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  <[email protected]>
  11  *   Avi Kivity   <[email protected]>
  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 #include "cpuid.h"
  24
  25 #include <linux/module.h>
  26 #include <linux/mod_devicetable.h>
  27 #include <linux/kernel.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/highmem.h>
  30 #include <linux/sched.h>
  31 #include <linux/ftrace_event.h>
  32 #include <linux/slab.h>
  33
  34 #include <asm/perf_event.h>
  35 #include <asm/tlbflush.h>
  36 #include <asm/desc.h>
  37 #include <asm/kvm_para.h>
  38
  39 #include <asm/virtext.h>
  40 #include "trace.h"
  41
  42 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  43
  44 MODULE_AUTHOR("Qumranet");
  45 MODULE_LICENSE("GPL");
  46
  47 static const struct x86_cpu_id svm_cpu_id[] = {
  48         X86_FEATURE_MATCH(X86_FEATURE_SVM),
  49         {}
  50 };
  51 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  52
  53 #define IOPM_ALLOC_ORDER 2
  54 #define MSRPM_ALLOC_ORDER 1
  55
  56 #define SEG_TYPE_LDT 2
  57 #define SEG_TYPE_BUSY_TSS16 3
  58
  59 #define SVM_FEATURE_NPT            (1 <<  0)
  60 #define SVM_FEATURE_LBRV           (1 <<  1)
  61 #define SVM_FEATURE_SVML           (1 <<  2)
  62 #define SVM_FEATURE_NRIP           (1 <<  3)
  63 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  64 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  65 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  66 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  67 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  68
  69 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  70 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  71 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  72
  73 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  74
  75 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  76 #define TSC_RATIO_MIN           0x0000000000000001ULL
  77 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  78
  79 static bool erratum_383_found __read_mostly;
  80
  81 static const u32 host_save_user_msrs[] = {
  82 #ifdef CONFIG_X86_64
  83         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  84         MSR_FS_BASE,
  85 #endif
  86         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  87 };
  88
  89 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  90
  91 struct kvm_vcpu;
  92
  93 struct nested_state {
  94         struct vmcb *hsave;
  95         u64 hsave_msr;
  96         u64 vm_cr_msr;
  97         u64 vmcb;
  98
  99         /* These are the merged vectors */
 100         u32 *msrpm;
 101
 102         /* gpa pointers to the real vectors */
 103         u64 vmcb_msrpm;
 104         u64 vmcb_iopm;
 105
 106         /* A VMEXIT is required but not yet emulated */
 107         bool exit_required;
 108
 109         /* cache for intercepts of the guest */
 110         u32 intercept_cr;
 111         u32 intercept_dr;
 112         u32 intercept_exceptions;
 113         u64 intercept;
 114
 115         /* Nested Paging related state */
 116         u64 nested_cr3;
 117 };
 118
 119 #define MSRPM_OFFSETS   16
 120 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 121
 122 /*
 123  * Set osvw_len to higher value when updated Revision Guides
 124  * are published and we know what the new status bits are
 125  */
 126 static uint64_t osvw_len = 4, osvw_status;
 127
 128 struct vcpu_svm {
 129         struct kvm_vcpu vcpu;
 130         struct vmcb *vmcb;
 131         unsigned long vmcb_pa;
 132         struct svm_cpu_data *svm_data;
 133         uint64_t asid_generation;
 134         uint64_t sysenter_esp;
 135         uint64_t sysenter_eip;
 136
 137         u64 next_rip;
 138
 139         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 140         struct {
 141                 u16 fs;
 142                 u16 gs;
 143                 u16 ldt;
 144                 u64 gs_base;
 145         } host;
 146
 147         u32 *msrpm;
 148
 149         ulong nmi_iret_rip;
 150
 151         struct nested_state nested;
 152
 153         bool nmi_singlestep;
 154
 155         unsigned int3_injected;
 156         unsigned long int3_rip;
 157         u32 apf_reason;
 158
 159         u64  tsc_ratio;
 160 };
 161
 162 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 163 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 164
 165 #define MSR_INVALID                     0xffffffffU
 166
 167 static const struct svm_direct_access_msrs {
 168         u32 index;   /* Index of the MSR */
 169         bool always; /* True if intercept is always on */
 170 } direct_access_msrs[] = {
 171         { .index = MSR_STAR,                            .always = true  },
 172         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 173 #ifdef CONFIG_X86_64
 174         { .index = MSR_GS_BASE,                         .always = true  },
 175         { .index = MSR_FS_BASE,                         .always = true  },
 176         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 177         { .index = MSR_LSTAR,                           .always = true  },
 178         { .index = MSR_CSTAR,                           .always = true  },
 179         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 180 #endif
 181         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 182         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 183         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 184         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 185         { .index = MSR_INVALID,                         .always = false },
 186 };
 187
 188 /* enable NPT for AMD64 and X86 with PAE */
 189 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 190 static bool npt_enabled = true;
 191 #else
 192 static bool npt_enabled;
 193 #endif
 194
 195 /* allow nested paging (virtualized MMU) for all guests */
 196 static int npt = true;
 197 module_param(npt, int, S_IRUGO);
 198
 199 /* allow nested virtualization in KVM/SVM */
 200 static int nested = true;
 201 module_param(nested, int, S_IRUGO);
 202
 203 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 204 static void svm_complete_interrupts(struct vcpu_svm *svm);
 205
 206 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 207 static int nested_svm_intercept(struct vcpu_svm *svm);
 208 static int nested_svm_vmexit(struct vcpu_svm *svm);
 209 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 210                                       bool has_error_code, u32 error_code);
 211 static u64 __scale_tsc(u64 ratio, u64 tsc);
 212
 213 enum {
 214         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 215                             pause filter count */
 216         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 217         VMCB_ASID,       /* ASID */
 218         VMCB_INTR,       /* int_ctl, int_vector */
 219         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 220         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 221         VMCB_DR,         /* DR6, DR7 */
 222         VMCB_DT,         /* GDT, IDT */
 223         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 224         VMCB_CR2,        /* CR2 only */
 225         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 226         VMCB_DIRTY_MAX,
 227 };
 228
 229 /* TPR and CR2 are always written before VMRUN */
 230 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 231
 232 static inline void mark_all_dirty(struct vmcb *vmcb)
 233 {
 234         vmcb->control.clean = 0;
 235 }
 236
 237 static inline void mark_all_clean(struct vmcb *vmcb)
 238 {
 239         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 240                                & ~VMCB_ALWAYS_DIRTY_MASK;
 241 }
 242
 243 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 244 {
 245         vmcb->control.clean &= ~(1 << bit);
 246 }
 247
 248 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 249 {
 250         return container_of(vcpu, struct vcpu_svm, vcpu);
 251 }
 252
 253 static void recalc_intercepts(struct vcpu_svm *svm)
 254 {
 255         struct vmcb_control_area *c, *h;
 256         struct nested_state *g;
 257
 258         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 259
 260         if (!is_guest_mode(&svm->vcpu))
 261                 return;
 262
 263         c = &svm->vmcb->control;
 264         h = &svm->nested.hsave->control;
 265         g = &svm->nested;
 266
 267         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 268         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 269         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 270         c->intercept = h->intercept | g->intercept;
 271 }
 272
 273 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 274 {
 275         if (is_guest_mode(&svm->vcpu))
 276                 return svm->nested.hsave;
 277         else
 278                 return svm->vmcb;
 279 }
 280
 281 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 282 {
 283         struct vmcb *vmcb = get_host_vmcb(svm);
 284
 285         vmcb->control.intercept_cr |= (1U << bit);
 286
 287         recalc_intercepts(svm);
 288 }
 289
 290 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 291 {
 292         struct vmcb *vmcb = get_host_vmcb(svm);
 293
 294         vmcb->control.intercept_cr &= ~(1U << bit);
 295
 296         recalc_intercepts(svm);
 297 }
 298
 299 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 300 {
 301         struct vmcb *vmcb = get_host_vmcb(svm);
 302
 303         return vmcb->control.intercept_cr & (1U << bit);
 304 }
 305
 306 static inline void set_dr_intercept(struct vcpu_svm *svm, int bit)
 307 {
 308         struct vmcb *vmcb = get_host_vmcb(svm);
 309
 310         vmcb->control.intercept_dr |= (1U << bit);
 311
 312         recalc_intercepts(svm);
 313 }
 314
 315 static inline void clr_dr_intercept(struct vcpu_svm *svm, int bit)
 316 {
 317         struct vmcb *vmcb = get_host_vmcb(svm);
 318
 319         vmcb->control.intercept_dr &= ~(1U << bit);
 320
 321         recalc_intercepts(svm);
 322 }
 323
 324 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 325 {
 326         struct vmcb *vmcb = get_host_vmcb(svm);
 327
 328         vmcb->control.intercept_exceptions |= (1U << bit);
 329
 330         recalc_intercepts(svm);
 331 }
 332
 333 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 334 {
 335         struct vmcb *vmcb = get_host_vmcb(svm);
 336
 337         vmcb->control.intercept_exceptions &= ~(1U << bit);
 338
 339         recalc_intercepts(svm);
 340 }
 341
 342 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 343 {
 344         struct vmcb *vmcb = get_host_vmcb(svm);
 345
 346         vmcb->control.intercept |= (1ULL << bit);
 347
 348         recalc_intercepts(svm);
 349 }
 350
 351 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 352 {
 353         struct vmcb *vmcb = get_host_vmcb(svm);
 354
 355         vmcb->control.intercept &= ~(1ULL << bit);
 356
 357         recalc_intercepts(svm);
 358 }
 359
 360 static inline void enable_gif(struct vcpu_svm *svm)
 361 {
 362         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 363 }
 364
 365 static inline void disable_gif(struct vcpu_svm *svm)
 366 {
 367         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 368 }
 369
 370 static inline bool gif_set(struct vcpu_svm *svm)
 371 {
 372         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 373 }
 374
 375 static unsigned long iopm_base;
 376
 377 struct kvm_ldttss_desc {
 378         u16 limit0;
 379         u16 base0;
 380         unsigned base1:8, type:5, dpl:2, p:1;
 381         unsigned limit1:4, zero0:3, g:1, base2:8;
 382         u32 base3;
 383         u32 zero1;
 384 } __attribute__((packed));
 385
 386 struct svm_cpu_data {
 387         int cpu;
 388
 389         u64 asid_generation;
 390         u32 max_asid;
 391         u32 next_asid;
 392         struct kvm_ldttss_desc *tss_desc;
 393
 394         struct page *save_area;
 395 };
 396
 397 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 398
 399 struct svm_init_data {
 400         int cpu;
 401         int r;
 402 };
 403
 404 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 405
 406 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 407 #define MSRS_RANGE_SIZE 2048
 408 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 409
 410 static u32 svm_msrpm_offset(u32 msr)
 411 {
 412         u32 offset;
 413         int i;
 414
 415         for (i = 0; i < NUM_MSR_MAPS; i++) {
 416                 if (msr < msrpm_ranges[i] ||
 417                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 418                         continue;
 419
 420                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 421                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 422
 423                 /* Now we have the u8 offset - but need the u32 offset */
 424                 return offset / 4;
 425         }
 426
 427         /* MSR not in any range */
 428         return MSR_INVALID;
 429 }
 430
 431 #define MAX_INST_SIZE 15
 432
 433 static inline void clgi(void)
 434 {
 435         asm volatile (__ex(SVM_CLGI));
 436 }
 437
 438 static inline void stgi(void)
 439 {
 440         asm volatile (__ex(SVM_STGI));
 441 }
 442
 443 static inline void invlpga(unsigned long addr, u32 asid)
 444 {
 445         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 446 }
 447
 448 static int get_npt_level(void)
 449 {
 450 #ifdef CONFIG_X86_64
 451         return PT64_ROOT_LEVEL;
 452 #else
 453         return PT32E_ROOT_LEVEL;
 454 #endif
 455 }
 456
 457 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 458 {
 459         vcpu->arch.efer = efer;
 460         if (!npt_enabled && !(efer & EFER_LMA))
 461                 efer &= ~EFER_LME;
 462
 463         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 464         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 465 }
 466
 467 static int is_external_interrupt(u32 info)
 468 {
 469         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 470         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 471 }
 472
 473 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 474 {
 475         struct vcpu_svm *svm = to_svm(vcpu);
 476         u32 ret = 0;
 477
 478         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 479                 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 480         return ret & mask;
 481 }
 482
 483 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 484 {
 485         struct vcpu_svm *svm = to_svm(vcpu);
 486
 487         if (mask == 0)
 488                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 489         else
 490                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 491
 492 }
 493
 494 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 495 {
 496         struct vcpu_svm *svm = to_svm(vcpu);
 497
 498         if (svm->vmcb->control.next_rip != 0)
 499                 svm->next_rip = svm->vmcb->control.next_rip;
 500
 501         if (!svm->next_rip) {
 502                 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 503                                 EMULATE_DONE)
 504                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 505                 return;
 506         }
 507         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 508                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 509                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 510
 511         kvm_rip_write(vcpu, svm->next_rip);
 512         svm_set_interrupt_shadow(vcpu, 0);
 513 }
 514
 515 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 516                                 bool has_error_code, u32 error_code,
 517                                 bool reinject)
 518 {
 519         struct vcpu_svm *svm = to_svm(vcpu);
 520
 521         /*
 522          * If we are within a nested VM we'd better #VMEXIT and let the guest
 523          * handle the exception
 524          */
 525         if (!reinject &&
 526             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 527                 return;
 528
 529         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 530                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 531
 532                 /*
 533                  * For guest debugging where we have to reinject #BP if some
 534                  * INT3 is guest-owned:
 535                  * Emulate nRIP by moving RIP forward. Will fail if injection
 536                  * raises a fault that is not intercepted. Still better than
 537                  * failing in all cases.
 538                  */
 539                 skip_emulated_instruction(&svm->vcpu);
 540                 rip = kvm_rip_read(&svm->vcpu);
 541                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 542                 svm->int3_injected = rip - old_rip;
 543         }
 544
 545         svm->vmcb->control.event_inj = nr
 546                 | SVM_EVTINJ_VALID
 547                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 548                 | SVM_EVTINJ_TYPE_EXEPT;
 549         svm->vmcb->control.event_inj_err = error_code;
 550 }
 551
 552 static void svm_init_erratum_383(void)
 553 {
 554         u32 low, high;
 555         int err;
 556         u64 val;
 557
 558         if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 559                 return;
 560
 561         /* Use _safe variants to not break nested virtualization */
 562         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 563         if (err)
 564                 return;
 565
 566         val |= (1ULL << 47);
 567
 568         low  = lower_32_bits(val);
 569         high = upper_32_bits(val);
 570
 571         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 572
 573         erratum_383_found = true;
 574 }
 575
 576 static void svm_init_osvw(struct kvm_vcpu *vcpu)
 577 {
 578         /*
 579          * Guests should see errata 400 and 415 as fixed (assuming that
 580          * HLT and IO instructions are intercepted).
 581          */
 582         vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 583         vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 584
 585         /*
 586          * By increasing VCPU's osvw.length to 3 we are telling the guest that
 587          * all osvw.status bits inside that length, including bit 0 (which is
 588          * reserved for erratum 298), are valid. However, if host processor's
 589          * osvw_len is 0 then osvw_status[0] carries no information. We need to
 590          * be conservative here and therefore we tell the guest that erratum 298
 591          * is present (because we really don't know).
 592          */
 593         if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 594                 vcpu->arch.osvw.status |= 1;
 595 }
 596
 597 static int has_svm(void)
 598 {
 599         const char *msg;
 600
 601         if (!cpu_has_svm(&msg)) {
 602                 printk(KERN_INFO "has_svm: %s\n", msg);
 603                 return 0;
 604         }
 605
 606         return 1;
 607 }
 608
 609 static void svm_hardware_disable(void *garbage)
 610 {
 611         /* Make sure we clean up behind us */
 612         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 613                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 614
 615         cpu_svm_disable();
 616
 617         amd_pmu_disable_virt();
 618 }
 619
 620 static int svm_hardware_enable(void *garbage)
 621 {
 622
 623         struct svm_cpu_data *sd;
 624         uint64_t efer;
 625         struct desc_ptr gdt_descr;
 626         struct desc_struct *gdt;
 627         int me = raw_smp_processor_id();
 628
 629         rdmsrl(MSR_EFER, efer);
 630         if (efer & EFER_SVME)
 631                 return -EBUSY;
 632
 633         if (!has_svm()) {
 634                 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 635                 return -EINVAL;
 636         }
 637         sd = per_cpu(svm_data, me);
 638         if (!sd) {
 639                 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 640                 return -EINVAL;
 641         }
 642
 643         sd->asid_generation = 1;
 644         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 645         sd->next_asid = sd->max_asid + 1;
 646
 647         native_store_gdt(&gdt_descr);
 648         gdt = (struct desc_struct *)gdt_descr.address;
 649         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 650
 651         wrmsrl(MSR_EFER, efer | EFER_SVME);
 652
 653         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 654
 655         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 656                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 657                 __get_cpu_var(current_tsc_ratio) = TSC_RATIO_DEFAULT;
 658         }
 659
 660
 661         /*
 662          * Get OSVW bits.
 663          *
 664          * Note that it is possible to have a system with mixed processor
 665          * revisions and therefore different OSVW bits. If bits are not the same
 666          * on different processors then choose the worst case (i.e. if erratum
 667          * is present on one processor and not on another then assume that the
 668          * erratum is present everywhere).
 669          */
 670         if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 671                 uint64_t len, status = 0;
 672                 int err;
 673
 674                 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 675                 if (!err)
 676                         status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 677                                                       &err);
 678
 679                 if (err)
 680                         osvw_status = osvw_len = 0;
 681                 else {
 682                         if (len < osvw_len)
 683                                 osvw_len = len;
 684                         osvw_status |= status;
 685                         osvw_status &= (1ULL << osvw_len) - 1;
 686                 }
 687         } else
 688                 osvw_status = osvw_len = 0;
 689
 690         svm_init_erratum_383();
 691
 692         amd_pmu_enable_virt();
 693
 694         return 0;
 695 }
 696
 697 static void svm_cpu_uninit(int cpu)
 698 {
 699         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 700
 701         if (!sd)
 702                 return;
 703
 704         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 705         __free_page(sd->save_area);
 706         kfree(sd);
 707 }
 708
 709 static int svm_cpu_init(int cpu)
 710 {
 711         struct svm_cpu_data *sd;
 712         int r;
 713
 714         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 715         if (!sd)
 716                 return -ENOMEM;
 717         sd->cpu = cpu;
 718         sd->save_area = alloc_page(GFP_KERNEL);
 719         r = -ENOMEM;
 720         if (!sd->save_area)
 721                 goto err_1;
 722
 723         per_cpu(svm_data, cpu) = sd;
 724
 725         return 0;
 726
 727 err_1:
 728         kfree(sd);
 729         return r;
 730
 731 }
 732
 733 static bool valid_msr_intercept(u32 index)
 734 {
 735         int i;
 736
 737         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 738                 if (direct_access_msrs[i].index == index)
 739                         return true;
 740
 741         return false;
 742 }
 743
 744 static void set_msr_interception(u32 *msrpm, unsigned msr,
 745                                  int read, int write)
 746 {
 747         u8 bit_read, bit_write;
 748         unsigned long tmp;
 749         u32 offset;
 750
 751         /*
 752          * If this warning triggers extend the direct_access_msrs list at the
 753          * beginning of the file
 754          */
 755         WARN_ON(!valid_msr_intercept(msr));
 756
 757         offset    = svm_msrpm_offset(msr);
 758         bit_read  = 2 * (msr & 0x0f);
 759         bit_write = 2 * (msr & 0x0f) + 1;
 760         tmp       = msrpm[offset];
 761
 762         BUG_ON(offset == MSR_INVALID);
 763
 764         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 765         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 766
 767         msrpm[offset] = tmp;
 768 }
 769
 770 static void svm_vcpu_init_msrpm(u32 *msrpm)
 771 {
 772         int i;
 773
 774         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 775
 776         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 777                 if (!direct_access_msrs[i].always)
 778                         continue;
 779
 780                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 781         }
 782 }
 783
 784 static void add_msr_offset(u32 offset)
 785 {
 786         int i;
 787
 788         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 789
 790                 /* Offset already in list? */
 791                 if (msrpm_offsets[i] == offset)
 792                         return;
 793
 794                 /* Slot used by another offset? */
 795                 if (msrpm_offsets[i] != MSR_INVALID)
 796                         continue;
 797
 798                 /* Add offset to list */
 799                 msrpm_offsets[i] = offset;
 800
 801                 return;
 802         }
 803
 804         /*
 805          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 806          * increase MSRPM_OFFSETS in this case.
 807          */
 808         BUG();
 809 }
 810
 811 static void init_msrpm_offsets(void)
 812 {
 813         int i;
 814
 815         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 816
 817         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 818                 u32 offset;
 819
 820                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 821                 BUG_ON(offset == MSR_INVALID);
 822
 823                 add_msr_offset(offset);
 824         }
 825 }
 826
 827 static void svm_enable_lbrv(struct vcpu_svm *svm)
 828 {
 829         u32 *msrpm = svm->msrpm;
 830
 831         svm->vmcb->control.lbr_ctl = 1;
 832         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 833         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 834         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 835         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 836 }
 837
 838 static void svm_disable_lbrv(struct vcpu_svm *svm)
 839 {
 840         u32 *msrpm = svm->msrpm;
 841
 842         svm->vmcb->control.lbr_ctl = 0;
 843         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 844         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 845         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 846         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 847 }
 848
 849 static __init int svm_hardware_setup(void)
 850 {
 851         int cpu;
 852         struct page *iopm_pages;
 853         void *iopm_va;
 854         int r;
 855
 856         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 857
 858         if (!iopm_pages)
 859                 return -ENOMEM;
 860
 861         iopm_va = page_address(iopm_pages);
 862         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 863         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 864
 865         init_msrpm_offsets();
 866
 867         if (boot_cpu_has(X86_FEATURE_NX))
 868                 kvm_enable_efer_bits(EFER_NX);
 869
 870         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 871                 kvm_enable_efer_bits(EFER_FFXSR);
 872
 873         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 874                 u64 max;
 875
 876                 kvm_has_tsc_control = true;
 877
 878                 /*
 879                  * Make sure the user can only configure tsc_khz values that
 880                  * fit into a signed integer.
 881                  * A min value is not calculated needed because it will always
 882                  * be 1 on all machines and a value of 0 is used to disable
 883                  * tsc-scaling for the vcpu.
 884                  */
 885                 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 886
 887                 kvm_max_guest_tsc_khz = max;
 888         }
 889
 890         if (nested) {
 891                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 892                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 893         }
 894
 895         for_each_possible_cpu(cpu) {
 896                 r = svm_cpu_init(cpu);
 897                 if (r)
 898                         goto err;
 899         }
 900
 901         if (!boot_cpu_has(X86_FEATURE_NPT))
 902                 npt_enabled = false;
 903
 904         if (npt_enabled && !npt) {
 905                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 906                 npt_enabled = false;
 907         }
 908
 909         if (npt_enabled) {
 910                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 911                 kvm_enable_tdp();
 912         } else
 913                 kvm_disable_tdp();
 914
 915         return 0;
 916
 917 err:
 918         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 919         iopm_base = 0;
 920         return r;
 921 }
 922
 923 static __exit void svm_hardware_unsetup(void)
 924 {
 925         int cpu;
 926
 927         for_each_possible_cpu(cpu)
 928                 svm_cpu_uninit(cpu);
 929
 930         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 931         iopm_base = 0;
 932 }
 933
 934 static void init_seg(struct vmcb_seg *seg)
 935 {
 936         seg->selector = 0;
 937         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 938                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 939         seg->limit = 0xffff;
 940         seg->base = 0;
 941 }
 942
 943 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 944 {
 945         seg->selector = 0;
 946         seg->attrib = SVM_SELECTOR_P_MASK | type;
 947         seg->limit = 0xffff;
 948         seg->base = 0;
 949 }
 950
 951 static u64 __scale_tsc(u64 ratio, u64 tsc)
 952 {
 953         u64 mult, frac, _tsc;
 954
 955         mult  = ratio >> 32;
 956         frac  = ratio & ((1ULL << 32) - 1);
 957
 958         _tsc  = tsc;
 959         _tsc *= mult;
 960         _tsc += (tsc >> 32) * frac;
 961         _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
 962
 963         return _tsc;
 964 }
 965
 966 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
 967 {
 968         struct vcpu_svm *svm = to_svm(vcpu);
 969         u64 _tsc = tsc;
 970
 971         if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
 972                 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
 973
 974         return _tsc;
 975 }
 976
 977 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
 978 {
 979         struct vcpu_svm *svm = to_svm(vcpu);
 980         u64 ratio;
 981         u64 khz;
 982
 983         /* Guest TSC same frequency as host TSC? */
 984         if (!scale) {
 985                 svm->tsc_ratio = TSC_RATIO_DEFAULT;
 986                 return;
 987         }
 988
 989         /* TSC scaling supported? */
 990         if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 991                 if (user_tsc_khz > tsc_khz) {
 992                         vcpu->arch.tsc_catchup = 1;
 993                         vcpu->arch.tsc_always_catchup = 1;
 994                 } else
 995                         WARN(1, "user requested TSC rate below hardware speed\n");
 996                 return;
 997         }
 998
 999         khz = user_tsc_khz;
1000
1001         /* TSC scaling required  - calculate ratio */
1002         ratio = khz << 32;
1003         do_div(ratio, tsc_khz);
1004
1005         if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1006                 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1007                                 user_tsc_khz);
1008                 return;
1009         }
1010         svm->tsc_ratio             = ratio;
1011 }
1012
1013 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1014 {
1015         struct vcpu_svm *svm = to_svm(vcpu);
1016
1017         return svm->vmcb->control.tsc_offset;
1018 }
1019
1020 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1021 {
1022         struct vcpu_svm *svm = to_svm(vcpu);
1023         u64 g_tsc_offset = 0;
1024
1025         if (is_guest_mode(vcpu)) {
1026                 g_tsc_offset = svm->vmcb->control.tsc_offset -
1027                                svm->nested.hsave->control.tsc_offset;
1028                 svm->nested.hsave->control.tsc_offset = offset;
1029         } else
1030                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1031                                            svm->vmcb->control.tsc_offset,
1032                                            offset);
1033
1034         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1035
1036         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1037 }
1038
1039 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1040 {
1041         struct vcpu_svm *svm = to_svm(vcpu);
1042
1043         WARN_ON(adjustment < 0);
1044         if (host)
1045                 adjustment = svm_scale_tsc(vcpu, adjustment);
1046
1047         svm->vmcb->control.tsc_offset += adjustment;
1048         if (is_guest_mode(vcpu))
1049                 svm->nested.hsave->control.tsc_offset += adjustment;
1050         else
1051                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1052                                      svm->vmcb->control.tsc_offset - adjustment,
1053                                      svm->vmcb->control.tsc_offset);
1054
1055         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1056 }
1057
1058 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1059 {
1060         u64 tsc;
1061
1062         tsc = svm_scale_tsc(vcpu, native_read_tsc());
1063
1064         return target_tsc - tsc;
1065 }
1066
1067 static void init_vmcb(struct vcpu_svm *svm)
1068 {
1069         struct vmcb_control_area *control = &svm->vmcb->control;
1070         struct vmcb_save_area *save = &svm->vmcb->save;
1071
1072         svm->vcpu.fpu_active = 1;
1073         svm->vcpu.arch.hflags = 0;
1074
1075         set_cr_intercept(svm, INTERCEPT_CR0_READ);
1076         set_cr_intercept(svm, INTERCEPT_CR3_READ);
1077         set_cr_intercept(svm, INTERCEPT_CR4_READ);
1078         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1079         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1080         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1081         set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1082
1083         set_dr_intercept(svm, INTERCEPT_DR0_READ);
1084         set_dr_intercept(svm, INTERCEPT_DR1_READ);
1085         set_dr_intercept(svm, INTERCEPT_DR2_READ);
1086         set_dr_intercept(svm, INTERCEPT_DR3_READ);
1087         set_dr_intercept(svm, INTERCEPT_DR4_READ);
1088         set_dr_intercept(svm, INTERCEPT_DR5_READ);
1089         set_dr_intercept(svm, INTERCEPT_DR6_READ);
1090         set_dr_intercept(svm, INTERCEPT_DR7_READ);
1091
1092         set_dr_intercept(svm, INTERCEPT_DR0_WRITE);
1093         set_dr_intercept(svm, INTERCEPT_DR1_WRITE);
1094         set_dr_intercept(svm, INTERCEPT_DR2_WRITE);
1095         set_dr_intercept(svm, INTERCEPT_DR3_WRITE);
1096         set_dr_intercept(svm, INTERCEPT_DR4_WRITE);
1097         set_dr_intercept(svm, INTERCEPT_DR5_WRITE);
1098         set_dr_intercept(svm, INTERCEPT_DR6_WRITE);
1099         set_dr_intercept(svm, INTERCEPT_DR7_WRITE);
1100
1101         set_exception_intercept(svm, PF_VECTOR);
1102         set_exception_intercept(svm, UD_VECTOR);
1103         set_exception_intercept(svm, MC_VECTOR);
1104
1105         set_intercept(svm, INTERCEPT_INTR);
1106         set_intercept(svm, INTERCEPT_NMI);
1107         set_intercept(svm, INTERCEPT_SMI);
1108         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1109         set_intercept(svm, INTERCEPT_RDPMC);
1110         set_intercept(svm, INTERCEPT_CPUID);
1111         set_intercept(svm, INTERCEPT_INVD);
1112         set_intercept(svm, INTERCEPT_HLT);
1113         set_intercept(svm, INTERCEPT_INVLPG);
1114         set_intercept(svm, INTERCEPT_INVLPGA);
1115         set_intercept(svm, INTERCEPT_IOIO_PROT);
1116         set_intercept(svm, INTERCEPT_MSR_PROT);
1117         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1118         set_intercept(svm, INTERCEPT_SHUTDOWN);
1119         set_intercept(svm, INTERCEPT_VMRUN);
1120         set_intercept(svm, INTERCEPT_VMMCALL);
1121         set_intercept(svm, INTERCEPT_VMLOAD);
1122         set_intercept(svm, INTERCEPT_VMSAVE);
1123         set_intercept(svm, INTERCEPT_STGI);
1124         set_intercept(svm, INTERCEPT_CLGI);
1125         set_intercept(svm, INTERCEPT_SKINIT);
1126         set_intercept(svm, INTERCEPT_WBINVD);
1127         set_intercept(svm, INTERCEPT_MONITOR);
1128         set_intercept(svm, INTERCEPT_MWAIT);
1129         set_intercept(svm, INTERCEPT_XSETBV);
1130
1131         control->iopm_base_pa = iopm_base;
1132         control->msrpm_base_pa = __pa(svm->msrpm);
1133         control->int_ctl = V_INTR_MASKING_MASK;
1134
1135         init_seg(&save->es);
1136         init_seg(&save->ss);
1137         init_seg(&save->ds);
1138         init_seg(&save->fs);
1139         init_seg(&save->gs);
1140
1141         save->cs.selector = 0xf000;
1142         save->cs.base = 0xffff0000;
1143         /* Executable/Readable Code Segment */
1144         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1145                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1146         save->cs.limit = 0xffff;
1147
1148         save->gdtr.limit = 0xffff;
1149         save->idtr.limit = 0xffff;
1150
1151         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1152         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1153
1154         svm_set_efer(&svm->vcpu, 0);
1155         save->dr6 = 0xffff0ff0;
1156         kvm_set_rflags(&svm->vcpu, 2);
1157         save->rip = 0x0000fff0;
1158         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1159
1160         /*
1161          * This is the guest-visible cr0 value.
1162          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1163          */
1164         svm->vcpu.arch.cr0 = 0;
1165         (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1166
1167         save->cr4 = X86_CR4_PAE;
1168         /* rdx = ?? */
1169
1170         if (npt_enabled) {
1171                 /* Setup VMCB for Nested Paging */
1172                 control->nested_ctl = 1;
1173                 clr_intercept(svm, INTERCEPT_INVLPG);
1174                 clr_exception_intercept(svm, PF_VECTOR);
1175                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1176                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1177                 save->g_pat = 0x0007040600070406ULL;
1178                 save->cr3 = 0;
1179                 save->cr4 = 0;
1180         }
1181         svm->asid_generation = 0;
1182
1183         svm->nested.vmcb = 0;
1184         svm->vcpu.arch.hflags = 0;
1185
1186         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1187                 control->pause_filter_count = 3000;
1188                 set_intercept(svm, INTERCEPT_PAUSE);
1189         }
1190
1191         mark_all_dirty(svm->vmcb);
1192
1193         enable_gif(svm);
1194 }
1195
1196 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
1197 {
1198         struct vcpu_svm *svm = to_svm(vcpu);
1199         u32 dummy;
1200         u32 eax = 1;
1201
1202         init_vmcb(svm);
1203
1204         kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1205         kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1206 }
1207
1208 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1209 {
1210         struct vcpu_svm *svm;
1211         struct page *page;
1212         struct page *msrpm_pages;
1213         struct page *hsave_page;
1214         struct page *nested_msrpm_pages;
1215         int err;
1216
1217         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1218         if (!svm) {
1219                 err = -ENOMEM;
1220                 goto out;
1221         }
1222
1223         svm->tsc_ratio = TSC_RATIO_DEFAULT;
1224
1225         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1226         if (err)
1227                 goto free_svm;
1228
1229         err = -ENOMEM;
1230         page = alloc_page(GFP_KERNEL);
1231         if (!page)
1232                 goto uninit;
1233
1234         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1235         if (!msrpm_pages)
1236                 goto free_page1;
1237
1238         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1239         if (!nested_msrpm_pages)
1240                 goto free_page2;
1241
1242         hsave_page = alloc_page(GFP_KERNEL);
1243         if (!hsave_page)
1244                 goto free_page3;
1245
1246         svm->nested.hsave = page_address(hsave_page);
1247
1248         svm->msrpm = page_address(msrpm_pages);
1249         svm_vcpu_init_msrpm(svm->msrpm);
1250
1251         svm->nested.msrpm = page_address(nested_msrpm_pages);
1252         svm_vcpu_init_msrpm(svm->nested.msrpm);
1253
1254         svm->vmcb = page_address(page);
1255         clear_page(svm->vmcb);
1256         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1257         svm->asid_generation = 0;
1258         init_vmcb(svm);
1259
1260         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1261         if (kvm_vcpu_is_bsp(&svm->vcpu))
1262                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1263
1264         svm_init_osvw(&svm->vcpu);
1265
1266         return &svm->vcpu;
1267
1268 free_page3:
1269         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1270 free_page2:
1271         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1272 free_page1:
1273         __free_page(page);
1274 uninit:
1275         kvm_vcpu_uninit(&svm->vcpu);
1276 free_svm:
1277         kmem_cache_free(kvm_vcpu_cache, svm);
1278 out:
1279         return ERR_PTR(err);
1280 }
1281
1282 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1283 {
1284         struct vcpu_svm *svm = to_svm(vcpu);
1285
1286         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1287         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1288         __free_page(virt_to_page(svm->nested.hsave));
1289         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1290         kvm_vcpu_uninit(vcpu);
1291         kmem_cache_free(kvm_vcpu_cache, svm);
1292 }
1293
1294 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1295 {
1296         struct vcpu_svm *svm = to_svm(vcpu);
1297         int i;
1298
1299         if (unlikely(cpu != vcpu->cpu)) {
1300                 svm->asid_generation = 0;
1301                 mark_all_dirty(svm->vmcb);
1302         }
1303
1304 #ifdef CONFIG_X86_64
1305         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1306 #endif
1307         savesegment(fs, svm->host.fs);
1308         savesegment(gs, svm->host.gs);
1309         svm->host.ldt = kvm_read_ldt();
1310
1311         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1312                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1313
1314         if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1315             svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1316                 __get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1317                 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1318         }
1319 }
1320
1321 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1322 {
1323         struct vcpu_svm *svm = to_svm(vcpu);
1324         int i;
1325
1326         ++vcpu->stat.host_state_reload;
1327         kvm_load_ldt(svm->host.ldt);
1328 #ifdef CONFIG_X86_64
1329         loadsegment(fs, svm->host.fs);
1330         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1331         load_gs_index(svm->host.gs);
1332 #else
1333 #ifdef CONFIG_X86_32_LAZY_GS
1334         loadsegment(gs, svm->host.gs);
1335 #endif
1336 #endif
1337         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1338                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1339 }
1340
1341 static void svm_update_cpl(struct kvm_vcpu *vcpu)
1342 {
1343         struct vcpu_svm *svm = to_svm(vcpu);
1344         int cpl;
1345
1346         if (!is_protmode(vcpu))
1347                 cpl = 0;
1348         else if (svm->vmcb->save.rflags & X86_EFLAGS_VM)
1349                 cpl = 3;
1350         else
1351                 cpl = svm->vmcb->save.cs.selector & 0x3;
1352
1353         svm->vmcb->save.cpl = cpl;
1354 }
1355
1356 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1357 {
1358         return to_svm(vcpu)->vmcb->save.rflags;
1359 }
1360
1361 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1362 {
1363         unsigned long old_rflags = to_svm(vcpu)->vmcb->save.rflags;
1364
1365         to_svm(vcpu)->vmcb->save.rflags = rflags;
1366         if ((old_rflags ^ rflags) & X86_EFLAGS_VM)
1367                 svm_update_cpl(vcpu);
1368 }
1369
1370 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1371 {
1372         switch (reg) {
1373         case VCPU_EXREG_PDPTR:
1374                 BUG_ON(!npt_enabled);
1375                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1376                 break;
1377         default:
1378                 BUG();
1379         }
1380 }
1381
1382 static void svm_set_vintr(struct vcpu_svm *svm)
1383 {
1384         set_intercept(svm, INTERCEPT_VINTR);
1385 }
1386
1387 static void svm_clear_vintr(struct vcpu_svm *svm)
1388 {
1389         clr_intercept(svm, INTERCEPT_VINTR);
1390 }
1391
1392 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1393 {
1394         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1395
1396         switch (seg) {
1397         case VCPU_SREG_CS: return &save->cs;
1398         case VCPU_SREG_DS: return &save->ds;
1399         case VCPU_SREG_ES: return &save->es;
1400         case VCPU_SREG_FS: return &save->fs;
1401         case VCPU_SREG_GS: return &save->gs;
1402         case VCPU_SREG_SS: return &save->ss;
1403         case VCPU_SREG_TR: return &save->tr;
1404         case VCPU_SREG_LDTR: return &save->ldtr;
1405         }
1406         BUG();
1407         return NULL;
1408 }
1409
1410 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1411 {
1412         struct vmcb_seg *s = svm_seg(vcpu, seg);
1413
1414         return s->base;
1415 }
1416
1417 static void svm_get_segment(struct kvm_vcpu *vcpu,
1418                             struct kvm_segment *var, int seg)
1419 {
1420         struct vmcb_seg *s = svm_seg(vcpu, seg);
1421
1422         var->base = s->base;
1423         var->limit = s->limit;
1424         var->selector = s->selector;
1425         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1426         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1427         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1428         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1429         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1430         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1431         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1432         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1433
1434         /*
1435          * AMD's VMCB does not have an explicit unusable field, so emulate it
1436          * for cross vendor migration purposes by "not present"
1437          */
1438         var->unusable = !var->present || (var->type == 0);
1439
1440         switch (seg) {
1441         case VCPU_SREG_CS:
1442                 /*
1443                  * SVM always stores 0 for the 'G' bit in the CS selector in
1444                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1445                  * Intel's VMENTRY has a check on the 'G' bit.
1446                  */
1447                 var->g = s->limit > 0xfffff;
1448                 break;
1449         case VCPU_SREG_TR:
1450                 /*
1451                  * Work around a bug where the busy flag in the tr selector
1452                  * isn't exposed
1453                  */
1454                 var->type |= 0x2;
1455                 break;
1456         case VCPU_SREG_DS:
1457         case VCPU_SREG_ES:
1458         case VCPU_SREG_FS:
1459         case VCPU_SREG_GS:
1460                 /*
1461                  * The accessed bit must always be set in the segment
1462                  * descriptor cache, although it can be cleared in the
1463                  * descriptor, the cached bit always remains at 1. Since
1464                  * Intel has a check on this, set it here to support
1465                  * cross-vendor migration.
1466                  */
1467                 if (!var->unusable)
1468                         var->type |= 0x1;
1469                 break;
1470         case VCPU_SREG_SS:
1471                 /*
1472                  * On AMD CPUs sometimes the DB bit in the segment
1473                  * descriptor is left as 1, although the whole segment has
1474                  * been made unusable. Clear it here to pass an Intel VMX
1475                  * entry check when cross vendor migrating.
1476                  */
1477                 if (var->unusable)
1478                         var->db = 0;
1479                 break;
1480         }
1481 }
1482
1483 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1484 {
1485         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1486
1487         return save->cpl;
1488 }
1489
1490 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1491 {
1492         struct vcpu_svm *svm = to_svm(vcpu);
1493
1494         dt->size = svm->vmcb->save.idtr.limit;
1495         dt->address = svm->vmcb->save.idtr.base;
1496 }
1497
1498 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1499 {
1500         struct vcpu_svm *svm = to_svm(vcpu);
1501
1502         svm->vmcb->save.idtr.limit = dt->size;
1503         svm->vmcb->save.idtr.base = dt->address ;
1504         mark_dirty(svm->vmcb, VMCB_DT);
1505 }
1506
1507 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1508 {
1509         struct vcpu_svm *svm = to_svm(vcpu);
1510
1511         dt->size = svm->vmcb->save.gdtr.limit;
1512         dt->address = svm->vmcb->save.gdtr.base;
1513 }
1514
1515 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1516 {
1517         struct vcpu_svm *svm = to_svm(vcpu);
1518
1519         svm->vmcb->save.gdtr.limit = dt->size;
1520         svm->vmcb->save.gdtr.base = dt->address ;
1521         mark_dirty(svm->vmcb, VMCB_DT);
1522 }
1523
1524 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1525 {
1526 }
1527
1528 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1529 {
1530 }
1531
1532 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1533 {
1534 }
1535
1536 static void update_cr0_intercept(struct vcpu_svm *svm)
1537 {
1538         ulong gcr0 = svm->vcpu.arch.cr0;
1539         u64 *hcr0 = &svm->vmcb->save.cr0;
1540
1541         if (!svm->vcpu.fpu_active)
1542                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1543         else
1544                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1545                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1546
1547         mark_dirty(svm->vmcb, VMCB_CR);
1548
1549         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1550                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1551                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1552         } else {
1553                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1554                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1555         }
1556 }
1557
1558 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1559 {
1560         struct vcpu_svm *svm = to_svm(vcpu);
1561
1562 #ifdef CONFIG_X86_64
1563         if (vcpu->arch.efer & EFER_LME) {
1564                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1565                         vcpu->arch.efer |= EFER_LMA;
1566                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1567                 }
1568
1569                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1570                         vcpu->arch.efer &= ~EFER_LMA;
1571                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1572                 }
1573         }
1574 #endif
1575         vcpu->arch.cr0 = cr0;
1576
1577         if (!npt_enabled)
1578                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1579
1580         if (!vcpu->fpu_active)
1581                 cr0 |= X86_CR0_TS;
1582         /*
1583          * re-enable caching here because the QEMU bios
1584          * does not do it - this results in some delay at
1585          * reboot
1586          */
1587         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1588         svm->vmcb->save.cr0 = cr0;
1589         mark_dirty(svm->vmcb, VMCB_CR);
1590         update_cr0_intercept(svm);
1591 }
1592
1593 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1594 {
1595         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1596         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1597
1598         if (cr4 & X86_CR4_VMXE)
1599                 return 1;
1600
1601         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1602                 svm_flush_tlb(vcpu);
1603
1604         vcpu->arch.cr4 = cr4;
1605         if (!npt_enabled)
1606                 cr4 |= X86_CR4_PAE;
1607         cr4 |= host_cr4_mce;
1608         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1609         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1610         return 0;
1611 }
1612
1613 static void svm_set_segment(struct kvm_vcpu *vcpu,
1614                             struct kvm_segment *var, int seg)
1615 {
1616         struct vcpu_svm *svm = to_svm(vcpu);
1617         struct vmcb_seg *s = svm_seg(vcpu, seg);
1618
1619         s->base = var->base;
1620         s->limit = var->limit;
1621         s->selector = var->selector;
1622         if (var->unusable)
1623                 s->attrib = 0;
1624         else {
1625                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1626                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1627                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1628                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1629                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1630                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1631                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1632                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1633         }
1634         if (seg == VCPU_SREG_CS)
1635                 svm_update_cpl(vcpu);
1636
1637         mark_dirty(svm->vmcb, VMCB_SEG);
1638 }
1639
1640 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1641 {
1642         struct vcpu_svm *svm = to_svm(vcpu);
1643
1644         clr_exception_intercept(svm, DB_VECTOR);
1645         clr_exception_intercept(svm, BP_VECTOR);
1646
1647         if (svm->nmi_singlestep)
1648                 set_exception_intercept(svm, DB_VECTOR);
1649
1650         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1651                 if (vcpu->guest_debug &
1652                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1653                         set_exception_intercept(svm, DB_VECTOR);
1654                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1655                         set_exception_intercept(svm, BP_VECTOR);
1656         } else
1657                 vcpu->guest_debug = 0;
1658 }
1659
1660 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1661 {
1662         if (sd->next_asid > sd->max_asid) {
1663                 ++sd->asid_generation;
1664                 sd->next_asid = 1;
1665                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1666         }
1667
1668         svm->asid_generation = sd->asid_generation;
1669         svm->vmcb->control.asid = sd->next_asid++;
1670
1671         mark_dirty(svm->vmcb, VMCB_ASID);
1672 }
1673
1674 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1675 {
1676         return to_svm(vcpu)->vmcb->save.dr6;
1677 }
1678
1679 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1680 {
1681         struct vcpu_svm *svm = to_svm(vcpu);
1682
1683         svm->vmcb->save.dr6 = value;
1684         mark_dirty(svm->vmcb, VMCB_DR);
1685 }
1686
1687 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1688 {
1689         struct vcpu_svm *svm = to_svm(vcpu);
1690
1691         svm->vmcb->save.dr7 = value;
1692         mark_dirty(svm->vmcb, VMCB_DR);
1693 }
1694
1695 static int pf_interception(struct vcpu_svm *svm)
1696 {
1697         u64 fault_address = svm->vmcb->control.exit_info_2;
1698         u32 error_code;
1699         int r = 1;
1700
1701         switch (svm->apf_reason) {
1702         default:
1703                 error_code = svm->vmcb->control.exit_info_1;
1704
1705                 trace_kvm_page_fault(fault_address, error_code);
1706                 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1707                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1708                 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1709                         svm->vmcb->control.insn_bytes,
1710                         svm->vmcb->control.insn_len);
1711                 break;
1712         case KVM_PV_REASON_PAGE_NOT_PRESENT:
1713                 svm->apf_reason = 0;
1714                 local_irq_disable();
1715                 kvm_async_pf_task_wait(fault_address);
1716                 local_irq_enable();
1717                 break;
1718         case KVM_PV_REASON_PAGE_READY:
1719                 svm->apf_reason = 0;
1720                 local_irq_disable();
1721                 kvm_async_pf_task_wake(fault_address);
1722                 local_irq_enable();
1723                 break;
1724         }
1725         return r;
1726 }
1727
1728 static int db_interception(struct vcpu_svm *svm)
1729 {
1730         struct kvm_run *kvm_run = svm->vcpu.run;
1731
1732         if (!(svm->vcpu.guest_debug &
1733               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1734                 !svm->nmi_singlestep) {
1735                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1736                 return 1;
1737         }
1738
1739         if (svm->nmi_singlestep) {
1740                 svm->nmi_singlestep = false;
1741                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1742                         svm->vmcb->save.rflags &=
1743                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1744                 update_db_bp_intercept(&svm->vcpu);
1745         }
1746
1747         if (svm->vcpu.guest_debug &
1748             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1749                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1750                 kvm_run->debug.arch.pc =
1751                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1752                 kvm_run->debug.arch.exception = DB_VECTOR;
1753                 return 0;
1754         }
1755
1756         return 1;
1757 }
1758
1759 static int bp_interception(struct vcpu_svm *svm)
1760 {
1761         struct kvm_run *kvm_run = svm->vcpu.run;
1762
1763         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1764         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1765         kvm_run->debug.arch.exception = BP_VECTOR;
1766         return 0;
1767 }
1768
1769 static int ud_interception(struct vcpu_svm *svm)
1770 {
1771         int er;
1772
1773         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1774         if (er != EMULATE_DONE)
1775                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1776         return 1;
1777 }
1778
1779 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1780 {
1781         struct vcpu_svm *svm = to_svm(vcpu);
1782
1783         clr_exception_intercept(svm, NM_VECTOR);
1784
1785         svm->vcpu.fpu_active = 1;
1786         update_cr0_intercept(svm);
1787 }
1788
1789 static int nm_interception(struct vcpu_svm *svm)
1790 {
1791         svm_fpu_activate(&svm->vcpu);
1792         return 1;
1793 }
1794
1795 static bool is_erratum_383(void)
1796 {
1797         int err, i;
1798         u64 value;
1799
1800         if (!erratum_383_found)
1801                 return false;
1802
1803         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1804         if (err)
1805                 return false;
1806
1807         /* Bit 62 may or may not be set for this mce */
1808         value &= ~(1ULL << 62);
1809
1810         if (value != 0xb600000000010015ULL)
1811                 return false;
1812
1813         /* Clear MCi_STATUS registers */
1814         for (i = 0; i < 6; ++i)
1815                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1816
1817         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1818         if (!err) {
1819                 u32 low, high;
1820
1821                 value &= ~(1ULL << 2);
1822                 low    = lower_32_bits(value);
1823                 high   = upper_32_bits(value);
1824
1825                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1826         }
1827
1828         /* Flush tlb to evict multi-match entries */
1829         __flush_tlb_all();
1830
1831         return true;
1832 }
1833
1834 static void svm_handle_mce(struct vcpu_svm *svm)
1835 {
1836         if (is_erratum_383()) {
1837                 /*
1838                  * Erratum 383 triggered. Guest state is corrupt so kill the
1839                  * guest.
1840                  */
1841                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1842
1843                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1844
1845                 return;
1846         }
1847
1848         /*
1849          * On an #MC intercept the MCE handler is not called automatically in
1850          * the host. So do it by hand here.
1851          */
1852         asm volatile (
1853                 "int $0x12\n");
1854         /* not sure if we ever come back to this point */
1855
1856         return;
1857 }
1858
1859 static int mc_interception(struct vcpu_svm *svm)
1860 {
1861         return 1;
1862 }
1863
1864 static int shutdown_interception(struct vcpu_svm *svm)
1865 {
1866         struct kvm_run *kvm_run = svm->vcpu.run;
1867
1868         /*
1869          * VMCB is undefined after a SHUTDOWN intercept
1870          * so reinitialize it.
1871          */
1872         clear_page(svm->vmcb);
1873         init_vmcb(svm);
1874
1875         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1876         return 0;
1877 }
1878
1879 static int io_interception(struct vcpu_svm *svm)
1880 {
1881         struct kvm_vcpu *vcpu = &svm->vcpu;
1882         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1883         int size, in, string;
1884         unsigned port;
1885
1886         ++svm->vcpu.stat.io_exits;
1887         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1888         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1889         if (string || in)
1890                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1891
1892         port = io_info >> 16;
1893         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1894         svm->next_rip = svm->vmcb->control.exit_info_2;
1895         skip_emulated_instruction(&svm->vcpu);
1896
1897         return kvm_fast_pio_out(vcpu, size, port);
1898 }
1899
1900 static int nmi_interception(struct vcpu_svm *svm)
1901 {
1902         return 1;
1903 }
1904
1905 static int intr_interception(struct vcpu_svm *svm)
1906 {
1907         ++svm->vcpu.stat.irq_exits;
1908         return 1;
1909 }
1910
1911 static int nop_on_interception(struct vcpu_svm *svm)
1912 {
1913         return 1;
1914 }
1915
1916 static int halt_interception(struct vcpu_svm *svm)
1917 {
1918         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1919         skip_emulated_instruction(&svm->vcpu);
1920         return kvm_emulate_halt(&svm->vcpu);
1921 }
1922
1923 static int vmmcall_interception(struct vcpu_svm *svm)
1924 {
1925         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1926         skip_emulated_instruction(&svm->vcpu);
1927         kvm_emulate_hypercall(&svm->vcpu);
1928         return 1;
1929 }
1930
1931 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1932 {
1933         struct vcpu_svm *svm = to_svm(vcpu);
1934
1935         return svm->nested.nested_cr3;
1936 }
1937
1938 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1939 {
1940         struct vcpu_svm *svm = to_svm(vcpu);
1941         u64 cr3 = svm->nested.nested_cr3;
1942         u64 pdpte;
1943         int ret;
1944
1945         ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1946                                   offset_in_page(cr3) + index * 8, 8);
1947         if (ret)
1948                 return 0;
1949         return pdpte;
1950 }
1951
1952 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1953                                    unsigned long root)
1954 {
1955         struct vcpu_svm *svm = to_svm(vcpu);
1956
1957         svm->vmcb->control.nested_cr3 = root;
1958         mark_dirty(svm->vmcb, VMCB_NPT);
1959         svm_flush_tlb(vcpu);
1960 }
1961
1962 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1963                                        struct x86_exception *fault)
1964 {
1965         struct vcpu_svm *svm = to_svm(vcpu);
1966
1967         svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1968         svm->vmcb->control.exit_code_hi = 0;
1969         svm->vmcb->control.exit_info_1 = fault->error_code;
1970         svm->vmcb->control.exit_info_2 = fault->address;
1971
1972         nested_svm_vmexit(svm);
1973 }
1974
1975 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1976 {
1977         kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1978
1979         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1980         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
1981         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
1982         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1983         vcpu->arch.mmu.shadow_root_level = get_npt_level();
1984         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
1985 }
1986
1987 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1988 {
1989         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1990 }
1991
1992 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1993 {
1994         if (!(svm->vcpu.arch.efer & EFER_SVME)
1995             || !is_paging(&svm->vcpu)) {
1996                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1997                 return 1;
1998         }
1999
2000         if (svm->vmcb->save.cpl) {
2001                 kvm_inject_gp(&svm->vcpu, 0);
2002                 return 1;
2003         }
2004
2005        return 0;
2006 }
2007
2008 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2009                                       bool has_error_code, u32 error_code)
2010 {
2011         int vmexit;
2012
2013         if (!is_guest_mode(&svm->vcpu))
2014                 return 0;
2015
2016         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2017         svm->vmcb->control.exit_code_hi = 0;
2018         svm->vmcb->control.exit_info_1 = error_code;
2019         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2020
2021         vmexit = nested_svm_intercept(svm);
2022         if (vmexit == NESTED_EXIT_DONE)
2023                 svm->nested.exit_required = true;
2024
2025         return vmexit;
2026 }
2027
2028 /* This function returns true if it is save to enable the irq window */
2029 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2030 {
2031         if (!is_guest_mode(&svm->vcpu))
2032                 return true;
2033
2034         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2035                 return true;
2036
2037         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2038                 return false;
2039
2040         /*
2041          * if vmexit was already requested (by intercepted exception
2042          * for instance) do not overwrite it with "external interrupt"
2043          * vmexit.
2044          */
2045         if (svm->nested.exit_required)
2046                 return false;
2047
2048         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
2049         svm->vmcb->control.exit_info_1 = 0;
2050         svm->vmcb->control.exit_info_2 = 0;
2051
2052         if (svm->nested.intercept & 1ULL) {
2053                 /*
2054                  * The #vmexit can't be emulated here directly because this
2055                  * code path runs with irqs and preemption disabled. A
2056                  * #vmexit emulation might sleep. Only signal request for
2057                  * the #vmexit here.
2058                  */
2059                 svm->nested.exit_required = true;
2060                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2061                 return false;
2062         }
2063
2064         return true;
2065 }
2066
2067 /* This function returns true if it is save to enable the nmi window */
2068 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2069 {
2070         if (!is_guest_mode(&svm->vcpu))
2071                 return true;
2072
2073         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2074                 return true;
2075
2076         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2077         svm->nested.exit_required = true;
2078
2079         return false;
2080 }
2081
2082 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2083 {
2084         struct page *page;
2085
2086         might_sleep();
2087
2088         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2089         if (is_error_page(page))
2090                 goto error;
2091
2092         *_page = page;
2093
2094         return kmap(page);
2095
2096 error:
2097         kvm_inject_gp(&svm->vcpu, 0);
2098
2099         return NULL;
2100 }
2101
2102 static void nested_svm_unmap(struct page *page)
2103 {
2104         kunmap(page);
2105         kvm_release_page_dirty(page);
2106 }
2107
2108 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2109 {
2110         unsigned port;
2111         u8 val, bit;
2112         u64 gpa;
2113
2114         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2115                 return NESTED_EXIT_HOST;
2116
2117         port = svm->vmcb->control.exit_info_1 >> 16;
2118         gpa  = svm->nested.vmcb_iopm + (port / 8);
2119         bit  = port % 8;
2120         val  = 0;
2121
2122         if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2123                 val &= (1 << bit);
2124
2125         return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2126 }
2127
2128 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2129 {
2130         u32 offset, msr, value;
2131         int write, mask;
2132
2133         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2134                 return NESTED_EXIT_HOST;
2135
2136         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2137         offset = svm_msrpm_offset(msr);
2138         write  = svm->vmcb->control.exit_info_1 & 1;
2139         mask   = 1 << ((2 * (msr & 0xf)) + write);
2140
2141         if (offset == MSR_INVALID)
2142                 return NESTED_EXIT_DONE;
2143
2144         /* Offset is in 32 bit units but need in 8 bit units */
2145         offset *= 4;
2146
2147         if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2148                 return NESTED_EXIT_DONE;
2149
2150         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2151 }
2152
2153 static int nested_svm_exit_special(struct vcpu_svm *svm)
2154 {
2155         u32 exit_code = svm->vmcb->control.exit_code;
2156
2157         switch (exit_code) {
2158         case SVM_EXIT_INTR:
2159         case SVM_EXIT_NMI:
2160         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2161                 return NESTED_EXIT_HOST;
2162         case SVM_EXIT_NPF:
2163                 /* For now we are always handling NPFs when using them */
2164                 if (npt_enabled)
2165                         return NESTED_EXIT_HOST;
2166                 break;
2167         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2168                 /* When we're shadowing, trap PFs, but not async PF */
2169                 if (!npt_enabled && svm->apf_reason == 0)
2170                         return NESTED_EXIT_HOST;
2171                 break;
2172         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2173                 nm_interception(svm);
2174                 break;
2175         default:
2176                 break;
2177         }
2178
2179         return NESTED_EXIT_CONTINUE;
2180 }
2181
2182 /*
2183  * If this function returns true, this #vmexit was already handled
2184  */
2185 static int nested_svm_intercept(struct vcpu_svm *svm)
2186 {
2187         u32 exit_code = svm->vmcb->control.exit_code;
2188         int vmexit = NESTED_EXIT_HOST;
2189
2190         switch (exit_code) {
2191         case SVM_EXIT_MSR:
2192                 vmexit = nested_svm_exit_handled_msr(svm);
2193                 break;
2194         case SVM_EXIT_IOIO:
2195                 vmexit = nested_svm_intercept_ioio(svm);
2196                 break;
2197         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2198                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2199                 if (svm->nested.intercept_cr & bit)
2200                         vmexit = NESTED_EXIT_DONE;
2201                 break;
2202         }
2203         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2204                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2205                 if (svm->nested.intercept_dr & bit)
2206                         vmexit = NESTED_EXIT_DONE;
2207                 break;
2208         }
2209         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2210                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2211                 if (svm->nested.intercept_exceptions & excp_bits)
2212                         vmexit = NESTED_EXIT_DONE;
2213                 /* async page fault always cause vmexit */
2214                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2215                          svm->apf_reason != 0)
2216                         vmexit = NESTED_EXIT_DONE;
2217                 break;
2218         }
2219         case SVM_EXIT_ERR: {
2220                 vmexit = NESTED_EXIT_DONE;
2221                 break;
2222         }
2223         default: {
2224                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2225                 if (svm->nested.intercept & exit_bits)
2226                         vmexit = NESTED_EXIT_DONE;
2227         }
2228         }
2229
2230         return vmexit;
2231 }
2232
2233 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2234 {
2235         int vmexit;
2236
2237         vmexit = nested_svm_intercept(svm);
2238
2239         if (vmexit == NESTED_EXIT_DONE)
2240                 nested_svm_vmexit(svm);
2241
2242         return vmexit;
2243 }
2244
2245 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2246 {
2247         struct vmcb_control_area *dst  = &dst_vmcb->control;
2248         struct vmcb_control_area *from = &from_vmcb->control;
2249
2250         dst->intercept_cr         = from->intercept_cr;
2251         dst->intercept_dr         = from->intercept_dr;
2252         dst->intercept_exceptions = from->intercept_exceptions;
2253         dst->intercept            = from->intercept;
2254         dst->iopm_base_pa         = from->iopm_base_pa;
2255         dst->msrpm_base_pa        = from->msrpm_base_pa;
2256         dst->tsc_offset           = from->tsc_offset;
2257         dst->asid                 = from->asid;
2258         dst->tlb_ctl              = from->tlb_ctl;
2259         dst->int_ctl              = from->int_ctl;
2260         dst->int_vector           = from->int_vector;
2261         dst->int_state            = from->int_state;
2262         dst->exit_code            = from->exit_code;
2263         dst->exit_code_hi         = from->exit_code_hi;
2264         dst->exit_info_1          = from->exit_info_1;
2265         dst->exit_info_2          = from->exit_info_2;
2266         dst->exit_int_info        = from->exit_int_info;
2267         dst->exit_int_info_err    = from->exit_int_info_err;
2268         dst->nested_ctl           = from->nested_ctl;
2269         dst->event_inj            = from->event_inj;
2270         dst->event_inj_err        = from->event_inj_err;
2271         dst->nested_cr3           = from->nested_cr3;
2272         dst->lbr_ctl              = from->lbr_ctl;
2273 }
2274
2275 static int nested_svm_vmexit(struct vcpu_svm *svm)
2276 {
2277         struct vmcb *nested_vmcb;
2278         struct vmcb *hsave = svm->nested.hsave;
2279         struct vmcb *vmcb = svm->vmcb;
2280         struct page *page;
2281
2282         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2283                                        vmcb->control.exit_info_1,
2284                                        vmcb->control.exit_info_2,
2285                                        vmcb->control.exit_int_info,
2286                                        vmcb->control.exit_int_info_err,
2287                                        KVM_ISA_SVM);
2288
2289         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2290         if (!nested_vmcb)
2291                 return 1;
2292
2293         /* Exit Guest-Mode */
2294         leave_guest_mode(&svm->vcpu);
2295         svm->nested.vmcb = 0;
2296
2297         /* Give the current vmcb to the guest */
2298         disable_gif(svm);
2299
2300         nested_vmcb->save.es     = vmcb->save.es;
2301         nested_vmcb->save.cs     = vmcb->save.cs;
2302         nested_vmcb->save.ss     = vmcb->save.ss;
2303         nested_vmcb->save.ds     = vmcb->save.ds;
2304         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2305         nested_vmcb->save.idtr   = vmcb->save.idtr;
2306         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2307         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2308         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2309         nested_vmcb->save.cr2    = vmcb->save.cr2;
2310         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2311         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2312         nested_vmcb->save.rip    = vmcb->save.rip;
2313         nested_vmcb->save.rsp    = vmcb->save.rsp;
2314         nested_vmcb->save.rax    = vmcb->save.rax;
2315         nested_vmcb->save.dr7    = vmcb->save.dr7;
2316         nested_vmcb->save.dr6    = vmcb->save.dr6;
2317         nested_vmcb->save.cpl    = vmcb->save.cpl;
2318
2319         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2320         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2321         nested_vmcb->control.int_state         = vmcb->control.int_state;
2322         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2323         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2324         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2325         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2326         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2327         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2328         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2329
2330         /*
2331          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2332          * to make sure that we do not lose injected events. So check event_inj
2333          * here and copy it to exit_int_info if it is valid.
2334          * Exit_int_info and event_inj can't be both valid because the case
2335          * below only happens on a VMRUN instruction intercept which has
2336          * no valid exit_int_info set.
2337          */
2338         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2339                 struct vmcb_control_area *nc = &nested_vmcb->control;
2340
2341                 nc->exit_int_info     = vmcb->control.event_inj;
2342                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2343         }
2344
2345         nested_vmcb->control.tlb_ctl           = 0;
2346         nested_vmcb->control.event_inj         = 0;
2347         nested_vmcb->control.event_inj_err     = 0;
2348
2349         /* We always set V_INTR_MASKING and remember the old value in hflags */
2350         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2351                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2352
2353         /* Restore the original control entries */
2354         copy_vmcb_control_area(vmcb, hsave);
2355
2356         kvm_clear_exception_queue(&svm->vcpu);
2357         kvm_clear_interrupt_queue(&svm->vcpu);
2358
2359         svm->nested.nested_cr3 = 0;
2360
2361         /* Restore selected save entries */
2362         svm->vmcb->save.es = hsave->save.es;
2363         svm->vmcb->save.cs = hsave->save.cs;
2364         svm->vmcb->save.ss = hsave->save.ss;
2365         svm->vmcb->save.ds = hsave->save.ds;
2366         svm->vmcb->save.gdtr = hsave->save.gdtr;
2367         svm->vmcb->save.idtr = hsave->save.idtr;
2368         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2369         svm_set_efer(&svm->vcpu, hsave->save.efer);
2370         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2371         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2372         if (npt_enabled) {
2373                 svm->vmcb->save.cr3 = hsave->save.cr3;
2374                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2375         } else {
2376                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2377         }
2378         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2379         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2380         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2381         svm->vmcb->save.dr7 = 0;
2382         svm->vmcb->save.cpl = 0;
2383         svm->vmcb->control.exit_int_info = 0;
2384
2385         mark_all_dirty(svm->vmcb);
2386
2387         nested_svm_unmap(page);
2388
2389         nested_svm_uninit_mmu_context(&svm->vcpu);
2390         kvm_mmu_reset_context(&svm->vcpu);
2391         kvm_mmu_load(&svm->vcpu);
2392
2393         return 0;
2394 }
2395
2396 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2397 {
2398         /*
2399          * This function merges the msr permission bitmaps of kvm and the
2400          * nested vmcb. It is optimized in that it only merges the parts where
2401          * the kvm msr permission bitmap may contain zero bits
2402          */
2403         int i;
2404
2405         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2406                 return true;
2407
2408         for (i = 0; i < MSRPM_OFFSETS; i++) {
2409                 u32 value, p;
2410                 u64 offset;
2411
2412                 if (msrpm_offsets[i] == 0xffffffff)
2413                         break;
2414
2415                 p      = msrpm_offsets[i];
2416                 offset = svm->nested.vmcb_msrpm + (p * 4);
2417
2418                 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2419                         return false;
2420
2421                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2422         }
2423
2424         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2425
2426         return true;
2427 }
2428
2429 static bool nested_vmcb_checks(struct vmcb *vmcb)
2430 {
2431         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2432                 return false;
2433
2434         if (vmcb->control.asid == 0)
2435                 return false;
2436
2437         if (vmcb->control.nested_ctl && !npt_enabled)
2438                 return false;
2439
2440         return true;
2441 }
2442
2443 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2444 {
2445         struct vmcb *nested_vmcb;
2446         struct vmcb *hsave = svm->nested.hsave;
2447         struct vmcb *vmcb = svm->vmcb;
2448         struct page *page;
2449         u64 vmcb_gpa;
2450
2451         vmcb_gpa = svm->vmcb->save.rax;
2452
2453         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2454         if (!nested_vmcb)
2455                 return false;
2456
2457         if (!nested_vmcb_checks(nested_vmcb)) {
2458                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2459                 nested_vmcb->control.exit_code_hi = 0;
2460                 nested_vmcb->control.exit_info_1  = 0;
2461                 nested_vmcb->control.exit_info_2  = 0;
2462
2463                 nested_svm_unmap(page);
2464
2465                 return false;
2466         }
2467
2468         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2469                                nested_vmcb->save.rip,
2470                                nested_vmcb->control.int_ctl,
2471                                nested_vmcb->control.event_inj,
2472                                nested_vmcb->control.nested_ctl);
2473
2474         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2475                                     nested_vmcb->control.intercept_cr >> 16,
2476                                     nested_vmcb->control.intercept_exceptions,
2477                                     nested_vmcb->control.intercept);
2478
2479         /* Clear internal status */
2480         kvm_clear_exception_queue(&svm->vcpu);
2481         kvm_clear_interrupt_queue(&svm->vcpu);
2482
2483         /*
2484          * Save the old vmcb, so we don't need to pick what we save, but can
2485          * restore everything when a VMEXIT occurs
2486          */
2487         hsave->save.es     = vmcb->save.es;
2488         hsave->save.cs     = vmcb->save.cs;
2489         hsave->save.ss     = vmcb->save.ss;
2490         hsave->save.ds     = vmcb->save.ds;
2491         hsave->save.gdtr   = vmcb->save.gdtr;
2492         hsave->save.idtr   = vmcb->save.idtr;
2493         hsave->save.efer   = svm->vcpu.arch.efer;
2494         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2495         hsave->save.cr4    = svm->vcpu.arch.cr4;
2496         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2497         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2498         hsave->save.rsp    = vmcb->save.rsp;
2499         hsave->save.rax    = vmcb->save.rax;
2500         if (npt_enabled)
2501                 hsave->save.cr3    = vmcb->save.cr3;
2502         else
2503                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2504
2505         copy_vmcb_control_area(hsave, vmcb);
2506
2507         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2508                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2509         else
2510                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2511
2512         if (nested_vmcb->control.nested_ctl) {
2513                 kvm_mmu_unload(&svm->vcpu);
2514                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2515                 nested_svm_init_mmu_context(&svm->vcpu);
2516         }
2517
2518         /* Load the nested guest state */
2519         svm->vmcb->save.es = nested_vmcb->save.es;
2520         svm->vmcb->save.cs = nested_vmcb->save.cs;
2521         svm->vmcb->save.ss = nested_vmcb->save.ss;
2522         svm->vmcb->save.ds = nested_vmcb->save.ds;
2523         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2524         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2525         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2526         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2527         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2528         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2529         if (npt_enabled) {
2530                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2531                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2532         } else
2533                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2534
2535         /* Guest paging mode is active - reset mmu */
2536         kvm_mmu_reset_context(&svm->vcpu);
2537
2538         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2539         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2540         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2541         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2542
2543         /* In case we don't even reach vcpu_run, the fields are not updated */
2544         svm->vmcb->save.rax = nested_vmcb->save.rax;
2545         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2546         svm->vmcb->save.rip = nested_vmcb->save.rip;
2547         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2548         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2549         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2550
2551         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2552         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2553
2554         /* cache intercepts */
2555         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2556         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2557         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2558         svm->nested.intercept            = nested_vmcb->control.intercept;
2559
2560         svm_flush_tlb(&svm->vcpu);
2561         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2562         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2563                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2564         else
2565                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2566
2567         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2568                 /* We only want the cr8 intercept bits of the guest */
2569                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2570                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2571         }
2572
2573         /* We don't want to see VMMCALLs from a nested guest */
2574         clr_intercept(svm, INTERCEPT_VMMCALL);
2575
2576         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2577         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2578         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2579         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2580         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2581         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2582
2583         nested_svm_unmap(page);
2584
2585         /* Enter Guest-Mode */
2586         enter_guest_mode(&svm->vcpu);
2587
2588         /*
2589          * Merge guest and host intercepts - must be called  with vcpu in
2590          * guest-mode to take affect here
2591          */
2592         recalc_intercepts(svm);
2593
2594         svm->nested.vmcb = vmcb_gpa;
2595
2596         enable_gif(svm);
2597
2598         mark_all_dirty(svm->vmcb);
2599
2600         return true;
2601 }
2602
2603 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2604 {
2605         to_vmcb->save.fs = from_vmcb->save.fs;
2606         to_vmcb->save.gs = from_vmcb->save.gs;
2607         to_vmcb->save.tr = from_vmcb->save.tr;
2608         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2609         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2610         to_vmcb->save.star = from_vmcb->save.star;
2611         to_vmcb->save.lstar = from_vmcb->save.lstar;
2612         to_vmcb->save.cstar = from_vmcb->save.cstar;
2613         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2614         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2615         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2616         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2617 }
2618
2619 static int vmload_interception(struct vcpu_svm *svm)
2620 {
2621         struct vmcb *nested_vmcb;
2622         struct page *page;
2623
2624         if (nested_svm_check_permissions(svm))
2625                 return 1;
2626
2627         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2628         if (!nested_vmcb)
2629                 return 1;
2630
2631         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2632         skip_emulated_instruction(&svm->vcpu);
2633
2634         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2635         nested_svm_unmap(page);
2636
2637         return 1;
2638 }
2639
2640 static int vmsave_interception(struct vcpu_svm *svm)
2641 {
2642         struct vmcb *nested_vmcb;
2643         struct page *page;
2644
2645         if (nested_svm_check_permissions(svm))
2646                 return 1;
2647
2648         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2649         if (!nested_vmcb)
2650                 return 1;
2651
2652         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2653         skip_emulated_instruction(&svm->vcpu);
2654
2655         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2656         nested_svm_unmap(page);
2657
2658         return 1;
2659 }
2660
2661 static int vmrun_interception(struct vcpu_svm *svm)
2662 {
2663         if (nested_svm_check_permissions(svm))
2664                 return 1;
2665
2666         /* Save rip after vmrun instruction */
2667         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2668
2669         if (!nested_svm_vmrun(svm))
2670                 return 1;
2671
2672         if (!nested_svm_vmrun_msrpm(svm))
2673                 goto failed;
2674
2675         return 1;
2676
2677 failed:
2678
2679         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2680         svm->vmcb->control.exit_code_hi = 0;
2681         svm->vmcb->control.exit_info_1  = 0;
2682         svm->vmcb->control.exit_info_2  = 0;
2683
2684         nested_svm_vmexit(svm);
2685
2686         return 1;
2687 }
2688
2689 static int stgi_interception(struct vcpu_svm *svm)
2690 {
2691         if (nested_svm_check_permissions(svm))
2692                 return 1;
2693
2694         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2695         skip_emulated_instruction(&svm->vcpu);
2696         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2697
2698         enable_gif(svm);
2699
2700         return 1;
2701 }
2702
2703 static int clgi_interception(struct vcpu_svm *svm)
2704 {
2705         if (nested_svm_check_permissions(svm))
2706                 return 1;
2707
2708         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2709         skip_emulated_instruction(&svm->vcpu);
2710
2711         disable_gif(svm);
2712
2713         /* After a CLGI no interrupts should come */
2714         svm_clear_vintr(svm);
2715         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2716
2717         mark_dirty(svm->vmcb, VMCB_INTR);
2718
2719         return 1;
2720 }
2721
2722 static int invlpga_interception(struct vcpu_svm *svm)
2723 {
2724         struct kvm_vcpu *vcpu = &svm->vcpu;
2725
2726         trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2727                           vcpu->arch.regs[VCPU_REGS_RAX]);
2728
2729         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2730         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2731
2732         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2733         skip_emulated_instruction(&svm->vcpu);
2734         return 1;
2735 }
2736
2737 static int skinit_interception(struct vcpu_svm *svm)
2738 {
2739         trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2740
2741         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2742         return 1;
2743 }
2744
2745 static int xsetbv_interception(struct vcpu_svm *svm)
2746 {
2747         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2748         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2749
2750         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2751                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2752                 skip_emulated_instruction(&svm->vcpu);
2753         }
2754
2755         return 1;
2756 }
2757
2758 static int invalid_op_interception(struct vcpu_svm *svm)
2759 {
2760         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2761         return 1;
2762 }
2763
2764 static int task_switch_interception(struct vcpu_svm *svm)
2765 {
2766         u16 tss_selector;
2767         int reason;
2768         int int_type = svm->vmcb->control.exit_int_info &
2769                 SVM_EXITINTINFO_TYPE_MASK;
2770         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2771         uint32_t type =
2772                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2773         uint32_t idt_v =
2774                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2775         bool has_error_code = false;
2776         u32 error_code = 0;
2777
2778         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2779
2780         if (svm->vmcb->control.exit_info_2 &
2781             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2782                 reason = TASK_SWITCH_IRET;
2783         else if (svm->vmcb->control.exit_info_2 &
2784                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2785                 reason = TASK_SWITCH_JMP;
2786         else if (idt_v)
2787                 reason = TASK_SWITCH_GATE;
2788         else
2789                 reason = TASK_SWITCH_CALL;
2790
2791         if (reason == TASK_SWITCH_GATE) {
2792                 switch (type) {
2793                 case SVM_EXITINTINFO_TYPE_NMI:
2794                         svm->vcpu.arch.nmi_injected = false;
2795                         break;
2796                 case SVM_EXITINTINFO_TYPE_EXEPT:
2797                         if (svm->vmcb->control.exit_info_2 &
2798                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2799                                 has_error_code = true;
2800                                 error_code =
2801                                         (u32)svm->vmcb->control.exit_info_2;
2802                         }
2803                         kvm_clear_exception_queue(&svm->vcpu);
2804                         break;
2805                 case SVM_EXITINTINFO_TYPE_INTR:
2806                         kvm_clear_interrupt_queue(&svm->vcpu);
2807                         break;
2808                 default:
2809                         break;
2810                 }
2811         }
2812
2813         if (reason != TASK_SWITCH_GATE ||
2814             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2815             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2816              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2817                 skip_emulated_instruction(&svm->vcpu);
2818
2819         if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2820                 int_vec = -1;
2821
2822         if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2823                                 has_error_code, error_code) == EMULATE_FAIL) {
2824                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2825                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2826                 svm->vcpu.run->internal.ndata = 0;
2827                 return 0;
2828         }
2829         return 1;
2830 }
2831
2832 static int cpuid_interception(struct vcpu_svm *svm)
2833 {
2834         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2835         kvm_emulate_cpuid(&svm->vcpu);
2836         return 1;
2837 }
2838
2839 static int iret_interception(struct vcpu_svm *svm)
2840 {
2841         ++svm->vcpu.stat.nmi_window_exits;
2842         clr_intercept(svm, INTERCEPT_IRET);
2843         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2844         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2845         return 1;
2846 }
2847
2848 static int invlpg_interception(struct vcpu_svm *svm)
2849 {
2850         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2851                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2852
2853         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2854         skip_emulated_instruction(&svm->vcpu);
2855         return 1;
2856 }
2857
2858 static int emulate_on_interception(struct vcpu_svm *svm)
2859 {
2860         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2861 }
2862
2863 static int rdpmc_interception(struct vcpu_svm *svm)
2864 {
2865         int err;
2866
2867         if (!static_cpu_has(X86_FEATURE_NRIPS))
2868                 return emulate_on_interception(svm);
2869
2870         err = kvm_rdpmc(&svm->vcpu);
2871         kvm_complete_insn_gp(&svm->vcpu, err);
2872
2873         return 1;
2874 }
2875
2876 bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2877 {
2878         unsigned long cr0 = svm->vcpu.arch.cr0;
2879         bool ret = false;
2880         u64 intercept;
2881
2882         intercept = svm->nested.intercept;
2883
2884         if (!is_guest_mode(&svm->vcpu) ||
2885             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2886                 return false;
2887
2888         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2889         val &= ~SVM_CR0_SELECTIVE_MASK;
2890
2891         if (cr0 ^ val) {
2892                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2893                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2894         }
2895
2896         return ret;
2897 }
2898
2899 #define CR_VALID (1ULL << 63)
2900
2901 static int cr_interception(struct vcpu_svm *svm)
2902 {
2903         int reg, cr;
2904         unsigned long val;
2905         int err;
2906
2907         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2908                 return emulate_on_interception(svm);
2909
2910         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2911                 return emulate_on_interception(svm);
2912
2913         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2914         cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2915
2916         err = 0;
2917         if (cr >= 16) { /* mov to cr */
2918                 cr -= 16;
2919                 val = kvm_register_read(&svm->vcpu, reg);
2920                 switch (cr) {
2921                 case 0:
2922                         if (!check_selective_cr0_intercepted(svm, val))
2923                                 err = kvm_set_cr0(&svm->vcpu, val);
2924                         else
2925                                 return 1;
2926
2927                         break;
2928                 case 3:
2929                         err = kvm_set_cr3(&svm->vcpu, val);
2930                         break;
2931                 case 4:
2932                         err = kvm_set_cr4(&svm->vcpu, val);
2933                         break;
2934                 case 8:
2935                         err = kvm_set_cr8(&svm->vcpu, val);
2936                         break;
2937                 default:
2938                         WARN(1, "unhandled write to CR%d", cr);
2939                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2940                         return 1;
2941                 }
2942         } else { /* mov from cr */
2943                 switch (cr) {
2944                 case 0:
2945                         val = kvm_read_cr0(&svm->vcpu);
2946                         break;
2947                 case 2:
2948                         val = svm->vcpu.arch.cr2;
2949                         break;
2950                 case 3:
2951                         val = kvm_read_cr3(&svm->vcpu);
2952                         break;
2953                 case 4:
2954                         val = kvm_read_cr4(&svm->vcpu);
2955                         break;
2956                 case 8:
2957                         val = kvm_get_cr8(&svm->vcpu);
2958                         break;
2959                 default:
2960                         WARN(1, "unhandled read from CR%d", cr);
2961                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2962                         return 1;
2963                 }
2964                 kvm_register_write(&svm->vcpu, reg, val);
2965         }
2966         kvm_complete_insn_gp(&svm->vcpu, err);
2967
2968         return 1;
2969 }
2970
2971 static int dr_interception(struct vcpu_svm *svm)
2972 {
2973         int reg, dr;
2974         unsigned long val;
2975         int err;
2976
2977         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2978                 return emulate_on_interception(svm);
2979
2980         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2981         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2982
2983         if (dr >= 16) { /* mov to DRn */
2984                 val = kvm_register_read(&svm->vcpu, reg);
2985                 kvm_set_dr(&svm->vcpu, dr - 16, val);
2986         } else {
2987                 err = kvm_get_dr(&svm->vcpu, dr, &val);
2988                 if (!err)
2989                         kvm_register_write(&svm->vcpu, reg, val);
2990         }
2991
2992         skip_emulated_instruction(&svm->vcpu);
2993
2994         return 1;
2995 }
2996
2997 static int cr8_write_interception(struct vcpu_svm *svm)
2998 {
2999         struct kvm_run *kvm_run = svm->vcpu.run;
3000         int r;
3001
3002         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3003         /* instruction emulation calls kvm_set_cr8() */
3004         r = cr_interception(svm);
3005         if (irqchip_in_kernel(svm->vcpu.kvm))
3006                 return r;
3007         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3008                 return r;
3009         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3010         return 0;
3011 }
3012
3013 u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3014 {
3015         struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3016         return vmcb->control.tsc_offset +
3017                 svm_scale_tsc(vcpu, host_tsc);
3018 }
3019
3020 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
3021 {
3022         struct vcpu_svm *svm = to_svm(vcpu);
3023
3024         switch (ecx) {
3025         case MSR_IA32_TSC: {
3026                 *data = svm->vmcb->control.tsc_offset +
3027                         svm_scale_tsc(vcpu, native_read_tsc());
3028
3029                 break;
3030         }
3031         case MSR_STAR:
3032                 *data = svm->vmcb->save.star;
3033                 break;
3034 #ifdef CONFIG_X86_64
3035         case MSR_LSTAR:
3036                 *data = svm->vmcb->save.lstar;
3037                 break;
3038         case MSR_CSTAR:
3039                 *data = svm->vmcb->save.cstar;
3040                 break;
3041         case MSR_KERNEL_GS_BASE:
3042                 *data = svm->vmcb->save.kernel_gs_base;
3043                 break;
3044         case MSR_SYSCALL_MASK:
3045                 *data = svm->vmcb->save.sfmask;
3046                 break;
3047 #endif
3048         case MSR_IA32_SYSENTER_CS:
3049                 *data = svm->vmcb->save.sysenter_cs;
3050                 break;
3051         case MSR_IA32_SYSENTER_EIP:
3052                 *data = svm->sysenter_eip;
3053                 break;
3054         case MSR_IA32_SYSENTER_ESP:
3055                 *data = svm->sysenter_esp;
3056                 break;
3057         /*
3058          * Nobody will change the following 5 values in the VMCB so we can
3059          * safely return them on rdmsr. They will always be 0 until LBRV is
3060          * implemented.
3061          */
3062         case MSR_IA32_DEBUGCTLMSR:
3063                 *data = svm->vmcb->save.dbgctl;
3064                 break;
3065         case MSR_IA32_LASTBRANCHFROMIP:
3066                 *data = svm->vmcb->save.br_from;
3067                 break;
3068         case MSR_IA32_LASTBRANCHTOIP:
3069                 *data = svm->vmcb->save.br_to;
3070                 break;
3071         case MSR_IA32_LASTINTFROMIP:
3072                 *data = svm->vmcb->save.last_excp_from;
3073                 break;
3074         case MSR_IA32_LASTINTTOIP:
3075                 *data = svm->vmcb->save.last_excp_to;
3076                 break;
3077         case MSR_VM_HSAVE_PA:
3078                 *data = svm->nested.hsave_msr;
3079                 break;
3080         case MSR_VM_CR:
3081                 *data = svm->nested.vm_cr_msr;
3082                 break;
3083         case MSR_IA32_UCODE_REV:
3084                 *data = 0x01000065;
3085                 break;
3086         default:
3087                 return kvm_get_msr_common(vcpu, ecx, data);
3088         }
3089         return 0;
3090 }
3091
3092 static int rdmsr_interception(struct vcpu_svm *svm)
3093 {
3094         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3095         u64 data;
3096
3097         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
3098                 trace_kvm_msr_read_ex(ecx);
3099                 kvm_inject_gp(&svm->vcpu, 0);
3100         } else {
3101                 trace_kvm_msr_read(ecx, data);
3102
3103                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
3104                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
3105                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3106                 skip_emulated_instruction(&svm->vcpu);
3107         }
3108         return 1;
3109 }
3110
3111 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3112 {
3113         struct vcpu_svm *svm = to_svm(vcpu);
3114         int svm_dis, chg_mask;
3115
3116         if (data & ~SVM_VM_CR_VALID_MASK)
3117                 return 1;
3118
3119         chg_mask = SVM_VM_CR_VALID_MASK;
3120
3121         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3122                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3123
3124         svm->nested.vm_cr_msr &= ~chg_mask;
3125         svm->nested.vm_cr_msr |= (data & chg_mask);
3126
3127         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3128
3129         /* check for svm_disable while efer.svme is set */
3130         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3131                 return 1;
3132
3133         return 0;
3134 }
3135
3136 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3137 {
3138         struct vcpu_svm *svm = to_svm(vcpu);
3139
3140         u32 ecx = msr->index;
3141         u64 data = msr->data;
3142         switch (ecx) {
3143         case MSR_IA32_TSC:
3144                 kvm_write_tsc(vcpu, msr);
3145                 break;
3146         case MSR_STAR:
3147                 svm->vmcb->save.star = data;
3148                 break;
3149 #ifdef CONFIG_X86_64
3150         case MSR_LSTAR:
3151                 svm->vmcb->save.lstar = data;
3152                 break;
3153         case MSR_CSTAR:
3154                 svm->vmcb->save.cstar = data;
3155                 break;
3156         case MSR_KERNEL_GS_BASE:
3157                 svm->vmcb->save.kernel_gs_base = data;
3158                 break;
3159         case MSR_SYSCALL_MASK:
3160                 svm->vmcb->save.sfmask = data;
3161                 break;
3162 #endif
3163         case MSR_IA32_SYSENTER_CS:
3164                 svm->vmcb->save.sysenter_cs = data;
3165                 break;
3166         case MSR_IA32_SYSENTER_EIP:
3167                 svm->sysenter_eip = data;
3168                 svm->vmcb->save.sysenter_eip = data;
3169                 break;
3170         case MSR_IA32_SYSENTER_ESP:
3171                 svm->sysenter_esp = data;
3172                 svm->vmcb->save.sysenter_esp = data;
3173                 break;
3174         case MSR_IA32_DEBUGCTLMSR:
3175                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3176                         vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3177                                     __func__, data);
3178                         break;
3179                 }
3180                 if (data & DEBUGCTL_RESERVED_BITS)
3181                         return 1;
3182
3183                 svm->vmcb->save.dbgctl = data;
3184                 mark_dirty(svm->vmcb, VMCB_LBR);
3185                 if (data & (1ULL<<0))
3186                         svm_enable_lbrv(svm);
3187                 else
3188                         svm_disable_lbrv(svm);
3189                 break;
3190         case MSR_VM_HSAVE_PA:
3191                 svm->nested.hsave_msr = data;
3192                 break;
3193         case MSR_VM_CR:
3194                 return svm_set_vm_cr(vcpu, data);
3195         case MSR_VM_IGNNE:
3196                 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3197                 break;
3198         default:
3199                 return kvm_set_msr_common(vcpu, msr);
3200         }
3201         return 0;
3202 }
3203
3204 static int wrmsr_interception(struct vcpu_svm *svm)
3205 {
3206         struct msr_data msr;
3207         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3208         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3209                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3210
3211         msr.data = data;
3212         msr.index = ecx;
3213         msr.host_initiated = false;
3214
3215         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3216         if (svm_set_msr(&svm->vcpu, &msr)) {
3217                 trace_kvm_msr_write_ex(ecx, data);
3218                 kvm_inject_gp(&svm->vcpu, 0);
3219         } else {
3220                 trace_kvm_msr_write(ecx, data);
3221                 skip_emulated_instruction(&svm->vcpu);
3222         }
3223         return 1;
3224 }
3225
3226 static int msr_interception(struct vcpu_svm *svm)
3227 {
3228         if (svm->vmcb->control.exit_info_1)
3229                 return wrmsr_interception(svm);
3230         else
3231                 return rdmsr_interception(svm);
3232 }
3233
3234 static int interrupt_window_interception(struct vcpu_svm *svm)
3235 {
3236         struct kvm_run *kvm_run = svm->vcpu.run;
3237
3238         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3239         svm_clear_vintr(svm);
3240         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3241         mark_dirty(svm->vmcb, VMCB_INTR);
3242         ++svm->vcpu.stat.irq_window_exits;
3243         /*
3244          * If the user space waits to inject interrupts, exit as soon as
3245          * possible
3246          */
3247         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3248             kvm_run->request_interrupt_window &&
3249             !kvm_cpu_has_interrupt(&svm->vcpu)) {
3250                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3251                 return 0;
3252         }
3253
3254         return 1;
3255 }
3256
3257 static int pause_interception(struct vcpu_svm *svm)
3258 {
3259         kvm_vcpu_on_spin(&(svm->vcpu));
3260         return 1;
3261 }
3262
3263 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3264         [SVM_EXIT_READ_CR0]                     = cr_interception,
3265         [SVM_EXIT_READ_CR3]                     = cr_interception,
3266         [SVM_EXIT_READ_CR4]                     = cr_interception,
3267         [SVM_EXIT_READ_CR8]                     = cr_interception,
3268         [SVM_EXIT_CR0_SEL_WRITE]                = emulate_on_interception,
3269         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
3270         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
3271         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
3272         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
3273         [SVM_EXIT_READ_DR0]                     = dr_interception,
3274         [SVM_EXIT_READ_DR1]                     = dr_interception,
3275         [SVM_EXIT_READ_DR2]                     = dr_interception,
3276         [SVM_EXIT_READ_DR3]                     = dr_interception,
3277         [SVM_EXIT_READ_DR4]                     = dr_interception,
3278         [SVM_EXIT_READ_DR5]                     = dr_interception,
3279         [SVM_EXIT_READ_DR6]                     = dr_interception,
3280         [SVM_EXIT_READ_DR7]                     = dr_interception,
3281         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
3282         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
3283         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
3284         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
3285         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
3286         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
3287         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
3288         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
3289         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
3290         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
3291         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
3292         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
3293         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
3294         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
3295         [SVM_EXIT_INTR]                         = intr_interception,
3296         [SVM_EXIT_NMI]                          = nmi_interception,
3297         [SVM_EXIT_SMI]                          = nop_on_interception,
3298         [SVM_EXIT_INIT]                         = nop_on_interception,
3299         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
3300         [SVM_EXIT_RDPMC]                        = rdpmc_interception,
3301         [SVM_EXIT_CPUID]                        = cpuid_interception,
3302         [SVM_EXIT_IRET]                         = iret_interception,
3303         [SVM_EXIT_INVD]                         = emulate_on_interception,
3304         [SVM_EXIT_PAUSE]                        = pause_interception,
3305         [SVM_EXIT_HLT]                          = halt_interception,
3306         [SVM_EXIT_INVLPG]                       = invlpg_interception,
3307         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
3308         [SVM_EXIT_IOIO]                         = io_interception,
3309         [SVM_EXIT_MSR]                          = msr_interception,
3310         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
3311         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
3312         [SVM_EXIT_VMRUN]                        = vmrun_interception,
3313         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
3314         [SVM_EXIT_VMLOAD]                       = vmload_interception,
3315         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
3316         [SVM_EXIT_STGI]                         = stgi_interception,
3317         [SVM_EXIT_CLGI]                         = clgi_interception,
3318         [SVM_EXIT_SKINIT]                       = skinit_interception,
3319         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
3320         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
3321         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
3322         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
3323         [SVM_EXIT_NPF]                          = pf_interception,
3324 };
3325
3326 static void dump_vmcb(struct kvm_vcpu *vcpu)
3327 {
3328         struct vcpu_svm *svm = to_svm(vcpu);
3329         struct vmcb_control_area *control = &svm->vmcb->control;
3330         struct vmcb_save_area *save = &svm->vmcb->save;
3331
3332         pr_err("VMCB Control Area:\n");
3333         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3334         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3335         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3336         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3337         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3338         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3339         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3340         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3341         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3342         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3343         pr_err("%-20s%d\n", "asid:", control->asid);
3344         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3345         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3346         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3347         pr_err("%-20s%08x\n", "int_state:", control->int_state);
3348         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3349         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3350         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3351         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3352         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3353         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3354         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3355         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3356         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3357         pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3358         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3359         pr_err("VMCB State Save Area:\n");
3360         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3361                "es:",
3362                save->es.selector, save->es.attrib,
3363                save->es.limit, save->es.base);
3364         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3365                "cs:",
3366                save->cs.selector, save->cs.attrib,
3367                save->cs.limit, save->cs.base);
3368         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3369                "ss:",
3370                save->ss.selector, save->ss.attrib,
3371                save->ss.limit, save->ss.base);
3372         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3373                "ds:",
3374                save->ds.selector, save->ds.attrib,
3375                save->ds.limit, save->ds.base);
3376         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3377                "fs:",
3378                save->fs.selector, save->fs.attrib,
3379                save->fs.limit, save->fs.base);
3380         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3381                "gs:",
3382                save->gs.selector, save->gs.attrib,
3383                save->gs.limit, save->gs.base);
3384         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3385                "gdtr:",
3386                save->gdtr.selector, save->gdtr.attrib,
3387                save->gdtr.limit, save->gdtr.base);
3388         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3389                "ldtr:",
3390                save->ldtr.selector, save->ldtr.attrib,
3391                save->ldtr.limit, save->ldtr.base);
3392         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3393                "idtr:",
3394                save->idtr.selector, save->idtr.attrib,
3395                save->idtr.limit, save->idtr.base);
3396         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3397                "tr:",
3398                save->tr.selector, save->tr.attrib,
3399                save->tr.limit, save->tr.base);
3400         pr_err("cpl:            %d                efer:         %016llx\n",
3401                 save->cpl, save->efer);
3402         pr_err("%-15s %016llx %-13s %016llx\n",
3403                "cr0:", save->cr0, "cr2:", save->cr2);
3404         pr_err("%-15s %016llx %-13s %016llx\n",
3405                "cr3:", save->cr3, "cr4:", save->cr4);
3406         pr_err("%-15s %016llx %-13s %016llx\n",
3407                "dr6:", save->dr6, "dr7:", save->dr7);
3408         pr_err("%-15s %016llx %-13s %016llx\n",
3409                "rip:", save->rip, "rflags:", save->rflags);
3410         pr_err("%-15s %016llx %-13s %016llx\n",
3411                "rsp:", save->rsp, "rax:", save->rax);
3412         pr_err("%-15s %016llx %-13s %016llx\n",
3413                "star:", save->star, "lstar:", save->lstar);
3414         pr_err("%-15s %016llx %-13s %016llx\n",
3415                "cstar:", save->cstar, "sfmask:", save->sfmask);
3416         pr_err("%-15s %016llx %-13s %016llx\n",
3417                "kernel_gs_base:", save->kernel_gs_base,
3418                "sysenter_cs:", save->sysenter_cs);
3419         pr_err("%-15s %016llx %-13s %016llx\n",
3420                "sysenter_esp:", save->sysenter_esp,
3421                "sysenter_eip:", save->sysenter_eip);
3422         pr_err("%-15s %016llx %-13s %016llx\n",
3423                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3424         pr_err("%-15s %016llx %-13s %016llx\n",
3425                "br_from:", save->br_from, "br_to:", save->br_to);
3426         pr_err("%-15s %016llx %-13s %016llx\n",
3427                "excp_from:", save->last_excp_from,
3428                "excp_to:", save->last_excp_to);
3429 }
3430
3431 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3432 {
3433         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3434
3435         *info1 = control->exit_info_1;
3436         *info2 = control->exit_info_2;
3437 }
3438
3439 static int handle_exit(struct kvm_vcpu *vcpu)
3440 {
3441         struct vcpu_svm *svm = to_svm(vcpu);
3442         struct kvm_run *kvm_run = vcpu->run;
3443         u32 exit_code = svm->vmcb->control.exit_code;
3444
3445         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3446                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3447         if (npt_enabled)
3448                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3449
3450         if (unlikely(svm->nested.exit_required)) {
3451                 nested_svm_vmexit(svm);
3452                 svm->nested.exit_required = false;
3453
3454                 return 1;
3455         }
3456
3457         if (is_guest_mode(vcpu)) {
3458                 int vmexit;
3459
3460                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3461                                         svm->vmcb->control.exit_info_1,
3462                                         svm->vmcb->control.exit_info_2,
3463                                         svm->vmcb->control.exit_int_info,
3464                                         svm->vmcb->control.exit_int_info_err,
3465                                         KVM_ISA_SVM);
3466
3467                 vmexit = nested_svm_exit_special(svm);
3468
3469                 if (vmexit == NESTED_EXIT_CONTINUE)
3470                         vmexit = nested_svm_exit_handled(svm);
3471
3472                 if (vmexit == NESTED_EXIT_DONE)
3473                         return 1;
3474         }
3475
3476         svm_complete_interrupts(svm);
3477
3478         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3479                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3480                 kvm_run->fail_entry.hardware_entry_failure_reason
3481                         = svm->vmcb->control.exit_code;
3482                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3483                 dump_vmcb(vcpu);
3484                 return 0;
3485         }
3486
3487         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3488             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3489             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3490             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3491                 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3492                        "exit_code 0x%x\n",
3493                        __func__, svm->vmcb->control.exit_int_info,
3494                        exit_code);
3495
3496         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3497             || !svm_exit_handlers[exit_code]) {
3498                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3499                 kvm_run->hw.hardware_exit_reason = exit_code;
3500                 return 0;
3501         }
3502
3503         return svm_exit_handlers[exit_code](svm);
3504 }
3505
3506 static void reload_tss(struct kvm_vcpu *vcpu)
3507 {
3508         int cpu = raw_smp_processor_id();
3509
3510         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3511         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3512         load_TR_desc();
3513 }
3514
3515 static void pre_svm_run(struct vcpu_svm *svm)
3516 {
3517         int cpu = raw_smp_processor_id();
3518
3519         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3520
3521         /* FIXME: handle wraparound of asid_generation */
3522         if (svm->asid_generation != sd->asid_generation)
3523                 new_asid(svm, sd);
3524 }
3525
3526 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3527 {
3528         struct vcpu_svm *svm = to_svm(vcpu);
3529
3530         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3531         vcpu->arch.hflags |= HF_NMI_MASK;
3532         set_intercept(svm, INTERCEPT_IRET);
3533         ++vcpu->stat.nmi_injections;
3534 }
3535
3536 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3537 {
3538         struct vmcb_control_area *control;
3539
3540         control = &svm->vmcb->control;
3541         control->int_vector = irq;
3542         control->int_ctl &= ~V_INTR_PRIO_MASK;
3543         control->int_ctl |= V_IRQ_MASK |
3544                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3545         mark_dirty(svm->vmcb, VMCB_INTR);
3546 }
3547
3548 static void svm_set_irq(struct kvm_vcpu *vcpu)
3549 {
3550         struct vcpu_svm *svm = to_svm(vcpu);
3551
3552         BUG_ON(!(gif_set(svm)));
3553
3554         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3555         ++vcpu->stat.irq_injections;
3556
3557         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3558                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3559 }
3560
3561 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3562 {
3563         struct vcpu_svm *svm = to_svm(vcpu);
3564
3565         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3566                 return;
3567
3568         clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3569
3570         if (irr == -1)
3571                 return;
3572
3573         if (tpr >= irr)
3574                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3575 }
3576
3577 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3578 {
3579         return;
3580 }
3581
3582 static int svm_vm_has_apicv(struct kvm *kvm)
3583 {
3584         return 0;
3585 }
3586
3587 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3588 {
3589         return;
3590 }
3591
3592 static void svm_hwapic_isr_update(struct kvm *kvm, int isr)
3593 {
3594         return;
3595 }
3596
3597 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3598 {
3599         return;
3600 }
3601
3602 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3603 {
3604         struct vcpu_svm *svm = to_svm(vcpu);
3605         struct vmcb *vmcb = svm->vmcb;
3606         int ret;
3607         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3608               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3609         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3610
3611         return ret;
3612 }
3613
3614 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3615 {
3616         struct vcpu_svm *svm = to_svm(vcpu);
3617
3618         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3619 }
3620
3621 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3622 {
3623         struct vcpu_svm *svm = to_svm(vcpu);
3624
3625         if (masked) {
3626                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3627                 set_intercept(svm, INTERCEPT_IRET);
3628         } else {
3629                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3630                 clr_intercept(svm, INTERCEPT_IRET);
3631         }
3632 }
3633
3634 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3635 {
3636         struct vcpu_svm *svm = to_svm(vcpu);
3637         struct vmcb *vmcb = svm->vmcb;
3638         int ret;
3639
3640         if (!gif_set(svm) ||
3641              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3642                 return 0;
3643
3644         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3645
3646         if (is_guest_mode(vcpu))
3647                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3648
3649         return ret;
3650 }
3651
3652 static int enable_irq_window(struct kvm_vcpu *vcpu)
3653 {
3654         struct vcpu_svm *svm = to_svm(vcpu);
3655
3656         /*
3657          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3658          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3659          * get that intercept, this function will be called again though and
3660          * we'll get the vintr intercept.
3661          */
3662         if (gif_set(svm) && nested_svm_intr(svm)) {
3663                 svm_set_vintr(svm);
3664                 svm_inject_irq(svm, 0x0);
3665         }
3666         return 0;
3667 }
3668
3669 static int enable_nmi_window(struct kvm_vcpu *vcpu)
3670 {
3671         struct vcpu_svm *svm = to_svm(vcpu);
3672
3673         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3674             == HF_NMI_MASK)
3675                 return 0; /* IRET will cause a vm exit */
3676
3677         /*
3678          * Something prevents NMI from been injected. Single step over possible
3679          * problem (IRET or exception injection or interrupt shadow)
3680          */
3681         svm->nmi_singlestep = true;
3682         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3683         update_db_bp_intercept(vcpu);
3684         return 0;
3685 }
3686
3687 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3688 {
3689         return 0;
3690 }
3691
3692 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3693 {
3694         struct vcpu_svm *svm = to_svm(vcpu);
3695
3696         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3697                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3698         else
3699                 svm->asid_generation--;
3700 }
3701
3702 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3703 {
3704 }
3705
3706 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3707 {
3708         struct vcpu_svm *svm = to_svm(vcpu);
3709
3710         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3711                 return;
3712
3713         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3714                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3715                 kvm_set_cr8(vcpu, cr8);
3716         }
3717 }
3718
3719 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3720 {
3721         struct vcpu_svm *svm = to_svm(vcpu);
3722         u64 cr8;
3723
3724         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3725                 return;
3726
3727         cr8 = kvm_get_cr8(vcpu);
3728         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3729         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3730 }
3731
3732 static void svm_complete_interrupts(struct vcpu_svm *svm)
3733 {
3734         u8 vector;
3735         int type;
3736         u32 exitintinfo = svm->vmcb->control.exit_int_info;
3737         unsigned int3_injected = svm->int3_injected;
3738
3739         svm->int3_injected = 0;
3740
3741         /*
3742          * If we've made progress since setting HF_IRET_MASK, we've
3743          * executed an IRET and can allow NMI injection.
3744          */
3745         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3746             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3747                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3748                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3749         }
3750
3751         svm->vcpu.arch.nmi_injected = false;
3752         kvm_clear_exception_queue(&svm->vcpu);
3753         kvm_clear_interrupt_queue(&svm->vcpu);
3754
3755         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3756                 return;
3757
3758         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3759
3760         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3761         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3762
3763         switch (type) {
3764         case SVM_EXITINTINFO_TYPE_NMI:
3765                 svm->vcpu.arch.nmi_injected = true;
3766                 break;
3767         case SVM_EXITINTINFO_TYPE_EXEPT:
3768                 /*
3769                  * In case of software exceptions, do not reinject the vector,
3770                  * but re-execute the instruction instead. Rewind RIP first
3771                  * if we emulated INT3 before.
3772                  */
3773                 if (kvm_exception_is_soft(vector)) {
3774                         if (vector == BP_VECTOR && int3_injected &&
3775                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3776                                 kvm_rip_write(&svm->vcpu,
3777                                               kvm_rip_read(&svm->vcpu) -
3778                                               int3_injected);
3779                         break;
3780                 }
3781                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3782                         u32 err = svm->vmcb->control.exit_int_info_err;
3783                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
3784
3785                 } else
3786                         kvm_requeue_exception(&svm->vcpu, vector);
3787                 break;
3788         case SVM_EXITINTINFO_TYPE_INTR:
3789                 kvm_queue_interrupt(&svm->vcpu, vector, false);
3790                 break;
3791         default:
3792                 break;
3793         }
3794 }
3795
3796 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3797 {
3798         struct vcpu_svm *svm = to_svm(vcpu);
3799         struct vmcb_control_area *control = &svm->vmcb->control;
3800
3801         control->exit_int_info = control->event_inj;
3802         control->exit_int_info_err = control->event_inj_err;
3803         control->event_inj = 0;
3804         svm_complete_interrupts(svm);
3805 }
3806
3807 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3808 {
3809         struct vcpu_svm *svm = to_svm(vcpu);
3810
3811         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3812         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3813         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3814
3815         /*
3816          * A vmexit emulation is required before the vcpu can be executed
3817          * again.
3818          */
3819         if (unlikely(svm->nested.exit_required))
3820                 return;
3821
3822         pre_svm_run(svm);
3823
3824         sync_lapic_to_cr8(vcpu);
3825
3826         svm->vmcb->save.cr2 = vcpu->arch.cr2;
3827
3828         clgi();
3829
3830         local_irq_enable();
3831
3832         asm volatile (
3833                 "push %%" _ASM_BP "; \n\t"
3834                 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
3835                 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
3836                 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
3837                 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
3838                 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
3839                 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
3840 #ifdef CONFIG_X86_64
3841                 "mov %c[r8](%[svm]),  %%r8  \n\t"
3842                 "mov %c[r9](%[svm]),  %%r9  \n\t"
3843                 "mov %c[r10](%[svm]), %%r10 \n\t"
3844                 "mov %c[r11](%[svm]), %%r11 \n\t"
3845                 "mov %c[r12](%[svm]), %%r12 \n\t"
3846                 "mov %c[r13](%[svm]), %%r13 \n\t"
3847                 "mov %c[r14](%[svm]), %%r14 \n\t"
3848                 "mov %c[r15](%[svm]), %%r15 \n\t"
3849 #endif
3850
3851                 /* Enter guest mode */
3852                 "push %%" _ASM_AX " \n\t"
3853                 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
3854                 __ex(SVM_VMLOAD) "\n\t"
3855                 __ex(SVM_VMRUN) "\n\t"
3856                 __ex(SVM_VMSAVE) "\n\t"
3857                 "pop %%" _ASM_AX " \n\t"
3858
3859                 /* Save guest registers, load host registers */
3860                 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
3861                 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
3862                 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
3863                 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
3864                 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
3865                 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
3866 #ifdef CONFIG_X86_64
3867                 "mov %%r8,  %c[r8](%[svm]) \n\t"
3868                 "mov %%r9,  %c[r9](%[svm]) \n\t"
3869                 "mov %%r10, %c[r10](%[svm]) \n\t"
3870                 "mov %%r11, %c[r11](%[svm]) \n\t"
3871                 "mov %%r12, %c[r12](%[svm]) \n\t"
3872                 "mov %%r13, %c[r13](%[svm]) \n\t"
3873                 "mov %%r14, %c[r14](%[svm]) \n\t"
3874                 "mov %%r15, %c[r15](%[svm]) \n\t"
3875 #endif
3876                 "pop %%" _ASM_BP
3877                 :
3878                 : [svm]"a"(svm),
3879                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3880                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3881                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3882                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3883                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3884                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3885                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3886 #ifdef CONFIG_X86_64
3887                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3888                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3889                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3890                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3891                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3892                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3893                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3894                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3895 #endif
3896                 : "cc", "memory"
3897 #ifdef CONFIG_X86_64
3898                 , "rbx", "rcx", "rdx", "rsi", "rdi"
3899                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3900 #else
3901                 , "ebx", "ecx", "edx", "esi", "edi"
3902 #endif
3903                 );
3904
3905 #ifdef CONFIG_X86_64
3906         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3907 #else
3908         loadsegment(fs, svm->host.fs);
3909 #ifndef CONFIG_X86_32_LAZY_GS
3910         loadsegment(gs, svm->host.gs);
3911 #endif
3912 #endif
3913
3914         reload_tss(vcpu);
3915
3916         local_irq_disable();
3917
3918         vcpu->arch.cr2 = svm->vmcb->save.cr2;
3919         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3920         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3921         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3922
3923         trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
3924
3925         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3926                 kvm_before_handle_nmi(&svm->vcpu);
3927
3928         stgi();
3929
3930         /* Any pending NMI will happen here */
3931
3932         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3933                 kvm_after_handle_nmi(&svm->vcpu);
3934
3935         sync_cr8_to_lapic(vcpu);
3936
3937         svm->next_rip = 0;
3938
3939         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3940
3941         /* if exit due to PF check for async PF */
3942         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3943                 svm->apf_reason = kvm_read_and_reset_pf_reason();
3944
3945         if (npt_enabled) {
3946                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3947                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3948         }
3949
3950         /*
3951          * We need to handle MC intercepts here before the vcpu has a chance to
3952          * change the physical cpu
3953          */
3954         if (unlikely(svm->vmcb->control.exit_code ==
3955                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
3956                 svm_handle_mce(svm);
3957
3958         mark_all_clean(svm->vmcb);
3959 }
3960
3961 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3962 {
3963         struct vcpu_svm *svm = to_svm(vcpu);
3964
3965         svm->vmcb->save.cr3 = root;
3966         mark_dirty(svm->vmcb, VMCB_CR);
3967         svm_flush_tlb(vcpu);
3968 }
3969
3970 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3971 {
3972         struct vcpu_svm *svm = to_svm(vcpu);
3973
3974         svm->vmcb->control.nested_cr3 = root;
3975         mark_dirty(svm->vmcb, VMCB_NPT);
3976
3977         /* Also sync guest cr3 here in case we live migrate */
3978         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
3979         mark_dirty(svm->vmcb, VMCB_CR);
3980
3981         svm_flush_tlb(vcpu);
3982 }
3983
3984 static int is_disabled(void)
3985 {
3986         u64 vm_cr;
3987
3988         rdmsrl(MSR_VM_CR, vm_cr);
3989         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3990                 return 1;
3991
3992         return 0;
3993 }
3994
3995 static void
3996 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3997 {
3998         /*
3999          * Patch in the VMMCALL instruction:
4000          */
4001         hypercall[0] = 0x0f;
4002         hypercall[1] = 0x01;
4003         hypercall[2] = 0xd9;
4004 }
4005
4006 static void svm_check_processor_compat(void *rtn)
4007 {
4008         *(int *)rtn = 0;
4009 }
4010
4011 static bool svm_cpu_has_accelerated_tpr(void)
4012 {
4013         return false;
4014 }
4015
4016 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4017 {
4018         return 0;
4019 }
4020
4021 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4022 {
4023 }
4024
4025 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4026 {
4027         switch (func) {
4028         case 0x80000001:
4029                 if (nested)
4030                         entry->ecx |= (1 << 2); /* Set SVM bit */
4031                 break;
4032         case 0x8000000A:
4033                 entry->eax = 1; /* SVM revision 1 */
4034                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4035                                    ASID emulation to nested SVM */
4036                 entry->ecx = 0; /* Reserved */
4037                 entry->edx = 0; /* Per default do not support any
4038                                    additional features */
4039
4040                 /* Support next_rip if host supports it */
4041                 if (boot_cpu_has(X86_FEATURE_NRIPS))
4042                         entry->edx |= SVM_FEATURE_NRIP;
4043
4044                 /* Support NPT for the guest if enabled */
4045                 if (npt_enabled)
4046                         entry->edx |= SVM_FEATURE_NPT;
4047
4048                 break;
4049         }
4050 }
4051
4052 static int svm_get_lpage_level(void)
4053 {
4054         return PT_PDPE_LEVEL;
4055 }
4056
4057 static bool svm_rdtscp_supported(void)
4058 {
4059         return false;
4060 }
4061
4062 static bool svm_invpcid_supported(void)
4063 {
4064         return false;
4065 }
4066
4067 static bool svm_has_wbinvd_exit(void)
4068 {
4069         return true;
4070 }
4071
4072 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4073 {
4074         struct vcpu_svm *svm = to_svm(vcpu);
4075
4076         set_exception_intercept(svm, NM_VECTOR);
4077         update_cr0_intercept(svm);
4078 }
4079
4080 #define PRE_EX(exit)  { .exit_code = (exit), \
4081                         .stage = X86_ICPT_PRE_EXCEPT, }
4082 #define POST_EX(exit) { .exit_code = (exit), \
4083                         .stage = X86_ICPT_POST_EXCEPT, }
4084 #define POST_MEM(exit) { .exit_code = (exit), \
4085                         .stage = X86_ICPT_POST_MEMACCESS, }
4086
4087 static const struct __x86_intercept {
4088         u32 exit_code;
4089         enum x86_intercept_stage stage;
4090 } x86_intercept_map[] = {
4091         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
4092         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
4093         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
4094         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
4095         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
4096         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
4097         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
4098         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
4099         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
4100         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
4101         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
4102         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
4103         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
4104         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
4105         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
4106         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
4107         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
4108         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
4109         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
4110         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
4111         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
4112         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
4113         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
4114         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
4115         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
4116         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
4117         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
4118         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
4119         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
4120         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
4121         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
4122         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
4123         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
4124         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
4125         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
4126         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
4127         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
4128         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
4129         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
4130         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
4131         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
4132         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
4133         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
4134         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
4135         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
4136         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
4137 };
4138
4139 #undef PRE_EX
4140 #undef POST_EX
4141 #undef POST_MEM
4142
4143 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4144                                struct x86_instruction_info *info,
4145                                enum x86_intercept_stage stage)
4146 {
4147         struct vcpu_svm *svm = to_svm(vcpu);
4148         int vmexit, ret = X86EMUL_CONTINUE;
4149         struct __x86_intercept icpt_info;
4150         struct vmcb *vmcb = svm->vmcb;
4151
4152         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4153                 goto out;
4154
4155         icpt_info = x86_intercept_map[info->intercept];
4156
4157         if (stage != icpt_info.stage)
4158                 goto out;
4159
4160         switch (icpt_info.exit_code) {
4161         case SVM_EXIT_READ_CR0:
4162                 if (info->intercept == x86_intercept_cr_read)
4163                         icpt_info.exit_code += info->modrm_reg;
4164                 break;
4165         case SVM_EXIT_WRITE_CR0: {
4166                 unsigned long cr0, val;
4167                 u64 intercept;
4168
4169                 if (info->intercept == x86_intercept_cr_write)
4170                         icpt_info.exit_code += info->modrm_reg;
4171
4172                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0)
4173                         break;
4174
4175                 intercept = svm->nested.intercept;
4176
4177                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4178                         break;
4179
4180                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4181                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4182
4183                 if (info->intercept == x86_intercept_lmsw) {
4184                         cr0 &= 0xfUL;
4185                         val &= 0xfUL;
4186                         /* lmsw can't clear PE - catch this here */
4187                         if (cr0 & X86_CR0_PE)
4188                                 val |= X86_CR0_PE;
4189                 }
4190
4191                 if (cr0 ^ val)
4192                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4193
4194                 break;
4195         }
4196         case SVM_EXIT_READ_DR0:
4197         case SVM_EXIT_WRITE_DR0:
4198                 icpt_info.exit_code += info->modrm_reg;
4199                 break;
4200         case SVM_EXIT_MSR:
4201                 if (info->intercept == x86_intercept_wrmsr)
4202                         vmcb->control.exit_info_1 = 1;
4203                 else
4204                         vmcb->control.exit_info_1 = 0;
4205                 break;
4206         case SVM_EXIT_PAUSE:
4207                 /*
4208                  * We get this for NOP only, but pause
4209                  * is rep not, check this here
4210                  */
4211                 if (info->rep_prefix != REPE_PREFIX)
4212                         goto out;
4213         case SVM_EXIT_IOIO: {
4214                 u64 exit_info;
4215                 u32 bytes;
4216
4217                 exit_info = (vcpu->arch.regs[VCPU_REGS_RDX] & 0xffff) << 16;
4218
4219                 if (info->intercept == x86_intercept_in ||
4220                     info->intercept == x86_intercept_ins) {
4221                         exit_info |= SVM_IOIO_TYPE_MASK;
4222                         bytes = info->src_bytes;
4223                 } else {
4224                         bytes = info->dst_bytes;
4225                 }
4226
4227                 if (info->intercept == x86_intercept_outs ||
4228                     info->intercept == x86_intercept_ins)
4229                         exit_info |= SVM_IOIO_STR_MASK;
4230
4231                 if (info->rep_prefix)
4232                         exit_info |= SVM_IOIO_REP_MASK;
4233
4234                 bytes = min(bytes, 4u);
4235
4236                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4237
4238                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4239
4240                 vmcb->control.exit_info_1 = exit_info;
4241                 vmcb->control.exit_info_2 = info->next_rip;
4242
4243                 break;
4244         }
4245         default:
4246                 break;
4247         }
4248
4249         vmcb->control.next_rip  = info->next_rip;
4250         vmcb->control.exit_code = icpt_info.exit_code;
4251         vmexit = nested_svm_exit_handled(svm);
4252
4253         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4254                                            : X86EMUL_CONTINUE;
4255
4256 out:
4257         return ret;
4258 }
4259
4260 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4261 {
4262         local_irq_enable();
4263 }
4264
4265 static struct kvm_x86_ops svm_x86_ops = {
4266         .cpu_has_kvm_support = has_svm,
4267         .disabled_by_bios = is_disabled,
4268         .hardware_setup = svm_hardware_setup,
4269         .hardware_unsetup = svm_hardware_unsetup,
4270         .check_processor_compatibility = svm_check_processor_compat,
4271         .hardware_enable = svm_hardware_enable,
4272         .hardware_disable = svm_hardware_disable,
4273         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4274
4275         .vcpu_create = svm_create_vcpu,
4276         .vcpu_free = svm_free_vcpu,
4277         .vcpu_reset = svm_vcpu_reset,
4278
4279         .prepare_guest_switch = svm_prepare_guest_switch,
4280         .vcpu_load = svm_vcpu_load,
4281         .vcpu_put = svm_vcpu_put,
4282
4283         .update_db_bp_intercept = update_db_bp_intercept,
4284         .get_msr = svm_get_msr,
4285         .set_msr = svm_set_msr,
4286         .get_segment_base = svm_get_segment_base,
4287         .get_segment = svm_get_segment,
4288         .set_segment = svm_set_segment,
4289         .get_cpl = svm_get_cpl,
4290         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4291         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4292         .decache_cr3 = svm_decache_cr3,
4293         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4294         .set_cr0 = svm_set_cr0,
4295         .set_cr3 = svm_set_cr3,
4296         .set_cr4 = svm_set_cr4,
4297         .set_efer = svm_set_efer,
4298         .get_idt = svm_get_idt,
4299         .set_idt = svm_set_idt,
4300         .get_gdt = svm_get_gdt,
4301         .set_gdt = svm_set_gdt,
4302         .get_dr6 = svm_get_dr6,
4303         .set_dr6 = svm_set_dr6,
4304         .set_dr7 = svm_set_dr7,
4305         .cache_reg = svm_cache_reg,
4306         .get_rflags = svm_get_rflags,
4307         .set_rflags = svm_set_rflags,
4308         .fpu_activate = svm_fpu_activate,
4309         .fpu_deactivate = svm_fpu_deactivate,
4310
4311         .tlb_flush = svm_flush_tlb,
4312
4313         .run = svm_vcpu_run,
4314         .handle_exit = handle_exit,
4315         .skip_emulated_instruction = skip_emulated_instruction,
4316         .set_interrupt_shadow = svm_set_interrupt_shadow,
4317         .get_interrupt_shadow = svm_get_interrupt_shadow,
4318         .patch_hypercall = svm_patch_hypercall,
4319         .set_irq = svm_set_irq,
4320         .set_nmi = svm_inject_nmi,
4321         .queue_exception = svm_queue_exception,
4322         .cancel_injection = svm_cancel_injection,
4323         .interrupt_allowed = svm_interrupt_allowed,
4324         .nmi_allowed = svm_nmi_allowed,
4325         .get_nmi_mask = svm_get_nmi_mask,
4326         .set_nmi_mask = svm_set_nmi_mask,
4327         .enable_nmi_window = enable_nmi_window,
4328         .enable_irq_window = enable_irq_window,
4329         .update_cr8_intercept = update_cr8_intercept,
4330         .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4331         .vm_has_apicv = svm_vm_has_apicv,
4332         .load_eoi_exitmap = svm_load_eoi_exitmap,
4333         .hwapic_isr_update = svm_hwapic_isr_update,
4334         .sync_pir_to_irr = svm_sync_pir_to_irr,
4335
4336         .set_tss_addr = svm_set_tss_addr,
4337         .get_tdp_level = get_npt_level,
4338         .get_mt_mask = svm_get_mt_mask,
4339
4340         .get_exit_info = svm_get_exit_info,
4341
4342         .get_lpage_level = svm_get_lpage_level,
4343
4344         .cpuid_update = svm_cpuid_update,
4345
4346         .rdtscp_supported = svm_rdtscp_supported,
4347         .invpcid_supported = svm_invpcid_supported,
4348
4349         .set_supported_cpuid = svm_set_supported_cpuid,
4350
4351         .has_wbinvd_exit = svm_has_wbinvd_exit,
4352
4353         .set_tsc_khz = svm_set_tsc_khz,
4354         .read_tsc_offset = svm_read_tsc_offset,
4355         .write_tsc_offset = svm_write_tsc_offset,
4356         .adjust_tsc_offset = svm_adjust_tsc_offset,
4357         .compute_tsc_offset = svm_compute_tsc_offset,
4358         .read_l1_tsc = svm_read_l1_tsc,
4359
4360         .set_tdp_cr3 = set_tdp_cr3,
4361
4362         .check_intercept = svm_check_intercept,
4363         .handle_external_intr = svm_handle_external_intr,
4364 };
4365
4366 static int __init svm_init(void)
4367 {
4368         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4369                         __alignof__(struct vcpu_svm), THIS_MODULE);
4370 }
4371
4372 static void __exit svm_exit(void)
4373 {
4374         kvm_exit();
4375 }
4376
4377 module_init(svm_init)
4378 module_exit(svm_exit)