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_TASK_SWITCH);
1088                 clr_intercept(svm, INTERCEPT_INVLPG);
1089                 clr_exception_intercept(svm, PF_VECTOR);
1090                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1091                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1092                 save->g_pat = 0x0007040600070406ULL;
1093                 save->cr3 = 0;
1094                 save->cr4 = 0;
1095         }
1096         svm->asid_generation = 0;
1097
1098         svm->nested.vmcb = 0;
1099         svm->vcpu.arch.hflags = 0;
1100
1101         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1102                 control->pause_filter_count = 3000;
1103                 set_intercept(svm, INTERCEPT_PAUSE);
1104         }
1105
1106         mark_all_dirty(svm->vmcb);
1107
1108         enable_gif(svm);
1109 }
1110
1111 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
1112 {
1113         struct vcpu_svm *svm = to_svm(vcpu);
1114
1115         init_vmcb(svm);
1116
1117         if (!kvm_vcpu_is_bsp(vcpu)) {
1118                 kvm_rip_write(vcpu, 0);
1119                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
1120                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
1121         }
1122         vcpu->arch.regs_avail = ~0;
1123         vcpu->arch.regs_dirty = ~0;
1124
1125         return 0;
1126 }
1127
1128 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1129 {
1130         struct vcpu_svm *svm;
1131         struct page *page;
1132         struct page *msrpm_pages;
1133         struct page *hsave_page;
1134         struct page *nested_msrpm_pages;
1135         int err;
1136
1137         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1138         if (!svm) {
1139                 err = -ENOMEM;
1140                 goto out;
1141         }
1142
1143         svm->tsc_ratio = TSC_RATIO_DEFAULT;
1144
1145         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1146         if (err)
1147                 goto free_svm;
1148
1149         err = -ENOMEM;
1150         page = alloc_page(GFP_KERNEL);
1151         if (!page)
1152                 goto uninit;
1153
1154         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1155         if (!msrpm_pages)
1156                 goto free_page1;
1157
1158         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1159         if (!nested_msrpm_pages)
1160                 goto free_page2;
1161
1162         hsave_page = alloc_page(GFP_KERNEL);
1163         if (!hsave_page)
1164                 goto free_page3;
1165
1166         svm->nested.hsave = page_address(hsave_page);
1167
1168         svm->msrpm = page_address(msrpm_pages);
1169         svm_vcpu_init_msrpm(svm->msrpm);
1170
1171         svm->nested.msrpm = page_address(nested_msrpm_pages);
1172         svm_vcpu_init_msrpm(svm->nested.msrpm);
1173
1174         svm->vmcb = page_address(page);
1175         clear_page(svm->vmcb);
1176         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1177         svm->asid_generation = 0;
1178         init_vmcb(svm);
1179         kvm_write_tsc(&svm->vcpu, 0);
1180
1181         err = fx_init(&svm->vcpu);
1182         if (err)
1183                 goto free_page4;
1184
1185         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1186         if (kvm_vcpu_is_bsp(&svm->vcpu))
1187                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1188
1189         return &svm->vcpu;
1190
1191 free_page4:
1192         __free_page(hsave_page);
1193 free_page3:
1194         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1195 free_page2:
1196         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1197 free_page1:
1198         __free_page(page);
1199 uninit:
1200         kvm_vcpu_uninit(&svm->vcpu);
1201 free_svm:
1202         kmem_cache_free(kvm_vcpu_cache, svm);
1203 out:
1204         return ERR_PTR(err);
1205 }
1206
1207 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1208 {
1209         struct vcpu_svm *svm = to_svm(vcpu);
1210
1211         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1212         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1213         __free_page(virt_to_page(svm->nested.hsave));
1214         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1215         kvm_vcpu_uninit(vcpu);
1216         kmem_cache_free(kvm_vcpu_cache, svm);
1217 }
1218
1219 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1220 {
1221         struct vcpu_svm *svm = to_svm(vcpu);
1222         int i;
1223
1224         if (unlikely(cpu != vcpu->cpu)) {
1225                 svm->asid_generation = 0;
1226                 mark_all_dirty(svm->vmcb);
1227         }
1228
1229 #ifdef CONFIG_X86_64
1230         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1231 #endif
1232         savesegment(fs, svm->host.fs);
1233         savesegment(gs, svm->host.gs);
1234         svm->host.ldt = kvm_read_ldt();
1235
1236         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1237                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1238
1239         if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1240             svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1241                 __get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1242                 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1243         }
1244 }
1245
1246 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1247 {
1248         struct vcpu_svm *svm = to_svm(vcpu);
1249         int i;
1250
1251         ++vcpu->stat.host_state_reload;
1252         kvm_load_ldt(svm->host.ldt);
1253 #ifdef CONFIG_X86_64
1254         loadsegment(fs, svm->host.fs);
1255         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1256         load_gs_index(svm->host.gs);
1257 #else
1258 #ifdef CONFIG_X86_32_LAZY_GS
1259         loadsegment(gs, svm->host.gs);
1260 #endif
1261 #endif
1262         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1263                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1264 }
1265
1266 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1267 {
1268         return to_svm(vcpu)->vmcb->save.rflags;
1269 }
1270
1271 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1272 {
1273         to_svm(vcpu)->vmcb->save.rflags = rflags;
1274 }
1275
1276 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1277 {
1278         switch (reg) {
1279         case VCPU_EXREG_PDPTR:
1280                 BUG_ON(!npt_enabled);
1281                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1282                 break;
1283         default:
1284                 BUG();
1285         }
1286 }
1287
1288 static void svm_set_vintr(struct vcpu_svm *svm)
1289 {
1290         set_intercept(svm, INTERCEPT_VINTR);
1291 }
1292
1293 static void svm_clear_vintr(struct vcpu_svm *svm)
1294 {
1295         clr_intercept(svm, INTERCEPT_VINTR);
1296 }
1297
1298 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1299 {
1300         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1301
1302         switch (seg) {
1303         case VCPU_SREG_CS: return &save->cs;
1304         case VCPU_SREG_DS: return &save->ds;
1305         case VCPU_SREG_ES: return &save->es;
1306         case VCPU_SREG_FS: return &save->fs;
1307         case VCPU_SREG_GS: return &save->gs;
1308         case VCPU_SREG_SS: return &save->ss;
1309         case VCPU_SREG_TR: return &save->tr;
1310         case VCPU_SREG_LDTR: return &save->ldtr;
1311         }
1312         BUG();
1313         return NULL;
1314 }
1315
1316 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1317 {
1318         struct vmcb_seg *s = svm_seg(vcpu, seg);
1319
1320         return s->base;
1321 }
1322
1323 static void svm_get_segment(struct kvm_vcpu *vcpu,
1324                             struct kvm_segment *var, int seg)
1325 {
1326         struct vmcb_seg *s = svm_seg(vcpu, seg);
1327
1328         var->base = s->base;
1329         var->limit = s->limit;
1330         var->selector = s->selector;
1331         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1332         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1333         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1334         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1335         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1336         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1337         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1338         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1339
1340         /*
1341          * AMD's VMCB does not have an explicit unusable field, so emulate it
1342          * for cross vendor migration purposes by "not present"
1343          */
1344         var->unusable = !var->present || (var->type == 0);
1345
1346         switch (seg) {
1347         case VCPU_SREG_CS:
1348                 /*
1349                  * SVM always stores 0 for the 'G' bit in the CS selector in
1350                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1351                  * Intel's VMENTRY has a check on the 'G' bit.
1352                  */
1353                 var->g = s->limit > 0xfffff;
1354                 break;
1355         case VCPU_SREG_TR:
1356                 /*
1357                  * Work around a bug where the busy flag in the tr selector
1358                  * isn't exposed
1359                  */
1360                 var->type |= 0x2;
1361                 break;
1362         case VCPU_SREG_DS:
1363         case VCPU_SREG_ES:
1364         case VCPU_SREG_FS:
1365         case VCPU_SREG_GS:
1366                 /*
1367                  * The accessed bit must always be set in the segment
1368                  * descriptor cache, although it can be cleared in the
1369                  * descriptor, the cached bit always remains at 1. Since
1370                  * Intel has a check on this, set it here to support
1371                  * cross-vendor migration.
1372                  */
1373                 if (!var->unusable)
1374                         var->type |= 0x1;
1375                 break;
1376         case VCPU_SREG_SS:
1377                 /*
1378                  * On AMD CPUs sometimes the DB bit in the segment
1379                  * descriptor is left as 1, although the whole segment has
1380                  * been made unusable. Clear it here to pass an Intel VMX
1381                  * entry check when cross vendor migrating.
1382                  */
1383                 if (var->unusable)
1384                         var->db = 0;
1385                 break;
1386         }
1387 }
1388
1389 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1390 {
1391         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1392
1393         return save->cpl;
1394 }
1395
1396 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1397 {
1398         struct vcpu_svm *svm = to_svm(vcpu);
1399
1400         dt->size = svm->vmcb->save.idtr.limit;
1401         dt->address = svm->vmcb->save.idtr.base;
1402 }
1403
1404 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1405 {
1406         struct vcpu_svm *svm = to_svm(vcpu);
1407
1408         svm->vmcb->save.idtr.limit = dt->size;
1409         svm->vmcb->save.idtr.base = dt->address ;
1410         mark_dirty(svm->vmcb, VMCB_DT);
1411 }
1412
1413 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1414 {
1415         struct vcpu_svm *svm = to_svm(vcpu);
1416
1417         dt->size = svm->vmcb->save.gdtr.limit;
1418         dt->address = svm->vmcb->save.gdtr.base;
1419 }
1420
1421 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1422 {
1423         struct vcpu_svm *svm = to_svm(vcpu);
1424
1425         svm->vmcb->save.gdtr.limit = dt->size;
1426         svm->vmcb->save.gdtr.base = dt->address ;
1427         mark_dirty(svm->vmcb, VMCB_DT);
1428 }
1429
1430 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1431 {
1432 }
1433
1434 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1435 {
1436 }
1437
1438 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1439 {
1440 }
1441
1442 static void update_cr0_intercept(struct vcpu_svm *svm)
1443 {
1444         ulong gcr0 = svm->vcpu.arch.cr0;
1445         u64 *hcr0 = &svm->vmcb->save.cr0;
1446
1447         if (!svm->vcpu.fpu_active)
1448                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1449         else
1450                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1451                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1452
1453         mark_dirty(svm->vmcb, VMCB_CR);
1454
1455         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1456                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1457                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1458         } else {
1459                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1460                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1461         }
1462 }
1463
1464 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1465 {
1466         struct vcpu_svm *svm = to_svm(vcpu);
1467
1468 #ifdef CONFIG_X86_64
1469         if (vcpu->arch.efer & EFER_LME) {
1470                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1471                         vcpu->arch.efer |= EFER_LMA;
1472                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1473                 }
1474
1475                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1476                         vcpu->arch.efer &= ~EFER_LMA;
1477                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1478                 }
1479         }
1480 #endif
1481         vcpu->arch.cr0 = cr0;
1482
1483         if (!npt_enabled)
1484                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1485
1486         if (!vcpu->fpu_active)
1487                 cr0 |= X86_CR0_TS;
1488         /*
1489          * re-enable caching here because the QEMU bios
1490          * does not do it - this results in some delay at
1491          * reboot
1492          */
1493         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1494         svm->vmcb->save.cr0 = cr0;
1495         mark_dirty(svm->vmcb, VMCB_CR);
1496         update_cr0_intercept(svm);
1497 }
1498
1499 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1500 {
1501         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1502         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1503
1504         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1505                 svm_flush_tlb(vcpu);
1506
1507         vcpu->arch.cr4 = cr4;
1508         if (!npt_enabled)
1509                 cr4 |= X86_CR4_PAE;
1510         cr4 |= host_cr4_mce;
1511         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1512         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1513 }
1514
1515 static void svm_set_segment(struct kvm_vcpu *vcpu,
1516                             struct kvm_segment *var, int seg)
1517 {
1518         struct vcpu_svm *svm = to_svm(vcpu);
1519         struct vmcb_seg *s = svm_seg(vcpu, seg);
1520
1521         s->base = var->base;
1522         s->limit = var->limit;
1523         s->selector = var->selector;
1524         if (var->unusable)
1525                 s->attrib = 0;
1526         else {
1527                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1528                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1529                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1530                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1531                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1532                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1533                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1534                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1535         }
1536         if (seg == VCPU_SREG_CS)
1537                 svm->vmcb->save.cpl
1538                         = (svm->vmcb->save.cs.attrib
1539                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
1540
1541         mark_dirty(svm->vmcb, VMCB_SEG);
1542 }
1543
1544 static void update_db_intercept(struct kvm_vcpu *vcpu)
1545 {
1546         struct vcpu_svm *svm = to_svm(vcpu);
1547
1548         clr_exception_intercept(svm, DB_VECTOR);
1549         clr_exception_intercept(svm, BP_VECTOR);
1550
1551         if (svm->nmi_singlestep)
1552                 set_exception_intercept(svm, DB_VECTOR);
1553
1554         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1555                 if (vcpu->guest_debug &
1556                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1557                         set_exception_intercept(svm, DB_VECTOR);
1558                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1559                         set_exception_intercept(svm, BP_VECTOR);
1560         } else
1561                 vcpu->guest_debug = 0;
1562 }
1563
1564 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1565 {
1566         struct vcpu_svm *svm = to_svm(vcpu);
1567
1568         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1569                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1570         else
1571                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1572
1573         mark_dirty(svm->vmcb, VMCB_DR);
1574
1575         update_db_intercept(vcpu);
1576 }
1577
1578 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1579 {
1580         if (sd->next_asid > sd->max_asid) {
1581                 ++sd->asid_generation;
1582                 sd->next_asid = 1;
1583                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1584         }
1585
1586         svm->asid_generation = sd->asid_generation;
1587         svm->vmcb->control.asid = sd->next_asid++;
1588
1589         mark_dirty(svm->vmcb, VMCB_ASID);
1590 }
1591
1592 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1593 {
1594         struct vcpu_svm *svm = to_svm(vcpu);
1595
1596         svm->vmcb->save.dr7 = value;
1597         mark_dirty(svm->vmcb, VMCB_DR);
1598 }
1599
1600 static int pf_interception(struct vcpu_svm *svm)
1601 {
1602         u64 fault_address = svm->vmcb->control.exit_info_2;
1603         u32 error_code;
1604         int r = 1;
1605
1606         switch (svm->apf_reason) {
1607         default:
1608                 error_code = svm->vmcb->control.exit_info_1;
1609
1610                 trace_kvm_page_fault(fault_address, error_code);
1611                 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1612                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1613                 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1614                         svm->vmcb->control.insn_bytes,
1615                         svm->vmcb->control.insn_len);
1616                 break;
1617         case KVM_PV_REASON_PAGE_NOT_PRESENT:
1618                 svm->apf_reason = 0;
1619                 local_irq_disable();
1620                 kvm_async_pf_task_wait(fault_address);
1621                 local_irq_enable();
1622                 break;
1623         case KVM_PV_REASON_PAGE_READY:
1624                 svm->apf_reason = 0;
1625                 local_irq_disable();
1626                 kvm_async_pf_task_wake(fault_address);
1627                 local_irq_enable();
1628                 break;
1629         }
1630         return r;
1631 }
1632
1633 static int db_interception(struct vcpu_svm *svm)
1634 {
1635         struct kvm_run *kvm_run = svm->vcpu.run;
1636
1637         if (!(svm->vcpu.guest_debug &
1638               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1639                 !svm->nmi_singlestep) {
1640                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1641                 return 1;
1642         }
1643
1644         if (svm->nmi_singlestep) {
1645                 svm->nmi_singlestep = false;
1646                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1647                         svm->vmcb->save.rflags &=
1648                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1649                 update_db_intercept(&svm->vcpu);
1650         }
1651
1652         if (svm->vcpu.guest_debug &
1653             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1654                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1655                 kvm_run->debug.arch.pc =
1656                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1657                 kvm_run->debug.arch.exception = DB_VECTOR;
1658                 return 0;
1659         }
1660
1661         return 1;
1662 }
1663
1664 static int bp_interception(struct vcpu_svm *svm)
1665 {
1666         struct kvm_run *kvm_run = svm->vcpu.run;
1667
1668         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1669         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1670         kvm_run->debug.arch.exception = BP_VECTOR;
1671         return 0;
1672 }
1673
1674 static int ud_interception(struct vcpu_svm *svm)
1675 {
1676         int er;
1677
1678         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1679         if (er != EMULATE_DONE)
1680                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1681         return 1;
1682 }
1683
1684 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1685 {
1686         struct vcpu_svm *svm = to_svm(vcpu);
1687
1688         clr_exception_intercept(svm, NM_VECTOR);
1689
1690         svm->vcpu.fpu_active = 1;
1691         update_cr0_intercept(svm);
1692 }
1693
1694 static int nm_interception(struct vcpu_svm *svm)
1695 {
1696         svm_fpu_activate(&svm->vcpu);
1697         return 1;
1698 }
1699
1700 static bool is_erratum_383(void)
1701 {
1702         int err, i;
1703         u64 value;
1704
1705         if (!erratum_383_found)
1706                 return false;
1707
1708         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1709         if (err)
1710                 return false;
1711
1712         /* Bit 62 may or may not be set for this mce */
1713         value &= ~(1ULL << 62);
1714
1715         if (value != 0xb600000000010015ULL)
1716                 return false;
1717
1718         /* Clear MCi_STATUS registers */
1719         for (i = 0; i < 6; ++i)
1720                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1721
1722         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1723         if (!err) {
1724                 u32 low, high;
1725
1726                 value &= ~(1ULL << 2);
1727                 low    = lower_32_bits(value);
1728                 high   = upper_32_bits(value);
1729
1730                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1731         }
1732
1733         /* Flush tlb to evict multi-match entries */
1734         __flush_tlb_all();
1735
1736         return true;
1737 }
1738
1739 static void svm_handle_mce(struct vcpu_svm *svm)
1740 {
1741         if (is_erratum_383()) {
1742                 /*
1743                  * Erratum 383 triggered. Guest state is corrupt so kill the
1744                  * guest.
1745                  */
1746                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1747
1748                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1749
1750                 return;
1751         }
1752
1753         /*
1754          * On an #MC intercept the MCE handler is not called automatically in
1755          * the host. So do it by hand here.
1756          */
1757         asm volatile (
1758                 "int $0x12\n");
1759         /* not sure if we ever come back to this point */
1760
1761         return;
1762 }
1763
1764 static int mc_interception(struct vcpu_svm *svm)
1765 {
1766         return 1;
1767 }
1768
1769 static int shutdown_interception(struct vcpu_svm *svm)
1770 {
1771         struct kvm_run *kvm_run = svm->vcpu.run;
1772
1773         /*
1774          * VMCB is undefined after a SHUTDOWN intercept
1775          * so reinitialize it.
1776          */
1777         clear_page(svm->vmcb);
1778         init_vmcb(svm);
1779
1780         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1781         return 0;
1782 }
1783
1784 static int io_interception(struct vcpu_svm *svm)
1785 {
1786         struct kvm_vcpu *vcpu = &svm->vcpu;
1787         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1788         int size, in, string;
1789         unsigned port;
1790
1791         ++svm->vcpu.stat.io_exits;
1792         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1793         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1794         if (string || in)
1795                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1796
1797         port = io_info >> 16;
1798         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1799         svm->next_rip = svm->vmcb->control.exit_info_2;
1800         skip_emulated_instruction(&svm->vcpu);
1801
1802         return kvm_fast_pio_out(vcpu, size, port);
1803 }
1804
1805 static int nmi_interception(struct vcpu_svm *svm)
1806 {
1807         return 1;
1808 }
1809
1810 static int intr_interception(struct vcpu_svm *svm)
1811 {
1812         ++svm->vcpu.stat.irq_exits;
1813         return 1;
1814 }
1815
1816 static int nop_on_interception(struct vcpu_svm *svm)
1817 {
1818         return 1;
1819 }
1820
1821 static int halt_interception(struct vcpu_svm *svm)
1822 {
1823         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1824         skip_emulated_instruction(&svm->vcpu);
1825         return kvm_emulate_halt(&svm->vcpu);
1826 }
1827
1828 static int vmmcall_interception(struct vcpu_svm *svm)
1829 {
1830         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1831         skip_emulated_instruction(&svm->vcpu);
1832         kvm_emulate_hypercall(&svm->vcpu);
1833         return 1;
1834 }
1835
1836 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1837 {
1838         struct vcpu_svm *svm = to_svm(vcpu);
1839
1840         return svm->nested.nested_cr3;
1841 }
1842
1843 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1844                                    unsigned long root)
1845 {
1846         struct vcpu_svm *svm = to_svm(vcpu);
1847
1848         svm->vmcb->control.nested_cr3 = root;
1849         mark_dirty(svm->vmcb, VMCB_NPT);
1850         svm_flush_tlb(vcpu);
1851 }
1852
1853 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1854                                        struct x86_exception *fault)
1855 {
1856         struct vcpu_svm *svm = to_svm(vcpu);
1857
1858         svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1859         svm->vmcb->control.exit_code_hi = 0;
1860         svm->vmcb->control.exit_info_1 = fault->error_code;
1861         svm->vmcb->control.exit_info_2 = fault->address;
1862
1863         nested_svm_vmexit(svm);
1864 }
1865
1866 static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1867 {
1868         int r;
1869
1870         r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1871
1872         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1873         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
1874         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1875         vcpu->arch.mmu.shadow_root_level = get_npt_level();
1876         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
1877
1878         return r;
1879 }
1880
1881 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1882 {
1883         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1884 }
1885
1886 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1887 {
1888         if (!(svm->vcpu.arch.efer & EFER_SVME)
1889             || !is_paging(&svm->vcpu)) {
1890                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1891                 return 1;
1892         }
1893
1894         if (svm->vmcb->save.cpl) {
1895                 kvm_inject_gp(&svm->vcpu, 0);
1896                 return 1;
1897         }
1898
1899        return 0;
1900 }
1901
1902 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1903                                       bool has_error_code, u32 error_code)
1904 {
1905         int vmexit;
1906
1907         if (!is_guest_mode(&svm->vcpu))
1908                 return 0;
1909
1910         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1911         svm->vmcb->control.exit_code_hi = 0;
1912         svm->vmcb->control.exit_info_1 = error_code;
1913         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1914
1915         vmexit = nested_svm_intercept(svm);
1916         if (vmexit == NESTED_EXIT_DONE)
1917                 svm->nested.exit_required = true;
1918
1919         return vmexit;
1920 }
1921
1922 /* This function returns true if it is save to enable the irq window */
1923 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1924 {
1925         if (!is_guest_mode(&svm->vcpu))
1926                 return true;
1927
1928         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1929                 return true;
1930
1931         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1932                 return false;
1933
1934         /*
1935          * if vmexit was already requested (by intercepted exception
1936          * for instance) do not overwrite it with "external interrupt"
1937          * vmexit.
1938          */
1939         if (svm->nested.exit_required)
1940                 return false;
1941
1942         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
1943         svm->vmcb->control.exit_info_1 = 0;
1944         svm->vmcb->control.exit_info_2 = 0;
1945
1946         if (svm->nested.intercept & 1ULL) {
1947                 /*
1948                  * The #vmexit can't be emulated here directly because this
1949                  * code path runs with irqs and preemtion disabled. A
1950                  * #vmexit emulation might sleep. Only signal request for
1951                  * the #vmexit here.
1952                  */
1953                 svm->nested.exit_required = true;
1954                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1955                 return false;
1956         }
1957
1958         return true;
1959 }
1960
1961 /* This function returns true if it is save to enable the nmi window */
1962 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1963 {
1964         if (!is_guest_mode(&svm->vcpu))
1965                 return true;
1966
1967         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1968                 return true;
1969
1970         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1971         svm->nested.exit_required = true;
1972
1973         return false;
1974 }
1975
1976 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1977 {
1978         struct page *page;
1979
1980         might_sleep();
1981
1982         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1983         if (is_error_page(page))
1984                 goto error;
1985
1986         *_page = page;
1987
1988         return kmap(page);
1989
1990 error:
1991         kvm_release_page_clean(page);
1992         kvm_inject_gp(&svm->vcpu, 0);
1993
1994         return NULL;
1995 }
1996
1997 static void nested_svm_unmap(struct page *page)
1998 {
1999         kunmap(page);
2000         kvm_release_page_dirty(page);
2001 }
2002
2003 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2004 {
2005         unsigned port;
2006         u8 val, bit;
2007         u64 gpa;
2008
2009         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2010                 return NESTED_EXIT_HOST;
2011
2012         port = svm->vmcb->control.exit_info_1 >> 16;
2013         gpa  = svm->nested.vmcb_iopm + (port / 8);
2014         bit  = port % 8;
2015         val  = 0;
2016
2017         if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2018                 val &= (1 << bit);
2019
2020         return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2021 }
2022
2023 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2024 {
2025         u32 offset, msr, value;
2026         int write, mask;
2027
2028         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2029                 return NESTED_EXIT_HOST;
2030
2031         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2032         offset = svm_msrpm_offset(msr);
2033         write  = svm->vmcb->control.exit_info_1 & 1;
2034         mask   = 1 << ((2 * (msr & 0xf)) + write);
2035
2036         if (offset == MSR_INVALID)
2037                 return NESTED_EXIT_DONE;
2038
2039         /* Offset is in 32 bit units but need in 8 bit units */
2040         offset *= 4;
2041
2042         if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2043                 return NESTED_EXIT_DONE;
2044
2045         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2046 }
2047
2048 static int nested_svm_exit_special(struct vcpu_svm *svm)
2049 {
2050         u32 exit_code = svm->vmcb->control.exit_code;
2051
2052         switch (exit_code) {
2053         case SVM_EXIT_INTR:
2054         case SVM_EXIT_NMI:
2055         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2056                 return NESTED_EXIT_HOST;
2057         case SVM_EXIT_NPF:
2058                 /* For now we are always handling NPFs when using them */
2059                 if (npt_enabled)
2060                         return NESTED_EXIT_HOST;
2061                 break;
2062         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2063                 /* When we're shadowing, trap PFs, but not async PF */
2064                 if (!npt_enabled && svm->apf_reason == 0)
2065                         return NESTED_EXIT_HOST;
2066                 break;
2067         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2068                 nm_interception(svm);
2069                 break;
2070         default:
2071                 break;
2072         }
2073
2074         return NESTED_EXIT_CONTINUE;
2075 }
2076
2077 /*
2078  * If this function returns true, this #vmexit was already handled
2079  */
2080 static int nested_svm_intercept(struct vcpu_svm *svm)
2081 {
2082         u32 exit_code = svm->vmcb->control.exit_code;
2083         int vmexit = NESTED_EXIT_HOST;
2084
2085         switch (exit_code) {
2086         case SVM_EXIT_MSR:
2087                 vmexit = nested_svm_exit_handled_msr(svm);
2088                 break;
2089         case SVM_EXIT_IOIO:
2090                 vmexit = nested_svm_intercept_ioio(svm);
2091                 break;
2092         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2093                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2094                 if (svm->nested.intercept_cr & bit)
2095                         vmexit = NESTED_EXIT_DONE;
2096                 break;
2097         }
2098         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2099                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2100                 if (svm->nested.intercept_dr & bit)
2101                         vmexit = NESTED_EXIT_DONE;
2102                 break;
2103         }
2104         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2105                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2106                 if (svm->nested.intercept_exceptions & excp_bits)
2107                         vmexit = NESTED_EXIT_DONE;
2108                 /* async page fault always cause vmexit */
2109                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2110                          svm->apf_reason != 0)
2111                         vmexit = NESTED_EXIT_DONE;
2112                 break;
2113         }
2114         case SVM_EXIT_ERR: {
2115                 vmexit = NESTED_EXIT_DONE;
2116                 break;
2117         }
2118         default: {
2119                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2120                 if (svm->nested.intercept & exit_bits)
2121                         vmexit = NESTED_EXIT_DONE;
2122         }
2123         }
2124
2125         return vmexit;
2126 }
2127
2128 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2129 {
2130         int vmexit;
2131
2132         vmexit = nested_svm_intercept(svm);
2133
2134         if (vmexit == NESTED_EXIT_DONE)
2135                 nested_svm_vmexit(svm);
2136
2137         return vmexit;
2138 }
2139
2140 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2141 {
2142         struct vmcb_control_area *dst  = &dst_vmcb->control;
2143         struct vmcb_control_area *from = &from_vmcb->control;
2144
2145         dst->intercept_cr         = from->intercept_cr;
2146         dst->intercept_dr         = from->intercept_dr;
2147         dst->intercept_exceptions = from->intercept_exceptions;
2148         dst->intercept            = from->intercept;
2149         dst->iopm_base_pa         = from->iopm_base_pa;
2150         dst->msrpm_base_pa        = from->msrpm_base_pa;
2151         dst->tsc_offset           = from->tsc_offset;
2152         dst->asid                 = from->asid;
2153         dst->tlb_ctl              = from->tlb_ctl;
2154         dst->int_ctl              = from->int_ctl;
2155         dst->int_vector           = from->int_vector;
2156         dst->int_state            = from->int_state;
2157         dst->exit_code            = from->exit_code;
2158         dst->exit_code_hi         = from->exit_code_hi;
2159         dst->exit_info_1          = from->exit_info_1;
2160         dst->exit_info_2          = from->exit_info_2;
2161         dst->exit_int_info        = from->exit_int_info;
2162         dst->exit_int_info_err    = from->exit_int_info_err;
2163         dst->nested_ctl           = from->nested_ctl;
2164         dst->event_inj            = from->event_inj;
2165         dst->event_inj_err        = from->event_inj_err;
2166         dst->nested_cr3           = from->nested_cr3;
2167         dst->lbr_ctl              = from->lbr_ctl;
2168 }
2169
2170 static int nested_svm_vmexit(struct vcpu_svm *svm)
2171 {
2172         struct vmcb *nested_vmcb;
2173         struct vmcb *hsave = svm->nested.hsave;
2174         struct vmcb *vmcb = svm->vmcb;
2175         struct page *page;
2176
2177         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2178                                        vmcb->control.exit_info_1,
2179                                        vmcb->control.exit_info_2,
2180                                        vmcb->control.exit_int_info,
2181                                        vmcb->control.exit_int_info_err);
2182
2183         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2184         if (!nested_vmcb)
2185                 return 1;
2186
2187         /* Exit Guest-Mode */
2188         leave_guest_mode(&svm->vcpu);
2189         svm->nested.vmcb = 0;
2190
2191         /* Give the current vmcb to the guest */
2192         disable_gif(svm);
2193
2194         nested_vmcb->save.es     = vmcb->save.es;
2195         nested_vmcb->save.cs     = vmcb->save.cs;
2196         nested_vmcb->save.ss     = vmcb->save.ss;
2197         nested_vmcb->save.ds     = vmcb->save.ds;
2198         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2199         nested_vmcb->save.idtr   = vmcb->save.idtr;
2200         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2201         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2202         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2203         nested_vmcb->save.cr2    = vmcb->save.cr2;
2204         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2205         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2206         nested_vmcb->save.rip    = vmcb->save.rip;
2207         nested_vmcb->save.rsp    = vmcb->save.rsp;
2208         nested_vmcb->save.rax    = vmcb->save.rax;
2209         nested_vmcb->save.dr7    = vmcb->save.dr7;
2210         nested_vmcb->save.dr6    = vmcb->save.dr6;
2211         nested_vmcb->save.cpl    = vmcb->save.cpl;
2212
2213         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2214         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2215         nested_vmcb->control.int_state         = vmcb->control.int_state;
2216         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2217         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2218         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2219         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2220         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2221         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2222         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2223
2224         /*
2225          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2226          * to make sure that we do not lose injected events. So check event_inj
2227          * here and copy it to exit_int_info if it is valid.
2228          * Exit_int_info and event_inj can't be both valid because the case
2229          * below only happens on a VMRUN instruction intercept which has
2230          * no valid exit_int_info set.
2231          */
2232         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2233                 struct vmcb_control_area *nc = &nested_vmcb->control;
2234
2235                 nc->exit_int_info     = vmcb->control.event_inj;
2236                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2237         }
2238
2239         nested_vmcb->control.tlb_ctl           = 0;
2240         nested_vmcb->control.event_inj         = 0;
2241         nested_vmcb->control.event_inj_err     = 0;
2242
2243         /* We always set V_INTR_MASKING and remember the old value in hflags */
2244         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2245                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2246
2247         /* Restore the original control entries */
2248         copy_vmcb_control_area(vmcb, hsave);
2249
2250         kvm_clear_exception_queue(&svm->vcpu);
2251         kvm_clear_interrupt_queue(&svm->vcpu);
2252
2253         svm->nested.nested_cr3 = 0;
2254
2255         /* Restore selected save entries */
2256         svm->vmcb->save.es = hsave->save.es;
2257         svm->vmcb->save.cs = hsave->save.cs;
2258         svm->vmcb->save.ss = hsave->save.ss;
2259         svm->vmcb->save.ds = hsave->save.ds;
2260         svm->vmcb->save.gdtr = hsave->save.gdtr;
2261         svm->vmcb->save.idtr = hsave->save.idtr;
2262         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2263         svm_set_efer(&svm->vcpu, hsave->save.efer);
2264         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2265         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2266         if (npt_enabled) {
2267                 svm->vmcb->save.cr3 = hsave->save.cr3;
2268                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2269         } else {
2270                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2271         }
2272         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2273         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2274         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2275         svm->vmcb->save.dr7 = 0;
2276         svm->vmcb->save.cpl = 0;
2277         svm->vmcb->control.exit_int_info = 0;
2278
2279         mark_all_dirty(svm->vmcb);
2280
2281         nested_svm_unmap(page);
2282
2283         nested_svm_uninit_mmu_context(&svm->vcpu);
2284         kvm_mmu_reset_context(&svm->vcpu);
2285         kvm_mmu_load(&svm->vcpu);
2286
2287         return 0;
2288 }
2289
2290 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2291 {
2292         /*
2293          * This function merges the msr permission bitmaps of kvm and the
2294          * nested vmcb. It is omptimized in that it only merges the parts where
2295          * the kvm msr permission bitmap may contain zero bits
2296          */
2297         int i;
2298
2299         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2300                 return true;
2301
2302         for (i = 0; i < MSRPM_OFFSETS; i++) {
2303                 u32 value, p;
2304                 u64 offset;
2305
2306                 if (msrpm_offsets[i] == 0xffffffff)
2307                         break;
2308
2309                 p      = msrpm_offsets[i];
2310                 offset = svm->nested.vmcb_msrpm + (p * 4);
2311
2312                 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2313                         return false;
2314
2315                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2316         }
2317
2318         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2319
2320         return true;
2321 }
2322
2323 static bool nested_vmcb_checks(struct vmcb *vmcb)
2324 {
2325         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2326                 return false;
2327
2328         if (vmcb->control.asid == 0)
2329                 return false;
2330
2331         if (vmcb->control.nested_ctl && !npt_enabled)
2332                 return false;
2333
2334         return true;
2335 }
2336
2337 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2338 {
2339         struct vmcb *nested_vmcb;
2340         struct vmcb *hsave = svm->nested.hsave;
2341         struct vmcb *vmcb = svm->vmcb;
2342         struct page *page;
2343         u64 vmcb_gpa;
2344
2345         vmcb_gpa = svm->vmcb->save.rax;
2346
2347         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2348         if (!nested_vmcb)
2349                 return false;
2350
2351         if (!nested_vmcb_checks(nested_vmcb)) {
2352                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2353                 nested_vmcb->control.exit_code_hi = 0;
2354                 nested_vmcb->control.exit_info_1  = 0;
2355                 nested_vmcb->control.exit_info_2  = 0;
2356
2357                 nested_svm_unmap(page);
2358
2359                 return false;
2360         }
2361
2362         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2363                                nested_vmcb->save.rip,
2364                                nested_vmcb->control.int_ctl,
2365                                nested_vmcb->control.event_inj,
2366                                nested_vmcb->control.nested_ctl);
2367
2368         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2369                                     nested_vmcb->control.intercept_cr >> 16,
2370                                     nested_vmcb->control.intercept_exceptions,
2371                                     nested_vmcb->control.intercept);
2372
2373         /* Clear internal status */
2374         kvm_clear_exception_queue(&svm->vcpu);
2375         kvm_clear_interrupt_queue(&svm->vcpu);
2376
2377         /*
2378          * Save the old vmcb, so we don't need to pick what we save, but can
2379          * restore everything when a VMEXIT occurs
2380          */
2381         hsave->save.es     = vmcb->save.es;
2382         hsave->save.cs     = vmcb->save.cs;
2383         hsave->save.ss     = vmcb->save.ss;
2384         hsave->save.ds     = vmcb->save.ds;
2385         hsave->save.gdtr   = vmcb->save.gdtr;
2386         hsave->save.idtr   = vmcb->save.idtr;
2387         hsave->save.efer   = svm->vcpu.arch.efer;
2388         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2389         hsave->save.cr4    = svm->vcpu.arch.cr4;
2390         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2391         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2392         hsave->save.rsp    = vmcb->save.rsp;
2393         hsave->save.rax    = vmcb->save.rax;
2394         if (npt_enabled)
2395                 hsave->save.cr3    = vmcb->save.cr3;
2396         else
2397                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2398
2399         copy_vmcb_control_area(hsave, vmcb);
2400
2401         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2402                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2403         else
2404                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2405
2406         if (nested_vmcb->control.nested_ctl) {
2407                 kvm_mmu_unload(&svm->vcpu);
2408                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2409                 nested_svm_init_mmu_context(&svm->vcpu);
2410         }
2411
2412         /* Load the nested guest state */
2413         svm->vmcb->save.es = nested_vmcb->save.es;
2414         svm->vmcb->save.cs = nested_vmcb->save.cs;
2415         svm->vmcb->save.ss = nested_vmcb->save.ss;
2416         svm->vmcb->save.ds = nested_vmcb->save.ds;
2417         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2418         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2419         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2420         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2421         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2422         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2423         if (npt_enabled) {
2424                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2425                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2426         } else
2427                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2428
2429         /* Guest paging mode is active - reset mmu */
2430         kvm_mmu_reset_context(&svm->vcpu);
2431
2432         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2433         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2434         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2435         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2436
2437         /* In case we don't even reach vcpu_run, the fields are not updated */
2438         svm->vmcb->save.rax = nested_vmcb->save.rax;
2439         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2440         svm->vmcb->save.rip = nested_vmcb->save.rip;
2441         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2442         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2443         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2444
2445         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2446         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2447
2448         /* cache intercepts */
2449         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2450         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2451         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2452         svm->nested.intercept            = nested_vmcb->control.intercept;
2453
2454         svm_flush_tlb(&svm->vcpu);
2455         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2456         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2457                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2458         else
2459                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2460
2461         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2462                 /* We only want the cr8 intercept bits of the guest */
2463                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2464                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2465         }
2466
2467         /* We don't want to see VMMCALLs from a nested guest */
2468         clr_intercept(svm, INTERCEPT_VMMCALL);
2469
2470         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2471         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2472         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2473         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2474         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2475         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2476
2477         nested_svm_unmap(page);
2478
2479         /* Enter Guest-Mode */
2480         enter_guest_mode(&svm->vcpu);
2481
2482         /*
2483          * Merge guest and host intercepts - must be called  with vcpu in
2484          * guest-mode to take affect here
2485          */
2486         recalc_intercepts(svm);
2487
2488         svm->nested.vmcb = vmcb_gpa;
2489
2490         enable_gif(svm);
2491
2492         mark_all_dirty(svm->vmcb);
2493
2494         return true;
2495 }
2496
2497 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2498 {
2499         to_vmcb->save.fs = from_vmcb->save.fs;
2500         to_vmcb->save.gs = from_vmcb->save.gs;
2501         to_vmcb->save.tr = from_vmcb->save.tr;
2502         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2503         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2504         to_vmcb->save.star = from_vmcb->save.star;
2505         to_vmcb->save.lstar = from_vmcb->save.lstar;
2506         to_vmcb->save.cstar = from_vmcb->save.cstar;
2507         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2508         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2509         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2510         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2511 }
2512
2513 static int vmload_interception(struct vcpu_svm *svm)
2514 {
2515         struct vmcb *nested_vmcb;
2516         struct page *page;
2517
2518         if (nested_svm_check_permissions(svm))
2519                 return 1;
2520
2521         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2522         if (!nested_vmcb)
2523                 return 1;
2524
2525         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2526         skip_emulated_instruction(&svm->vcpu);
2527
2528         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2529         nested_svm_unmap(page);
2530
2531         return 1;
2532 }
2533
2534 static int vmsave_interception(struct vcpu_svm *svm)
2535 {
2536         struct vmcb *nested_vmcb;
2537         struct page *page;
2538
2539         if (nested_svm_check_permissions(svm))
2540                 return 1;
2541
2542         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2543         if (!nested_vmcb)
2544                 return 1;
2545
2546         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2547         skip_emulated_instruction(&svm->vcpu);
2548
2549         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2550         nested_svm_unmap(page);
2551
2552         return 1;
2553 }
2554
2555 static int vmrun_interception(struct vcpu_svm *svm)
2556 {
2557         if (nested_svm_check_permissions(svm))
2558                 return 1;
2559
2560         /* Save rip after vmrun instruction */
2561         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2562
2563         if (!nested_svm_vmrun(svm))
2564                 return 1;
2565
2566         if (!nested_svm_vmrun_msrpm(svm))
2567                 goto failed;
2568
2569         return 1;
2570
2571 failed:
2572
2573         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2574         svm->vmcb->control.exit_code_hi = 0;
2575         svm->vmcb->control.exit_info_1  = 0;
2576         svm->vmcb->control.exit_info_2  = 0;
2577
2578         nested_svm_vmexit(svm);
2579
2580         return 1;
2581 }
2582
2583 static int stgi_interception(struct vcpu_svm *svm)
2584 {
2585         if (nested_svm_check_permissions(svm))
2586                 return 1;
2587
2588         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2589         skip_emulated_instruction(&svm->vcpu);
2590         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2591
2592         enable_gif(svm);
2593
2594         return 1;
2595 }
2596
2597 static int clgi_interception(struct vcpu_svm *svm)
2598 {
2599         if (nested_svm_check_permissions(svm))
2600                 return 1;
2601
2602         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2603         skip_emulated_instruction(&svm->vcpu);
2604
2605         disable_gif(svm);
2606
2607         /* After a CLGI no interrupts should come */
2608         svm_clear_vintr(svm);
2609         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2610
2611         mark_dirty(svm->vmcb, VMCB_INTR);
2612
2613         return 1;
2614 }
2615
2616 static int invlpga_interception(struct vcpu_svm *svm)
2617 {
2618         struct kvm_vcpu *vcpu = &svm->vcpu;
2619
2620         trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2621                           vcpu->arch.regs[VCPU_REGS_RAX]);
2622
2623         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2624         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2625
2626         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2627         skip_emulated_instruction(&svm->vcpu);
2628         return 1;
2629 }
2630
2631 static int skinit_interception(struct vcpu_svm *svm)
2632 {
2633         trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2634
2635         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2636         return 1;
2637 }
2638
2639 static int xsetbv_interception(struct vcpu_svm *svm)
2640 {
2641         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2642         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2643
2644         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2645                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2646                 skip_emulated_instruction(&svm->vcpu);
2647         }
2648
2649         return 1;
2650 }
2651
2652 static int invalid_op_interception(struct vcpu_svm *svm)
2653 {
2654         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2655         return 1;
2656 }
2657
2658 static int task_switch_interception(struct vcpu_svm *svm)
2659 {
2660         u16 tss_selector;
2661         int reason;
2662         int int_type = svm->vmcb->control.exit_int_info &
2663                 SVM_EXITINTINFO_TYPE_MASK;
2664         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2665         uint32_t type =
2666                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2667         uint32_t idt_v =
2668                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2669         bool has_error_code = false;
2670         u32 error_code = 0;
2671
2672         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2673
2674         if (svm->vmcb->control.exit_info_2 &
2675             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2676                 reason = TASK_SWITCH_IRET;
2677         else if (svm->vmcb->control.exit_info_2 &
2678                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2679                 reason = TASK_SWITCH_JMP;
2680         else if (idt_v)
2681                 reason = TASK_SWITCH_GATE;
2682         else
2683                 reason = TASK_SWITCH_CALL;
2684
2685         if (reason == TASK_SWITCH_GATE) {
2686                 switch (type) {
2687                 case SVM_EXITINTINFO_TYPE_NMI:
2688                         svm->vcpu.arch.nmi_injected = false;
2689                         break;
2690                 case SVM_EXITINTINFO_TYPE_EXEPT:
2691                         if (svm->vmcb->control.exit_info_2 &
2692                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2693                                 has_error_code = true;
2694                                 error_code =
2695                                         (u32)svm->vmcb->control.exit_info_2;
2696                         }
2697                         kvm_clear_exception_queue(&svm->vcpu);
2698                         break;
2699                 case SVM_EXITINTINFO_TYPE_INTR:
2700                         kvm_clear_interrupt_queue(&svm->vcpu);
2701                         break;
2702                 default:
2703                         break;
2704                 }
2705         }
2706
2707         if (reason != TASK_SWITCH_GATE ||
2708             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2709             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2710              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2711                 skip_emulated_instruction(&svm->vcpu);
2712
2713         if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
2714                                 has_error_code, error_code) == EMULATE_FAIL) {
2715                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2716                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2717                 svm->vcpu.run->internal.ndata = 0;
2718                 return 0;
2719         }
2720         return 1;
2721 }
2722
2723 static int cpuid_interception(struct vcpu_svm *svm)
2724 {
2725         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2726         kvm_emulate_cpuid(&svm->vcpu);
2727         return 1;
2728 }
2729
2730 static int iret_interception(struct vcpu_svm *svm)
2731 {
2732         ++svm->vcpu.stat.nmi_window_exits;
2733         clr_intercept(svm, INTERCEPT_IRET);
2734         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2735         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2736         return 1;
2737 }
2738
2739 static int invlpg_interception(struct vcpu_svm *svm)
2740 {
2741         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2742                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2743
2744         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2745         skip_emulated_instruction(&svm->vcpu);
2746         return 1;
2747 }
2748
2749 static int emulate_on_interception(struct vcpu_svm *svm)
2750 {
2751         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2752 }
2753
2754 bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2755 {
2756         unsigned long cr0 = svm->vcpu.arch.cr0;
2757         bool ret = false;
2758         u64 intercept;
2759
2760         intercept = svm->nested.intercept;
2761
2762         if (!is_guest_mode(&svm->vcpu) ||
2763             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2764                 return false;
2765
2766         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2767         val &= ~SVM_CR0_SELECTIVE_MASK;
2768
2769         if (cr0 ^ val) {
2770                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2771                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2772         }
2773
2774         return ret;
2775 }
2776
2777 #define CR_VALID (1ULL << 63)
2778
2779 static int cr_interception(struct vcpu_svm *svm)
2780 {
2781         int reg, cr;
2782         unsigned long val;
2783         int err;
2784
2785         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2786                 return emulate_on_interception(svm);
2787
2788         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2789                 return emulate_on_interception(svm);
2790
2791         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2792         cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2793
2794         err = 0;
2795         if (cr >= 16) { /* mov to cr */
2796                 cr -= 16;
2797                 val = kvm_register_read(&svm->vcpu, reg);
2798                 switch (cr) {
2799                 case 0:
2800                         if (!check_selective_cr0_intercepted(svm, val))
2801                                 err = kvm_set_cr0(&svm->vcpu, val);
2802                         else
2803                                 return 1;
2804
2805                         break;
2806                 case 3:
2807                         err = kvm_set_cr3(&svm->vcpu, val);
2808                         break;
2809                 case 4:
2810                         err = kvm_set_cr4(&svm->vcpu, val);
2811                         break;
2812                 case 8:
2813                         err = kvm_set_cr8(&svm->vcpu, val);
2814                         break;
2815                 default:
2816                         WARN(1, "unhandled write to CR%d", cr);
2817                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2818                         return 1;
2819                 }
2820         } else { /* mov from cr */
2821                 switch (cr) {
2822                 case 0:
2823                         val = kvm_read_cr0(&svm->vcpu);
2824                         break;
2825                 case 2:
2826                         val = svm->vcpu.arch.cr2;
2827                         break;
2828                 case 3:
2829                         val = kvm_read_cr3(&svm->vcpu);
2830                         break;
2831                 case 4:
2832                         val = kvm_read_cr4(&svm->vcpu);
2833                         break;
2834                 case 8:
2835                         val = kvm_get_cr8(&svm->vcpu);
2836                         break;
2837                 default:
2838                         WARN(1, "unhandled read from CR%d", cr);
2839                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2840                         return 1;
2841                 }
2842                 kvm_register_write(&svm->vcpu, reg, val);
2843         }
2844         kvm_complete_insn_gp(&svm->vcpu, err);
2845
2846         return 1;
2847 }
2848
2849 static int dr_interception(struct vcpu_svm *svm)
2850 {
2851         int reg, dr;
2852         unsigned long val;
2853         int err;
2854
2855         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2856                 return emulate_on_interception(svm);
2857
2858         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2859         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2860
2861         if (dr >= 16) { /* mov to DRn */
2862                 val = kvm_register_read(&svm->vcpu, reg);
2863                 kvm_set_dr(&svm->vcpu, dr - 16, val);
2864         } else {
2865                 err = kvm_get_dr(&svm->vcpu, dr, &val);
2866                 if (!err)
2867                         kvm_register_write(&svm->vcpu, reg, val);
2868         }
2869
2870         skip_emulated_instruction(&svm->vcpu);
2871
2872         return 1;
2873 }
2874
2875 static int cr8_write_interception(struct vcpu_svm *svm)
2876 {
2877         struct kvm_run *kvm_run = svm->vcpu.run;
2878         int r;
2879
2880         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2881         /* instruction emulation calls kvm_set_cr8() */
2882         r = cr_interception(svm);
2883         if (irqchip_in_kernel(svm->vcpu.kvm)) {
2884                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2885                 return r;
2886         }
2887         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2888                 return r;
2889         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2890         return 0;
2891 }
2892
2893 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2894 {
2895         struct vcpu_svm *svm = to_svm(vcpu);
2896
2897         switch (ecx) {
2898         case MSR_IA32_TSC: {
2899                 struct vmcb *vmcb = get_host_vmcb(svm);
2900
2901                 *data = vmcb->control.tsc_offset +
2902                         svm_scale_tsc(vcpu, native_read_tsc());
2903
2904                 break;
2905         }
2906         case MSR_STAR:
2907                 *data = svm->vmcb->save.star;
2908                 break;
2909 #ifdef CONFIG_X86_64
2910         case MSR_LSTAR:
2911                 *data = svm->vmcb->save.lstar;
2912                 break;
2913         case MSR_CSTAR:
2914                 *data = svm->vmcb->save.cstar;
2915                 break;
2916         case MSR_KERNEL_GS_BASE:
2917                 *data = svm->vmcb->save.kernel_gs_base;
2918                 break;
2919         case MSR_SYSCALL_MASK:
2920                 *data = svm->vmcb->save.sfmask;
2921                 break;
2922 #endif
2923         case MSR_IA32_SYSENTER_CS:
2924                 *data = svm->vmcb->save.sysenter_cs;
2925                 break;
2926         case MSR_IA32_SYSENTER_EIP:
2927                 *data = svm->sysenter_eip;
2928                 break;
2929         case MSR_IA32_SYSENTER_ESP:
2930                 *data = svm->sysenter_esp;
2931                 break;
2932         /*
2933          * Nobody will change the following 5 values in the VMCB so we can
2934          * safely return them on rdmsr. They will always be 0 until LBRV is
2935          * implemented.
2936          */
2937         case MSR_IA32_DEBUGCTLMSR:
2938                 *data = svm->vmcb->save.dbgctl;
2939                 break;
2940         case MSR_IA32_LASTBRANCHFROMIP:
2941                 *data = svm->vmcb->save.br_from;
2942                 break;
2943         case MSR_IA32_LASTBRANCHTOIP:
2944                 *data = svm->vmcb->save.br_to;
2945                 break;
2946         case MSR_IA32_LASTINTFROMIP:
2947                 *data = svm->vmcb->save.last_excp_from;
2948                 break;
2949         case MSR_IA32_LASTINTTOIP:
2950                 *data = svm->vmcb->save.last_excp_to;
2951                 break;
2952         case MSR_VM_HSAVE_PA:
2953                 *data = svm->nested.hsave_msr;
2954                 break;
2955         case MSR_VM_CR:
2956                 *data = svm->nested.vm_cr_msr;
2957                 break;
2958         case MSR_IA32_UCODE_REV:
2959                 *data = 0x01000065;
2960                 break;
2961         default:
2962                 return kvm_get_msr_common(vcpu, ecx, data);
2963         }
2964         return 0;
2965 }
2966
2967 static int rdmsr_interception(struct vcpu_svm *svm)
2968 {
2969         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2970         u64 data;
2971
2972         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2973                 trace_kvm_msr_read_ex(ecx);
2974                 kvm_inject_gp(&svm->vcpu, 0);
2975         } else {
2976                 trace_kvm_msr_read(ecx, data);
2977
2978                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2979                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2980                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2981                 skip_emulated_instruction(&svm->vcpu);
2982         }
2983         return 1;
2984 }
2985
2986 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2987 {
2988         struct vcpu_svm *svm = to_svm(vcpu);
2989         int svm_dis, chg_mask;
2990
2991         if (data & ~SVM_VM_CR_VALID_MASK)
2992                 return 1;
2993
2994         chg_mask = SVM_VM_CR_VALID_MASK;
2995
2996         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2997                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2998
2999         svm->nested.vm_cr_msr &= ~chg_mask;
3000         svm->nested.vm_cr_msr |= (data & chg_mask);
3001
3002         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3003
3004         /* check for svm_disable while efer.svme is set */
3005         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3006                 return 1;
3007
3008         return 0;
3009 }
3010
3011 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
3012 {
3013         struct vcpu_svm *svm = to_svm(vcpu);
3014
3015         switch (ecx) {
3016         case MSR_IA32_TSC:
3017                 kvm_write_tsc(vcpu, data);
3018                 break;
3019         case MSR_STAR:
3020                 svm->vmcb->save.star = data;
3021                 break;
3022 #ifdef CONFIG_X86_64
3023         case MSR_LSTAR:
3024                 svm->vmcb->save.lstar = data;
3025                 break;
3026         case MSR_CSTAR:
3027                 svm->vmcb->save.cstar = data;
3028                 break;
3029         case MSR_KERNEL_GS_BASE:
3030                 svm->vmcb->save.kernel_gs_base = data;
3031                 break;
3032         case MSR_SYSCALL_MASK:
3033                 svm->vmcb->save.sfmask = data;
3034                 break;
3035 #endif
3036         case MSR_IA32_SYSENTER_CS:
3037                 svm->vmcb->save.sysenter_cs = data;
3038                 break;
3039         case MSR_IA32_SYSENTER_EIP:
3040                 svm->sysenter_eip = data;
3041                 svm->vmcb->save.sysenter_eip = data;
3042                 break;
3043         case MSR_IA32_SYSENTER_ESP:
3044                 svm->sysenter_esp = data;
3045                 svm->vmcb->save.sysenter_esp = data;
3046                 break;
3047         case MSR_IA32_DEBUGCTLMSR:
3048                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3049                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3050                                         __func__, data);
3051                         break;
3052                 }
3053                 if (data & DEBUGCTL_RESERVED_BITS)
3054                         return 1;
3055
3056                 svm->vmcb->save.dbgctl = data;
3057                 mark_dirty(svm->vmcb, VMCB_LBR);
3058                 if (data & (1ULL<<0))
3059                         svm_enable_lbrv(svm);
3060                 else
3061                         svm_disable_lbrv(svm);
3062                 break;
3063         case MSR_VM_HSAVE_PA:
3064                 svm->nested.hsave_msr = data;
3065                 break;
3066         case MSR_VM_CR:
3067                 return svm_set_vm_cr(vcpu, data);
3068         case MSR_VM_IGNNE:
3069                 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3070                 break;
3071         default:
3072                 return kvm_set_msr_common(vcpu, ecx, data);
3073         }
3074         return 0;
3075 }
3076
3077 static int wrmsr_interception(struct vcpu_svm *svm)
3078 {
3079         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3080         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3081                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3082
3083
3084         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3085         if (svm_set_msr(&svm->vcpu, ecx, data)) {
3086                 trace_kvm_msr_write_ex(ecx, data);
3087                 kvm_inject_gp(&svm->vcpu, 0);
3088         } else {
3089                 trace_kvm_msr_write(ecx, data);
3090                 skip_emulated_instruction(&svm->vcpu);
3091         }
3092         return 1;
3093 }
3094
3095 static int msr_interception(struct vcpu_svm *svm)
3096 {
3097         if (svm->vmcb->control.exit_info_1)
3098                 return wrmsr_interception(svm);
3099         else
3100                 return rdmsr_interception(svm);
3101 }
3102
3103 static int interrupt_window_interception(struct vcpu_svm *svm)
3104 {
3105         struct kvm_run *kvm_run = svm->vcpu.run;
3106
3107         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3108         svm_clear_vintr(svm);
3109         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3110         mark_dirty(svm->vmcb, VMCB_INTR);
3111         /*
3112          * If the user space waits to inject interrupts, exit as soon as
3113          * possible
3114          */
3115         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3116             kvm_run->request_interrupt_window &&
3117             !kvm_cpu_has_interrupt(&svm->vcpu)) {
3118                 ++svm->vcpu.stat.irq_window_exits;
3119                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3120                 return 0;
3121         }
3122
3123         return 1;
3124 }
3125
3126 static int pause_interception(struct vcpu_svm *svm)
3127 {
3128         kvm_vcpu_on_spin(&(svm->vcpu));
3129         return 1;
3130 }
3131
3132 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
3133         [SVM_EXIT_READ_CR0]                     = cr_interception,
3134         [SVM_EXIT_READ_CR3]                     = cr_interception,
3135         [SVM_EXIT_READ_CR4]                     = cr_interception,
3136         [SVM_EXIT_READ_CR8]                     = cr_interception,
3137         [SVM_EXIT_CR0_SEL_WRITE]                = emulate_on_interception,
3138         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
3139         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
3140         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
3141         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
3142         [SVM_EXIT_READ_DR0]                     = dr_interception,
3143         [SVM_EXIT_READ_DR1]                     = dr_interception,
3144         [SVM_EXIT_READ_DR2]                     = dr_interception,
3145         [SVM_EXIT_READ_DR3]                     = dr_interception,
3146         [SVM_EXIT_READ_DR4]                     = dr_interception,
3147         [SVM_EXIT_READ_DR5]                     = dr_interception,
3148         [SVM_EXIT_READ_DR6]                     = dr_interception,
3149         [SVM_EXIT_READ_DR7]                     = dr_interception,
3150         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
3151         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
3152         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
3153         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
3154         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
3155         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
3156         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
3157         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
3158         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
3159         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
3160         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
3161         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
3162         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
3163         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
3164         [SVM_EXIT_INTR]                         = intr_interception,
3165         [SVM_EXIT_NMI]                          = nmi_interception,
3166         [SVM_EXIT_SMI]                          = nop_on_interception,
3167         [SVM_EXIT_INIT]                         = nop_on_interception,
3168         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
3169         [SVM_EXIT_CPUID]                        = cpuid_interception,
3170         [SVM_EXIT_IRET]                         = iret_interception,
3171         [SVM_EXIT_INVD]                         = emulate_on_interception,
3172         [SVM_EXIT_PAUSE]                        = pause_interception,
3173         [SVM_EXIT_HLT]                          = halt_interception,
3174         [SVM_EXIT_INVLPG]                       = invlpg_interception,
3175         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
3176         [SVM_EXIT_IOIO]                         = io_interception,
3177         [SVM_EXIT_MSR]                          = msr_interception,
3178         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
3179         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
3180         [SVM_EXIT_VMRUN]                        = vmrun_interception,
3181         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
3182         [SVM_EXIT_VMLOAD]                       = vmload_interception,
3183         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
3184         [SVM_EXIT_STGI]                         = stgi_interception,
3185         [SVM_EXIT_CLGI]                         = clgi_interception,
3186         [SVM_EXIT_SKINIT]                       = skinit_interception,
3187         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
3188         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
3189         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
3190         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
3191         [SVM_EXIT_NPF]                          = pf_interception,
3192 };
3193
3194 static void dump_vmcb(struct kvm_vcpu *vcpu)
3195 {
3196         struct vcpu_svm *svm = to_svm(vcpu);
3197         struct vmcb_control_area *control = &svm->vmcb->control;
3198         struct vmcb_save_area *save = &svm->vmcb->save;
3199
3200         pr_err("VMCB Control Area:\n");
3201         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3202         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3203         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3204         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3205         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3206         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3207         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3208         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3209         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3210         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3211         pr_err("%-20s%d\n", "asid:", control->asid);
3212         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3213         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3214         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3215         pr_err("%-20s%08x\n", "int_state:", control->int_state);
3216         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3217         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3218         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3219         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3220         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3221         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3222         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3223         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3224         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3225         pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3226         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3227         pr_err("VMCB State Save Area:\n");
3228         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3229                "es:",
3230                save->es.selector, save->es.attrib,
3231                save->es.limit, save->es.base);
3232         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3233                "cs:",
3234                save->cs.selector, save->cs.attrib,
3235                save->cs.limit, save->cs.base);
3236         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3237                "ss:",
3238                save->ss.selector, save->ss.attrib,
3239                save->ss.limit, save->ss.base);
3240         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3241                "ds:",
3242                save->ds.selector, save->ds.attrib,
3243                save->ds.limit, save->ds.base);
3244         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3245                "fs:",
3246                save->fs.selector, save->fs.attrib,
3247                save->fs.limit, save->fs.base);
3248         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3249                "gs:",
3250                save->gs.selector, save->gs.attrib,
3251                save->gs.limit, save->gs.base);
3252         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3253                "gdtr:",
3254                save->gdtr.selector, save->gdtr.attrib,
3255                save->gdtr.limit, save->gdtr.base);
3256         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3257                "ldtr:",
3258                save->ldtr.selector, save->ldtr.attrib,
3259                save->ldtr.limit, save->ldtr.base);
3260         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3261                "idtr:",
3262                save->idtr.selector, save->idtr.attrib,
3263                save->idtr.limit, save->idtr.base);
3264         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3265                "tr:",
3266                save->tr.selector, save->tr.attrib,
3267                save->tr.limit, save->tr.base);
3268         pr_err("cpl:            %d                efer:         %016llx\n",
3269                 save->cpl, save->efer);
3270         pr_err("%-15s %016llx %-13s %016llx\n",
3271                "cr0:", save->cr0, "cr2:", save->cr2);
3272         pr_err("%-15s %016llx %-13s %016llx\n",
3273                "cr3:", save->cr3, "cr4:", save->cr4);
3274         pr_err("%-15s %016llx %-13s %016llx\n",
3275                "dr6:", save->dr6, "dr7:", save->dr7);
3276         pr_err("%-15s %016llx %-13s %016llx\n",
3277                "rip:", save->rip, "rflags:", save->rflags);
3278         pr_err("%-15s %016llx %-13s %016llx\n",
3279                "rsp:", save->rsp, "rax:", save->rax);
3280         pr_err("%-15s %016llx %-13s %016llx\n",
3281                "star:", save->star, "lstar:", save->lstar);
3282         pr_err("%-15s %016llx %-13s %016llx\n",
3283                "cstar:", save->cstar, "sfmask:", save->sfmask);
3284         pr_err("%-15s %016llx %-13s %016llx\n",
3285                "kernel_gs_base:", save->kernel_gs_base,
3286                "sysenter_cs:", save->sysenter_cs);
3287         pr_err("%-15s %016llx %-13s %016llx\n",
3288                "sysenter_esp:", save->sysenter_esp,
3289                "sysenter_eip:", save->sysenter_eip);
3290         pr_err("%-15s %016llx %-13s %016llx\n",
3291                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3292         pr_err("%-15s %016llx %-13s %016llx\n",
3293                "br_from:", save->br_from, "br_to:", save->br_to);
3294         pr_err("%-15s %016llx %-13s %016llx\n",
3295                "excp_from:", save->last_excp_from,
3296                "excp_to:", save->last_excp_to);
3297 }
3298
3299 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3300 {
3301         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3302
3303         *info1 = control->exit_info_1;
3304         *info2 = control->exit_info_2;
3305 }
3306
3307 static int handle_exit(struct kvm_vcpu *vcpu)
3308 {
3309         struct vcpu_svm *svm = to_svm(vcpu);
3310         struct kvm_run *kvm_run = vcpu->run;
3311         u32 exit_code = svm->vmcb->control.exit_code;
3312
3313         trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
3314
3315         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3316                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3317         if (npt_enabled)
3318                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3319
3320         if (unlikely(svm->nested.exit_required)) {
3321                 nested_svm_vmexit(svm);
3322                 svm->nested.exit_required = false;
3323
3324                 return 1;
3325         }
3326
3327         if (is_guest_mode(vcpu)) {
3328                 int vmexit;
3329
3330                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3331                                         svm->vmcb->control.exit_info_1,
3332                                         svm->vmcb->control.exit_info_2,
3333                                         svm->vmcb->control.exit_int_info,
3334                                         svm->vmcb->control.exit_int_info_err);
3335
3336                 vmexit = nested_svm_exit_special(svm);
3337
3338                 if (vmexit == NESTED_EXIT_CONTINUE)
3339                         vmexit = nested_svm_exit_handled(svm);
3340
3341                 if (vmexit == NESTED_EXIT_DONE)
3342                         return 1;
3343         }
3344
3345         svm_complete_interrupts(svm);
3346
3347         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3348                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3349                 kvm_run->fail_entry.hardware_entry_failure_reason
3350                         = svm->vmcb->control.exit_code;
3351                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3352                 dump_vmcb(vcpu);
3353                 return 0;
3354         }
3355
3356         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3357             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3358             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3359             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3360                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
3361                        "exit_code 0x%x\n",
3362                        __func__, svm->vmcb->control.exit_int_info,
3363                        exit_code);
3364
3365         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3366             || !svm_exit_handlers[exit_code]) {
3367                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3368                 kvm_run->hw.hardware_exit_reason = exit_code;
3369                 return 0;
3370         }
3371
3372         return svm_exit_handlers[exit_code](svm);
3373 }
3374
3375 static void reload_tss(struct kvm_vcpu *vcpu)
3376 {
3377         int cpu = raw_smp_processor_id();
3378
3379         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3380         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3381         load_TR_desc();
3382 }
3383
3384 static void pre_svm_run(struct vcpu_svm *svm)
3385 {
3386         int cpu = raw_smp_processor_id();
3387
3388         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3389
3390         /* FIXME: handle wraparound of asid_generation */
3391         if (svm->asid_generation != sd->asid_generation)
3392                 new_asid(svm, sd);
3393 }
3394
3395 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3396 {
3397         struct vcpu_svm *svm = to_svm(vcpu);
3398
3399         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3400         vcpu->arch.hflags |= HF_NMI_MASK;
3401         set_intercept(svm, INTERCEPT_IRET);
3402         ++vcpu->stat.nmi_injections;
3403 }
3404
3405 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3406 {
3407         struct vmcb_control_area *control;
3408
3409         control = &svm->vmcb->control;
3410         control->int_vector = irq;
3411         control->int_ctl &= ~V_INTR_PRIO_MASK;
3412         control->int_ctl |= V_IRQ_MASK |
3413                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3414         mark_dirty(svm->vmcb, VMCB_INTR);
3415 }
3416
3417 static void svm_set_irq(struct kvm_vcpu *vcpu)
3418 {
3419         struct vcpu_svm *svm = to_svm(vcpu);
3420
3421         BUG_ON(!(gif_set(svm)));
3422
3423         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3424         ++vcpu->stat.irq_injections;
3425
3426         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3427                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3428 }
3429
3430 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3431 {
3432         struct vcpu_svm *svm = to_svm(vcpu);
3433
3434         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3435                 return;
3436
3437         if (irr == -1)
3438                 return;
3439
3440         if (tpr >= irr)
3441                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3442 }
3443
3444 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3445 {
3446         struct vcpu_svm *svm = to_svm(vcpu);
3447         struct vmcb *vmcb = svm->vmcb;
3448         int ret;
3449         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3450               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3451         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3452
3453         return ret;
3454 }
3455
3456 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3457 {
3458         struct vcpu_svm *svm = to_svm(vcpu);
3459
3460         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3461 }
3462
3463 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3464 {
3465         struct vcpu_svm *svm = to_svm(vcpu);
3466
3467         if (masked) {
3468                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3469                 set_intercept(svm, INTERCEPT_IRET);
3470         } else {
3471                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3472                 clr_intercept(svm, INTERCEPT_IRET);
3473         }
3474 }
3475
3476 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3477 {
3478         struct vcpu_svm *svm = to_svm(vcpu);
3479         struct vmcb *vmcb = svm->vmcb;
3480         int ret;
3481
3482         if (!gif_set(svm) ||
3483              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3484                 return 0;
3485
3486         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3487
3488         if (is_guest_mode(vcpu))
3489                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3490
3491         return ret;
3492 }
3493
3494 static void enable_irq_window(struct kvm_vcpu *vcpu)
3495 {
3496         struct vcpu_svm *svm = to_svm(vcpu);
3497
3498         /*
3499          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3500          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3501          * get that intercept, this function will be called again though and
3502          * we'll get the vintr intercept.
3503          */
3504         if (gif_set(svm) && nested_svm_intr(svm)) {
3505                 svm_set_vintr(svm);
3506                 svm_inject_irq(svm, 0x0);
3507         }
3508 }
3509
3510 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3511 {
3512         struct vcpu_svm *svm = to_svm(vcpu);
3513
3514         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3515             == HF_NMI_MASK)
3516                 return; /* IRET will cause a vm exit */
3517
3518         /*
3519          * Something prevents NMI from been injected. Single step over possible
3520          * problem (IRET or exception injection or interrupt shadow)
3521          */
3522         svm->nmi_singlestep = true;
3523         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3524         update_db_intercept(vcpu);
3525 }
3526
3527 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3528 {
3529         return 0;
3530 }
3531
3532 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3533 {
3534         struct vcpu_svm *svm = to_svm(vcpu);
3535
3536         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3537                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3538         else
3539                 svm->asid_generation--;
3540 }
3541
3542 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3543 {
3544 }
3545
3546 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3547 {
3548         struct vcpu_svm *svm = to_svm(vcpu);
3549
3550         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3551                 return;
3552
3553         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3554                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3555                 kvm_set_cr8(vcpu, cr8);
3556         }
3557 }
3558
3559 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3560 {
3561         struct vcpu_svm *svm = to_svm(vcpu);
3562         u64 cr8;
3563
3564         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3565                 return;
3566
3567         cr8 = kvm_get_cr8(vcpu);
3568         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3569         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3570 }
3571
3572 static void svm_complete_interrupts(struct vcpu_svm *svm)
3573 {
3574         u8 vector;
3575         int type;
3576         u32 exitintinfo = svm->vmcb->control.exit_int_info;
3577         unsigned int3_injected = svm->int3_injected;
3578
3579         svm->int3_injected = 0;
3580
3581         /*
3582          * If we've made progress since setting HF_IRET_MASK, we've
3583          * executed an IRET and can allow NMI injection.
3584          */
3585         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3586             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3587                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3588                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3589         }
3590
3591         svm->vcpu.arch.nmi_injected = false;
3592         kvm_clear_exception_queue(&svm->vcpu);
3593         kvm_clear_interrupt_queue(&svm->vcpu);
3594
3595         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3596                 return;
3597
3598         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3599
3600         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3601         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3602
3603         switch (type) {
3604         case SVM_EXITINTINFO_TYPE_NMI:
3605                 svm->vcpu.arch.nmi_injected = true;
3606                 break;
3607         case SVM_EXITINTINFO_TYPE_EXEPT:
3608                 /*
3609                  * In case of software exceptions, do not reinject the vector,
3610                  * but re-execute the instruction instead. Rewind RIP first
3611                  * if we emulated INT3 before.
3612                  */
3613                 if (kvm_exception_is_soft(vector)) {
3614                         if (vector == BP_VECTOR && int3_injected &&
3615                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3616                                 kvm_rip_write(&svm->vcpu,
3617                                               kvm_rip_read(&svm->vcpu) -
3618                                               int3_injected);
3619                         break;
3620                 }
3621                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3622                         u32 err = svm->vmcb->control.exit_int_info_err;
3623                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
3624
3625                 } else
3626                         kvm_requeue_exception(&svm->vcpu, vector);
3627                 break;
3628         case SVM_EXITINTINFO_TYPE_INTR:
3629                 kvm_queue_interrupt(&svm->vcpu, vector, false);
3630                 break;
3631         default:
3632                 break;
3633         }
3634 }
3635
3636 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3637 {
3638         struct vcpu_svm *svm = to_svm(vcpu);
3639         struct vmcb_control_area *control = &svm->vmcb->control;
3640
3641         control->exit_int_info = control->event_inj;
3642         control->exit_int_info_err = control->event_inj_err;
3643         control->event_inj = 0;
3644         svm_complete_interrupts(svm);
3645 }
3646
3647 #ifdef CONFIG_X86_64
3648 #define R "r"
3649 #else
3650 #define R "e"
3651 #endif
3652
3653 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3654 {
3655         struct vcpu_svm *svm = to_svm(vcpu);
3656
3657         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3658         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3659         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3660
3661         /*
3662          * A vmexit emulation is required before the vcpu can be executed
3663          * again.
3664          */
3665         if (unlikely(svm->nested.exit_required))
3666                 return;
3667
3668         pre_svm_run(svm);
3669
3670         sync_lapic_to_cr8(vcpu);
3671
3672         svm->vmcb->save.cr2 = vcpu->arch.cr2;
3673
3674         clgi();
3675
3676         local_irq_enable();
3677
3678         asm volatile (
3679                 "push %%"R"bp; \n\t"
3680                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
3681                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
3682                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
3683                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
3684                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
3685                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
3686 #ifdef CONFIG_X86_64
3687                 "mov %c[r8](%[svm]),  %%r8  \n\t"
3688                 "mov %c[r9](%[svm]),  %%r9  \n\t"
3689                 "mov %c[r10](%[svm]), %%r10 \n\t"
3690                 "mov %c[r11](%[svm]), %%r11 \n\t"
3691                 "mov %c[r12](%[svm]), %%r12 \n\t"
3692                 "mov %c[r13](%[svm]), %%r13 \n\t"
3693                 "mov %c[r14](%[svm]), %%r14 \n\t"
3694                 "mov %c[r15](%[svm]), %%r15 \n\t"
3695 #endif
3696
3697                 /* Enter guest mode */
3698                 "push %%"R"ax \n\t"
3699                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
3700                 __ex(SVM_VMLOAD) "\n\t"
3701                 __ex(SVM_VMRUN) "\n\t"
3702                 __ex(SVM_VMSAVE) "\n\t"
3703                 "pop %%"R"ax \n\t"
3704
3705                 /* Save guest registers, load host registers */
3706                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
3707                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
3708                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
3709                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
3710                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
3711                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
3712 #ifdef CONFIG_X86_64
3713                 "mov %%r8,  %c[r8](%[svm]) \n\t"
3714                 "mov %%r9,  %c[r9](%[svm]) \n\t"
3715                 "mov %%r10, %c[r10](%[svm]) \n\t"
3716                 "mov %%r11, %c[r11](%[svm]) \n\t"
3717                 "mov %%r12, %c[r12](%[svm]) \n\t"
3718                 "mov %%r13, %c[r13](%[svm]) \n\t"
3719                 "mov %%r14, %c[r14](%[svm]) \n\t"
3720                 "mov %%r15, %c[r15](%[svm]) \n\t"
3721 #endif
3722                 "pop %%"R"bp"
3723                 :
3724                 : [svm]"a"(svm),
3725                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3726                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3727                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3728                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3729                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3730                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3731                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3732 #ifdef CONFIG_X86_64
3733                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3734                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3735                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3736                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3737                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3738                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3739                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3740                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3741 #endif
3742                 : "cc", "memory"
3743                 , R"bx", R"cx", R"dx", R"si", R"di"
3744 #ifdef CONFIG_X86_64
3745                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3746 #endif
3747                 );
3748
3749 #ifdef CONFIG_X86_64
3750         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3751 #else
3752         loadsegment(fs, svm->host.fs);
3753 #ifndef CONFIG_X86_32_LAZY_GS
3754         loadsegment(gs, svm->host.gs);
3755 #endif
3756 #endif
3757
3758         reload_tss(vcpu);
3759
3760         local_irq_disable();
3761
3762         vcpu->arch.cr2 = svm->vmcb->save.cr2;
3763         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3764         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3765         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3766
3767         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3768                 kvm_before_handle_nmi(&svm->vcpu);
3769
3770         stgi();
3771
3772         /* Any pending NMI will happen here */
3773
3774         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3775                 kvm_after_handle_nmi(&svm->vcpu);
3776
3777         sync_cr8_to_lapic(vcpu);
3778
3779         svm->next_rip = 0;
3780
3781         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3782
3783         /* if exit due to PF check for async PF */
3784         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3785                 svm->apf_reason = kvm_read_and_reset_pf_reason();
3786
3787         if (npt_enabled) {
3788                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3789                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3790         }
3791
3792         /*
3793          * We need to handle MC intercepts here before the vcpu has a chance to
3794          * change the physical cpu
3795          */
3796         if (unlikely(svm->vmcb->control.exit_code ==
3797                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
3798                 svm_handle_mce(svm);
3799
3800         mark_all_clean(svm->vmcb);
3801 }
3802
3803 #undef R
3804
3805 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3806 {
3807         struct vcpu_svm *svm = to_svm(vcpu);
3808
3809         svm->vmcb->save.cr3 = root;
3810         mark_dirty(svm->vmcb, VMCB_CR);
3811         svm_flush_tlb(vcpu);
3812 }
3813
3814 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3815 {
3816         struct vcpu_svm *svm = to_svm(vcpu);
3817
3818         svm->vmcb->control.nested_cr3 = root;
3819         mark_dirty(svm->vmcb, VMCB_NPT);
3820
3821         /* Also sync guest cr3 here in case we live migrate */
3822         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
3823         mark_dirty(svm->vmcb, VMCB_CR);
3824
3825         svm_flush_tlb(vcpu);
3826 }
3827
3828 static int is_disabled(void)
3829 {
3830         u64 vm_cr;
3831
3832         rdmsrl(MSR_VM_CR, vm_cr);
3833         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3834                 return 1;
3835
3836         return 0;
3837 }
3838
3839 static void
3840 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3841 {
3842         /*
3843          * Patch in the VMMCALL instruction:
3844          */
3845         hypercall[0] = 0x0f;
3846         hypercall[1] = 0x01;
3847         hypercall[2] = 0xd9;
3848 }
3849
3850 static void svm_check_processor_compat(void *rtn)
3851 {
3852         *(int *)rtn = 0;
3853 }
3854
3855 static bool svm_cpu_has_accelerated_tpr(void)
3856 {
3857         return false;
3858 }
3859
3860 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3861 {
3862         return 0;
3863 }
3864
3865 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3866 {
3867 }
3868
3869 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
3870 {
3871         switch (func) {
3872         case 0x80000001:
3873                 if (nested)
3874                         entry->ecx |= (1 << 2); /* Set SVM bit */
3875                 break;
3876         case 0x8000000A:
3877                 entry->eax = 1; /* SVM revision 1 */
3878                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
3879                                    ASID emulation to nested SVM */
3880                 entry->ecx = 0; /* Reserved */
3881                 entry->edx = 0; /* Per default do not support any
3882                                    additional features */
3883
3884                 /* Support next_rip if host supports it */
3885                 if (boot_cpu_has(X86_FEATURE_NRIPS))
3886                         entry->edx |= SVM_FEATURE_NRIP;
3887
3888                 /* Support NPT for the guest if enabled */
3889                 if (npt_enabled)
3890                         entry->edx |= SVM_FEATURE_NPT;
3891
3892                 break;
3893         }
3894 }
3895
3896 static const struct trace_print_flags svm_exit_reasons_str[] = {
3897         { SVM_EXIT_READ_CR0,                    "read_cr0" },
3898         { SVM_EXIT_READ_CR3,                    "read_cr3" },
3899         { SVM_EXIT_READ_CR4,                    "read_cr4" },
3900         { SVM_EXIT_READ_CR8,                    "read_cr8" },
3901         { SVM_EXIT_WRITE_CR0,                   "write_cr0" },
3902         { SVM_EXIT_WRITE_CR3,                   "write_cr3" },
3903         { SVM_EXIT_WRITE_CR4,                   "write_cr4" },
3904         { SVM_EXIT_WRITE_CR8,                   "write_cr8" },
3905         { SVM_EXIT_READ_DR0,                    "read_dr0" },
3906         { SVM_EXIT_READ_DR1,                    "read_dr1" },
3907         { SVM_EXIT_READ_DR2,                    "read_dr2" },
3908         { SVM_EXIT_READ_DR3,                    "read_dr3" },
3909         { SVM_EXIT_WRITE_DR0,                   "write_dr0" },
3910         { SVM_EXIT_WRITE_DR1,                   "write_dr1" },
3911         { SVM_EXIT_WRITE_DR2,                   "write_dr2" },
3912         { SVM_EXIT_WRITE_DR3,                   "write_dr3" },
3913         { SVM_EXIT_WRITE_DR5,                   "write_dr5" },
3914         { SVM_EXIT_WRITE_DR7,                   "write_dr7" },
3915         { SVM_EXIT_EXCP_BASE + DB_VECTOR,       "DB excp" },
3916         { SVM_EXIT_EXCP_BASE + BP_VECTOR,       "BP excp" },
3917         { SVM_EXIT_EXCP_BASE + UD_VECTOR,       "UD excp" },
3918         { SVM_EXIT_EXCP_BASE + PF_VECTOR,       "PF excp" },
3919         { SVM_EXIT_EXCP_BASE + NM_VECTOR,       "NM excp" },
3920         { SVM_EXIT_EXCP_BASE + MC_VECTOR,       "MC excp" },
3921         { SVM_EXIT_INTR,                        "interrupt" },
3922         { SVM_EXIT_NMI,                         "nmi" },
3923         { SVM_EXIT_SMI,                         "smi" },
3924         { SVM_EXIT_INIT,                        "init" },
3925         { SVM_EXIT_VINTR,                       "vintr" },
3926         { SVM_EXIT_CPUID,                       "cpuid" },
3927         { SVM_EXIT_INVD,                        "invd" },
3928         { SVM_EXIT_HLT,                         "hlt" },
3929         { SVM_EXIT_INVLPG,                      "invlpg" },
3930         { SVM_EXIT_INVLPGA,                     "invlpga" },
3931         { SVM_EXIT_IOIO,                        "io" },
3932         { SVM_EXIT_MSR,                         "msr" },
3933         { SVM_EXIT_TASK_SWITCH,                 "task_switch" },
3934         { SVM_EXIT_SHUTDOWN,                    "shutdown" },
3935         { SVM_EXIT_VMRUN,                       "vmrun" },
3936         { SVM_EXIT_VMMCALL,                     "hypercall" },
3937         { SVM_EXIT_VMLOAD,                      "vmload" },
3938         { SVM_EXIT_VMSAVE,                      "vmsave" },
3939         { SVM_EXIT_STGI,                        "stgi" },
3940         { SVM_EXIT_CLGI,                        "clgi" },
3941         { SVM_EXIT_SKINIT,                      "skinit" },
3942         { SVM_EXIT_WBINVD,                      "wbinvd" },
3943         { SVM_EXIT_MONITOR,                     "monitor" },
3944         { SVM_EXIT_MWAIT,                       "mwait" },
3945         { SVM_EXIT_XSETBV,                      "xsetbv" },
3946         { SVM_EXIT_NPF,                         "npf" },
3947         { -1, NULL }
3948 };
3949
3950 static int svm_get_lpage_level(void)
3951 {
3952         return PT_PDPE_LEVEL;
3953 }
3954
3955 static bool svm_rdtscp_supported(void)
3956 {
3957         return false;
3958 }
3959
3960 static bool svm_has_wbinvd_exit(void)
3961 {
3962         return true;
3963 }
3964
3965 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3966 {
3967         struct vcpu_svm *svm = to_svm(vcpu);
3968
3969         set_exception_intercept(svm, NM_VECTOR);
3970         update_cr0_intercept(svm);
3971 }
3972
3973 #define PRE_EX(exit)  { .exit_code = (exit), \
3974                         .stage = X86_ICPT_PRE_EXCEPT, }
3975 #define POST_EX(exit) { .exit_code = (exit), \
3976                         .stage = X86_ICPT_POST_EXCEPT, }
3977 #define POST_MEM(exit) { .exit_code = (exit), \
3978                         .stage = X86_ICPT_POST_MEMACCESS, }
3979
3980 static struct __x86_intercept {
3981         u32 exit_code;
3982         enum x86_intercept_stage stage;
3983 } x86_intercept_map[] = {
3984         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
3985         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
3986         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
3987         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
3988         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
3989         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
3990         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
3991         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
3992         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
3993         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
3994         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
3995         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
3996         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
3997         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
3998         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
3999         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
4000         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
4001         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
4002         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
4003         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
4004         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
4005         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
4006         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
4007         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
4008         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
4009         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
4010         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
4011         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
4012         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
4013         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
4014         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
4015         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
4016         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
4017         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
4018         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
4019         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
4020         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
4021         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
4022         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
4023         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
4024         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
4025         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
4026         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
4027         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
4028         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
4029         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
4030 };
4031
4032 #undef PRE_EX
4033 #undef POST_EX
4034 #undef POST_MEM
4035
4036 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4037                                struct x86_instruction_info *info,
4038                                enum x86_intercept_stage stage)
4039 {
4040         struct vcpu_svm *svm = to_svm(vcpu);
4041         int vmexit, ret = X86EMUL_CONTINUE;
4042         struct __x86_intercept icpt_info;
4043         struct vmcb *vmcb = svm->vmcb;
4044
4045         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4046                 goto out;
4047
4048         icpt_info = x86_intercept_map[info->intercept];
4049
4050         if (stage != icpt_info.stage)
4051                 goto out;
4052
4053         switch (icpt_info.exit_code) {
4054         case SVM_EXIT_READ_CR0:
4055                 if (info->intercept == x86_intercept_cr_read)
4056                         icpt_info.exit_code += info->modrm_reg;
4057                 break;
4058         case SVM_EXIT_WRITE_CR0: {
4059                 unsigned long cr0, val;
4060                 u64 intercept;
4061
4062                 if (info->intercept == x86_intercept_cr_write)
4063                         icpt_info.exit_code += info->modrm_reg;
4064
4065                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0)
4066                         break;
4067
4068                 intercept = svm->nested.intercept;
4069
4070                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4071                         break;
4072
4073                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4074                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4075
4076                 if (info->intercept == x86_intercept_lmsw) {
4077                         cr0 &= 0xfUL;
4078                         val &= 0xfUL;
4079                         /* lmsw can't clear PE - catch this here */
4080                         if (cr0 & X86_CR0_PE)
4081                                 val |= X86_CR0_PE;
4082                 }
4083
4084                 if (cr0 ^ val)
4085                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4086
4087                 break;
4088         }
4089         case SVM_EXIT_READ_DR0:
4090         case SVM_EXIT_WRITE_DR0:
4091                 icpt_info.exit_code += info->modrm_reg;
4092                 break;
4093         case SVM_EXIT_MSR:
4094                 if (info->intercept == x86_intercept_wrmsr)
4095                         vmcb->control.exit_info_1 = 1;
4096                 else
4097                         vmcb->control.exit_info_1 = 0;
4098                 break;
4099         case SVM_EXIT_PAUSE:
4100                 /*
4101                  * We get this for NOP only, but pause
4102                  * is rep not, check this here
4103                  */
4104                 if (info->rep_prefix != REPE_PREFIX)
4105                         goto out;
4106         case SVM_EXIT_IOIO: {
4107                 u64 exit_info;
4108                 u32 bytes;
4109
4110                 exit_info = (vcpu->arch.regs[VCPU_REGS_RDX] & 0xffff) << 16;
4111
4112                 if (info->intercept == x86_intercept_in ||
4113                     info->intercept == x86_intercept_ins) {
4114                         exit_info |= SVM_IOIO_TYPE_MASK;
4115                         bytes = info->src_bytes;
4116                 } else {
4117                         bytes = info->dst_bytes;
4118                 }
4119
4120                 if (info->intercept == x86_intercept_outs ||
4121                     info->intercept == x86_intercept_ins)
4122                         exit_info |= SVM_IOIO_STR_MASK;
4123
4124                 if (info->rep_prefix)
4125                         exit_info |= SVM_IOIO_REP_MASK;
4126
4127                 bytes = min(bytes, 4u);
4128
4129                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4130
4131                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4132
4133                 vmcb->control.exit_info_1 = exit_info;
4134                 vmcb->control.exit_info_2 = info->next_rip;
4135
4136                 break;
4137         }
4138         default:
4139                 break;
4140         }
4141
4142         vmcb->control.next_rip  = info->next_rip;
4143         vmcb->control.exit_code = icpt_info.exit_code;
4144         vmexit = nested_svm_exit_handled(svm);
4145
4146         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4147                                            : X86EMUL_CONTINUE;
4148
4149 out:
4150         return ret;
4151 }
4152
4153 static struct kvm_x86_ops svm_x86_ops = {
4154         .cpu_has_kvm_support = has_svm,
4155         .disabled_by_bios = is_disabled,
4156         .hardware_setup = svm_hardware_setup,
4157         .hardware_unsetup = svm_hardware_unsetup,
4158         .check_processor_compatibility = svm_check_processor_compat,
4159         .hardware_enable = svm_hardware_enable,
4160         .hardware_disable = svm_hardware_disable,
4161         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4162
4163         .vcpu_create = svm_create_vcpu,
4164         .vcpu_free = svm_free_vcpu,
4165         .vcpu_reset = svm_vcpu_reset,
4166
4167         .prepare_guest_switch = svm_prepare_guest_switch,
4168         .vcpu_load = svm_vcpu_load,
4169         .vcpu_put = svm_vcpu_put,
4170
4171         .set_guest_debug = svm_guest_debug,
4172         .get_msr = svm_get_msr,
4173         .set_msr = svm_set_msr,
4174         .get_segment_base = svm_get_segment_base,
4175         .get_segment = svm_get_segment,
4176         .set_segment = svm_set_segment,
4177         .get_cpl = svm_get_cpl,
4178         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4179         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4180         .decache_cr3 = svm_decache_cr3,
4181         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4182         .set_cr0 = svm_set_cr0,
4183         .set_cr3 = svm_set_cr3,
4184         .set_cr4 = svm_set_cr4,
4185         .set_efer = svm_set_efer,
4186         .get_idt = svm_get_idt,
4187         .set_idt = svm_set_idt,
4188         .get_gdt = svm_get_gdt,
4189         .set_gdt = svm_set_gdt,
4190         .set_dr7 = svm_set_dr7,
4191         .cache_reg = svm_cache_reg,
4192         .get_rflags = svm_get_rflags,
4193         .set_rflags = svm_set_rflags,
4194         .fpu_activate = svm_fpu_activate,
4195         .fpu_deactivate = svm_fpu_deactivate,
4196
4197         .tlb_flush = svm_flush_tlb,
4198
4199         .run = svm_vcpu_run,
4200         .handle_exit = handle_exit,
4201         .skip_emulated_instruction = skip_emulated_instruction,
4202         .set_interrupt_shadow = svm_set_interrupt_shadow,
4203         .get_interrupt_shadow = svm_get_interrupt_shadow,
4204         .patch_hypercall = svm_patch_hypercall,
4205         .set_irq = svm_set_irq,
4206         .set_nmi = svm_inject_nmi,
4207         .queue_exception = svm_queue_exception,
4208         .cancel_injection = svm_cancel_injection,
4209         .interrupt_allowed = svm_interrupt_allowed,
4210         .nmi_allowed = svm_nmi_allowed,
4211         .get_nmi_mask = svm_get_nmi_mask,
4212         .set_nmi_mask = svm_set_nmi_mask,
4213         .enable_nmi_window = enable_nmi_window,
4214         .enable_irq_window = enable_irq_window,
4215         .update_cr8_intercept = update_cr8_intercept,
4216
4217         .set_tss_addr = svm_set_tss_addr,
4218         .get_tdp_level = get_npt_level,
4219         .get_mt_mask = svm_get_mt_mask,
4220
4221         .get_exit_info = svm_get_exit_info,
4222         .exit_reasons_str = svm_exit_reasons_str,
4223
4224         .get_lpage_level = svm_get_lpage_level,
4225
4226         .cpuid_update = svm_cpuid_update,
4227
4228         .rdtscp_supported = svm_rdtscp_supported,
4229
4230         .set_supported_cpuid = svm_set_supported_cpuid,
4231
4232         .has_wbinvd_exit = svm_has_wbinvd_exit,
4233
4234         .set_tsc_khz = svm_set_tsc_khz,
4235         .write_tsc_offset = svm_write_tsc_offset,
4236         .adjust_tsc_offset = svm_adjust_tsc_offset,
4237         .compute_tsc_offset = svm_compute_tsc_offset,
4238
4239         .set_tdp_cr3 = set_tdp_cr3,
4240
4241         .check_intercept = svm_check_intercept,
4242 };
4243
4244 static int __init svm_init(void)
4245 {
4246         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4247                         __alignof__(struct vcpu_svm), THIS_MODULE);
4248 }
4249
4250 static void __exit svm_exit(void)
4251 {
4252         kvm_exit();
4253 }
4254
4255 module_init(svm_init)
4256 module_exit(svm_exit)