cpuid_arg->entries);
1310                 if (r)
1311                         goto out;
1312                 r = -EFAULT;
1313                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1314                         goto out;
1315                 r = 0;
1316                 break;
1317         }
1318         case KVM_GET_MSRS:
1319                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1320                 break;
1321         case KVM_SET_MSRS:
1322                 r = msr_io(vcpu, argp, do_set_msr, 0);
1323                 break;
1324         case KVM_TPR_ACCESS_REPORTING: {
1325                 struct kvm_tpr_access_ctl tac;
1326
1327                 r = -EFAULT;
1328                 if (copy_from_user(&tac, argp, sizeof tac))
1329                         goto out;
1330                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1331                 if (r)
1332                         goto out;
1333                 r = -EFAULT;
1334                 if (copy_to_user(argp, &tac, sizeof tac))
1335                         goto out;
1336                 r = 0;
1337                 break;
1338         };
1339         case KVM_SET_VAPIC_ADDR: {
1340                 struct kvm_vapic_addr va;
1341
1342                 r = -EINVAL;
1343                 if (!irqchip_in_kernel(vcpu->kvm))
1344                         goto out;
1345                 r = -EFAULT;
1346                 if (copy_from_user(&va, argp, sizeof va))
1347                         goto out;
1348                 r = 0;
1349                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1350                 break;
1351         }
1352         default:
1353                 r = -EINVAL;
1354         }
1355 out:
1356         return r;
1357 }
1358
1359 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1360 {
1361         int ret;
1362
1363         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1364                 return -1;
1365         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1366         return ret;
1367 }
1368
1369 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1370                                           u32 kvm_nr_mmu_pages)
1371 {
1372         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1373                 return -EINVAL;
1374
1375         down_write(&kvm->slots_lock);
1376
1377         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1378         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1379
1380         up_write(&kvm->slots_lock);
1381         return 0;
1382 }
1383
1384 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1385 {
1386         return kvm->arch.n_alloc_mmu_pages;
1387 }
1388
1389 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1390 {
1391         int i;
1392         struct kvm_mem_alias *alias;
1393
1394         for (i = 0; i < kvm->arch.naliases; ++i) {
1395                 alias = &kvm->arch.aliases[i];
1396                 if (gfn >= alias->base_gfn
1397                     && gfn < alias->base_gfn + alias->npages)
1398                         return alias->target_gfn + gfn - alias->base_gfn;
1399         }
1400         return gfn;
1401 }
1402
1403 /*
1404  * Set a new alias region.  Aliases map a portion of physical memory into
1405  * another portion.  This is useful for memory windows, for example the PC
1406  * VGA region.
1407  */
1408 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1409                                          struct kvm_memory_alias *alias)
1410 {
1411         int r, n;
1412         struct kvm_mem_alias *p;
1413
1414         r = -EINVAL;
1415         /* General sanity checks */
1416         if (alias->memory_size & (PAGE_SIZE - 1))
1417                 goto out;
1418         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1419                 goto out;
1420         if (alias->slot >= KVM_ALIAS_SLOTS)
1421                 goto out;
1422         if (alias->guest_phys_addr + alias->memory_size
1423             < alias->guest_phys_addr)
1424                 goto out;
1425         if (alias->target_phys_addr + alias->memory_size
1426             < alias->target_phys_addr)
1427                 goto out;
1428
1429         down_write(&kvm->slots_lock);
1430
1431         p = &kvm->arch.aliases[alias->slot];
1432         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1433         p->npages = alias->memory_size >> PAGE_SHIFT;
1434         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1435
1436         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1437                 if (kvm->arch.aliases[n - 1].npages)
1438                         break;
1439         kvm->arch.naliases = n;
1440
1441         kvm_mmu_zap_all(kvm);
1442
1443         up_write(&kvm->slots_lock);
1444
1445         return 0;
1446
1447 out:
1448         return r;
1449 }
1450
1451 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1452 {
1453         int r;
1454
1455         r = 0;
1456         switch (chip->chip_id) {
1457         case KVM_IRQCHIP_PIC_MASTER:
1458                 memcpy(&chip->chip.pic,
1459                         &pic_irqchip(kvm)->pics[0],
1460                         sizeof(struct kvm_pic_state));
1461                 break;
1462         case KVM_IRQCHIP_PIC_SLAVE:
1463                 memcpy(&chip->chip.pic,
1464                         &pic_irqchip(kvm)->pics[1],
1465                         sizeof(struct kvm_pic_state));
1466                 break;
1467         case KVM_IRQCHIP_IOAPIC:
1468                 memcpy(&chip->chip.ioapic,
1469                         ioapic_irqchip(kvm),
1470                         sizeof(struct kvm_ioapic_state));
1471                 break;
1472         default:
1473                 r = -EINVAL;
1474                 break;
1475         }
1476         return r;
1477 }
1478
1479 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1480 {
1481         int r;
1482
1483         r = 0;
1484         switch (chip->chip_id) {
1485         case KVM_IRQCHIP_PIC_MASTER:
1486                 memcpy(&pic_irqchip(kvm)->pics[0],
1487                         &chip->chip.pic,
1488                         sizeof(struct kvm_pic_state));
1489                 break;
1490         case KVM_IRQCHIP_PIC_SLAVE:
1491                 memcpy(&pic_irqchip(kvm)->pics[1],
1492                         &chip->chip.pic,
1493                         sizeof(struct kvm_pic_state));
1494                 break;
1495         case KVM_IRQCHIP_IOAPIC:
1496                 memcpy(ioapic_irqchip(kvm),
1497                         &chip->chip.ioapic,
1498                         sizeof(struct kvm_ioapic_state));
1499                 break;
1500         default:
1501                 r = -EINVAL;
1502                 break;
1503         }
1504         kvm_pic_update_irq(pic_irqchip(kvm));
1505         return r;
1506 }
1507
1508 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1509 {
1510         int r = 0;
1511
1512         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1513         return r;
1514 }
1515
1516 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1517 {
1518         int r = 0;
1519
1520         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1521         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1522         return r;
1523 }
1524
1525 /*
1526  * Get (and clear) the dirty memory log for a memory slot.
1527  */
1528 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1529                                       struct kvm_dirty_log *log)
1530 {
1531         int r;
1532         int n;
1533         struct kvm_memory_slot *memslot;
1534         int is_dirty = 0;
1535
1536         down_write(&kvm->slots_lock);
1537
1538         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1539         if (r)
1540                 goto out;
1541
1542         /* If nothing is dirty, don't bother messing with page tables. */
1543         if (is_dirty) {
1544                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1545                 kvm_flush_remote_tlbs(kvm);
1546                 memslot = &kvm->memslots[log->slot];
1547                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1548                 memset(memslot->dirty_bitmap, 0, n);
1549         }
1550         r = 0;
1551 out:
1552         up_write(&kvm->slots_lock);
1553         return r;
1554 }
1555
1556 long kvm_arch_vm_ioctl(struct file *filp,
1557                        unsigned int ioctl, unsigned long arg)
1558 {
1559         struct kvm *kvm = filp->private_data;
1560         void __user *argp = (void __user *)arg;
1561         int r = -EINVAL;
1562
1563         switch (ioctl) {
1564         case KVM_SET_TSS_ADDR:
1565                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1566                 if (r < 0)
1567                         goto out;
1568                 break;
1569         case KVM_SET_MEMORY_REGION: {
1570                 struct kvm_memory_region kvm_mem;
1571                 struct kvm_userspace_memory_region kvm_userspace_mem;
1572
1573                 r = -EFAULT;
1574                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1575                         goto out;
1576                 kvm_userspace_mem.slot = kvm_mem.slot;
1577                 kvm_userspace_mem.flags = kvm_mem.flags;
1578                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1579                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1580                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1581                 if (r)
1582                         goto out;
1583                 break;
1584         }
1585         case KVM_SET_NR_MMU_PAGES:
1586                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1587                 if (r)
1588                         goto out;
1589                 break;
1590         case KVM_GET_NR_MMU_PAGES:
1591                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1592                 break;
1593         case KVM_SET_MEMORY_ALIAS: {
1594                 struct kvm_memory_alias alias;
1595
1596                 r = -EFAULT;
1597                 if (copy_from_user(&alias, argp, sizeof alias))
1598                         goto out;
1599                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1600                 if (r)
1601                         goto out;
1602                 break;
1603         }
1604         case KVM_CREATE_IRQCHIP:
1605                 r = -ENOMEM;
1606                 kvm->arch.vpic = kvm_create_pic(kvm);
1607                 if (kvm->arch.vpic) {
1608                         r = kvm_ioapic_init(kvm);
1609                         if (r) {
1610                                 kfree(kvm->arch.vpic);
1611                                 kvm->arch.vpic = NULL;
1612                                 goto out;
1613                         }
1614                 } else
1615                         goto out;
1616                 break;
1617         case KVM_CREATE_PIT:
1618                 r = -ENOMEM;
1619                 kvm->arch.vpit = kvm_create_pit(kvm);
1620                 if (kvm->arch.vpit)
1621                         r = 0;
1622                 break;
1623         case KVM_IRQ_LINE: {
1624                 struct kvm_irq_level irq_event;
1625
1626                 r = -EFAULT;
1627                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1628                         goto out;
1629                 if (irqchip_in_kernel(kvm)) {
1630                         mutex_lock(&kvm->lock);
1631                         if (irq_event.irq < 16)
1632                                 kvm_pic_set_irq(pic_irqchip(kvm),
1633                                         irq_event.irq,
1634                                         irq_event.level);
1635                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1636                                         irq_event.irq,
1637                                         irq_event.level);
1638                         mutex_unlock(&kvm->lock);
1639                         r = 0;
1640                 }
1641                 break;
1642         }
1643         case KVM_GET_IRQCHIP: {
1644                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1645                 struct kvm_irqchip chip;
1646
1647                 r = -EFAULT;
1648                 if (copy_from_user(&chip, argp, sizeof chip))
1649                         goto out;
1650                 r = -ENXIO;
1651                 if (!irqchip_in_kernel(kvm))
1652                         goto out;
1653                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1654                 if (r)
1655                         goto out;
1656                 r = -EFAULT;
1657                 if (copy_to_user(argp, &chip, sizeof chip))
1658                         goto out;
1659                 r = 0;
1660                 break;
1661         }
1662         case KVM_SET_IRQCHIP: {
1663                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1664                 struct kvm_irqchip chip;
1665
1666                 r = -EFAULT;
1667                 if (copy_from_user(&chip, argp, sizeof chip))
1668                         goto out;
1669                 r = -ENXIO;
1670                 if (!irqchip_in_kernel(kvm))
1671                         goto out;
1672                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1673                 if (r)
1674                         goto out;
1675                 r = 0;
1676                 break;
1677         }
1678         case KVM_GET_PIT: {
1679                 struct kvm_pit_state ps;
1680                 r = -EFAULT;
1681                 if (copy_from_user(&ps, argp, sizeof ps))
1682                         goto out;
1683                 r = -ENXIO;
1684                 if (!kvm->arch.vpit)
1685                         goto out;
1686                 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1687                 if (r)
1688                         goto out;
1689                 r = -EFAULT;
1690                 if (copy_to_user(argp, &ps, sizeof ps))
1691                         goto out;
1692                 r = 0;
1693                 break;
1694         }
1695         case KVM_SET_PIT: {
1696                 struct kvm_pit_state ps;
1697                 r = -EFAULT;
1698                 if (copy_from_user(&ps, argp, sizeof ps))
1699                         goto out;
1700                 r = -ENXIO;
1701                 if (!kvm->arch.vpit)
1702                         goto out;
1703                 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1704                 if (r)
1705                         goto out;
1706                 r = 0;
1707                 break;
1708         }
1709         default:
1710                 ;
1711         }
1712 out:
1713         return r;
1714 }
1715
1716 static void kvm_init_msr_list(void)
1717 {
1718         u32 dummy[2];
1719         unsigned i, j;
1720
1721         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1722                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1723                         continue;
1724                 if (j < i)
1725                         msrs_to_save[j] = msrs_to_save[i];
1726                 j++;
1727         }
1728         num_msrs_to_save = j;
1729 }
1730
1731 /*
1732  * Only apic need an MMIO device hook, so shortcut now..
1733  */
1734 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1735                                                 gpa_t addr)
1736 {
1737         struct kvm_io_device *dev;
1738
1739         if (vcpu->arch.apic) {
1740                 dev = &vcpu->arch.apic->dev;
1741                 if (dev->in_range(dev, addr))
1742                         return dev;
1743         }
1744         return NULL;
1745 }
1746
1747
1748 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1749                                                 gpa_t addr)
1750 {
1751         struct kvm_io_device *dev;
1752
1753         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1754         if (dev == NULL)
1755                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1756         return dev;
1757 }
1758
1759 int emulator_read_std(unsigned long addr,
1760                              void *val,
1761                              unsigned int bytes,
1762                              struct kvm_vcpu *vcpu)
1763 {
1764         void *data = val;
1765         int r = X86EMUL_CONTINUE;
1766
1767         down_read(&vcpu->kvm->slots_lock);
1768         while (bytes) {
1769                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1770                 unsigned offset = addr & (PAGE_SIZE-1);
1771                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1772                 int ret;
1773
1774                 if (gpa == UNMAPPED_GVA) {
1775                         r = X86EMUL_PROPAGATE_FAULT;
1776                         goto out;
1777                 }
1778                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1779                 if (ret < 0) {
1780                         r = X86EMUL_UNHANDLEABLE;
1781                         goto out;
1782                 }
1783
1784                 bytes -= tocopy;
1785                 data += tocopy;
1786                 addr += tocopy;
1787         }
1788 out:
1789         up_read(&vcpu->kvm->slots_lock);
1790         return r;
1791 }
1792 EXPORT_SYMBOL_GPL(emulator_read_std);
1793
1794 static int emulator_read_emulated(unsigned long addr,
1795                                   void *val,
1796                                   unsigned int bytes,
1797                                   struct kvm_vcpu *vcpu)
1798 {
1799         struct kvm_io_device *mmio_dev;
1800         gpa_t                 gpa;
1801
1802         if (vcpu->mmio_read_completed) {
1803                 memcpy(val, vcpu->mmio_data, bytes);
1804                 vcpu->mmio_read_completed = 0;
1805                 return X86EMUL_CONTINUE;
1806         }
1807
1808         down_read(&vcpu->kvm->slots_lock);
1809         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1810         up_read(&vcpu->kvm->slots_lock);
1811
1812         /* For APIC access vmexit */
1813         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1814                 goto mmio;
1815
1816         if (emulator_read_std(addr, val, bytes, vcpu)
1817                         == X86EMUL_CONTINUE)
1818                 return X86EMUL_CONTINUE;
1819         if (gpa == UNMAPPED_GVA)
1820                 return X86EMUL_PROPAGATE_FAULT;
1821
1822 mmio:
1823         /*
1824          * Is this MMIO handled locally?
1825          */
1826         mutex_lock(&vcpu->kvm->lock);
1827         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1828         if (mmio_dev) {
1829                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1830                 mutex_unlock(&vcpu->kvm->lock);
1831                 return X86EMUL_CONTINUE;
1832         }
1833         mutex_unlock(&vcpu->kvm->lock);
1834
1835         vcpu->mmio_needed = 1;
1836         vcpu->mmio_phys_addr = gpa;
1837         vcpu->mmio_size = bytes;
1838         vcpu->mmio_is_write = 0;
1839
1840         return X86EMUL_UNHANDLEABLE;
1841 }
1842
1843 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1844                                const void *val, int bytes)
1845 {
1846         int ret;
1847
1848         down_read(&vcpu->kvm->slots_lock);
1849         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1850         if (ret < 0) {
1851                 up_read(&vcpu->kvm->slots_lock);
1852                 return 0;
1853         }
1854         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1855         up_read(&vcpu->kvm->slots_lock);
1856         return 1;
1857 }
1858
1859 static int emulator_write_emulated_onepage(unsigned long addr,
1860                                            const void *val,
1861                                            unsigned int bytes,
1862                                            struct kvm_vcpu *vcpu)
1863 {
1864         struct kvm_io_device *mmio_dev;
1865         gpa_t                 gpa;
1866
1867         down_read(&vcpu->kvm->slots_lock);
1868         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1869         up_read(&vcpu->kvm->slots_lock);
1870
1871         if (gpa == UNMAPPED_GVA) {
1872                 kvm_inject_page_fault(vcpu, addr, 2);
1873                 return X86EMUL_PROPAGATE_FAULT;
1874         }
1875
1876         /* For APIC access vmexit */
1877         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1878                 goto mmio;
1879
1880         if (emulator_write_phys(vcpu, gpa, val, bytes))
1881                 return X86EMUL_CONTINUE;
1882
1883 mmio:
1884         /*
1885          * Is this MMIO handled locally?
1886          */
1887         mutex_lock(&vcpu->kvm->lock);
1888         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1889         if (mmio_dev) {
1890                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1891                 mutex_unlock(&vcpu->kvm->lock);
1892                 return X86EMUL_CONTINUE;
1893         }
1894         mutex_unlock(&vcpu->kvm->lock);
1895
1896         vcpu->mmio_needed = 1;
1897         vcpu->mmio_phys_addr = gpa;
1898         vcpu->mmio_size = bytes;
1899         vcpu->mmio_is_write = 1;
1900         memcpy(vcpu->mmio_data, val, bytes);
1901
1902         return X86EMUL_CONTINUE;
1903 }
1904
1905 int emulator_write_emulated(unsigned long addr,
1906                                    const void *val,
1907                                    unsigned int bytes,
1908                                    struct kvm_vcpu *vcpu)
1909 {
1910         /* Crossing a page boundary? */
1911         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1912                 int rc, now;
1913
1914                 now = -addr & ~PAGE_MASK;
1915                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1916                 if (rc != X86EMUL_CONTINUE)
1917                         return rc;
1918                 addr += now;
1919                 val += now;
1920                 bytes -= now;
1921         }
1922         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1923 }
1924 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1925
1926 static int emulator_cmpxchg_emulated(unsigned long addr,
1927                                      const void *old,
1928                                      const void *new,
1929                                      unsigned int bytes,
1930                                      struct kvm_vcpu *vcpu)
1931 {
1932         static int reported;
1933
1934         if (!reported) {
1935                 reported = 1;
1936                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1937         }
1938 #ifndef CONFIG_X86_64
1939         /* guests cmpxchg8b have to be emulated atomically */
1940         if (bytes == 8) {
1941                 gpa_t gpa;
1942                 struct page *page;
1943                 char *kaddr;
1944                 u64 val;
1945
1946                 down_read(&vcpu->kvm->slots_lock);
1947                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1948
1949                 if (gpa == UNMAPPED_GVA ||
1950                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1951                         goto emul_write;
1952
1953                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1954                         goto emul_write;
1955
1956                 val = *(u64 *)new;
1957
1958                 down_read(&current->mm->mmap_sem);
1959                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1960                 up_read(&current->mm->mmap_sem);
1961
1962                 kaddr = kmap_atomic(page, KM_USER0);
1963                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1964                 kunmap_atomic(kaddr, KM_USER0);
1965                 kvm_release_page_dirty(page);
1966         emul_write:
1967                 up_read(&vcpu->kvm->slots_lock);
1968         }
1969 #endif
1970
1971         return emulator_write_emulated(addr, new, bytes, vcpu);
1972 }
1973
1974 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1975 {
1976         return kvm_x86_ops->get_segment_base(vcpu, seg);
1977 }
1978
1979 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1980 {
1981         return X86EMUL_CONTINUE;
1982 }
1983
1984 int emulate_clts(struct kvm_vcpu *vcpu)
1985 {
1986         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1987         return X86EMUL_CONTINUE;
1988 }
1989
1990 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1991 {
1992         struct kvm_vcpu *vcpu = ctxt->vcpu;
1993
1994         switch (dr) {
1995         case 0 ... 3:
1996                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1997                 return X86EMUL_CONTINUE;
1998         default:
1999                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2000                 return X86EMUL_UNHANDLEABLE;
2001         }
2002 }
2003
2004 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2005 {
2006         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2007         int exception;
2008
2009         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2010         if (exception) {
2011                 /* FIXME: better handling */
2012                 return X86EMUL_UNHANDLEABLE;
2013         }
2014         return X86EMUL_CONTINUE;
2015 }
2016
2017 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2018 {
2019         static int reported;
2020         u8 opcodes[4];
2021         unsigned long rip = vcpu->arch.rip;
2022         unsigned long rip_linear;
2023
2024         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2025
2026         if (reported)
2027                 return;
2028
2029         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2030
2031         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2032                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2033         reported = 1;
2034 }
2035 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2036
2037 static struct x86_emulate_ops emulate_ops = {
2038         .read_std            = emulator_read_std,
2039         .read_emulated       = emulator_read_emulated,
2040         .write_emulated      = emulator_write_emulated,
2041         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2042 };
2043
2044 int emulate_instruction(struct kvm_vcpu *vcpu,
2045                         struct kvm_run *run,
2046                         unsigned long cr2,
2047                         u16 error_code,
2048                         int emulation_type)
2049 {
2050         int r;
2051         struct decode_cache *c;
2052
2053         vcpu->arch.mmio_fault_cr2 = cr2;
2054         kvm_x86_ops->cache_regs(vcpu);
2055
2056         vcpu->mmio_is_write = 0;
2057         vcpu->arch.pio.string = 0;
2058
2059         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2060                 int cs_db, cs_l;
2061                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2062
2063                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2064                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2065                 vcpu->arch.emulate_ctxt.mode =
2066                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2067                         ? X86EMUL_MODE_REAL : cs_l
2068                         ? X86EMUL_MODE_PROT64 : cs_db
2069                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2070
2071                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2072                         vcpu->arch.emulate_ctxt.cs_base = 0;
2073                         vcpu->arch.emulate_ctxt.ds_base = 0;
2074                         vcpu->arch.emulate_ctxt.es_base = 0;
2075                         vcpu->arch.emulate_ctxt.ss_base = 0;
2076                 } else {
2077                         vcpu->arch.emulate_ctxt.cs_base =
2078                                         get_segment_base(vcpu, VCPU_SREG_CS);
2079                         vcpu->arch.emulate_ctxt.ds_base =
2080                                         get_segment_base(vcpu, VCPU_SREG_DS);
2081                         vcpu->arch.emulate_ctxt.es_base =
2082                                         get_segment_base(vcpu, VCPU_SREG_ES);
2083                         vcpu->arch.emulate_ctxt.ss_base =
2084                                         get_segment_base(vcpu, VCPU_SREG_SS);
2085                 }
2086
2087                 vcpu->arch.emulate_ctxt.gs_base =
2088                                         get_segment_base(vcpu, VCPU_SREG_GS);
2089                 vcpu->arch.emulate_ctxt.fs_base =
2090                                         get_segment_base(vcpu, VCPU_SREG_FS);
2091
2092                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2093
2094                 /* Reject the instructions other than VMCALL/VMMCALL when
2095                  * try to emulate invalid opcode */
2096                 c = &vcpu->arch.emulate_ctxt.decode;
2097                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2098                     (!(c->twobyte && c->b == 0x01 &&
2099                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2100                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2101                         return EMULATE_FAIL;
2102
2103                 ++vcpu->stat.insn_emulation;
2104                 if (r)  {
2105                         ++vcpu->stat.insn_emulation_fail;
2106                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2107                                 return EMULATE_DONE;
2108                         return EMULATE_FAIL;
2109                 }
2110         }
2111
2112         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2113
2114         if (vcpu->arch.pio.string)
2115                 return EMULATE_DO_MMIO;
2116
2117         if ((r || vcpu->mmio_is_write) && run) {
2118                 run->exit_reason = KVM_EXIT_MMIO;
2119                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2120                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2121                 run->mmio.len = vcpu->mmio_size;
2122                 run->mmio.is_write = vcpu->mmio_is_write;
2123         }
2124
2125         if (r) {
2126                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2127                         return EMULATE_DONE;
2128                 if (!vcpu->mmio_needed) {
2129                         kvm_report_emulation_failure(vcpu, "mmio");
2130                         return EMULATE_FAIL;
2131                 }
2132                 return EMULATE_DO_MMIO;
2133         }
2134
2135         kvm_x86_ops->decache_regs(vcpu);
2136         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2137
2138         if (vcpu->mmio_is_write) {
2139                 vcpu->mmio_needed = 0;
2140                 return EMULATE_DO_MMIO;
2141         }
2142
2143         return EMULATE_DONE;
2144 }
2145 EXPORT_SYMBOL_GPL(emulate_instruction);
2146
2147 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2148 {
2149         int i;
2150
2151         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2152                 if (vcpu->arch.pio.guest_pages[i]) {
2153                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2154                         vcpu->arch.pio.guest_pages[i] = NULL;
2155                 }
2156 }
2157
2158 static int pio_copy_data(struct kvm_vcpu *vcpu)
2159 {
2160         void *p = vcpu->arch.pio_data;
2161         void *q;
2162         unsigned bytes;
2163         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2164
2165         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2166                  PAGE_KERNEL);
2167         if (!q) {
2168                 free_pio_guest_pages(vcpu);
2169                 return -ENOMEM;
2170         }
2171         q += vcpu->arch.pio.guest_page_offset;
2172         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2173         if (vcpu->arch.pio.in)
2174                 memcpy(q, p, bytes);
2175         else
2176                 memcpy(p, q, bytes);
2177         q -= vcpu->arch.pio.guest_page_offset;
2178         vunmap(q);
2179         free_pio_guest_pages(vcpu);
2180         return 0;
2181 }
2182
2183 int complete_pio(struct kvm_vcpu *vcpu)
2184 {
2185         struct kvm_pio_request *io = &vcpu->arch.pio;
2186         long delta;
2187         int r;
2188
2189         kvm_x86_ops->cache_regs(vcpu);
2190
2191         if (!io->string) {
2192                 if (io->in)
2193                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2194                                io->size);
2195         } else {
2196                 if (io->in) {
2197                         r = pio_copy_data(vcpu);
2198                         if (r) {
2199                                 kvm_x86_ops->cache_regs(vcpu);
2200                                 return r;
2201                         }
2202                 }
2203
2204                 delta = 1;
2205                 if (io->rep) {
2206                         delta *= io->cur_count;
2207                         /*
2208                          * The size of the register should really depend on
2209                          * current address size.
2210                          */
2211                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2212                 }
2213                 if (io->down)
2214                         delta = -delta;
2215                 delta *= io->size;
2216                 if (io->in)
2217                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2218                 else
2219                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2220         }
2221
2222         kvm_x86_ops->decache_regs(vcpu);
2223
2224         io->count -= io->cur_count;
2225         io->cur_count = 0;
2226
2227         return 0;
2228 }
2229
2230 static void kernel_pio(struct kvm_io_device *pio_dev,
2231                        struct kvm_vcpu *vcpu,
2232                        void *pd)
2233 {
2234         /* TODO: String I/O for in kernel device */
2235
2236         mutex_lock(&vcpu->kvm->lock);
2237         if (vcpu->arch.pio.in)
2238                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2239                                   vcpu->arch.pio.size,
2240                                   pd);
2241         else
2242                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2243                                    vcpu->arch.pio.size,
2244                                    pd);
2245         mutex_unlock(&vcpu->kvm->lock);
2246 }
2247
2248 static void pio_string_write(struct kvm_io_device *pio_dev,
2249                              struct kvm_vcpu *vcpu)
2250 {
2251         struct kvm_pio_request *io = &vcpu->arch.pio;
2252         void *pd = vcpu->arch.pio_data;
2253         int i;
2254
2255         mutex_lock(&vcpu->kvm->lock);
2256         for (i = 0; i < io->cur_count; i++) {
2257                 kvm_iodevice_write(pio_dev, io->port,
2258                                    io->size,
2259                                    pd);
2260                 pd += io->size;
2261         }
2262         mutex_unlock(&vcpu->kvm->lock);
2263 }
2264
2265 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2266                                                gpa_t addr)
2267 {
2268         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2269 }
2270
2271 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2272                   int size, unsigned port)
2273 {
2274         struct kvm_io_device *pio_dev;
2275
2276         vcpu->run->exit_reason = KVM_EXIT_IO;
2277         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2278         vcpu->run->io.size = vcpu->arch.pio.size = size;
2279         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2280         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2281         vcpu->run->io.port = vcpu->arch.pio.port = port;
2282         vcpu->arch.pio.in = in;
2283         vcpu->arch.pio.string = 0;
2284         vcpu->arch.pio.down = 0;
2285         vcpu->arch.pio.guest_page_offset = 0;
2286         vcpu->arch.pio.rep = 0;
2287
2288         kvm_x86_ops->cache_regs(vcpu);
2289         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2290         kvm_x86_ops->decache_regs(vcpu);
2291
2292         kvm_x86_ops->skip_emulated_instruction(vcpu);
2293
2294         pio_dev = vcpu_find_pio_dev(vcpu, port);
2295         if (pio_dev) {
2296                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2297                 complete_pio(vcpu);
2298                 return 1;
2299         }
2300         return 0;
2301 }
2302 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2303
2304 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2305                   int size, unsigned long count, int down,
2306                   gva_t address, int rep, unsigned port)
2307 {
2308         unsigned now, in_page;
2309         int i, ret = 0;
2310         int nr_pages = 1;
2311         struct page *page;
2312         struct kvm_io_device *pio_dev;
2313
2314         vcpu->run->exit_reason = KVM_EXIT_IO;
2315         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2316         vcpu->run->io.size = vcpu->arch.pio.size = size;
2317         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2318         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2319         vcpu->run->io.port = vcpu->arch.pio.port = port;
2320         vcpu->arch.pio.in = in;
2321         vcpu->arch.pio.string = 1;
2322         vcpu->arch.pio.down = down;
2323         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2324         vcpu->arch.pio.rep = rep;
2325
2326         if (!count) {
2327                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2328                 return 1;
2329         }
2330
2331         if (!down)
2332                 in_page = PAGE_SIZE - offset_in_page(address);
2333         else
2334                 in_page = offset_in_page(address) + size;
2335         now = min(count, (unsigned long)in_page / size);
2336         if (!now) {
2337                 /*
2338                  * String I/O straddles page boundary.  Pin two guest pages
2339                  * so that we satisfy atomicity constraints.  Do just one
2340                  * transaction to avoid complexity.
2341                  */
2342                 nr_pages = 2;
2343                 now = 1;
2344         }
2345         if (down) {
2346                 /*
2347                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2348                  */
2349                 pr_unimpl(vcpu, "guest string pio down\n");
2350                 kvm_inject_gp(vcpu, 0);
2351                 return 1;
2352         }
2353         vcpu->run->io.count = now;
2354         vcpu->arch.pio.cur_count = now;
2355
2356         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2357                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2358
2359         for (i = 0; i < nr_pages; ++i) {
2360                 down_read(&vcpu->kvm->slots_lock);
2361                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2362                 vcpu->arch.pio.guest_pages[i] = page;
2363                 up_read(&vcpu->kvm->slots_lock);
2364                 if (!page) {
2365                         kvm_inject_gp(vcpu, 0);
2366                         free_pio_guest_pages(vcpu);
2367                         return 1;
2368                 }
2369         }
2370
2371         pio_dev = vcpu_find_pio_dev(vcpu, port);
2372         if (!vcpu->arch.pio.in) {
2373                 /* string PIO write */
2374                 ret = pio_copy_data(vcpu);
2375                 if (ret >= 0 && pio_dev) {
2376                         pio_string_write(pio_dev, vcpu);
2377                         complete_pio(vcpu);
2378                         if (vcpu->arch.pio.count == 0)
2379                                 ret = 1;
2380                 }
2381         } else if (pio_dev)
2382                 pr_unimpl(vcpu, "no string pio read support yet, "
2383                        "port %x size %d count %ld\n",
2384                         port, size, count);
2385
2386         return ret;
2387 }
2388 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2389
2390 int kvm_arch_init(void *opaque)
2391 {
2392         int r;
2393         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2394
2395         if (kvm_x86_ops) {
2396                 printk(KERN_ERR "kvm: already loaded the other module\n");
2397                 r = -EEXIST;
2398                 goto out;
2399         }
2400
2401         if (!ops->cpu_has_kvm_support()) {
2402                 printk(KERN_ERR "kvm: no hardware support\n");
2403                 r = -EOPNOTSUPP;
2404                 goto out;
2405         }
2406         if (ops->disabled_by_bios()) {
2407                 printk(KERN_ERR "kvm: disabled by bios\n");
2408                 r = -EOPNOTSUPP;
2409                 goto out;
2410         }
2411
2412         r = kvm_mmu_module_init();
2413         if (r)
2414                 goto out;
2415
2416         kvm_init_msr_list();
2417
2418         kvm_x86_ops = ops;
2419         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2420         return 0;
2421
2422 out:
2423         return r;
2424 }
2425
2426 void kvm_arch_exit(void)
2427 {
2428         kvm_x86_ops = NULL;
2429         kvm_mmu_module_exit();
2430 }
2431
2432 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2433 {
2434         ++vcpu->stat.halt_exits;
2435         if (irqchip_in_kernel(vcpu->kvm)) {
2436                 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2437                 kvm_vcpu_block(vcpu);
2438                 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2439                         return -EINTR;
2440                 return 1;
2441         } else {
2442                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2443                 return 0;
2444         }
2445 }
2446 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2447
2448 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2449 {
2450         unsigned long nr, a0, a1, a2, a3, ret;
2451
2452         kvm_x86_ops->cache_regs(vcpu);
2453
2454         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2455         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2456         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2457         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2458         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2459
2460         if (!is_long_mode(vcpu)) {
2461                 nr &= 0xFFFFFFFF;
2462                 a0 &= 0xFFFFFFFF;
2463                 a1 &= 0xFFFFFFFF;
2464                 a2 &= 0xFFFFFFFF;
2465                 a3 &= 0xFFFFFFFF;
2466         }
2467
2468         switch (nr) {
2469         case KVM_HC_VAPIC_POLL_IRQ:
2470                 ret = 0;
2471                 break;
2472         default:
2473                 ret = -KVM_ENOSYS;
2474                 break;
2475         }
2476         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2477         kvm_x86_ops->decache_regs(vcpu);
2478         ++vcpu->stat.hypercalls;
2479         return 0;
2480 }
2481 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2482
2483 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2484 {
2485         char instruction[3];
2486         int ret = 0;
2487
2488
2489         /*
2490          * Blow out the MMU to ensure that no other VCPU has an active mapping
2491          * to ensure that the updated hypercall appears atomically across all
2492          * VCPUs.
2493          */
2494         kvm_mmu_zap_all(vcpu->kvm);
2495
2496         kvm_x86_ops->cache_regs(vcpu);
2497         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2498         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2499             != X86EMUL_CONTINUE)
2500                 ret = -EFAULT;
2501
2502         return ret;
2503 }
2504
2505 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2506 {
2507         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2508 }
2509
2510 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2511 {
2512         struct descriptor_table dt = { limit, base };
2513
2514         kvm_x86_ops->set_gdt(vcpu, &dt);
2515 }
2516
2517 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2518 {
2519         struct descriptor_table dt = { limit, base };
2520
2521         kvm_x86_ops->set_idt(vcpu, &dt);
2522 }
2523
2524 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2525                    unsigned long *rflags)
2526 {
2527         kvm_lmsw(vcpu, msw);
2528         *rflags = kvm_x86_ops->get_rflags(vcpu);
2529 }
2530
2531 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2532 {
2533         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2534         switch (cr) {
2535         case 0:
2536                 return vcpu->arch.cr0;
2537         case 2:
2538                 return vcpu->arch.cr2;
2539         case 3:
2540                 return vcpu->arch.cr3;
2541         case 4:
2542                 return vcpu->arch.cr4;
2543         case 8:
2544                 return kvm_get_cr8(vcpu);
2545         default:
2546                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2547                 return 0;
2548         }
2549 }
2550
2551 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2552                      unsigned long *rflags)
2553 {
2554         switch (cr) {
2555         case 0:
2556                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2557                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2558                 break;
2559         case 2:
2560                 vcpu->arch.cr2 = val;
2561                 break;
2562         case 3:
2563                 kvm_set_cr3(vcpu, val);
2564                 break;
2565         case 4:
2566                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2567                 break;
2568         case 8:
2569                 kvm_set_cr8(vcpu, val & 0xfUL);
2570                 break;
2571         default:
2572                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2573         }
2574 }
2575
2576 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2577 {
2578         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2579         int j, nent = vcpu->arch.cpuid_nent;
2580
2581         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2582         /* when no next entry is found, the current entry[i] is reselected */
2583         for (j = i + 1; j == i; j = (j + 1) % nent) {
2584                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2585                 if (ej->function == e->function) {
2586                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2587                         return j;
2588                 }
2589         }
2590         return 0; /* silence gcc, even though control never reaches here */
2591 }
2592
2593 /* find an entry with matching function, matching index (if needed), and that
2594  * should be read next (if it's stateful) */
2595 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2596         u32 function, u32 index)
2597 {
2598         if (e->function != function)
2599                 return 0;
2600         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2601                 return 0;
2602         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2603                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2604                 return 0;
2605         return 1;
2606 }
2607
2608 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2609 {
2610         int i;
2611         u32 function, index;
2612         struct kvm_cpuid_entry2 *e, *best;
2613
2614         kvm_x86_ops->cache_regs(vcpu);
2615         function = vcpu->arch.regs[VCPU_REGS_RAX];
2616         index = vcpu->arch.regs[VCPU_REGS_RCX];
2617         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2618         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2619         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2620         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2621         best = NULL;
2622         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2623                 e = &vcpu->arch.cpuid_entries[i];
2624                 if (is_matching_cpuid_entry(e, function, index)) {
2625                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2626                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2627                         best = e;
2628                         break;
2629                 }
2630                 /*
2631                  * Both basic or both extended?
2632                  */
2633                 if (((e->function ^ function) & 0x80000000) == 0)
2634                         if (!best || e->function > best->function)
2635                                 best = e;
2636         }
2637         if (best) {
2638                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2639                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2640                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2641                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2642         }
2643         kvm_x86_ops->decache_regs(vcpu);
2644         kvm_x86_ops->skip_emulated_instruction(vcpu);
2645 }
2646 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2647
2648 /*
2649  * Check if userspace requested an interrupt window, and that the
2650  * interrupt window is open.
2651  *
2652  * No need to exit to userspace if we already have an interrupt queued.
2653  */
2654 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2655                                           struct kvm_run *kvm_run)
2656 {
2657         return (!vcpu->arch.irq_summary &&
2658                 kvm_run->request_interrupt_window &&
2659                 vcpu->arch.interrupt_window_open &&
2660                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2661 }
2662
2663 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2664                               struct kvm_run *kvm_run)
2665 {
2666         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2667         kvm_run->cr8 = kvm_get_cr8(vcpu);
2668         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2669         if (irqchip_in_kernel(vcpu->kvm))
2670                 kvm_run->ready_for_interrupt_injection = 1;
2671         else
2672                 kvm_run->ready_for_interrupt_injection =
2673                                         (vcpu->arch.interrupt_window_open &&
2674                                          vcpu->arch.irq_summary == 0);
2675 }
2676
2677 static void vapic_enter(struct kvm_vcpu *vcpu)
2678 {
2679         struct kvm_lapic *apic = vcpu->arch.apic;
2680         struct page *page;
2681
2682         if (!apic || !apic->vapic_addr)
2683                 return;
2684
2685         down_read(&current->mm->mmap_sem);
2686         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2687         up_read(&current->mm->mmap_sem);
2688
2689         vcpu->arch.apic->vapic_page = page;
2690 }
2691
2692 static void vapic_exit(struct kvm_vcpu *vcpu)
2693 {
2694         struct kvm_lapic *apic = vcpu->arch.apic;
2695
2696         if (!apic || !apic->vapic_addr)
2697                 return;
2698
2699         kvm_release_page_dirty(apic->vapic_page);
2700         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2701 }
2702
2703 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2704 {
2705         int r;
2706
2707         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2708                 pr_debug("vcpu %d received sipi with vector # %x\n",
2709                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2710                 kvm_lapic_reset(vcpu);
2711                 r = kvm_x86_ops->vcpu_reset(vcpu);
2712                 if (r)
2713                         return r;
2714                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2715         }
2716
2717         vapic_enter(vcpu);
2718
2719 preempted:
2720         if (vcpu->guest_debug.enabled)
2721                 kvm_x86_ops->guest_debug_pre(vcpu);
2722
2723 again:
2724         if (vcpu->requests)
2725                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2726                         kvm_mmu_unload(vcpu);
2727
2728         r = kvm_mmu_reload(vcpu);
2729         if (unlikely(r))
2730                 goto out;
2731
2732         if (vcpu->requests) {
2733                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2734                         __kvm_migrate_apic_timer(vcpu);
2735                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2736                                        &vcpu->requests)) {
2737                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2738                         r = 0;
2739                         goto out;
2740                 }
2741                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2742                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2743                         r = 0;
2744                         goto out;
2745                 }
2746         }
2747
2748         kvm_inject_pending_timer_irqs(vcpu);
2749
2750         preempt_disable();
2751
2752         kvm_x86_ops->prepare_guest_switch(vcpu);
2753         kvm_load_guest_fpu(vcpu);
2754
2755         local_irq_disable();
2756
2757         if (need_resched()) {
2758                 local_irq_enable();
2759                 preempt_enable();
2760                 r = 1;
2761                 goto out;
2762         }
2763
2764         if (vcpu->requests)
2765                 if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
2766                         local_irq_enable();
2767                         preempt_enable();
2768                         r = 1;
2769                         goto out;
2770                 }
2771
2772         if (signal_pending(current)) {
2773                 local_irq_enable();
2774                 preempt_enable();
2775                 r = -EINTR;
2776                 kvm_run->exit_reason = KVM_EXIT_INTR;
2777                 ++vcpu->stat.signal_exits;
2778                 goto out;
2779         }
2780
2781         if (vcpu->arch.exception.pending)
2782                 __queue_exception(vcpu);
2783         else if (irqchip_in_kernel(vcpu->kvm))
2784                 kvm_x86_ops->inject_pending_irq(vcpu);
2785         else
2786                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2787
2788         kvm_lapic_sync_to_vapic(vcpu);
2789
2790         vcpu->guest_mode = 1;
2791         kvm_guest_enter();
2792
2793         if (vcpu->requests)
2794                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2795                         kvm_x86_ops->tlb_flush(vcpu);
2796
2797         kvm_x86_ops->run(vcpu, kvm_run);
2798
2799         vcpu->guest_mode = 0;
2800         local_irq_enable();
2801
2802         ++vcpu->stat.exits;
2803
2804         /*
2805          * We must have an instruction between local_irq_enable() and
2806          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2807          * the interrupt shadow.  The stat.exits increment will do nicely.
2808          * But we need to prevent reordering, hence this barrier():
2809          */
2810         barrier();
2811
2812         kvm_guest_exit();
2813
2814         preempt_enable();
2815
2816         /*
2817          * Profile KVM exit RIPs:
2818          */
2819         if (unlikely(prof_on == KVM_PROFILING)) {
2820                 kvm_x86_ops->cache_regs(vcpu);
2821                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2822         }
2823
2824         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2825                 vcpu->arch.exception.pending = false;
2826
2827         kvm_lapic_sync_from_vapic(vcpu);
2828
2829         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2830
2831         if (r > 0) {
2832                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2833                         r = -EINTR;
2834                         kvm_run->exit_reason = KVM_EXIT_INTR;
2835                         ++vcpu->stat.request_irq_exits;
2836                         goto out;
2837                 }
2838                 if (!need_resched())
2839                         goto again;
2840         }
2841
2842 out:
2843         if (r > 0) {
2844                 kvm_resched(vcpu);
2845                 goto preempted;
2846         }
2847
2848         post_kvm_run_save(vcpu, kvm_run);
2849
2850         vapic_exit(vcpu);
2851
2852         return r;
2853 }
2854
2855 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2856 {
2857         int r;
2858         sigset_t sigsaved;
2859
2860         vcpu_load(vcpu);
2861
2862         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2863                 kvm_vcpu_block(vcpu);
2864                 vcpu_put(vcpu);
2865                 return -EAGAIN;
2866         }
2867
2868         if (vcpu->sigset_active)
2869                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2870
2871         /* re-sync apic's tpr */
2872         if (!irqchip_in_kernel(vcpu->kvm))
2873                 kvm_set_cr8(vcpu, kvm_run->cr8);
2874
2875         if (vcpu->arch.pio.cur_count) {
2876                 r = complete_pio(vcpu);
2877                 if (r)
2878                         goto out;
2879         }
2880 #if CONFIG_HAS_IOMEM
2881         if (vcpu->mmio_needed) {
2882                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2883                 vcpu->mmio_read_completed = 1;
2884                 vcpu->mmio_needed = 0;
2885                 r = emulate_instruction(vcpu, kvm_run,
2886                                         vcpu->arch.mmio_fault_cr2, 0,
2887                                         EMULTYPE_NO_DECODE);
2888                 if (r == EMULATE_DO_MMIO) {
2889                         /*
2890                          * Read-modify-write.  Back to userspace.
2891                          */
2892                         r = 0;
2893                         goto out;
2894                 }
2895         }
2896 #endif
2897         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2898                 kvm_x86_ops->cache_regs(vcpu);
2899                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2900                 kvm_x86_ops->decache_regs(vcpu);
2901         }
2902
2903         r = __vcpu_run(vcpu, kvm_run);
2904
2905 out:
2906         if (vcpu->sigset_active)
2907                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2908
2909         vcpu_put(vcpu);
2910         return r;
2911 }
2912
2913 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2914 {
2915         vcpu_load(vcpu);
2916
2917         kvm_x86_ops->cache_regs(vcpu);
2918
2919         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2920         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2921         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2922         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2923         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2924         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2925         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2926         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2927 #ifdef CONFIG_X86_64
2928         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2929         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2930         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2931         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2932         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2933         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2934         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2935         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2936 #endif
2937
2938         regs->rip = vcpu->arch.rip;
2939         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2940
2941         /*
2942          * Don't leak debug flags in case they were set for guest debugging
2943          */
2944         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2945                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2946
2947         vcpu_put(vcpu);
2948
2949         return 0;
2950 }
2951
2952 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2953 {
2954         vcpu_load(vcpu);
2955
2956         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2957         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2958         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2959         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2960         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2961         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2962         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2963         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2964 #ifdef CONFIG_X86_64
2965         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2966         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2967         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2968         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2969         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2970         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2971         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2972         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2973 #endif
2974
2975         vcpu->arch.rip = regs->rip;
2976         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2977
2978         kvm_x86_ops->decache_regs(vcpu);
2979
2980         vcpu_put(vcpu);
2981
2982         return 0;
2983 }
2984
2985 static void get_segment(struct kvm_vcpu *vcpu,
2986                         struct kvm_segment *var, int seg)
2987 {
2988         kvm_x86_ops->get_segment(vcpu, var, seg);
2989 }
2990
2991 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2992 {
2993         struct kvm_segment cs;
2994
2995         get_segment(vcpu, &cs, VCPU_SREG_CS);
2996         *db = cs.db;
2997         *l = cs.l;
2998 }
2999 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3000
3001 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3002                                   struct kvm_sregs *sregs)
3003 {
3004         struct descriptor_table dt;
3005         int pending_vec;
3006
3007         vcpu_load(vcpu);
3008
3009         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3010         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3011         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3012         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3013         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3014         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3015
3016         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3017         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3018
3019         kvm_x86_ops->get_idt(vcpu, &dt);
3020         sregs->idt.limit = dt.limit;
3021         sregs->idt.base = dt.base;
3022         kvm_x86_ops->get_gdt(vcpu, &dt);
3023         sregs->gdt.limit = dt.limit;
3024         sregs->gdt.base = dt.base;
3025
3026         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3027         sregs->cr0 = vcpu->arch.cr0;
3028         sregs->cr2 = vcpu->arch.cr2;
3029         sregs->cr3 = vcpu->arch.cr3;
3030         sregs->cr4 = vcpu->arch.cr4;
3031         sregs->cr8 = kvm_get_cr8(vcpu);
3032         sregs->efer = vcpu->arch.shadow_efer;
3033         sregs->apic_base = kvm_get_apic_base(vcpu);
3034
3035         if (irqchip_in_kernel(vcpu->kvm)) {
3036                 memset(sregs->interrupt_bitmap, 0,
3037                        sizeof sregs->interrupt_bitmap);
3038                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3039                 if (pending_vec >= 0)
3040                         set_bit(pending_vec,
3041                                 (unsigned long *)sregs->interrupt_bitmap);
3042         } else
3043                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3044                        sizeof sregs->interrupt_bitmap);
3045
3046         vcpu_put(vcpu);
3047
3048         return 0;
3049 }
3050
3051 static void set_segment(struct kvm_vcpu *vcpu,
3052                         struct kvm_segment *var, int seg)
3053 {
3054         kvm_x86_ops->set_segment(vcpu, var, seg);
3055 }
3056
3057 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3058                                   struct kvm_sregs *sregs)
3059 {
3060         int mmu_reset_needed = 0;
3061         int i, pending_vec, max_bits;
3062         struct descriptor_table dt;
3063
3064         vcpu_load(vcpu);
3065
3066         dt.limit = sregs->idt.limit;
3067         dt.base = sregs->idt.base;
3068         kvm_x86_ops->set_idt(vcpu, &dt);
3069         dt.limit = sregs->gdt.limit;
3070         dt.base = sregs->gdt.base;
3071         kvm_x86_ops->set_gdt(vcpu, &dt);
3072
3073         vcpu->arch.cr2 = sregs->cr2;
3074         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3075         vcpu->arch.cr3 = sregs->cr3;
3076
3077         kvm_set_cr8(vcpu, sregs->cr8);
3078
3079         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3080         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3081         kvm_set_apic_base(vcpu, sregs->apic_base);
3082
3083         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3084
3085         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3086         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3087         vcpu->arch.cr0 = sregs->cr0;
3088
3089         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3090         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3091         if (!is_long_mode(vcpu) && is_pae(vcpu))
3092                 load_pdptrs(vcpu, vcpu->arch.cr3);
3093
3094         if (mmu_reset_needed)
3095                 kvm_mmu_reset_context(vcpu);
3096
3097         if (!irqchip_in_kernel(vcpu->kvm)) {
3098                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3099                        sizeof vcpu->arch.irq_pending);
3100                 vcpu->arch.irq_summary = 0;
3101                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3102                         if (vcpu->arch.irq_pending[i])
3103                                 __set_bit(i, &vcpu->arch.irq_summary);
3104         } else {
3105                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3106                 pending_vec = find_first_bit(
3107                         (const unsigned long *)sregs->interrupt_bitmap,
3108                         max_bits);
3109                 /* Only pending external irq is handled here */
3110                 if (pending_vec < max_bits) {
3111                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3112                         pr_debug("Set back pending irq %d\n",
3113                                  pending_vec);
3114                 }
3115         }
3116
3117         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3118         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3119         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3120         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3121         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3122         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3123
3124         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3125         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3126
3127         vcpu_put(vcpu);
3128
3129         return 0;
3130 }
3131
3132 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3133                                     struct kvm_debug_guest *dbg)
3134 {
3135         int r;
3136
3137         vcpu_load(vcpu);
3138
3139         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3140
3141         vcpu_put(vcpu);
3142
3143         return r;
3144 }
3145
3146 /*
3147  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3148  * we have asm/x86/processor.h
3149  */
3150 struct fxsave {
3151         u16     cwd;
3152         u16     swd;
3153         u16     twd;
3154         u16     fop;
3155         u64     rip;
3156         u64     rdp;
3157         u32     mxcsr;
3158         u32     mxcsr_mask;
3159         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3160 #ifdef CONFIG_X86_64
3161         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3162 #else
3163         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3164 #endif
3165 };
3166
3167 /*
3168  * Translate a guest virtual address to a guest physical address.
3169  */
3170 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3171                                     struct kvm_translation *tr)
3172 {
3173         unsigned long vaddr = tr->linear_address;
3174         gpa_t gpa;
3175
3176         vcpu_load(vcpu);
3177         down_read(&vcpu->kvm->slots_lock);
3178         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3179         up_read(&vcpu->kvm->slots_lock);
3180         tr->physical_address = gpa;
3181         tr->valid = gpa != UNMAPPED_GVA;
3182         tr->writeable = 1;
3183         tr->usermode = 0;
3184         vcpu_put(vcpu);
3185
3186         return 0;
3187 }
3188
3189 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3190 {
3191         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3192
3193         vcpu_load(vcpu);
3194
3195         memcpy(fpu->fpr, fxsave->st_space, 128);
3196         fpu->fcw = fxsave->cwd;
3197         fpu->fsw = fxsave->swd;
3198         fpu->ftwx = fxsave->twd;
3199         fpu->last_opcode = fxsave->fop;
3200         fpu->last_ip = fxsave->rip;
3201         fpu->last_dp = fxsave->rdp;
3202         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3203
3204         vcpu_put(vcpu);
3205
3206         return 0;
3207 }
3208
3209 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3210 {
3211         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3212
3213         vcpu_load(vcpu);
3214
3215         memcpy(fxsave->st_space, fpu->fpr, 128);
3216         fxsave->cwd = fpu->fcw;
3217         fxsave->swd = fpu->fsw;
3218         fxsave->twd = fpu->ftwx;
3219         fxsave->fop = fpu->last_opcode;
3220         fxsave->rip = fpu->last_ip;
3221         fxsave->rdp = fpu->last_dp;
3222         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3223
3224         vcpu_put(vcpu);
3225
3226         return 0;
3227 }
3228
3229 void fx_init(struct kvm_vcpu *vcpu)
3230 {
3231         unsigned after_mxcsr_mask;
3232
3233         /* Initialize guest FPU by resetting ours and saving into guest's */
3234         preempt_disable();
3235         fx_save(&vcpu->arch.host_fx_image);
3236         fpu_init();
3237         fx_save(&vcpu->arch.guest_fx_image);
3238         fx_restore(&vcpu->arch.host_fx_image);
3239         preempt_enable();
3240
3241         vcpu->arch.cr0 |= X86_CR0_ET;
3242         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3243         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3244         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3245                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3246 }
3247 EXPORT_SYMBOL_GPL(fx_init);
3248
3249 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3250 {
3251         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3252                 return;
3253
3254         vcpu->guest_fpu_loaded = 1;
3255         fx_save(&vcpu->arch.host_fx_image);
3256         fx_restore(&vcpu->arch.guest_fx_image);
3257 }
3258 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3259
3260 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3261 {
3262         if (!vcpu->guest_fpu_loaded)
3263                 return;
3264
3265         vcpu->guest_fpu_loaded = 0;
3266         fx_save(&vcpu->arch.guest_fx_image);
3267         fx_restore(&vcpu->arch.host_fx_image);
3268         ++vcpu->stat.fpu_reload;
3269 }
3270 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3271
3272 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3273 {
3274         kvm_x86_ops->vcpu_free(vcpu);
3275 }
3276
3277 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3278                                                 unsigned int id)
3279 {
3280         return kvm_x86_ops->vcpu_create(kvm, id);
3281 }
3282
3283 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3284 {
3285         int r;
3286
3287         /* We do fxsave: this must be aligned. */
3288         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3289
3290         vcpu_load(vcpu);
3291         r = kvm_arch_vcpu_reset(vcpu);
3292         if (r == 0)
3293                 r = kvm_mmu_setup(vcpu);
3294         vcpu_put(vcpu);
3295         if (r < 0)
3296                 goto free_vcpu;
3297
3298         return 0;
3299 free_vcpu:
3300         kvm_x86_ops->vcpu_free(vcpu);
3301         return r;
3302 }
3303
3304 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3305 {
3306         vcpu_load(vcpu);
3307         kvm_mmu_unload(vcpu);
3308         vcpu_put(vcpu);
3309
3310         kvm_x86_ops->vcpu_free(vcpu);
3311 }
3312
3313 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3314 {
3315         return kvm_x86_ops->vcpu_reset(vcpu);
3316 }
3317
3318 void kvm_arch_hardware_enable(void *garbage)
3319 {
3320         kvm_x86_ops->hardware_enable(garbage);
3321 }
3322
3323 void kvm_arch_hardware_disable(void *garbage)
3324 {
3325         kvm_x86_ops->hardware_disable(garbage);
3326 }
3327
3328 int kvm_arch_hardware_setup(void)
3329 {
3330         return kvm_x86_ops->hardware_setup();
3331 }
3332
3333 void kvm_arch_hardware_unsetup(void)
3334 {
3335         kvm_x86_ops->hardware_unsetup();
3336 }
3337
3338 void kvm_arch_check_processor_compat(void *rtn)
3339 {
3340         kvm_x86_ops->check_processor_compatibility(rtn);
3341 }
3342
3343 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3344 {
3345         struct page *page;
3346         struct kvm *kvm;
3347         int r;
3348
3349         BUG_ON(vcpu->kvm == NULL);
3350         kvm = vcpu->kvm;
3351
3352         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3353         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3354                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3355         else
3356                 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3357
3358         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3359         if (!page) {
3360                 r = -ENOMEM;
3361                 goto fail;
3362         }
3363         vcpu->arch.pio_data = page_address(page);
3364
3365         r = kvm_mmu_create(vcpu);
3366         if (r < 0)
3367                 goto fail_free_pio_data;
3368
3369         if (irqchip_in_kernel(kvm)) {
3370                 r = kvm_create_lapic(vcpu);
3371                 if (r < 0)
3372                         goto fail_mmu_destroy;
3373         }
3374
3375         return 0;
3376
3377 fail_mmu_destroy:
3378         kvm_mmu_destroy(vcpu);
3379 fail_free_pio_data:
3380         free_page((unsigned long)vcpu->arch.pio_data);
3381 fail:
3382         return r;
3383 }
3384
3385 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3386 {
3387         kvm_free_lapic(vcpu);
3388         kvm_mmu_destroy(vcpu);
3389         free_page((unsigned long)vcpu->arch.pio_data);
3390 }
3391
3392 struct  kvm *kvm_arch_create_vm(void)
3393 {
3394         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3395
3396         if (!kvm)
3397                 return ERR_PTR(-ENOMEM);
3398
3399         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3400
3401         return kvm;
3402 }
3403
3404 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3405 {
3406         vcpu_load(vcpu);
3407         kvm_mmu_unload(vcpu);
3408         vcpu_put(vcpu);
3409 }
3410
3411 static void kvm_free_vcpus(struct kvm *kvm)
3412 {
3413         unsigned int i;
3414
3415         /*
3416          * Unpin any mmu pages first.
3417          */
3418         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3419                 if (kvm->vcpus[i])
3420                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3421         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3422                 if (kvm->vcpus[i]) {
3423                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3424                         kvm->vcpus[i] = NULL;
3425                 }
3426         }
3427
3428 }
3429
3430 void kvm_arch_destroy_vm(struct kvm *kvm)
3431 {
3432         kvm_free_pit(kvm);
3433         kfree(kvm->arch.vpic);
3434         kfree(kvm->arch.vioapic);
3435         kvm_free_vcpus(kvm);
3436         kvm_free_physmem(kvm);
3437         kfree(kvm);
3438 }
3439
3440 int kvm_arch_set_memory_region(struct kvm *kvm,
3441                                 struct kvm_userspace_memory_region *mem,
3442                                 struct kvm_memory_slot old,
3443                                 int user_alloc)
3444 {
3445         int npages = mem->memory_size >> PAGE_SHIFT;
3446         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3447
3448         /*To keep backward compatibility with older userspace,
3449          *x86 needs to hanlde !user_alloc case.
3450          */
3451         if (!user_alloc) {
3452                 if (npages && !old.rmap) {
3453                         down_write(&current->mm->mmap_sem);
3454                         memslot->userspace_addr = do_mmap(NULL, 0,
3455                                                      npages * PAGE_SIZE,
3456                                                      PROT_READ | PROT_WRITE,
3457                                                      MAP_SHARED | MAP_ANONYMOUS,
3458                                                      0);
3459                         up_write(&current->mm->mmap_sem);
3460
3461                         if (IS_ERR((void *)memslot->userspace_addr))
3462                                 return PTR_ERR((void *)memslot->userspace_addr);
3463                 } else {
3464                         if (!old.user_alloc && old.rmap) {
3465                                 int ret;
3466
3467                                 down_write(&current->mm->mmap_sem);
3468                                 ret = do_munmap(current->mm, old.userspace_addr,
3469                                                 old.npages * PAGE_SIZE);
3470                                 up_write(&current->mm->mmap_sem);
3471                                 if (ret < 0)
3472                                         printk(KERN_WARNING
3473                                        "kvm_vm_ioctl_set_memory_region: "
3474                                        "failed to munmap memory\n");
3475                         }
3476                 }
3477         }
3478
3479         if (!kvm->arch.n_requested_mmu_pages) {
3480                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3481                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3482         }
3483
3484         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3485         kvm_flush_remote_tlbs(kvm);
3486
3487         return 0;
3488 }
3489
3490 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3491 {
3492         return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3493                || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3494 }
3495
3496 static void vcpu_kick_intr(void *info)
3497 {
3498 #ifdef DEBUG
3499         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3500         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3501 #endif
3502 }
3503
3504 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3505 {
3506         int ipi_pcpu = vcpu->cpu;
3507
3508         if (waitqueue_active(&vcpu->wq)) {
3509                 wake_up_interruptible(&vcpu->wq);
3510                 ++vcpu->stat.halt_wakeup;
3511         }
3512         if (vcpu->guest_mode)
3513                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
3514 }