arch/x86/kvm/svm.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * AMD SVM support
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8  *
   9  * Authors:
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17 #include <linux/kvm_host.h>
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "kvm_cache_regs.h"
  22 #include "x86.h"
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/highmem.h>
  28 #include <linux/sched.h>
  29 #include <linux/ftrace_event.h>
  30 #include <linux/slab.h>
  31
  32 #include <asm/tlbflush.h>
  33 #include <asm/desc.h>
  34 #include <asm/kvm_para.h>
  35
  36 #include <asm/virtext.h>
  37 #include "trace.h"
  38
  39 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  40
  41 MODULE_AUTHOR("Qumranet");
  42 MODULE_LICENSE("GPL");
  43
  44 #define IOPM_ALLOC_ORDER 2
  45 #define MSRPM_ALLOC_ORDER 1
  46
  47 #define SEG_TYPE_LDT 2
  48 #define SEG_TYPE_BUSY_TSS16 3
  49
  50 #define SVM_FEATURE_NPT            (1 <<  0)
  51 #define SVM_FEATURE_LBRV           (1 <<  1)
  52 #define SVM_FEATURE_SVML           (1 <<  2)
  53 #define SVM_FEATURE_NRIP           (1 <<  3)
  54 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  55 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  56 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  57 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  58 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  59
  60 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  61 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  62 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  63
  64 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  65
  66 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  67 #define TSC_RATIO_MIN           0x0000000000000001ULL
  68 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  69
  70 static bool erratum_383_found __read_mostly;
  71
  72 static const u32 host_save_user_msrs[] = {
  73 #ifdef CONFIG_X86_64
  74         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  75         MSR_FS_BASE,
  76 #endif
  77         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  78 };
  79
  80 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  81
  82 struct kvm_vcpu;
  83
  84 struct nested_state {
  85         struct vmcb *hsave;
  86         u64 hsave_msr;
  87         u64 vm_cr_msr;
  88         u64 vmcb;
  89
  90         /* These are the merged vectors */
  91         u32 *msrpm;
  92
  93         /* gpa pointers to the real vectors */
  94         u64 vmcb_msrpm;
  95         u64 vmcb_iopm;
  96
  97         /* A VMEXIT is required but not yet emulated */
  98         bool exit_required;
  99
 100         /* cache for intercepts of the guest */
 101         u32 intercept_cr;
 102         u32 intercept_dr;
 103         u32 intercept_exceptions;
 104         u64 intercept;
 105
 106         /* Nested Paging related state */
 107         u64 nested_cr3;
 108 };
 109
 110 #define MSRPM_OFFSETS   16
 111 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 112
 113 struct vcpu_svm {
 114         struct kvm_vcpu vcpu;
 115         struct vmcb *vmcb;
 116         unsigned long vmcb_pa;
 117         struct svm_cpu_data *svm_data;
 118         uint64_t asid_generation;
 119         uint64_t sysenter_esp;
 120         uint64_t sysenter_eip;
 121
 122         u64 next_rip;
 123
 124         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 125         struct {
 126                 u16 fs;
 127                 u16 gs;
 128                 u16 ldt;
 129                 u64 gs_base;
 130         } host;
 131
 132         u32 *msrpm;
 133
 134         ulong nmi_iret_rip;
 135
 136         struct nested_state nested;
 137
 138         bool nmi_singlestep;
 139
 140         unsigned int3_injected;
 141         unsigned long int3_rip;
 142         u32 apf_reason;
 143
 144         u64  tsc_ratio;
 145 };
 146
 147 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 148 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 149
 150 #define MSR_INVALID                     0xffffffffU
 151
 152 static struct svm_direct_access_msrs {
 153         u32 index;   /* Index of the MSR */
 154         bool always; /* True if intercept is always on */
 155 } direct_access_msrs[] = {
 156         { .index = MSR_STAR,                            .always = true  },
 157         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 158 #ifdef CONFIG_X86_64
 159         { .index = MSR_GS_BASE,                         .always = true  },
 160         { .index = MSR_FS_BASE,                         .always = true  },
 161         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 162         { .index = MSR_LSTAR,                           .always = true  },
 163         { .index = MSR_CSTAR,                           .always = true  },
 164         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 165 #endif
 166         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 167         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 168         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 169         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 170         { .index = MSR_INVALID,                         .always = false },
 171 };
 172
 173 /* enable NPT for AMD64 and X86 with PAE */
 174 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 175 static bool npt_enabled = true;
 176 #else
 177 static bool npt_enabled;
 178 #endif
 179 static int npt = 1;
 180
 181 module_param(npt, int, S_IRUGO);
 182
 183 static int nested = 1;
 184 module_param(nested, int, S_IRUGO);
 185
 186 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 187 static void svm_complete_interrupts(struct vcpu_svm *svm);
 188
 189 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 190 static int nested_svm_intercept(struct vcpu_svm *svm);
 191 static int nested_svm_vmexit(struct vcpu_svm *svm);
 192 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 193                                       bool has_error_code, u32 error_code);
 194 static u64 __scale_tsc(u64 ratio, u64 tsc);
 195
 196 enum {
 197         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 198                             pause filter count */
 199         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 200         VMCB_ASID,       /* ASID */
 201         VMCB_INTR,       /* int_ctl, int_vector */
 202         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 203         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 204         VMCB_DR,         /* DR6, DR7 */
 205         VMCB_DT,         /* GDT, IDT */
 206         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 207         VMCB_CR2,        /* CR2 only */
 208         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 209         VMCB_DIRTY_MAX,
 210 };
 211
 212 /* TPR and CR2 are always written before VMRUN */
 213 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 214
 215 static inline void mark_all_dirty(struct vmcb *vmcb)
 216 {
 217         vmcb->control.clean = 0;
 218 }
 219
 220 static inline void mark_all_clean(struct vmcb *vmcb)
 221 {
 222         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 223                                & ~VMCB_ALWAYS_DIRTY_MASK;
 224 }
 225
 226 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 227 {
 228         vmcb->control.clean &= ~(1 << bit);
 229 }
 230
 231 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 232 {
 233         return container_of(vcpu, struct vcpu_svm, vcpu);
 234 }
 235
 236 static void recalc_intercepts(struct vcpu_svm *svm)
 237 {
 238         struct vmcb_control_area *c, *h;
 239         struct nested_state *g;
 240
 241         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 242
 243         if (!is_guest_mode(&svm->vcpu))
 244                 return;
 245
 246         c = &svm->vmcb->control;
 247         h = &svm->nested.hsave->control;
 248         g = &svm->nested;
 249
 250         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 251         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 252         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 253         c->intercept = h->intercept | g->intercept;
 254 }
 255
 256 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 257 {
 258         if (is_guest_mode(&svm->vcpu))
 259                 return svm->nested.hsave;
 260         else
 261                 return svm->vmcb;
 262 }
 263
 264 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 265 {
 266         struct vmcb *vmcb = get_host_vmcb(svm);
 267
 268         vmcb->control.intercept_cr |= (1U << bit);
 269
 270         recalc_intercepts(svm);
 271 }
 272
 273 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 274 {
 275         struct vmcb *vmcb = get_host_vmcb(svm);
 276
 277         vmcb->control.intercept_cr &= ~(1U << bit);
 278
 279         recalc_intercepts(svm);
 280 }
 281
 282 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 283 {
 284         struct vmcb *vmcb = get_host_vmcb(svm);
 285
 286         return vmcb->control.intercept_cr & (1U << bit);
 287 }
 288
 289 static inline void set_dr_intercept(struct vcpu_svm *svm, int bit)
 290 {
 291         struct vmcb *vmcb = get_host_vmcb(svm);
 292
 293         vmcb->control.intercept_dr |= (1U << bit);
 294
 295         recalc_intercepts(svm);
 296 }
 297
 298 static inline void clr_dr_intercept(struct vcpu_svm *svm, int bit)
 299 {
 300         struct vmcb *vmcb = get_host_vmcb(svm);
 301
 302         vmcb->control.intercept_dr &= ~(1U << bit);
 303
 304         recalc_intercepts(svm);
 305 }
 306
 307 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 308 {
 309         struct vmcb *vmcb = get_host_vmcb(svm);
 310
 311         vmcb->control.intercept_exceptions |= (1U << bit);
 312
 313         recalc_intercepts(svm);
 314 }
 315
 316 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 317 {
 318         struct vmcb *vmcb = get_host_vmcb(svm);
 319
 320         vmcb->control.intercept_exceptions &= ~(1U << bit);
 321
 322         recalc_intercepts(svm);
 323 }
 324
 325 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 326 {
 327         struct vmcb *vmcb = get_host_vmcb(svm);
 328
 329         vmcb->control.intercept |= (1ULL << bit);
 330
 331         recalc_intercepts(svm);
 332 }
 333
 334 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 335 {
 336         struct vmcb *vmcb = get_host_vmcb(svm);
 337
 338         vmcb->control.intercept &= ~(1ULL << bit);
 339
 340         recalc_intercepts(svm);
 341 }
 342
 343 static inline void enable_gif(struct vcpu_svm *svm)
 344 {
 345         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 346 }
 347
 348 static inline void disable_gif(struct vcpu_svm *svm)
 349 {
 350         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 351 }
 352
 353 static inline bool gif_set(struct vcpu_svm *svm)
 354 {
 355         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 356 }
 357
 358 static unsigned long iopm_base;
 359
 360 struct kvm_ldttss_desc {
 361         u16 limit0;
 362         u16 base0;
 363         unsigned base1:8, type:5, dpl:2, p:1;
 364         unsigned limit1:4, zero0:3, g:1, base2:8;
 365         u32 base3;
 366         u32 zero1;
 367 } __attribute__((packed));
 368
 369 struct svm_cpu_data {
 370         int cpu;
 371
 372         u64 asid_generation;
 373         u32 max_asid;
 374         u32 next_asid;
 375         struct kvm_ldttss_desc *tss_desc;
 376
 377         struct page *save_area;
 378 };
 379
 380 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 381
 382 struct svm_init_data {
 383         int cpu;
 384         int r;
 385 };
 386
 387 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 388
 389 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 390 #define MSRS_RANGE_SIZE 2048
 391 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 392
 393 static u32 svm_msrpm_offset(u32 msr)
 394 {
 395         u32 offset;
 396         int i;
 397
 398         for (i = 0; i < NUM_MSR_MAPS; i++) {
 399                 if (msr < msrpm_ranges[i] ||
 400                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 401                         continue;
 402
 403                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 404                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 405
 406                 /* Now we have the u8 offset - but need the u32 offset */
 407                 return offset / 4;
 408         }
 409
 410         /* MSR not in any range */
 411         return MSR_INVALID;
 412 }
 413
 414 #define MAX_INST_SIZE 15
 415
 416 static inline void clgi(void)
 417 {
 418         asm volatile (__ex(SVM_CLGI));
 419 }
 420
 421 static inline void stgi(void)
 422 {
 423         asm volatile (__ex(SVM_STGI));
 424 }
 425
 426 static inline void invlpga(unsigned long addr, u32 asid)
 427 {
 428         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 429 }
 430
 431 static int get_npt_level(void)
 432 {
 433 #ifdef CONFIG_X86_64
 434         return PT64_ROOT_LEVEL;
 435 #else
 436         return PT32E_ROOT_LEVEL;
 437 #endif
 438 }
 439
 440 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 441 {
 442         vcpu->arch.efer = efer;
 443         if (!npt_enabled && !(efer & EFER_LMA))
 444                 efer &= ~EFER_LME;
 445
 446         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 447         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 448 }
 449
 450 static int is_external_interrupt(u32 info)
 451 {
 452         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 453         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 454 }
 455
 456 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 457 {
 458         struct vcpu_svm *svm = to_svm(vcpu);
 459         u32 ret = 0;
 460
 461         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 462                 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 463         return ret & mask;
 464 }
 465
 466 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 467 {
 468         struct vcpu_svm *svm = to_svm(vcpu);
 469
 470         if (mask == 0)
 471                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 472         else
 473                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 474
 475 }
 476
 477 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 478 {
 479         struct vcpu_svm *svm = to_svm(vcpu);
 480
 481         if (svm->vmcb->control.next_rip != 0)
 482                 svm->next_rip = svm->vmcb->control.next_rip;
 483
 484         if (!svm->next_rip) {
 485                 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 486                                 EMULATE_DONE)
 487                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 488                 return;
 489         }
 490         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 491                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 492                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 493
 494         kvm_rip_write(vcpu, svm->next_rip);
 495         svm_set_interrupt_shadow(vcpu, 0);
 496 }
 497
 498 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 499                                 bool has_error_code, u32 error_code,
 500                                 bool reinject)
 501 {
 502         struct vcpu_svm *svm = to_svm(vcpu);
 503
 504         /*
 505          * If we are within a nested VM we'd better #VMEXIT and let the guest
 506          * handle the exception
 507          */
 508         if (!reinject &&
 509             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 510                 return;
 511
 512         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 513                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 514
 515                 /*
 516                  * For guest debugging where we have to reinject #BP if some
 517                  * INT3 is guest-owned:
 518                  * Emulate nRIP by moving RIP forward. Will fail if injection
 519                  * raises a fault that is not intercepted. Still better than
 520                  * failing in all cases.
 521                  */
 522                 skip_emulated_instruction(&svm->vcpu);
 523                 rip = kvm_rip_read(&svm->vcpu);
 524                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 525                 svm->int3_injected = rip - old_rip;
 526         }
 527
 528         svm->vmcb->control.event_inj = nr
 529                 | SVM_EVTINJ_VALID
 530                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 531                 | SVM_EVTINJ_TYPE_EXEPT;
 532         svm->vmcb->control.event_inj_err = error_code;
 533 }
 534
 535 static void svm_init_erratum_383(void)
 536 {
 537         u32 low, high;
 538         int err;
 539         u64 val;
 540
 541         if (!cpu_has_amd_erratum(amd_erratum_383))
 542                 return;
 543
 544         /* Use _safe variants to not break nested virtualization */
 545         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 546         if (err)
 547                 return;
 548
 549         val |= (1ULL << 47);
 550
 551         low  = lower_32_bits(val);
 552         high = upper_32_bits(val);
 553
 554         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 555
 556         erratum_383_found = true;
 557 }
 558
 559 static int has_svm(void)
 560 {
 561         const char *msg;
 562
 563         if (!cpu_has_svm(&msg)) {
 564                 printk(KERN_INFO "has_svm: %s\n", msg);
 565                 return 0;
 566         }
 567
 568         return 1;
 569 }
 570
 571 static void svm_hardware_disable(void *garbage)
 572 {
 573         /* Make sure we clean up behind us */
 574         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 575                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 576
 577         cpu_svm_disable();
 578 }
 579
 580 static int svm_hardware_enable(void *garbage)
 581 {
 582
 583         struct svm_cpu_data *sd;
 584         uint64_t efer;
 585         struct desc_ptr gdt_descr;
 586         struct desc_struct *gdt;
 587         int me = raw_smp_processor_id();
 588
 589         rdmsrl(MSR_EFER, efer);
 590         if (efer & EFER_SVME)
 591                 return -EBUSY;
 592
 593         if (!has_svm()) {
 594                 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
 595                        me);
 596                 return -EINVAL;
 597         }
 598         sd = per_cpu(svm_data, me);
 599
 600         if (!sd) {
 601                 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
 602                        me);
 603                 return -EINVAL;
 604         }
 605
 606         sd->asid_generation = 1;
 607         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 608         sd->next_asid = sd->max_asid + 1;
 609
 610         native_store_gdt(&gdt_descr);
 611         gdt = (struct desc_struct *)gdt_descr.address;
 612         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 613
 614         wrmsrl(MSR_EFER, efer | EFER_SVME);
 615
 616         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 617
 618         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 619                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 620                 __get_cpu_var(current_tsc_ratio) = TSC_RATIO_DEFAULT;
 621         }
 622
 623         svm_init_erratum_383();
 624
 625         return 0;
 626 }
 627
 628 static void svm_cpu_uninit(int cpu)
 629 {
 630         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 631
 632         if (!sd)
 633                 return;
 634
 635         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 636         __free_page(sd->save_area);
 637         kfree(sd);
 638 }
 639
 640 static int svm_cpu_init(int cpu)
 641 {
 642         struct svm_cpu_data *sd;
 643         int r;
 644
 645         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 646         if (!sd)
 647                 return -ENOMEM;
 648         sd->cpu = cpu;
 649         sd->save_area = alloc_page(GFP_KERNEL);
 650         r = -ENOMEM;
 651         if (!sd->save_area)
 652                 goto err_1;
 653
 654         per_cpu(svm_data, cpu) = sd;
 655
 656         return 0;
 657
 658 err_1:
 659         kfree(sd);
 660         return r;
 661
 662 }
 663
 664 static bool valid_msr_intercept(u32 index)
 665 {
 666         int i;
 667
 668         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 669                 if (direct_access_msrs[i].index == index)
 670                         return true;
 671
 672         return false;
 673 }
 674
 675 static void set_msr_interception(u32 *msrpm, unsigned msr,
 676                                  int read, int write)
 677 {
 678         u8 bit_read, bit_write;
 679         unsigned long tmp;
 680         u32 offset;
 681
 682         /*
 683          * If this warning triggers extend the direct_access_msrs list at the
 684          * beginning of the file
 685          */
 686         WARN_ON(!valid_msr_intercept(msr));
 687
 688         offset    = svm_msrpm_offset(msr);
 689         bit_read  = 2 * (msr & 0x0f);
 690         bit_write = 2 * (msr & 0x0f) + 1;
 691         tmp       = msrpm[offset];
 692
 693         BUG_ON(offset == MSR_INVALID);
 694
 695         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 696         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 697
 698         msrpm[offset] = tmp;
 699 }
 700
 701 static void svm_vcpu_init_msrpm(u32 *msrpm)
 702 {
 703         int i;
 704
 705         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 706
 707         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 708                 if (!direct_access_msrs[i].always)
 709                         continue;
 710
 711                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 712         }
 713 }
 714
 715 static void add_msr_offset(u32 offset)
 716 {
 717         int i;
 718
 719         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 720
 721                 /* Offset already in list? */
 722                 if (msrpm_offsets[i] == offset)
 723                         return;
 724
 725                 /* Slot used by another offset? */
 726                 if (msrpm_offsets[i] != MSR_INVALID)
 727                         continue;
 728
 729                 /* Add offset to list */
 730                 msrpm_offsets[i] = offset;
 731
 732                 return;
 733         }
 734
 735         /*
 736          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 737          * increase MSRPM_OFFSETS in this case.
 738          */
 739         BUG();
 740 }
 741
 742 static void init_msrpm_offsets(void)
 743 {
 744         int i;
 745
 746         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 747
 748         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 749                 u32 offset;
 750
 751                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 752                 BUG_ON(offset == MSR_INVALID);
 753
 754                 add_msr_offset(offset);
 755         }
 756 }
 757
 758 static void svm_enable_lbrv(struct vcpu_svm *svm)
 759 {
 760         u32 *msrpm = svm->msrpm;
 761
 762         svm->vmcb->control.lbr_ctl = 1;
 763         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 764         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 765         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 766         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 767 }
 768
 769 static void svm_disable_lbrv(struct vcpu_svm *svm)
 770 {
 771         u32 *msrpm = svm->msrpm;
 772
 773         svm->vmcb->control.lbr_ctl = 0;
 774         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 775         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 776         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 777         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 778 }
 779
 780 static __init int svm_hardware_setup(void)
 781 {
 782         int cpu;
 783         struct page *iopm_pages;
 784         void *iopm_va;
 785         int r;
 786
 787         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 788
 789         if (!iopm_pages)
 790                 return -ENOMEM;
 791
 792         iopm_va = page_address(iopm_pages);
 793         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 794         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 795
 796         init_msrpm_offsets();
 797
 798         if (boot_cpu_has(X86_FEATURE_NX))
 799                 kvm_enable_efer_bits(EFER_NX);
 800
 801         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 802                 kvm_enable_efer_bits(EFER_FFXSR);
 803
 804         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 805                 u64 max;
 806
 807                 kvm_has_tsc_control = true;
 808
 809                 /*
 810                  * Make sure the user can only configure tsc_khz values that
 811                  * fit into a signed integer.
 812                  * A min value is not calculated needed because it will always
 813                  * be 1 on all machines and a value of 0 is used to disable
 814                  * tsc-scaling for the vcpu.
 815                  */
 816                 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 817
 818                 kvm_max_guest_tsc_khz = max;
 819         }
 820
 821         if (nested) {
 822                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 823                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 824         }
 825
 826         for_each_possible_cpu(cpu) {
 827                 r = svm_cpu_init(cpu);
 828                 if (r)
 829                         goto err;
 830         }
 831
 832         if (!boot_cpu_has(X86_FEATURE_NPT))
 833                 npt_enabled = false;
 834
 835         if (npt_enabled && !npt) {
 836                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 837                 npt_enabled = false;
 838         }
 839
 840         if (npt_enabled) {
 841                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 842                 kvm_enable_tdp();
 843         } else
 844                 kvm_disable_tdp();
 845
 846         return 0;
 847
 848 err:
 849         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 850         iopm_base = 0;
 851         return r;
 852 }
 853
 854 static __exit void svm_hardware_unsetup(void)
 855 {
 856         int cpu;
 857
 858         for_each_possible_cpu(cpu)
 859                 svm_cpu_uninit(cpu);
 860
 861         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 862         iopm_base = 0;
 863 }
 864
 865 static void init_seg(struct vmcb_seg *seg)
 866 {
 867         seg->selector = 0;
 868         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 869                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 870         seg->limit = 0xffff;
 871         seg->base = 0;
 872 }
 873
 874 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 875 {
 876         seg->selector = 0;
 877         seg->attrib = SVM_SELECTOR_P_MASK | type;
 878         seg->limit = 0xffff;
 879         seg->base = 0;
 880 }
 881
 882 static u64 __scale_tsc(u64 ratio, u64 tsc)
 883 {
 884         u64 mult, frac, _tsc;
 885
 886         mult  = ratio >> 32;
 887         frac  = ratio & ((1ULL << 32) - 1);
 888
 889         _tsc  = tsc;
 890         _tsc *= mult;
 891         _tsc += (tsc >> 32) * frac;
 892         _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
 893
 894         return _tsc;
 895 }
 896
 897 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
 898 {
 899         struct vcpu_svm *svm = to_svm(vcpu);
 900         u64 _tsc = tsc;
 901
 902         if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
 903                 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
 904
 905         return _tsc;
 906 }
 907
 908 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
 909 {
 910         struct vcpu_svm *svm = to_svm(vcpu);
 911         u64 ratio;
 912         u64 khz;
 913
 914         /* TSC scaling supported? */
 915         if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR))
 916                 return;
 917
 918         /* TSC-Scaling disabled or guest TSC same frequency as host TSC? */
 919         if (user_tsc_khz == 0) {
 920                 vcpu->arch.virtual_tsc_khz = 0;
 921                 svm->tsc_ratio = TSC_RATIO_DEFAULT;
 922                 return;
 923         }
 924
 925         khz = user_tsc_khz;
 926
 927         /* TSC scaling required  - calculate ratio */
 928         ratio = khz << 32;
 929         do_div(ratio, tsc_khz);
 930
 931         if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
 932                 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
 933                                 user_tsc_khz);
 934                 return;
 935         }
 936         vcpu->arch.virtual_tsc_khz = user_tsc_khz;
 937         svm->tsc_ratio             = ratio;
 938 }
 939
 940 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 941 {
 942         struct vcpu_svm *svm = to_svm(vcpu);
 943         u64 g_tsc_offset = 0;
 944
 945         if (is_guest_mode(vcpu)) {
 946                 g_tsc_offset = svm->vmcb->control.tsc_offset -
 947                                svm->nested.hsave->control.tsc_offset;
 948                 svm->nested.hsave->control.tsc_offset = offset;
 949         }
 950
 951         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
 952
 953         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 954 }
 955
 956 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
 957 {
 958         struct vcpu_svm *svm = to_svm(vcpu);
 959
 960         svm->vmcb->control.tsc_offset += adjustment;
 961         if (is_guest_mode(vcpu))
 962                 svm->nested.hsave->control.tsc_offset += adjustment;
 963         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 964 }
 965
 966 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
 967 {
 968         u64 tsc;
 969
 970         tsc = svm_scale_tsc(vcpu, native_read_tsc());
 971
 972         return target_tsc - tsc;
 973 }
 974
 975 static void init_vmcb(struct vcpu_svm *svm)
 976 {
 977         struct vmcb_control_area *control = &svm->vmcb->control;
 978         struct vmcb_save_area *save = &svm->vmcb->save;
 979
 980         svm->vcpu.fpu_active = 1;
 981         svm->vcpu.arch.hflags = 0;
 982
 983         set_cr_intercept(svm, INTERCEPT_CR0_READ);
 984         set_cr_intercept(svm, INTERCEPT_CR3_READ);
 985         set_cr_intercept(svm, INTERCEPT_CR4_READ);
 986         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
 987         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
 988         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
 989         set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
 990
 991         set_dr_intercept(svm, INTERCEPT_DR0_READ);
 992         set_dr_intercept(svm, INTERCEPT_DR1_READ);
 993         set_dr_intercept(svm, INTERCEPT_DR2_READ);
 994         set_dr_intercept(svm, INTERCEPT_DR3_READ);
 995         set_dr_intercept(svm, INTERCEPT_DR4_READ);
 996         set_dr_intercept(svm, INTERCEPT_DR5_READ);
 997         set_dr_intercept(svm, INTERCEPT_DR6_READ);
 998         set_dr_intercept(svm, INTERCEPT_DR7_READ);
 999
1000         set_dr_intercept(svm, INTERCEPT_DR0_WRITE);
1001         set_dr_intercept(svm, INTERCEPT_DR1_WRITE);
1002         set_dr_intercept(svm, INTERCEPT_DR2_WRITE);
1003         set_dr_intercept(svm, INTERCEPT_DR3_WRITE);
1004         set_dr_intercept(svm, INTERCEPT_DR4_WRITE);
1005         set_dr_intercept(svm, INTERCEPT_DR5_WRITE);
1006         set_dr_intercept(svm, INTERCEPT_DR6_WRITE);
1007         set_dr_intercept(svm, INTERCEPT_DR7_WRITE);
1008
1009         set_exception_intercept(svm, PF_VECTOR);
1010         set_exception_intercept(svm, UD_VECTOR);
1011         set_exception_intercept(svm, MC_VECTOR);
1012
1013         set_intercept(svm, INTERCEPT_INTR);
1014         set_intercept(svm, INTERCEPT_NMI);
1015         set_intercept(svm, INTERCEPT_SMI);
1016         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1017         set_intercept(svm, INTERCEPT_CPUID);
1018         set_intercept(svm, INTERCEPT_INVD);
1019         set_intercept(svm, INTERCEPT_HLT);
1020         set_intercept(svm, INTERCEPT_INVLPG);
1021         set_intercept(svm, INTERCEPT_INVLPGA);
1022         set_intercept(svm, INTERCEPT_IOIO_PROT);
1023         set_intercept(svm, INTERCEPT_MSR_PROT);
1024         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1025         set_intercept(svm, INTERCEPT_SHUTDOWN);
1026         set_intercept(svm, INTERCEPT_VMRUN);
1027         set_intercept(svm, INTERCEPT_VMMCALL);
1028         set_intercept(svm, INTERCEPT_VMLOAD);
1029         set_intercept(svm, INTERCEPT_VMSAVE);
1030         set_intercept(svm, INTERCEPT_STGI);
1031         set_intercept(svm, INTERCEPT_CLGI);
1032         set_intercept(svm, INTERCEPT_SKINIT);
1033         set_intercept(svm, INTERCEPT_WBINVD);
1034         set_intercept(svm, INTERCEPT_MONITOR);
1035         set_intercept(svm, INTERCEPT_MWAIT);
1036         set_intercept(svm, INTERCEPT_XSETBV);
1037
1038         control->iopm_base_pa = iopm_base;
1039         control->msrpm_base_pa = __pa(svm->msrpm);
1040         control->int_ctl = V_INTR_MASKING_MASK;
1041
1042         init_seg(&save->es);
1043         init_seg(&save->ss);
1044         init_seg(&save->ds);
1045         init_seg(&save->fs);
1046         init_seg(&save->gs);
1047
1048         save->cs.selector = 0xf000;
1049         /* Executable/Readable Code Segment */
1050         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1051                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1052         save->cs.limit = 0xffff;
1053         /*
1054          * cs.base should really be 0xffff0000, but vmx can't handle that, so
1055          * be consistent with it.
1056          *
1057          * Replace when we have real mode working for vmx.
1058          */
1059         save->cs.base = 0xf0000;
1060
1061         save->gdtr.limit = 0xffff;
1062         save->idtr.limit = 0xffff;
1063
1064         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1065         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1066
1067         svm_set_efer(&svm->vcpu, 0);
1068         save->dr6 = 0xffff0ff0;
1069         save->dr7 = 0x400;
1070         kvm_set_rflags(&svm->vcpu, 2);
1071         save->rip = 0x0000fff0;
1072         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1073
1074         /*
1075          * This is the guest-visible cr0 value.
1076          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1077          */
1078         svm->vcpu.arch.cr0 = 0;
1079         (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1080
1081         save->cr4 = X86_CR4_PAE;
1082         /* rdx = ?? */
1083
1084         if (npt_enabled) {
1085                 /* Setup VMCB for Nested Paging */
1086                 control->nested_ctl = 1;
1087                 clr_intercept(svm, INTERCEPT_INVLPG);
1088                 clr_exception_intercept(svm, PF_VECTOR);
1089                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1090                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1091                 save->g_pat = 0x0007040600070406ULL;
1092                 save->cr3 = 0;
1093                 save->cr4 = 0;
1094         }
1095         svm->asid_generation = 0;
1096
1097         svm->nested.vmcb = 0;
1098         svm->vcpu.arch.hflags = 0;
1099
1100         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1101                 control->pause_filter_count = 3000;
1102                 set_intercept(svm, INTERCEPT_PAUSE);
1103         }
1104
1105         mark_all_dirty(svm->vmcb);
1106
1107         enable_gif(svm);
1108 }
1109
1110 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
1111 {
1112         struct vcpu_svm *svm = to_svm(vcpu);
1113
1114         init_vmcb(svm);
1115
1116         if (!kvm_vcpu_is_bsp(vcpu)) {
1117                 kvm_rip_write(vcpu, 0);
1118                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
1119                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
1120         }
1121         vcpu->arch.regs_avail = ~0;
1122         vcpu->arch.regs_dirty = ~0;
1123
1124         return 0;
1125 }
1126
1127 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1128 {
1129         struct vcpu_svm *svm;
1130         struct page *page;
1131         struct page *msrpm_pages;
1132         struct page *hsave_page;
1133         struct page *nested_msrpm_pages;
1134         int err;
1135
1136         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1137         if (!svm) {
1138                 err = -ENOMEM;
1139                 goto out;
1140         }
1141
1142         svm->tsc_ratio = TSC_RATIO_DEFAULT;
1143
1144         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1145         if (err)
1146                 goto free_svm;
1147
1148         err = -ENOMEM;
1149         page = alloc_page(GFP_KERNEL);
1150         if (!page)
1151                 goto uninit;
1152
1153         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1154         if (!msrpm_pages)
1155                 goto free_page1;
1156
1157         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1158         if (!nested_msrpm_pages)
1159                 goto free_page2;
1160
1161         hsave_page = alloc_page(GFP_KERNEL);
1162         if (!hsave_page)
1163                 goto free_page3;
1164
1165         svm->nested.hsave = page_address(hsave_page);
1166
1167         svm->msrpm = page_address(msrpm_pages);
1168         svm_vcpu_init_msrpm(svm->msrpm);
1169
1170         svm->nested.msrpm = page_address(nested_msrpm_pages);
1171         svm_vcpu_init_msrpm(svm->nested.msrpm);
1172
1173         svm->vmcb = page_address(page);
1174         clear_page(svm->vmcb);
1175         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1176         svm->asid_generation = 0;
1177         init_vmcb(svm);
1178         kvm_write_tsc(&svm->vcpu, 0);
1179
1180         err = fx_init(&svm->vcpu);
1181         if (err)
1182                 goto free_page4;
1183
1184         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1185         if (kvm_vcpu_is_bsp(&svm->vcpu))
1186                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1187
1188         return &svm->vcpu;
1189
1190 free_page4:
1191         __free_page(hsave_page);
1192 free_page3:
1193         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1194 free_page2:
1195         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1196 free_page1:
1197         __free_page(page);
1198 uninit:
1199         kvm_vcpu_uninit(&svm->vcpu);
1200 free_svm:
1201         kmem_cache_free(kvm_vcpu_cache, svm);
1202 out:
1203         return ERR_PTR(err);
1204 }
1205
1206 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1207 {
1208         struct vcpu_svm *svm = to_svm(vcpu);
1209
1210         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1211         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1212         __free_page(virt_to_page(svm->nested.hsave));
1213         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1214         kvm_vcpu_uninit(vcpu);
1215         kmem_cache_free(kvm_vcpu_cache, svm);
1216 }
1217
1218 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1219 {
1220         struct vcpu_svm *svm = to_svm(vcpu);
1221         int i;
1222
1223         if (unlikely(cpu != vcpu->cpu)) {
1224                 svm->asid_generation = 0;
1225                 mark_all_dirty(svm->vmcb);
1226         }
1227
1228 #ifdef CONFIG_X86_64
1229         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1230 #endif
1231         savesegment(fs, svm->host.fs);
1232         savesegment(gs, svm->host.gs);
1233         svm->host.ldt = kvm_read_ldt();
1234
1235         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1236                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1237
1238         if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1239             svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1240                 __get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1241                 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1242         }
1243 }
1244
1245 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1246 {
1247         struct vcpu_svm *svm = to_svm(vcpu);
1248         int i;
1249
1250         ++vcpu->stat.host_state_reload;
1251         kvm_load_ldt(svm->host.ldt);
1252 #ifdef CONFIG_X86_64
1253         loadsegment(fs, svm->host.fs);
1254         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1255         load_gs_index(svm->host.gs);
1256 #else
1257 #ifdef CONFIG_X86_32_LAZY_GS
1258         loadsegment(gs, svm->host.gs);
1259 #endif
1260 #endif
1261         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1262                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1263 }
1264
1265 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1266 {
1267         return to_svm(vcpu)->vmcb->save.rflags;
1268 }
1269
1270 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1271 {
1272         to_svm(vcpu)->vmcb->save.rflags = rflags;
1273 }
1274
1275 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1276 {
1277         switch (reg) {
1278         case VCPU_EXREG_PDPTR:
1279                 BUG_ON(!npt_enabled);
1280                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1281                 break;
1282         default:
1283                 BUG();
1284         }
1285 }
1286
1287 static void svm_set_vintr(struct vcpu_svm *svm)
1288 {
1289         set_intercept(svm, INTERCEPT_VINTR);
1290 }
1291
1292 static void svm_clear_vintr(struct vcpu_svm *svm)
1293 {
1294         clr_intercept(svm, INTERCEPT_VINTR);
1295 }
1296
1297 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1298 {
1299         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1300
1301         switch (seg) {
1302         case VCPU_SREG_CS: return &save->cs;
1303         case VCPU_SREG_DS: return &save->ds;
1304         case VCPU_SREG_ES: return &save->es;
1305         case VCPU_SREG_FS: return &save->fs;
1306         case VCPU_SREG_GS: return &save->gs;
1307         case VCPU_SREG_SS: return &save->ss;
1308         case VCPU_SREG_TR: return &save->tr;
1309         case VCPU_SREG_LDTR: return &save->ldtr;
1310         }
1311         BUG();
1312         return NULL;
1313 }
1314
1315 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1316 {
1317         struct vmcb_seg *s = svm_seg(vcpu, seg);
1318
1319         return s->base;
1320 }
1321
1322 static void svm_get_segment(struct kvm_vcpu *vcpu,
1323                             struct kvm_segment *var, int seg)
1324 {
1325         struct vmcb_seg *s = svm_seg(vcpu, seg);
1326
1327         var->base = s->base;
1328         var->limit = s->limit;
1329         var->selector = s->selector;
1330         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1331         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1332         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1333         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1334         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1335         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1336         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1337         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1338
1339         /*
1340          * AMD's VMCB does not have an explicit unusable field, so emulate it
1341          * for cross vendor migration purposes by "not present"
1342          */
1343         var->unusable = !var->present || (var->type == 0);
1344
1345         switch (seg) {
1346         case VCPU_SREG_CS:
1347                 /*
1348                  * SVM always stores 0 for the 'G' bit in the CS selector in
1349                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1350                  * Intel's VMENTRY has a check on the 'G' bit.
1351                  */
1352                 var->g = s->limit > 0xfffff;
1353                 break;
1354         case VCPU_SREG_TR:
1355                 /*
1356                  * Work around a bug where the busy flag in the tr selector
1357                  * isn't exposed
1358                  */
1359                 var->type |= 0x2;
1360                 break;
1361         case VCPU_SREG_DS:
1362         case VCPU_SREG_ES:
1363         case VCPU_SREG_FS:
1364         case VCPU_SREG_GS:
1365                 /*
1366                  * The accessed bit must always be set in the segment
1367                  * descriptor cache, although it can be cleared in the
1368                  * descriptor, the cached bit always remains at 1. Since
1369                  * Intel has a check on this, set it here to support
1370                  * cross-vendor migration.
1371                  */
1372                 if (!var->unusable)
1373                         var->type |= 0x1;
1374                 break;
1375         case VCPU_SREG_SS:
1376                 /*
1377                  * On AMD CPUs sometimes the DB bit in the segment
1378                  * descriptor is left as 1, although the whole segment has
1379                  * been made unusable. Clear it here to pass an Intel VMX
1380                  * entry check when cross vendor migrating.
1381                  */
1382                 if (var->unusable)
1383                         var->db = 0;
1384                 break;
1385         }
1386 }
1387
1388 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1389 {
1390         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1391
1392         return save->cpl;
1393 }
1394
1395 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1396 {
1397         struct vcpu_svm *svm = to_svm(vcpu);
1398
1399         dt->size = svm->vmcb->save.idtr.limit;
1400         dt->address = svm->vmcb->save.idtr.base;
1401 }
1402
1403 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1404 {
1405         struct vcpu_svm *svm = to_svm(vcpu);
1406
1407         svm->vmcb->save.idtr.limit = dt->size;
1408         svm->vmcb->save.idtr.base = dt->address ;
1409         mark_dirty(svm->vmcb, VMCB_DT);
1410 }
1411
1412 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1413 {
1414         struct vcpu_svm *svm = to_svm(vcpu);
1415
1416         dt->size = svm->vmcb->save.gdtr.limit;
1417         dt->address = svm->vmcb->save.gdtr.base;
1418 }
1419
1420 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1421 {
1422         struct vcpu_svm *svm = to_svm(vcpu);
1423
1424         svm->vmcb->save.gdtr.limit = dt->size;
1425         svm->vmcb->save.gdtr.base = dt->address ;
1426         mark_dirty(svm->vmcb, VMCB_DT);
1427 }
1428
1429 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1430 {
1431 }
1432
1433 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1434 {
1435 }
1436
1437 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1438 {
1439 }
1440
1441 static void update_cr0_intercept(struct vcpu_svm *svm)
1442 {
1443         ulong gcr0 = svm->vcpu.arch.cr0;
1444         u64 *hcr0 = &svm->vmcb->save.cr0;
1445
1446         if (!svm->vcpu.fpu_active)
1447                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1448         else
1449                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1450                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1451
1452         mark_dirty(svm->vmcb, VMCB_CR);
1453
1454         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1455                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1456                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1457         } else {
1458                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1459                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1460         }
1461 }
1462
1463 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1464 {
1465         struct vcpu_svm *svm = to_svm(vcpu);
1466
1467 #ifdef CONFIG_X86_64
1468         if (vcpu->arch.efer & EFER_LME) {
1469                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1470                         vcpu->arch.efer |= EFER_LMA;
1471                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1472                 }
1473
1474                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1475                         vcpu->arch.efer &= ~EFER_LMA;
1476                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1477                 }
1478         }
1479 #endif
1480         vcpu->arch.cr0 = cr0;
1481
1482         if (!npt_enabled)
1483                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1484
1485         if (!vcpu->fpu_active)
1486                 cr0 |= X86_CR0_TS;
1487         /*
1488          * re-enable caching here because the QEMU bios
1489          * does not do it - this results in some delay at
1490          * reboot
1491          */
1492         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1493         svm->vmcb->save.cr0 = cr0;
1494         mark_dirty(svm->vmcb, VMCB_CR);
1495         update_cr0_intercept(svm);
1496 }
1497
1498 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1499 {
1500         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1501         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1502
1503         if (cr4 & X86_CR4_VMXE)
1504                 return 1;
1505
1506         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1507                 svm_flush_tlb(vcpu);
1508
1509         vcpu->arch.cr4 = cr4;
1510         if (!npt_enabled)
1511                 cr4 |= X86_CR4_PAE;
1512         cr4 |= host_cr4_mce;
1513         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1514         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1515         return 0;
1516 }
1517
1518 static void svm_set_segment(struct kvm_vcpu *vcpu,
1519                             struct kvm_segment *var, int seg)
1520 {
1521         struct vcpu_svm *svm = to_svm(vcpu);
1522         struct vmcb_seg *s = svm_seg(vcpu, seg);
1523
1524         s->base = var->base;
1525         s->limit = var->limit;
1526         s->selector = var->selector;
1527         if (var->unusable)
1528                 s->attrib = 0;
1529         else {
1530                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1531                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1532                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1533                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1534                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1535                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1536                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1537                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1538         }
1539         if (seg == VCPU_SREG_CS)
1540                 svm->vmcb->save.cpl
1541                         = (svm->vmcb->save.cs.attrib
1542                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
1543
1544         mark_dirty(svm->vmcb, VMCB_SEG);
1545 }
1546
1547 static void update_db_intercept(struct kvm_vcpu *vcpu)
1548 {
1549         struct vcpu_svm *svm = to_svm(vcpu);
1550
1551         clr_exception_intercept(svm, DB_VECTOR);
1552         clr_exception_intercept(svm, BP_VECTOR);
1553
1554         if (svm->nmi_singlestep)
1555                 set_exception_intercept(svm, DB_VECTOR);
1556
1557         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1558                 if (vcpu->guest_debug &
1559                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1560                         set_exception_intercept(svm, DB_VECTOR);
1561                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1562                         set_exception_intercept(svm, BP_VECTOR);
1563         } else
1564                 vcpu->guest_debug = 0;
1565 }
1566
1567 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1568 {
1569         struct vcpu_svm *svm = to_svm(vcpu);
1570
1571         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1572                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1573         else
1574                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1575
1576         mark_dirty(svm->vmcb, VMCB_DR);
1577
1578         update_db_intercept(vcpu);
1579 }
1580
1581 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1582 {
1583         if (sd->next_asid > sd->max_asid) {
1584                 ++sd->asid_generation;
1585                 sd->next_asid = 1;
1586                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1587         }
1588
1589         svm->asid_generation = sd->asid_generation;
1590         svm->vmcb->control.asid = sd->next_asid++;
1591
1592         mark_dirty(svm->vmcb, VMCB_ASID);
1593 }
1594
1595 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1596 {
1597         struct vcpu_svm *svm = to_svm(vcpu);
1598
1599         svm->vmcb->save.dr7 = value;
1600         mark_dirty(svm->vmcb, VMCB_DR);
1601 }
1602
1603 static int pf_interception(struct vcpu_svm *svm)
1604 {
1605         u64 fault_address = svm->vmcb->control.exit_info_2;
1606         u32 error_code;
1607         int r = 1;
1608
1609         switch (svm->apf_reason) {
1610         default:
1611                 error_code = svm->vmcb->control.exit_info_1;
1612
1613                 trace_kvm_page_fault(fault_address, error_code);
1614                 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1615                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1616                 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1617                         svm->vmcb->control.insn_bytes,
1618                         svm->vmcb->control.insn_len);
1619                 break;
1620         case KVM_PV_REASON_PAGE_NOT_PRESENT:
1621                 svm->apf_reason = 0;
1622                 local_irq_disable();
1623                 kvm_async_pf_task_wait(fault_address);
1624                 local_irq_enable();
1625                 break;
1626         case KVM_PV_REASON_PAGE_READY:
1627                 svm->apf_reason = 0;
1628                 local_irq_disable();
1629                 kvm_async_pf_task_wake(fault_address);
1630                 local_irq_enable();
1631                 break;
1632         }
1633         return r;
1634 }
1635
1636 static int db_interception(struct vcpu_svm *svm)
1637 {
1638         struct kvm_run *kvm_run = svm->vcpu.run;
1639
1640         if (!(svm->vcpu.guest_debug &
1641               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1642                 !svm->nmi_singlestep) {
1643                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1644                 return 1;
1645         }
1646
1647         if (svm->nmi_singlestep) {
1648                 svm->nmi_singlestep = false;
1649                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1650                         svm->vmcb->save.rflags &=
1651                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1652                 update_db_intercept(&svm->vcpu);
1653         }
1654
1655         if (svm->vcpu.guest_debug &
1656             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1657                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1658                 kvm_run->debug.arch.pc =
1659                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1660                 kvm_run->debug.arch.exception = DB_VECTOR;
1661                 return 0;
1662         }
1663
1664         return 1;
1665 }
1666
1667 static int bp_interception(struct vcpu_svm *svm)
1668 {
1669         struct kvm_run *kvm_run = svm->vcpu.run;
1670
1671         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1672         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1673         kvm_run->debug.arch.exception = BP_VECTOR;
1674         return 0;
1675 }
1676
1677 static int ud_interception(struct vcpu_svm *svm)
1678 {
1679         int er;
1680
1681         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1682         if (er != EMULATE_DONE)
1683                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1684         return 1;
1685 }
1686
1687 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1688 {
1689         struct vcpu_svm *svm = to_svm(vcpu);
1690
1691         clr_exception_intercept(svm, NM_VECTOR);
1692
1693         svm->vcpu.fpu_active = 1;
1694         update_cr0_intercept(svm);
1695 }
1696
1697 static int nm_interception(struct vcpu_svm *svm)
1698 {
1699         svm_fpu_activate(&svm->vcpu);
1700         return 1;
1701 }
1702
1703 static bool is_erratum_383(void)
1704 {
1705         int err, i;
1706         u64 value;
1707
1708         if (!erratum_383_found)
1709                 return false;
1710
1711         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1712         if (err)
1713                 return false;
1714
1715         /* Bit 62 may or may not be set for this mce */
1716         value &= ~(1ULL << 62);
1717
1718         if (value != 0xb600000000010015ULL)
1719                 return false;
1720
1721         /* Clear MCi_STATUS registers */
1722         for (i = 0; i < 6; ++i)
1723                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1724
1725         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1726         if (!err) {
1727                 u32 low, high;
1728
1729                 value &= ~(1ULL << 2);
1730                 low    = lower_32_bits(value);
1731                 high   = upper_32_bits(value);
1732
1733                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1734         }
1735
1736         /* Flush tlb to evict multi-match entries */
1737         __flush_tlb_all();
1738
1739         return true;
1740 }
1741
1742 static void svm_handle_mce(struct vcpu_svm *svm)
1743 {
1744         if (is_erratum_383()) {
1745                 /*
1746                  * Erratum 383 triggered. Guest state is corrupt so kill the
1747                  * guest.
1748                  */
1749                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1750
1751                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1752
1753                 return;
1754         }
1755
1756         /*
1757          * On an #MC intercept the MCE handler is not called automatically in
1758          * the host. So do it by hand here.
1759          */
1760         asm volatile (
1761                 "int $0x12\n");
1762         /* not sure if we ever come back to this point */
1763
1764         return;
1765 }
1766
1767 static int mc_interception(struct vcpu_svm *svm)
1768 {
1769         return 1;
1770 }
1771
1772 static int shutdown_interception(struct vcpu_svm *svm)
1773 {
1774         struct kvm_run *kvm_run = svm->vcpu.run;
1775
1776         /*
1777          * VMCB is undefined after a SHUTDOWN intercept
1778          * so reinitialize it.
1779          */
1780         clear_page(svm->vmcb);
1781         init_vmcb(svm);
1782
1783         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1784         return 0;
1785 }
1786
1787 static int io_interception(struct vcpu_svm *svm)
1788 {
1789         struct kvm_vcpu *vcpu = &svm->vcpu;
1790         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1791         int size, in, string;
1792         unsigned port;
1793
1794         ++svm->vcpu.stat.io_exits;
1795         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1796         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1797         if (string || in)
1798                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1799
1800         port = io_info >> 16;
1801         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1802         svm->next_rip = svm->vmcb->control.exit_info_2;
1803         skip_emulated_instruction(&svm->vcpu);
1804
1805         return kvm_fast_pio_out(vcpu, size, port);
1806 }
1807
1808 static int nmi_interception(struct vcpu_svm *svm)
1809 {
1810         return 1;
1811 }
1812
1813 static int intr_interception(struct vcpu_svm *svm)
1814 {
1815         ++svm->vcpu.stat.irq_exits;
1816         return 1;
1817 }
1818
1819 static int nop_on_interception(struct vcpu_svm *svm)
1820 {
1821         return 1;
1822 }
1823
1824 static int halt_interception(struct vcpu_svm *svm)
1825 {
1826         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1827         skip_emulated_instruction(&svm->vcpu);
1828         return kvm_emulate_halt(&svm->vcpu);
1829 }
1830
1831 static int vmmcall_interception(struct vcpu_svm *svm)
1832 {
1833         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1834         skip_emulated_instruction(&svm->vcpu);
1835         kvm_emulate_hypercall(&svm->vcpu);
1836         return 1;
1837 }
1838
1839 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1840 {
1841         struct vcpu_svm *svm = to_svm(vcpu);
1842
1843         return svm->nested.nested_cr3;
1844 }
1845
1846 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1847 {
1848         struct vcpu_svm *svm = to_svm(vcpu);
1849         u64 cr3 = svm->nested.nested_cr3;
1850         u64 pdpte;
1851         int ret;
1852
1853         ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1854                                   offset_in_page(cr3) + index * 8, 8);
1855         if (ret)
1856                 return 0;
1857         return pdpte;
1858 }
1859
1860 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1861                                    unsigned long root)
1862 {
1863         struct vcpu_svm *svm = to_svm(vcpu);
1864
1865         svm->vmcb->control.nested_cr3 = root;
1866         mark_dirty(svm->vmcb, VMCB_NPT);
1867         svm_flush_tlb(vcpu);
1868 }
1869
1870 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1871                                        struct x86_exception *fault)
1872 {
1873         struct vcpu_svm *svm = to_svm(vcpu);
1874
1875         svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1876         svm->vmcb->control.exit_code_hi = 0;
1877         svm->vmcb->control.exit_info_1 = fault->error_code;
1878         svm->vmcb->control.exit_info_2 = fault->address;
1879
1880         nested_svm_vmexit(svm);
1881 }
1882
1883 static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1884 {
1885         int r;
1886
1887         r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1888
1889         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1890         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
1891         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
1892         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1893         vcpu->arch.mmu.shadow_root_level = get_npt_level();
1894         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
1895
1896         return r;
1897 }
1898
1899 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1900 {
1901         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1902 }
1903
1904 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1905 {
1906         if (!(svm->vcpu.arch.efer & EFER_SVME)
1907             || !is_paging(&svm->vcpu)) {
1908                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1909                 return 1;
1910         }
1911
1912         if (svm->vmcb->save.cpl) {
1913                 kvm_inject_gp(&svm->vcpu, 0);
1914                 return 1;
1915         }
1916
1917        return 0;
1918 }
1919
1920 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1921                                       bool has_error_code, u32 error_code)
1922 {
1923         int vmexit;
1924
1925         if (!is_guest_mode(&svm->vcpu))
1926                 return 0;
1927
1928         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1929         svm->vmcb->control.exit_code_hi = 0;
1930         svm->vmcb->control.exit_info_1 = error_code;
1931         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1932
1933         vmexit = nested_svm_intercept(svm);
1934         if (vmexit == NESTED_EXIT_DONE)
1935                 svm->nested.exit_required = true;
1936
1937         return vmexit;
1938 }
1939
1940 /* This function returns true if it is save to enable the irq window */
1941 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1942 {
1943         if (!is_guest_mode(&svm->vcpu))
1944                 return true;
1945
1946         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1947                 return true;
1948
1949         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1950                 return false;
1951
1952         /*
1953          * if vmexit was already requested (by intercepted exception
1954          * for instance) do not overwrite it with "external interrupt"
1955          * vmexit.
1956          */
1957         if (svm->nested.exit_required)
1958                 return false;
1959
1960         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
1961         svm->vmcb->control.exit_info_1 = 0;
1962         svm->vmcb->control.exit_info_2 = 0;
1963
1964         if (svm->nested.intercept & 1ULL) {
1965                 /*
1966                  * The #vmexit can't be emulated here directly because this
1967                  * code path runs with irqs and preemtion disabled. A
1968                  * #vmexit emulation might sleep. Only signal request for
1969                  * the #vmexit here.
1970                  */
1971                 svm->nested.exit_required = true;
1972                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1973                 return false;
1974         }
1975
1976         return true;
1977 }
1978
1979 /* This function returns true if it is save to enable the nmi window */
1980 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1981 {
1982         if (!is_guest_mode(&svm->vcpu))
1983                 return true;
1984
1985         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1986                 return true;
1987
1988         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1989         svm->nested.exit_required = true;
1990
1991         return false;
1992 }
1993
1994 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1995 {
1996         struct page *page;
1997
1998         might_sleep();
1999
2000         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2001         if (is_error_page(page))
2002                 goto error;
2003
2004         *_page = page;
2005
2006         return kmap(page);
2007
2008 error:
2009         kvm_release_page_clean(page);
2010         kvm_inject_gp(&svm->vcpu, 0);
2011
2012         return NULL;
2013 }
2014
2015 static void nested_svm_unmap(struct page *page)
2016 {
2017         kunmap(page);
2018         kvm_release_page_dirty(page);
2019 }
2020
2021 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2022 {
2023         unsigned port;
2024         u8 val, bit;
2025         u64 gpa;
2026
2027         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2028                 return NESTED_EXIT_HOST;
2029
2030         port = svm->vmcb->control.exit_info_1 >> 16;
2031         gpa  = svm->nested.vmcb_iopm + (port / 8);
2032         bit  = port % 8;
2033         val  = 0;
2034
2035         if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2036                 val &= (1 << bit);
2037
2038         return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2039 }
2040
2041 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2042 {
2043         u32 offset, msr, value;
2044         int write, mask;
2045
2046         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2047                 return NESTED_EXIT_HOST;
2048
2049         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2050         offset = svm_msrpm_offset(msr);
2051         write  = svm->vmcb->control.exit_info_1 & 1;
2052         mask   = 1 << ((2 * (msr & 0xf)) + write);
2053
2054         if (offset == MSR_INVALID)
2055                 return NESTED_EXIT_DONE;
2056
2057         /* Offset is in 32 bit units but need in 8 bit units */
2058         offset *= 4;
2059
2060         if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2061                 return NESTED_EXIT_DONE;
2062
2063         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2064 }
2065
2066 static int nested_svm_exit_special(struct vcpu_svm *svm)
2067 {
2068         u32 exit_code = svm->vmcb->control.exit_code;
2069
2070         switch (exit_code) {
2071         case SVM_EXIT_INTR:
2072         case SVM_EXIT_NMI:
2073         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2074                 return NESTED_EXIT_HOST;
2075         case SVM_EXIT_NPF:
2076                 /* For now we are always handling NPFs when using them */
2077                 if (npt_enabled)
2078                         return NESTED_EXIT_HOST;
2079                 break;
2080         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2081                 /* When we're shadowing, trap PFs, but not async PF */
2082                 if (!npt_enabled && svm->apf_reason == 0)
2083                         return NESTED_EXIT_HOST;
2084                 break;
2085         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2086                 nm_interception(svm);
2087                 break;
2088         default:
2089                 break;
2090         }
2091
2092         return NESTED_EXIT_CONTINUE;
2093 }
2094
2095 /*
2096  * If this function returns true, this #vmexit was already handled
2097  */
2098 static int nested_svm_intercept(struct vcpu_svm *svm)
2099 {
2100         u32 exit_code = svm->vmcb->control.exit_code;
2101         int vmexit = NESTED_EXIT_HOST;
2102
2103         switch (exit_code) {
2104         case SVM_EXIT_MSR:
2105                 vmexit = nested_svm_exit_handled_msr(svm);
2106                 break;
2107         case SVM_EXIT_IOIO:
2108                 vmexit = nested_svm_intercept_ioio(svm);
2109                 break;
2110         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2111                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2112                 if (svm->nested.intercept_cr & bit)
2113                         vmexit = NESTED_EXIT_DONE;
2114                 break;
2115         }
2116         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2117                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2118                 if (svm->nested.intercept_dr & bit)
2119                         vmexit = NESTED_EXIT_DONE;
2120                 break;
2121         }
2122         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2123                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2124                 if (svm->nested.intercept_exceptions & excp_bits)
2125                         vmexit = NESTED_EXIT_DONE;
2126                 /* async page fault always cause vmexit */
2127                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2128                          svm->apf_reason != 0)
2129                         vmexit = NESTED_EXIT_DONE;
2130                 break;
2131         }
2132         case SVM_EXIT_ERR: {
2133                 vmexit = NESTED_EXIT_DONE;
2134                 break;
2135         }
2136         default: {
2137                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2138                 if (svm->nested.intercept & exit_bits)
2139                         vmexit = NESTED_EXIT_DONE;
2140         }
2141         }
2142
2143         return vmexit;
2144 }
2145
2146 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2147 {
2148         int vmexit;
2149
2150         vmexit = nested_svm_intercept(svm);
2151
2152         if (vmexit == NESTED_EXIT_DONE)
2153                 nested_svm_vmexit(svm);
2154
2155         return vmexit;
2156 }
2157
2158 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2159 {
2160         struct vmcb_control_area *dst  = &dst_vmcb->control;
2161         struct vmcb_control_area *from = &from_vmcb->control;
2162
2163         dst->intercept_cr         = from->intercept_cr;
2164         dst->intercept_dr         = from->intercept_dr;
2165         dst->intercept_exceptions = from->intercept_exceptions;
2166         dst->intercept            = from->intercept;
2167         dst->iopm_base_pa         = from->iopm_base_pa;
2168         dst->msrpm_base_pa        = from->msrpm_base_pa;
2169         dst->tsc_offset           = from->tsc_offset;
2170         dst->asid                 = from->asid;
2171         dst->tlb_ctl              = from->tlb_ctl;
2172         dst->int_ctl              = from->int_ctl;
2173         dst->int_vector           = from->int_vector;
2174         dst->int_state            = from->int_state;
2175         dst->exit_code            = from->exit_code;
2176         dst->exit_code_hi         = from->exit_code_hi;
2177         dst->exit_info_1          = from->exit_info_1;
2178         dst->exit_info_2          = from->exit_info_2;
2179         dst->exit_int_info        = from->exit_int_info;
2180         dst->exit_int_info_err    = from->exit_int_info_err;
2181         dst->nested_ctl           = from->nested_ctl;
2182         dst->event_inj            = from->event_inj;
2183         dst->event_inj_err        = from->event_inj_err;
2184         dst->nested_cr3           = from->nested_cr3;
2185         dst->lbr_ctl              = from->lbr_ctl;
2186 }
2187
2188 static int nested_svm_vmexit(struct vcpu_svm *svm)
2189 {
2190         struct vmcb *nested_vmcb;
2191         struct vmcb *hsave = svm->nested.hsave;
2192         struct vmcb *vmcb = svm->vmcb;
2193         struct page *page;
2194
2195         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2196                                        vmcb->control.exit_info_1,
2197                                        vmcb->control.exit_info_2,
2198                                        vmcb->control.exit_int_info,
2199                                        vmcb->control.exit_int_info_err,
2200                                        KVM_ISA_SVM);
2201
2202         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2203         if (!nested_vmcb)
2204                 return 1;
2205
2206         /* Exit Guest-Mode */
2207         leave_guest_mode(&svm->vcpu);
2208         svm->nested.vmcb = 0;
2209
2210         /* Give the current vmcb to the guest */
2211         disable_gif(svm);
2212
2213         nested_vmcb->save.es     = vmcb->save.es;
2214         nested_vmcb->save.cs     = vmcb->save.cs;
2215         nested_vmcb->save.ss     = vmcb->save.ss;
2216         nested_vmcb->save.ds     = vmcb->save.ds;
2217         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2218         nested_vmcb->save.idtr   = vmcb->save.idtr;
2219         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2220         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2221         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2222         nested_vmcb->save.cr2    = vmcb->save.cr2;
2223         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2224         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2225         nested_vmcb->save.rip    = vmcb->save.rip;
2226         nested_vmcb->save.rsp    = vmcb->save.rsp;
2227         nested_vmcb->save.rax    = vmcb->save.rax;
2228         nested_vmcb->save.dr7    = vmcb->save.dr7;
2229         nested_vmcb->save.dr6    = vmcb->save.dr6;
2230         nested_vmcb->save.cpl    = vmcb->save.cpl;
2231
2232         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2233         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2234         nested_vmcb->control.int_state         = vmcb->control.int_state;
2235         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2236         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2237         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2238         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2239         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2240         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2241         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2242
2243         /*
2244          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2245          * to make sure that we do not lose injected events. So check event_inj
2246          * here and copy it to exit_int_info if it is valid.
2247          * Exit_int_info and event_inj can't be both valid because the case
2248          * below only happens on a VMRUN instruction intercept which has
2249          * no valid exit_int_info set.
2250          */
2251         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2252                 struct vmcb_control_area *nc = &nested_vmcb->control;
2253
2254                 nc->exit_int_info     = vmcb->control.event_inj;
2255                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2256         }
2257
2258         nested_vmcb->control.tlb_ctl           = 0;
2259         nested_vmcb->control.event_inj         = 0;
2260         nested_vmcb->control.event_inj_err     = 0;
2261
2262         /* We always set V_INTR_MASKING and remember the old value in hflags */
2263         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2264                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2265
2266         /* Restore the original control entries */
2267         copy_vmcb_control_area(vmcb, hsave);
2268
2269         kvm_clear_exception_queue(&svm->vcpu);
2270         kvm_clear_interrupt_queue(&svm->vcpu);
2271
2272         svm->nested.nested_cr3 = 0;
2273
2274         /* Restore selected save entries */
2275         svm->vmcb->save.es = hsave->save.es;
2276         svm->vmcb->save.cs = hsave->save.cs;
2277         svm->vmcb->save.ss = hsave->save.ss;
2278         svm->vmcb->save.ds = hsave->save.ds;
2279         svm->vmcb->save.gdtr = hsave->save.gdtr;
2280         svm->vmcb->save.idtr = hsave->save.idtr;
2281         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2282         svm_set_efer(&svm->vcpu, hsave->save.efer);
2283         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2284         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2285         if (npt_enabled) {
2286                 svm->vmcb->save.cr3 = hsave->save.cr3;
2287                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2288         } else {
2289                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2290         }
2291         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2292         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2293         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2294         svm->vmcb->save.dr7 = 0;
2295         svm->vmcb->save.cpl = 0;
2296         svm->vmcb->control.exit_int_info = 0;
2297
2298         mark_all_dirty(svm->vmcb);
2299
2300         nested_svm_unmap(page);
2301
2302         nested_svm_uninit_mmu_context(&svm->vcpu);
2303         kvm_mmu_reset_context(&svm->vcpu);
2304         kvm_mmu_load(&svm->vcpu);
2305
2306         return 0;
2307 }
2308
2309 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2310 {
2311         /*
2312          * This function merges the msr permission bitmaps of kvm and the
2313          * nested vmcb. It is omptimized in that it only merges the parts where
2314          * the kvm msr permission bitmap may contain zero bits
2315          */
2316         int i;
2317
2318         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2319                 return true;
2320
2321         for (i = 0; i < MSRPM_OFFSETS; i++) {
2322                 u32 value, p;
2323                 u64 offset;
2324
2325                 if (msrpm_offsets[i] == 0xffffffff)
2326                         break;
2327
2328                 p      = msrpm_offsets[i];
2329                 offset = svm->nested.vmcb_msrpm + (p * 4);
2330
2331                 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2332                         return false;
2333
2334                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2335         }
2336
2337         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2338
2339         return true;
2340 }
2341
2342 static bool nested_vmcb_checks(struct vmcb *vmcb)
2343 {
2344         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2345                 return false;
2346
2347         if (vmcb->control.asid == 0)
2348                 return false;
2349
2350         if (vmcb->control.nested_ctl && !npt_enabled)
2351                 return false;
2352
2353         return true;
2354 }
2355
2356 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2357 {
2358         struct vmcb *nested_vmcb;
2359         struct vmcb *hsave = svm->nested.hsave;
2360         struct vmcb *vmcb = svm->vmcb;
2361         struct page *page;
2362         u64 vmcb_gpa;
2363
2364         vmcb_gpa = svm->vmcb->save.rax;
2365
2366         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2367         if (!nested_vmcb)
2368                 return false;
2369
2370         if (!nested_vmcb_checks(nested_vmcb)) {
2371                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2372                 nested_vmcb->control.exit_code_hi = 0;
2373                 nested_vmcb->control.exit_info_1  = 0;
2374                 nested_vmcb->control.exit_info_2  = 0;
2375
2376                 nested_svm_unmap(page);
2377
2378                 return false;
2379         }
2380
2381         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2382                                nested_vmcb->save.rip,
2383                                nested_vmcb->control.int_ctl,
2384                                nested_vmcb->control.event_inj,
2385                                nested_vmcb->control.nested_ctl);
2386
2387         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2388                                     nested_vmcb->control.intercept_cr >> 16,
2389                                     nested_vmcb->control.intercept_exceptions,
2390                                     nested_vmcb->control.intercept);
2391
2392         /* Clear internal status */
2393         kvm_clear_exception_queue(&svm->vcpu);
2394         kvm_clear_interrupt_queue(&svm->vcpu);
2395
2396         /*
2397          * Save the old vmcb, so we don't need to pick what we save, but can
2398          * restore everything when a VMEXIT occurs
2399          */
2400         hsave->save.es     = vmcb->save.es;
2401         hsave->save.cs     = vmcb->save.cs;
2402         hsave->save.ss     = vmcb->save.ss;
2403         hsave->save.ds     = vmcb->save.ds;
2404         hsave->save.gdtr   = vmcb->save.gdtr;
2405         hsave->save.idtr   = vmcb->save.idtr;
2406         hsave->save.efer   = svm->vcpu.arch.efer;
2407         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2408         hsave->save.cr4    = svm->vcpu.arch.cr4;
2409         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2410         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2411         hsave->save.rsp    = vmcb->save.rsp;
2412         hsave->save.rax    = vmcb->save.rax;
2413         if (npt_enabled)
2414                 hsave->save.cr3    = vmcb->save.cr3;
2415         else
2416                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2417
2418         copy_vmcb_control_area(hsave, vmcb);
2419
2420         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2421                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2422         else
2423                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2424
2425         if (nested_vmcb->control.nested_ctl) {
2426                 kvm_mmu_unload(&svm->vcpu);
2427                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2428                 nested_svm_init_mmu_context(&svm->vcpu);
2429         }
2430
2431         /* Load the nested guest state */
2432         svm->vmcb->save.es = nested_vmcb->save.es;
2433         svm->vmcb->save.cs = nested_vmcb->save.cs;
2434         svm->vmcb->save.ss = nested_vmcb->save.ss;
2435         svm->vmcb->save.ds = nested_vmcb->save.ds;
2436         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2437         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2438         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2439         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2440         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2441         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2442         if (npt_enabled) {
2443                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2444                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2445         } else
2446                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2447
2448         /* Guest paging mode is active - reset mmu */
2449         kvm_mmu_reset_context(&svm->vcpu);
2450
2451         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2452         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2453         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2454         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2455
2456         /* In case we don't even reach vcpu_run, the fields are not updated */
2457         svm->vmcb->save.rax = nested_vmcb->save.rax;
2458         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2459         svm->vmcb->save.rip = nested_vmcb->save.rip;
2460         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2461         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2462         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2463
2464         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2465         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2466
2467         /* cache intercepts */
2468         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2469         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2470         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2471         svm->nested.intercept            = nested_vmcb->control.intercept;
2472
2473         svm_flush_tlb(&svm->vcpu);
2474         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2475         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2476                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2477         else
2478                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2479
2480         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2481                 /* We only want the cr8 intercept bits of the guest */
2482                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2483                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2484         }
2485
2486         /* We don't want to see VMMCALLs from a nested guest */
2487         clr_intercept(svm, INTERCEPT_VMMCALL);
2488
2489         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2490         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2491         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2492         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2493         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2494         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2495
2496         nested_svm_unmap(page);
2497
2498         /* Enter Guest-Mode */
2499         enter_guest_mode(&svm->vcpu);
2500
2501         /*
2502          * Merge guest and host intercepts - must be called  with vcpu in
2503          * guest-mode to take affect here
2504          */
2505         recalc_intercepts(svm);
2506
2507         svm->nested.vmcb = vmcb_gpa;
2508
2509         enable_gif(svm);
2510
2511         mark_all_dirty(svm->vmcb);
2512
2513         return true;
2514 }
2515
2516 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2517 {
2518         to_vmcb->save.fs = from_vmcb->save.fs;
2519         to_vmcb->save.gs = from_vmcb->save.gs;
2520         to_vmcb->save.tr = from_vmcb->save.tr;
2521         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2522         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2523         to_vmcb->save.star = from_vmcb->save.star;
2524         to_vmcb->save.lstar = from_vmcb->save.lstar;
2525         to_vmcb->save.cstar = from_vmcb->save.cstar;
2526         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2527         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2528         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2529         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2530 }
2531
2532 static int vmload_interception(struct vcpu_svm *svm)
2533 {
2534         struct vmcb *nested_vmcb;
2535         struct page *page;
2536
2537         if (nested_svm_check_permissions(svm))
2538                 return 1;
2539
2540         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2541         if (!nested_vmcb)
2542                 return 1;
2543
2544         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2545         skip_emulated_instruction(&svm->vcpu);
2546
2547         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2548         nested_svm_unmap(page);
2549
2550         return 1;
2551 }
2552
2553 static int vmsave_interception(struct vcpu_svm *svm)
2554 {
2555         struct vmcb *nested_vmcb;
2556         struct page *page;
2557
2558         if (nested_svm_check_permissions(svm))
2559                 return 1;
2560
2561         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2562         if (!nested_vmcb)
2563                 return 1;
2564
2565         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2566         skip_emulated_instruction(&svm->vcpu);
2567
2568         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2569         nested_svm_unmap(page);
2570
2571         return 1;
2572 }
2573
2574 static int vmrun_interception(struct vcpu_svm *svm)
2575 {
2576         if (nested_svm_check_permissions(svm))
2577                 return 1;
2578
2579         /* Save rip after vmrun instruction */
2580         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2581
2582         if (!nested_svm_vmrun(svm))
2583                 return 1;
2584
2585         if (!nested_svm_vmrun_msrpm(svm))
2586                 goto failed;
2587
2588         return 1;
2589
2590 failed:
2591
2592         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2593         svm->vmcb->control.exit_code_hi = 0;
2594         svm->vmcb->control.exit_info_1  = 0;
2595         svm->vmcb->control.exit_info_2  = 0;
2596
2597         nested_svm_vmexit(svm);
2598
2599         return 1;
2600 }
2601
2602 static int stgi_interception(struct vcpu_svm *svm)
2603 {
2604         if (nested_svm_check_permissions(svm))
2605                 return 1;
2606
2607         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2608         skip_emulated_instruction(&svm->vcpu);
2609         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2610
2611         enable_gif(svm);
2612
2613         return 1;
2614 }
2615
2616 static int clgi_interception(struct vcpu_svm *svm)
2617 {
2618         if (nested_svm_check_permissions(svm))
2619                 return 1;
2620
2621         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2622         skip_emulated_instruction(&svm->vcpu);
2623
2624         disable_gif(svm);
2625
2626         /* After a CLGI no interrupts should come */
2627         svm_clear_vintr(svm);
2628         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2629
2630         mark_dirty(svm->vmcb, VMCB_INTR);
2631
2632         return 1;
2633 }
2634
2635 static int invlpga_interception(struct vcpu_svm *svm)
2636 {
2637         struct kvm_vcpu *vcpu = &svm->vcpu;
2638
2639         trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2640                           vcpu->arch.regs[VCPU_REGS_RAX]);
2641
2642         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2643         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2644
2645         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2646         skip_emulated_instruction(&svm->vcpu);
2647         return 1;
2648 }
2649
2650 static int skinit_interception(struct vcpu_svm *svm)
2651 {
2652         trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2653
2654         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2655         return 1;
2656 }
2657
2658 static int xsetbv_interception(struct vcpu_svm *svm)
2659 {
2660         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2661         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2662
2663         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2664                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2665                 skip_emulated_instruction(&svm->vcpu);
2666         }
2667
2668         return 1;
2669 }
2670
2671 static int invalid_op_interception(struct vcpu_svm *svm)
2672 {
2673         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2674         return 1;
2675 }
2676
2677 static int task_switch_interception(struct vcpu_svm *svm)
2678 {
2679         u16 tss_selector;
2680         int reason;
2681         int int_type = svm->vmcb->control.exit_int_info &
2682                 SVM_EXITINTINFO_TYPE_MASK;
2683         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2684         uint32_t type =
2685                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2686         uint32_t idt_v =
2687                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2688         bool has_error_code = false;
2689         u32 error_code = 0;
2690
2691         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2692
2693         if (svm->vmcb->control.exit_info_2 &
2694             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2695                 reason = TASK_SWITCH_IRET;
2696         else if (svm->vmcb->control.exit_info_2 &
2697                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2698                 reason = TASK_SWITCH_JMP;
2699         else if (idt_v)
2700                 reason = TASK_SWITCH_GATE;
2701         else
2702                 reason = TASK_SWITCH_CALL;
2703
2704         if (reason == TASK_SWITCH_GATE) {
2705                 switch (type) {
2706                 case SVM_EXITINTINFO_TYPE_NMI:
2707                         svm->vcpu.arch.nmi_injected = false;
2708                         break;
2709                 case SVM_EXITINTINFO_TYPE_EXEPT:
2710                         if (svm->vmcb->control.exit_info_2 &
2711                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2712                                 has_error_code = true;
2713                                 error_code =
2714                                         (u32)svm->vmcb->control.exit_info_2;
2715                         }
2716                         kvm_clear_exception_queue(&svm->vcpu);
2717                         break;
2718                 case SVM_EXITINTINFO_TYPE_INTR:
2719                         kvm_clear_interrupt_queue(&svm->vcpu);
2720                         break;
2721                 default:
2722                         break;
2723                 }
2724         }
2725
2726         if (reason != TASK_SWITCH_GATE ||
2727             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2728             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2729              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2730                 skip_emulated_instruction(&svm->vcpu);
2731
2732         if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
2733                                 has_error_code, error_code) == EMULATE_FAIL) {
2734                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2735                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2736                 svm->vcpu.run->internal.ndata = 0;
2737                 return 0;
2738         }
2739         return 1;
2740 }
2741
2742 static int cpuid_interception(struct vcpu_svm *svm)
2743 {
2744         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2745         kvm_emulate_cpuid(&svm->vcpu);
2746         return 1;
2747 }
2748
2749 static int iret_interception(struct vcpu_svm *svm)
2750 {
2751         ++svm->vcpu.stat.nmi_window_exits;
2752         clr_intercept(svm, INTERCEPT_IRET);
2753         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2754         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2755         return 1;
2756 }
2757
2758 static int invlpg_interception(struct vcpu_svm *svm)
2759 {
2760         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2761                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2762
2763         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2764         skip_emulated_instruction(&svm->vcpu);
2765         return 1;
2766 }
2767
2768 static int emulate_on_interception(struct vcpu_svm *svm)
2769 {
2770         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2771 }
2772
2773 bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2774 {
2775         unsigned long cr0 = svm->vcpu.arch.cr0;
2776         bool ret = false;
2777         u64 intercept;
2778
2779         intercept = svm->nested.intercept;
2780
2781         if (!is_guest_mode(&svm->vcpu) ||
2782             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2783                 return false;
2784
2785         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2786         val &= ~SVM_CR0_SELECTIVE_MASK;
2787
2788         if (cr0 ^ val) {
2789                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2790                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2791         }
2792
2793         return ret;
2794 }
2795
2796 #define CR_VALID (1ULL << 63)
2797
2798 static int cr_interception(struct vcpu_svm *svm)
2799 {
2800         int reg, cr;
2801         unsigned long val;
2802         int err;
2803
2804         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2805                 return emulate_on_interception(svm);
2806
2807         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2808                 return emulate_on_interception(svm);
2809
2810         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2811         cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2812
2813         err = 0;
2814         if (cr >= 16) { /* mov to cr */
2815                 cr -= 16;
2816                 val = kvm_register_read(&svm->vcpu, reg);
2817                 switch (cr) {
2818                 case 0:
2819                         if (!check_selective_cr0_intercepted(svm, val))
2820                                 err = kvm_set_cr0(&svm->vcpu, val);
2821                         else
2822                                 return 1;
2823
2824                         break;
2825                 case 3:
2826                         err = kvm_set_cr3(&svm->vcpu, val);
2827                         break;
2828                 case 4:
2829                         err = kvm_set_cr4(&svm->vcpu, val);
2830                         break;
2831                 case 8:
2832                         err = kvm_set_cr8(&svm->vcpu, val);
2833                         break;
2834                 default:
2835                         WARN(1, "unhandled write to CR%d", cr);
2836                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2837                         return 1;
2838                 }
2839         } else { /* mov from cr */
2840                 switch (cr) {
2841                 case 0:
2842                         val = kvm_read_cr0(&svm->vcpu);
2843                         break;
2844                 case 2:
2845                         val = svm->vcpu.arch.cr2;
2846                         break;
2847                 case 3:
2848                         val = kvm_read_cr3(&svm->vcpu);
2849                         break;
2850                 case 4:
2851                         val = kvm_read_cr4(&svm->vcpu);
2852                         break;
2853                 case 8:
2854                         val = kvm_get_cr8(&svm->vcpu);
2855                         break;
2856                 default:
2857                         WARN(1, "unhandled read from CR%d", cr);
2858                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2859                         return 1;
2860                 }
2861                 kvm_register_write(&svm->vcpu, reg, val);
2862         }
2863         kvm_complete_insn_gp(&svm->vcpu, err);
2864
2865         return 1;
2866 }
2867
2868 static int dr_interception(struct vcpu_svm *svm)
2869 {
2870         int reg, dr;
2871         unsigned long val;
2872         int err;
2873
2874         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2875                 return emulate_on_interception(svm);
2876
2877         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2878         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2879
2880         if (dr >= 16) { /* mov to DRn */
2881                 val = kvm_register_read(&svm->vcpu, reg);
2882                 kvm_set_dr(&svm->vcpu, dr - 16, val);
2883         } else {
2884                 err = kvm_get_dr(&svm->vcpu, dr, &val);
2885                 if (!err)
2886                         kvm_register_write(&svm->vcpu, reg, val);
2887         }
2888
2889         skip_emulated_instruction(&svm->vcpu);
2890
2891         return 1;
2892 }
2893
2894 static int cr8_write_interception(struct vcpu_svm *svm)
2895 {
2896         struct kvm_run *kvm_run = svm->vcpu.run;
2897         int r;
2898
2899         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2900         /* instruction emulation calls kvm_set_cr8() */
2901         r = cr_interception(svm);
2902         if (irqchip_in_kernel(svm->vcpu.kvm)) {
2903                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2904                 return r;
2905         }
2906         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2907                 return r;
2908         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2909         return 0;
2910 }
2911
2912 u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu)
2913 {
2914         struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
2915         return vmcb->control.tsc_offset +
2916                 svm_scale_tsc(vcpu, native_read_tsc());
2917 }
2918
2919 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2920 {
2921         struct vcpu_svm *svm = to_svm(vcpu);
2922
2923         switch (ecx) {
2924         case MSR_IA32_TSC: {
2925                 *data = svm->vmcb->control.tsc_offset +
2926                         svm_scale_tsc(vcpu, native_read_tsc());
2927
2928                 break;
2929         }
2930         case MSR_STAR:
2931                 *data = svm->vmcb->save.star;
2932                 break;
2933 #ifdef CONFIG_X86_64
2934         case MSR_LSTAR:
2935                 *data = svm->vmcb->save.lstar;
2936                 break;
2937         case MSR_CSTAR:
2938                 *data = svm->vmcb->save.cstar;
2939                 break;
2940         case MSR_KERNEL_GS_BASE:
2941                 *data = svm->vmcb->save.kernel_gs_base;
2942                 break;
2943         case MSR_SYSCALL_MASK:
2944                 *data = svm->vmcb->save.sfmask;
2945                 break;
2946 #endif
2947         case MSR_IA32_SYSENTER_CS:
2948                 *data = svm->vmcb->save.sysenter_cs;
2949                 break;
2950         case MSR_IA32_SYSENTER_EIP:
2951                 *data = svm->sysenter_eip;
2952                 break;
2953         case MSR_IA32_SYSENTER_ESP:
2954                 *data = svm->sysenter_esp;
2955                 break;
2956         /*
2957          * Nobody will change the following 5 values in the VMCB so we can
2958          * safely return them on rdmsr. They will always be 0 until LBRV is
2959          * implemented.
2960          */
2961         case MSR_IA32_DEBUGCTLMSR:
2962                 *data = svm->vmcb->save.dbgctl;
2963                 break;
2964         case MSR_IA32_LASTBRANCHFROMIP:
2965                 *data = svm->vmcb->save.br_from;
2966                 break;
2967         case MSR_IA32_LASTBRANCHTOIP:
2968                 *data = svm->vmcb->save.br_to;
2969                 break;
2970         case MSR_IA32_LASTINTFROMIP:
2971                 *data = svm->vmcb->save.last_excp_from;
2972                 break;
2973         case MSR_IA32_LASTINTTOIP:
2974                 *data = svm->vmcb->save.last_excp_to;
2975                 break;
2976         case MSR_VM_HSAVE_PA:
2977                 *data = svm->nested.hsave_msr;
2978                 break;
2979         case MSR_VM_CR:
2980                 *data = svm->nested.vm_cr_msr;
2981                 break;
2982         case MSR_IA32_UCODE_REV:
2983                 *data = 0x01000065;
2984                 break;
2985         default:
2986                 return kvm_get_msr_common(vcpu, ecx, data);
2987         }
2988         return 0;
2989 }
2990
2991 static int rdmsr_interception(struct vcpu_svm *svm)
2992 {
2993         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2994         u64 data;
2995
2996         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2997                 trace_kvm_msr_read_ex(ecx);
2998                 kvm_inject_gp(&svm->vcpu, 0);
2999         } else {
3000                 trace_kvm_msr_read(ecx, data);
3001
3002                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
3003                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
3004                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3005                 skip_emulated_instruction(&svm->vcpu);
3006         }
3007         return 1;
3008 }
3009
3010 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3011 {
3012         struct vcpu_svm *svm = to_svm(vcpu);
3013         int svm_dis, chg_mask;
3014
3015         if (data & ~SVM_VM_CR_VALID_MASK)
3016                 return 1;
3017
3018         chg_mask = SVM_VM_CR_VALID_MASK;
3019
3020         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3021                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3022
3023         svm->nested.vm_cr_msr &= ~chg_mask;
3024         svm->nested.vm_cr_msr |= (data & chg_mask);
3025
3026         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3027
3028         /* check for svm_disable while efer.svme is set */
3029         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3030                 return 1;
3031
3032         return 0;
3033 }
3034
3035 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
3036 {
3037         struct vcpu_svm *svm = to_svm(vcpu);
3038
3039         switch (ecx) {
3040         case MSR_IA32_TSC:
3041                 kvm_write_tsc(vcpu, data);
3042                 break;
3043         case MSR_STAR:
3044                 svm->vmcb->save.star = data;
3045                 break;
3046 #ifdef CONFIG_X86_64
3047         case MSR_LSTAR:
3048                 svm->vmcb->save.lstar = data;
3049                 break;
3050         case MSR_CSTAR:
3051                 svm->vmcb->save.cstar = data;
3052                 break;
3053         case MSR_KERNEL_GS_BASE:
3054                 svm->vmcb->save.kernel_gs_base = data;
3055                 break;
3056         case MSR_SYSCALL_MASK:
3057                 svm->vmcb->save.sfmask = data;
3058                 break;
3059 #endif
3060         case MSR_IA32_SYSENTER_CS:
3061                 svm->vmcb->save.sysenter_cs = data;
3062                 break;
3063         case MSR_IA32_SYSENTER_EIP:
3064                 svm->sysenter_eip = data;
3065                 svm->vmcb->save.sysenter_eip = data;
3066                 break;
3067         case MSR_IA32_SYSENTER_ESP:
3068                 svm->sysenter_esp = data;
3069                 svm->vmcb->save.sysenter_esp = data;
3070                 break;
3071         case MSR_IA32_DEBUGCTLMSR:
3072                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3073                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3074                                         __func__, data);
3075                         break;
3076                 }
3077                 if (data & DEBUGCTL_RESERVED_BITS)
3078                         return 1;
3079
3080                 svm->vmcb->save.dbgctl = data;
3081                 mark_dirty(svm->vmcb, VMCB_LBR);
3082                 if (data & (1ULL<<0))
3083                         svm_enable_lbrv(svm);
3084                 else
3085                         svm_disable_lbrv(svm);
3086                 break;
3087         case MSR_VM_HSAVE_PA:
3088                 svm->nested.hsave_msr = data;
3089                 break;
3090         case MSR_VM_CR:
3091                 return svm_set_vm_cr(vcpu, data);
3092         case MSR_VM_IGNNE:
3093                 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3094                 break;
3095         default:
3096                 return kvm_set_msr_common(vcpu, ecx, data);
3097         }
3098         return 0;
3099 }
3100
3101 static int wrmsr_interception(struct vcpu_svm *svm)
3102 {
3103         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3104         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3105                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3106
3107
3108         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3109         if (svm_set_msr(&svm->vcpu, ecx, data)) {
3110                 trace_kvm_msr_write_ex(ecx, data);
3111                 kvm_inject_gp(&svm->vcpu, 0);
3112         } else {
3113                 trace_kvm_msr_write(ecx, data);
3114                 skip_emulated_instruction(&svm->vcpu);
3115         }
3116         return 1;
3117 }
3118
3119 static int msr_interception(struct vcpu_svm *svm)
3120 {
3121         if (svm->vmcb->control.exit_info_1)
3122                 return wrmsr_interception(svm);
3123         else
3124                 return rdmsr_interception(svm);
3125 }
3126
3127 static int interrupt_window_interception(struct vcpu_svm *svm)
3128 {
3129         struct kvm_run *kvm_run = svm->vcpu.run;
3130
3131         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3132         svm_clear_vintr(svm);
3133         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3134         mark_dirty(svm->vmcb, VMCB_INTR);
3135         /*
3136          * If the user space waits to inject interrupts, exit as soon as
3137          * possible
3138          */
3139         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3140             kvm_run->request_interrupt_window &&
3141             !kvm_cpu_has_interrupt(&svm->vcpu)) {
3142                 ++svm->vcpu.stat.irq_window_exits;
3143                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3144                 return 0;
3145         }
3146
3147         return 1;
3148 }
3149
3150 static int pause_interception(struct vcpu_svm *svm)
3151 {
3152         kvm_vcpu_on_spin(&(svm->vcpu));
3153         return 1;
3154 }
3155
3156 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
3157         [SVM_EXIT_READ_CR0]                     = cr_interception,
3158         [SVM_EXIT_READ_CR3]                     = cr_interception,
3159         [SVM_EXIT_READ_CR4]                     = cr_interception,
3160         [SVM_EXIT_READ_CR8]                     = cr_interception,
3161         [SVM_EXIT_CR0_SEL_WRITE]                = emulate_on_interception,
3162         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
3163         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
3164         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
3165         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
3166         [SVM_EXIT_READ_DR0]                     = dr_interception,
3167         [SVM_EXIT_READ_DR1]                     = dr_interception,
3168         [SVM_EXIT_READ_DR2]                     = dr_interception,
3169         [SVM_EXIT_READ_DR3]                     = dr_interception,
3170         [SVM_EXIT_READ_DR4]                     = dr_interception,
3171         [SVM_EXIT_READ_DR5]                     = dr_interception,
3172         [SVM_EXIT_READ_DR6]                     = dr_interception,
3173         [SVM_EXIT_READ_DR7]                     = dr_interception,
3174         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
3175         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
3176         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
3177         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
3178         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
3179         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
3180         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
3181         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
3182         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
3183         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
3184         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
3185         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
3186         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
3187         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
3188         [SVM_EXIT_INTR]                         = intr_interception,
3189         [SVM_EXIT_NMI]                          = nmi_interception,
3190         [SVM_EXIT_SMI]                          = nop_on_interception,
3191         [SVM_EXIT_INIT]                         = nop_on_interception,
3192         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
3193         [SVM_EXIT_CPUID]                        = cpuid_interception,
3194         [SVM_EXIT_IRET]                         = iret_interception,
3195         [SVM_EXIT_INVD]                         = emulate_on_interception,
3196         [SVM_EXIT_PAUSE]                        = pause_interception,
3197         [SVM_EXIT_HLT]                          = halt_interception,
3198         [SVM_EXIT_INVLPG]                       = invlpg_interception,
3199         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
3200         [SVM_EXIT_IOIO]                         = io_interception,
3201         [SVM_EXIT_MSR]                          = msr_interception,
3202         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
3203         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
3204         [SVM_EXIT_VMRUN]                        = vmrun_interception,
3205         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
3206         [SVM_EXIT_VMLOAD]                       = vmload_interception,
3207         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
3208         [SVM_EXIT_STGI]                         = stgi_interception,
3209         [SVM_EXIT_CLGI]                         = clgi_interception,
3210         [SVM_EXIT_SKINIT]                       = skinit_interception,
3211         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
3212         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
3213         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
3214         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
3215         [SVM_EXIT_NPF]                          = pf_interception,
3216 };
3217
3218 static void dump_vmcb(struct kvm_vcpu *vcpu)
3219 {
3220         struct vcpu_svm *svm = to_svm(vcpu);
3221         struct vmcb_control_area *control = &svm->vmcb->control;
3222         struct vmcb_save_area *save = &svm->vmcb->save;
3223
3224         pr_err("VMCB Control Area:\n");
3225         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3226         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3227         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3228         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3229         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3230         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3231         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3232         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3233         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3234         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3235         pr_err("%-20s%d\n", "asid:", control->asid);
3236         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3237         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3238         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3239         pr_err("%-20s%08x\n", "int_state:", control->int_state);
3240         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3241         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3242         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3243         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3244         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3245         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3246         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3247         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3248         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3249         pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3250         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3251         pr_err("VMCB State Save Area:\n");
3252         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3253                "es:",
3254                save->es.selector, save->es.attrib,
3255                save->es.limit, save->es.base);
3256         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3257                "cs:",
3258                save->cs.selector, save->cs.attrib,
3259                save->cs.limit, save->cs.base);
3260         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3261                "ss:",
3262                save->ss.selector, save->ss.attrib,
3263                save->ss.limit, save->ss.base);
3264         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3265                "ds:",
3266                save->ds.selector, save->ds.attrib,
3267                save->ds.limit, save->ds.base);
3268         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3269                "fs:",
3270                save->fs.selector, save->fs.attrib,
3271                save->fs.limit, save->fs.base);
3272         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3273                "gs:",
3274                save->gs.selector, save->gs.attrib,
3275                save->gs.limit, save->gs.base);
3276         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3277                "gdtr:",
3278                save->gdtr.selector, save->gdtr.attrib,
3279                save->gdtr.limit, save->gdtr.base);
3280         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3281                "ldtr:",
3282                save->ldtr.selector, save->ldtr.attrib,
3283                save->ldtr.limit, save->ldtr.base);
3284         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3285                "idtr:",
3286                save->idtr.selector, save->idtr.attrib,
3287                save->idtr.limit, save->idtr.base);
3288         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3289                "tr:",
3290                save->tr.selector, save->tr.attrib,
3291                save->tr.limit, save->tr.base);
3292         pr_err("cpl:            %d                efer:         %016llx\n",
3293                 save->cpl, save->efer);
3294         pr_err("%-15s %016llx %-13s %016llx\n",
3295                "cr0:", save->cr0, "cr2:", save->cr2);
3296         pr_err("%-15s %016llx %-13s %016llx\n",
3297                "cr3:", save->cr3, "cr4:", save->cr4);
3298         pr_err("%-15s %016llx %-13s %016llx\n",
3299                "dr6:", save->dr6, "dr7:", save->dr7);
3300         pr_err("%-15s %016llx %-13s %016llx\n",
3301                "rip:", save->rip, "rflags:", save->rflags);
3302         pr_err("%-15s %016llx %-13s %016llx\n",
3303                "rsp:", save->rsp, "rax:", save->rax);
3304         pr_err("%-15s %016llx %-13s %016llx\n",
3305                "star:", save->star, "lstar:", save->lstar);
3306         pr_err("%-15s %016llx %-13s %016llx\n",
3307                "cstar:", save->cstar, "sfmask:", save->sfmask);
3308         pr_err("%-15s %016llx %-13s %016llx\n",
3309                "kernel_gs_base:", save->kernel_gs_base,
3310                "sysenter_cs:", save->sysenter_cs);
3311         pr_err("%-15s %016llx %-13s %016llx\n",
3312                "sysenter_esp:", save->sysenter_esp,
3313                "sysenter_eip:", save->sysenter_eip);
3314         pr_err("%-15s %016llx %-13s %016llx\n",
3315                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3316         pr_err("%-15s %016llx %-13s %016llx\n",
3317                "br_from:", save->br_from, "br_to:", save->br_to);
3318         pr_err("%-15s %016llx %-13s %016llx\n",
3319                "excp_from:", save->last_excp_from,
3320                "excp_to:", save->last_excp_to);
3321 }
3322
3323 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3324 {
3325         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3326
3327         *info1 = control->exit_info_1;
3328         *info2 = control->exit_info_2;
3329 }
3330
3331 static int handle_exit(struct kvm_vcpu *vcpu)
3332 {
3333         struct vcpu_svm *svm = to_svm(vcpu);
3334         struct kvm_run *kvm_run = vcpu->run;
3335         u32 exit_code = svm->vmcb->control.exit_code;
3336
3337         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3338                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3339         if (npt_enabled)
3340                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3341
3342         if (unlikely(svm->nested.exit_required)) {
3343                 nested_svm_vmexit(svm);
3344                 svm->nested.exit_required = false;
3345
3346                 return 1;
3347         }
3348
3349         if (is_guest_mode(vcpu)) {
3350                 int vmexit;
3351
3352                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,