arch/x86/kvm/svm.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * AMD SVM support
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8  *
   9  * Authors:
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17 #include <linux/kvm_host.h>
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "kvm_cache_regs.h"
  22 #include "x86.h"
  23 #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/debugreg.h>
  38 #include <asm/kvm_para.h>
  39
  40 #include <asm/virtext.h>
  41 #include "trace.h"
  42
  43 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  44
  45 MODULE_AUTHOR("Qumranet");
  46 MODULE_LICENSE("GPL");
  47
  48 static const struct x86_cpu_id svm_cpu_id[] = {
  49         X86_FEATURE_MATCH(X86_FEATURE_SVM),
  50         {}
  51 };
  52 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  53
  54 #define IOPM_ALLOC_ORDER 2
  55 #define MSRPM_ALLOC_ORDER 1
  56
  57 #define SEG_TYPE_LDT 2
  58 #define SEG_TYPE_BUSY_TSS16 3
  59
  60 #define SVM_FEATURE_NPT            (1 <<  0)
  61 #define SVM_FEATURE_LBRV           (1 <<  1)
  62 #define SVM_FEATURE_SVML           (1 <<  2)
  63 #define SVM_FEATURE_NRIP           (1 <<  3)
  64 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  65 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  66 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  67 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  68 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  69
  70 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  71 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  72 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  73
  74 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  75
  76 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  77 #define TSC_RATIO_MIN           0x0000000000000001ULL
  78 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  79
  80 static bool erratum_383_found __read_mostly;
  81
  82 static const u32 host_save_user_msrs[] = {
  83 #ifdef CONFIG_X86_64
  84         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  85         MSR_FS_BASE,
  86 #endif
  87         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  88 };
  89
  90 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  91
  92 struct kvm_vcpu;
  93
  94 struct nested_state {
  95         struct vmcb *hsave;
  96         u64 hsave_msr;
  97         u64 vm_cr_msr;
  98         u64 vmcb;
  99
 100         /* These are the merged vectors */
 101         u32 *msrpm;
 102
 103         /* gpa pointers to the real vectors */
 104         u64 vmcb_msrpm;
 105         u64 vmcb_iopm;
 106
 107         /* A VMEXIT is required but not yet emulated */
 108         bool exit_required;
 109
 110         /* cache for intercepts of the guest */
 111         u32 intercept_cr;
 112         u32 intercept_dr;
 113         u32 intercept_exceptions;
 114         u64 intercept;
 115
 116         /* Nested Paging related state */
 117         u64 nested_cr3;
 118 };
 119
 120 #define MSRPM_OFFSETS   16
 121 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 122
 123 /*
 124  * Set osvw_len to higher value when updated Revision Guides
 125  * are published and we know what the new status bits are
 126  */
 127 static uint64_t osvw_len = 4, osvw_status;
 128
 129 struct vcpu_svm {
 130         struct kvm_vcpu vcpu;
 131         struct vmcb *vmcb;
 132         unsigned long vmcb_pa;
 133         struct svm_cpu_data *svm_data;
 134         uint64_t asid_generation;
 135         uint64_t sysenter_esp;
 136         uint64_t sysenter_eip;
 137
 138         u64 next_rip;
 139
 140         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 141         struct {
 142                 u16 fs;
 143                 u16 gs;
 144                 u16 ldt;
 145                 u64 gs_base;
 146         } host;
 147
 148         u32 *msrpm;
 149
 150         ulong nmi_iret_rip;
 151
 152         struct nested_state nested;
 153
 154         bool nmi_singlestep;
 155
 156         unsigned int3_injected;
 157         unsigned long int3_rip;
 158         u32 apf_reason;
 159
 160         u64  tsc_ratio;
 161 };
 162
 163 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 164 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 165
 166 #define MSR_INVALID                     0xffffffffU
 167
 168 static const struct svm_direct_access_msrs {
 169         u32 index;   /* Index of the MSR */
 170         bool always; /* True if intercept is always on */
 171 } direct_access_msrs[] = {
 172         { .index = MSR_STAR,                            .always = true  },
 173         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 174 #ifdef CONFIG_X86_64
 175         { .index = MSR_GS_BASE,                         .always = true  },
 176         { .index = MSR_FS_BASE,                         .always = true  },
 177         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 178         { .index = MSR_LSTAR,                           .always = true  },
 179         { .index = MSR_CSTAR,                           .always = true  },
 180         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 181 #endif
 182         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 183         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 184         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 185         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 186         { .index = MSR_INVALID,                         .always = false },
 187 };
 188
 189 /* enable NPT for AMD64 and X86 with PAE */
 190 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 191 static bool npt_enabled = true;
 192 #else
 193 static bool npt_enabled;
 194 #endif
 195
 196 /* allow nested paging (virtualized MMU) for all guests */
 197 static int npt = true;
 198 module_param(npt, int, S_IRUGO);
 199
 200 /* allow nested virtualization in KVM/SVM */
 201 static int nested = true;
 202 module_param(nested, int, S_IRUGO);
 203
 204 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 205 static void svm_complete_interrupts(struct vcpu_svm *svm);
 206
 207 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 208 static int nested_svm_intercept(struct vcpu_svm *svm);
 209 static int nested_svm_vmexit(struct vcpu_svm *svm);
 210 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 211                                       bool has_error_code, u32 error_code);
 212 static u64 __scale_tsc(u64 ratio, u64 tsc);
 213
 214 enum {
 215         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 216                             pause filter count */
 217         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 218         VMCB_ASID,       /* ASID */
 219         VMCB_INTR,       /* int_ctl, int_vector */
 220         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 221         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 222         VMCB_DR,         /* DR6, DR7 */
 223         VMCB_DT,         /* GDT, IDT */
 224         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 225         VMCB_CR2,        /* CR2 only */
 226         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 227         VMCB_DIRTY_MAX,
 228 };
 229
 230 /* TPR and CR2 are always written before VMRUN */
 231 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 232
 233 static inline void mark_all_dirty(struct vmcb *vmcb)
 234 {
 235         vmcb->control.clean = 0;
 236 }
 237
 238 static inline void mark_all_clean(struct vmcb *vmcb)
 239 {
 240         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 241                                & ~VMCB_ALWAYS_DIRTY_MASK;
 242 }
 243
 244 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 245 {
 246         vmcb->control.clean &= ~(1 << bit);
 247 }
 248
 249 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 250 {
 251         return container_of(vcpu, struct vcpu_svm, vcpu);
 252 }
 253
 254 static void recalc_intercepts(struct vcpu_svm *svm)
 255 {
 256         struct vmcb_control_area *c, *h;
 257         struct nested_state *g;
 258
 259         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 260
 261         if (!is_guest_mode(&svm->vcpu))
 262                 return;
 263
 264         c = &svm->vmcb->control;
 265         h = &svm->nested.hsave->control;
 266         g = &svm->nested;
 267
 268         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 269         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 270         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 271         c->intercept = h->intercept | g->intercept;
 272 }
 273
 274 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 275 {
 276         if (is_guest_mode(&svm->vcpu))
 277                 return svm->nested.hsave;
 278         else
 279                 return svm->vmcb;
 280 }
 281
 282 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 283 {
 284         struct vmcb *vmcb = get_host_vmcb(svm);
 285
 286         vmcb->control.intercept_cr |= (1U << bit);
 287
 288         recalc_intercepts(svm);
 289 }
 290
 291 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 292 {
 293         struct vmcb *vmcb = get_host_vmcb(svm);
 294
 295         vmcb->control.intercept_cr &= ~(1U << bit);
 296
 297         recalc_intercepts(svm);
 298 }
 299
 300 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 301 {
 302         struct vmcb *vmcb = get_host_vmcb(svm);
 303
 304         return vmcb->control.intercept_cr & (1U << bit);
 305 }
 306
 307 static inline void set_dr_intercepts(struct vcpu_svm *svm)
 308 {
 309         struct vmcb *vmcb = get_host_vmcb(svm);
 310
 311         vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
 312                 | (1 << INTERCEPT_DR1_READ)
 313                 | (1 << INTERCEPT_DR2_READ)
 314                 | (1 << INTERCEPT_DR3_READ)
 315                 | (1 << INTERCEPT_DR4_READ)
 316                 | (1 << INTERCEPT_DR5_READ)
 317                 | (1 << INTERCEPT_DR6_READ)
 318                 | (1 << INTERCEPT_DR7_READ)
 319                 | (1 << INTERCEPT_DR0_WRITE)
 320                 | (1 << INTERCEPT_DR1_WRITE)
 321                 | (1 << INTERCEPT_DR2_WRITE)
 322                 | (1 << INTERCEPT_DR3_WRITE)
 323                 | (1 << INTERCEPT_DR4_WRITE)
 324                 | (1 << INTERCEPT_DR5_WRITE)
 325                 | (1 << INTERCEPT_DR6_WRITE)
 326                 | (1 << INTERCEPT_DR7_WRITE);
 327
 328         recalc_intercepts(svm);
 329 }
 330
 331 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
 332 {
 333         struct vmcb *vmcb = get_host_vmcb(svm);
 334
 335         vmcb->control.intercept_dr = 0;
 336
 337         recalc_intercepts(svm);
 338 }
 339
 340 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 341 {
 342         struct vmcb *vmcb = get_host_vmcb(svm);
 343
 344         vmcb->control.intercept_exceptions |= (1U << bit);
 345
 346         recalc_intercepts(svm);
 347 }
 348
 349 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 350 {
 351         struct vmcb *vmcb = get_host_vmcb(svm);
 352
 353         vmcb->control.intercept_exceptions &= ~(1U << bit);
 354
 355         recalc_intercepts(svm);
 356 }
 357
 358 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 359 {
 360         struct vmcb *vmcb = get_host_vmcb(svm);
 361
 362         vmcb->control.intercept |= (1ULL << bit);
 363
 364         recalc_intercepts(svm);
 365 }
 366
 367 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 368 {
 369         struct vmcb *vmcb = get_host_vmcb(svm);
 370
 371         vmcb->control.intercept &= ~(1ULL << bit);
 372
 373         recalc_intercepts(svm);
 374 }
 375
 376 static inline void enable_gif(struct vcpu_svm *svm)
 377 {
 378         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 379 }
 380
 381 static inline void disable_gif(struct vcpu_svm *svm)
 382 {
 383         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 384 }
 385
 386 static inline bool gif_set(struct vcpu_svm *svm)
 387 {
 388         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 389 }
 390
 391 static unsigned long iopm_base;
 392
 393 struct kvm_ldttss_desc {
 394         u16 limit0;
 395         u16 base0;
 396         unsigned base1:8, type:5, dpl:2, p:1;
 397         unsigned limit1:4, zero0:3, g:1, base2:8;
 398         u32 base3;
 399         u32 zero1;
 400 } __attribute__((packed));
 401
 402 struct svm_cpu_data {
 403         int cpu;
 404
 405         u64 asid_generation;
 406         u32 max_asid;
 407         u32 next_asid;
 408         struct kvm_ldttss_desc *tss_desc;
 409
 410         struct page *save_area;
 411 };
 412
 413 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 414
 415 struct svm_init_data {
 416         int cpu;
 417         int r;
 418 };
 419
 420 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 421
 422 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 423 #define MSRS_RANGE_SIZE 2048
 424 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 425
 426 static u32 svm_msrpm_offset(u32 msr)
 427 {
 428         u32 offset;
 429         int i;
 430
 431         for (i = 0; i < NUM_MSR_MAPS; i++) {
 432                 if (msr < msrpm_ranges[i] ||
 433                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 434                         continue;
 435
 436                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 437                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 438
 439                 /* Now we have the u8 offset - but need the u32 offset */
 440                 return offset / 4;
 441         }
 442
 443         /* MSR not in any range */
 444         return MSR_INVALID;
 445 }
 446
 447 #define MAX_INST_SIZE 15
 448
 449 static inline void clgi(void)
 450 {
 451         asm volatile (__ex(SVM_CLGI));
 452 }
 453
 454 static inline void stgi(void)
 455 {
 456         asm volatile (__ex(SVM_STGI));
 457 }
 458
 459 static inline void invlpga(unsigned long addr, u32 asid)
 460 {
 461         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 462 }
 463
 464 static int get_npt_level(void)
 465 {
 466 #ifdef CONFIG_X86_64
 467         return PT64_ROOT_LEVEL;
 468 #else
 469         return PT32E_ROOT_LEVEL;
 470 #endif
 471 }
 472
 473 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 474 {
 475         vcpu->arch.efer = efer;
 476         if (!npt_enabled && !(efer & EFER_LMA))
 477                 efer &= ~EFER_LME;
 478
 479         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 480         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 481 }
 482
 483 static int is_external_interrupt(u32 info)
 484 {
 485         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 486         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 487 }
 488
 489 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
 490 {
 491         struct vcpu_svm *svm = to_svm(vcpu);
 492         u32 ret = 0;
 493
 494         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 495                 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 496         return ret;
 497 }
 498
 499 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 500 {
 501         struct vcpu_svm *svm = to_svm(vcpu);
 502
 503         if (mask == 0)
 504                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 505         else
 506                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 507
 508 }
 509
 510 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 511 {
 512         struct vcpu_svm *svm = to_svm(vcpu);
 513
 514         if (svm->vmcb->control.next_rip != 0)
 515                 svm->next_rip = svm->vmcb->control.next_rip;
 516
 517         if (!svm->next_rip) {
 518                 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 519                                 EMULATE_DONE)
 520                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 521                 return;
 522         }
 523         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 524                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 525                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 526
 527         kvm_rip_write(vcpu, svm->next_rip);
 528         svm_set_interrupt_shadow(vcpu, 0);
 529 }
 530
 531 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 532                                 bool has_error_code, u32 error_code,
 533                                 bool reinject)
 534 {
 535         struct vcpu_svm *svm = to_svm(vcpu);
 536
 537         /*
 538          * If we are within a nested VM we'd better #VMEXIT and let the guest
 539          * handle the exception
 540          */
 541         if (!reinject &&
 542             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 543                 return;
 544
 545         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 546                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 547
 548                 /*
 549                  * For guest debugging where we have to reinject #BP if some
 550                  * INT3 is guest-owned:
 551                  * Emulate nRIP by moving RIP forward. Will fail if injection
 552                  * raises a fault that is not intercepted. Still better than
 553                  * failing in all cases.
 554                  */
 555                 skip_emulated_instruction(&svm->vcpu);
 556                 rip = kvm_rip_read(&svm->vcpu);
 557                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 558                 svm->int3_injected = rip - old_rip;
 559         }
 560
 561         svm->vmcb->control.event_inj = nr
 562                 | SVM_EVTINJ_VALID
 563                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 564                 | SVM_EVTINJ_TYPE_EXEPT;
 565         svm->vmcb->control.event_inj_err = error_code;
 566 }
 567
 568 static void svm_init_erratum_383(void)
 569 {
 570         u32 low, high;
 571         int err;
 572         u64 val;
 573
 574         if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 575                 return;
 576
 577         /* Use _safe variants to not break nested virtualization */
 578         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 579         if (err)
 580                 return;
 581
 582         val |= (1ULL << 47);
 583
 584         low  = lower_32_bits(val);
 585         high = upper_32_bits(val);
 586
 587         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 588
 589         erratum_383_found = true;
 590 }
 591
 592 static void svm_init_osvw(struct kvm_vcpu *vcpu)
 593 {
 594         /*
 595          * Guests should see errata 400 and 415 as fixed (assuming that
 596          * HLT and IO instructions are intercepted).
 597          */
 598         vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 599         vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 600
 601         /*
 602          * By increasing VCPU's osvw.length to 3 we are telling the guest that
 603          * all osvw.status bits inside that length, including bit 0 (which is
 604          * reserved for erratum 298), are valid. However, if host processor's
 605          * osvw_len is 0 then osvw_status[0] carries no information. We need to
 606          * be conservative here and therefore we tell the guest that erratum 298
 607          * is present (because we really don't know).
 608          */
 609         if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 610                 vcpu->arch.osvw.status |= 1;
 611 }
 612
 613 static int has_svm(void)
 614 {
 615         const char *msg;
 616
 617         if (!cpu_has_svm(&msg)) {
 618                 printk(KERN_INFO "has_svm: %s\n", msg);
 619                 return 0;
 620         }
 621
 622         return 1;
 623 }
 624
 625 static void svm_hardware_disable(void)
 626 {
 627         /* Make sure we clean up behind us */
 628         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 629                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 630
 631         cpu_svm_disable();
 632
 633         amd_pmu_disable_virt();
 634 }
 635
 636 static int svm_hardware_enable(void)
 637 {
 638
 639         struct svm_cpu_data *sd;
 640         uint64_t efer;
 641         struct desc_ptr gdt_descr;
 642         struct desc_struct *gdt;
 643         int me = raw_smp_processor_id();
 644
 645         rdmsrl(MSR_EFER, efer);
 646         if (efer & EFER_SVME)
 647                 return -EBUSY;
 648
 649         if (!has_svm()) {
 650                 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 651                 return -EINVAL;
 652         }
 653         sd = per_cpu(svm_data, me);
 654         if (!sd) {
 655                 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 656                 return -EINVAL;
 657         }
 658
 659         sd->asid_generation = 1;
 660         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 661         sd->next_asid = sd->max_asid + 1;
 662
 663         native_store_gdt(&gdt_descr);
 664         gdt = (struct desc_struct *)gdt_descr.address;
 665         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 666
 667         wrmsrl(MSR_EFER, efer | EFER_SVME);
 668
 669         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 670
 671         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 672                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 673                 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
 674         }
 675
 676
 677         /*
 678          * Get OSVW bits.
 679          *
 680          * Note that it is possible to have a system with mixed processor
 681          * revisions and therefore different OSVW bits. If bits are not the same
 682          * on different processors then choose the worst case (i.e. if erratum
 683          * is present on one processor and not on another then assume that the
 684          * erratum is present everywhere).
 685          */
 686         if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 687                 uint64_t len, status = 0;
 688                 int err;
 689
 690                 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 691                 if (!err)
 692                         status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 693                                                       &err);
 694
 695                 if (err)
 696                         osvw_status = osvw_len = 0;
 697                 else {
 698                         if (len < osvw_len)
 699                                 osvw_len = len;
 700                         osvw_status |= status;
 701                         osvw_status &= (1ULL << osvw_len) - 1;
 702                 }
 703         } else
 704                 osvw_status = osvw_len = 0;
 705
 706         svm_init_erratum_383();
 707
 708         amd_pmu_enable_virt();
 709
 710         return 0;
 711 }
 712
 713 static void svm_cpu_uninit(int cpu)
 714 {
 715         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 716
 717         if (!sd)
 718                 return;
 719
 720         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 721         __free_page(sd->save_area);
 722         kfree(sd);
 723 }
 724
 725 static int svm_cpu_init(int cpu)
 726 {
 727         struct svm_cpu_data *sd;
 728         int r;
 729
 730         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 731         if (!sd)
 732                 return -ENOMEM;
 733         sd->cpu = cpu;
 734         sd->save_area = alloc_page(GFP_KERNEL);
 735         r = -ENOMEM;
 736         if (!sd->save_area)
 737                 goto err_1;
 738
 739         per_cpu(svm_data, cpu) = sd;
 740
 741         return 0;
 742
 743 err_1:
 744         kfree(sd);
 745         return r;
 746
 747 }
 748
 749 static bool valid_msr_intercept(u32 index)
 750 {
 751         int i;
 752
 753         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 754                 if (direct_access_msrs[i].index == index)
 755                         return true;
 756
 757         return false;
 758 }
 759
 760 static void set_msr_interception(u32 *msrpm, unsigned msr,
 761                                  int read, int write)
 762 {
 763         u8 bit_read, bit_write;
 764         unsigned long tmp;
 765         u32 offset;
 766
 767         /*
 768          * If this warning triggers extend the direct_access_msrs list at the
 769          * beginning of the file
 770          */
 771         WARN_ON(!valid_msr_intercept(msr));
 772
 773         offset    = svm_msrpm_offset(msr);
 774         bit_read  = 2 * (msr & 0x0f);
 775         bit_write = 2 * (msr & 0x0f) + 1;
 776         tmp       = msrpm[offset];
 777
 778         BUG_ON(offset == MSR_INVALID);
 779
 780         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 781         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 782
 783         msrpm[offset] = tmp;
 784 }
 785
 786 static void svm_vcpu_init_msrpm(u32 *msrpm)
 787 {
 788         int i;
 789
 790         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 791
 792         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 793                 if (!direct_access_msrs[i].always)
 794                         continue;
 795
 796                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 797         }
 798 }
 799
 800 static void add_msr_offset(u32 offset)
 801 {
 802         int i;
 803
 804         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 805
 806                 /* Offset already in list? */
 807                 if (msrpm_offsets[i] == offset)
 808                         return;
 809
 810                 /* Slot used by another offset? */
 811                 if (msrpm_offsets[i] != MSR_INVALID)
 812                         continue;
 813
 814                 /* Add offset to list */
 815                 msrpm_offsets[i] = offset;
 816
 817                 return;
 818         }
 819
 820         /*
 821          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 822          * increase MSRPM_OFFSETS in this case.
 823          */
 824         BUG();
 825 }
 826
 827 static void init_msrpm_offsets(void)
 828 {
 829         int i;
 830
 831         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 832
 833         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 834                 u32 offset;
 835
 836                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 837                 BUG_ON(offset == MSR_INVALID);
 838
 839                 add_msr_offset(offset);
 840         }
 841 }
 842
 843 static void svm_enable_lbrv(struct vcpu_svm *svm)
 844 {
 845         u32 *msrpm = svm->msrpm;
 846
 847         svm->vmcb->control.lbr_ctl = 1;
 848         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 849         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 850         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 851         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 852 }
 853
 854 static void svm_disable_lbrv(struct vcpu_svm *svm)
 855 {
 856         u32 *msrpm = svm->msrpm;
 857
 858         svm->vmcb->control.lbr_ctl = 0;
 859         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 860         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 861         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 862         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 863 }
 864
 865 static __init int svm_hardware_setup(void)
 866 {
 867         int cpu;
 868         struct page *iopm_pages;
 869         void *iopm_va;
 870         int r;
 871
 872         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 873
 874         if (!iopm_pages)
 875                 return -ENOMEM;
 876
 877         iopm_va = page_address(iopm_pages);
 878         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 879         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 880
 881         init_msrpm_offsets();
 882
 883         if (boot_cpu_has(X86_FEATURE_NX))
 884                 kvm_enable_efer_bits(EFER_NX);
 885
 886         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 887                 kvm_enable_efer_bits(EFER_FFXSR);
 888
 889         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 890                 u64 max;
 891
 892                 kvm_has_tsc_control = true;
 893
 894                 /*
 895                  * Make sure the user can only configure tsc_khz values that
 896                  * fit into a signed integer.
 897                  * A min value is not calculated needed because it will always
 898                  * be 1 on all machines and a value of 0 is used to disable
 899                  * tsc-scaling for the vcpu.
 900                  */
 901                 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 902
 903                 kvm_max_guest_tsc_khz = max;
 904         }
 905
 906         if (nested) {
 907                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 908                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 909         }
 910
 911         for_each_possible_cpu(cpu) {
 912                 r = svm_cpu_init(cpu);
 913                 if (r)
 914                         goto err;
 915         }
 916
 917         if (!boot_cpu_has(X86_FEATURE_NPT))
 918                 npt_enabled = false;
 919
 920         if (npt_enabled && !npt) {
 921                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 922                 npt_enabled = false;
 923         }
 924
 925         if (npt_enabled) {
 926                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 927                 kvm_enable_tdp();
 928         } else
 929                 kvm_disable_tdp();
 930
 931         return 0;
 932
 933 err:
 934         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 935         iopm_base = 0;
 936         return r;
 937 }
 938
 939 static __exit void svm_hardware_unsetup(void)
 940 {
 941         int cpu;
 942
 943         for_each_possible_cpu(cpu)
 944                 svm_cpu_uninit(cpu);
 945
 946         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 947         iopm_base = 0;
 948 }
 949
 950 static void init_seg(struct vmcb_seg *seg)
 951 {
 952         seg->selector = 0;
 953         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 954                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 955         seg->limit = 0xffff;
 956         seg->base = 0;
 957 }
 958
 959 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 960 {
 961         seg->selector = 0;
 962         seg->attrib = SVM_SELECTOR_P_MASK | type;
 963         seg->limit = 0xffff;
 964         seg->base = 0;
 965 }
 966
 967 static u64 __scale_tsc(u64 ratio, u64 tsc)
 968 {
 969         u64 mult, frac, _tsc;
 970
 971         mult  = ratio >> 32;
 972         frac  = ratio & ((1ULL << 32) - 1);
 973
 974         _tsc  = tsc;
 975         _tsc *= mult;
 976         _tsc += (tsc >> 32) * frac;
 977         _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
 978
 979         return _tsc;
 980 }
 981
 982 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
 983 {
 984         struct vcpu_svm *svm = to_svm(vcpu);
 985         u64 _tsc = tsc;
 986
 987         if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
 988                 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
 989
 990         return _tsc;
 991 }
 992
 993 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
 994 {
 995         struct vcpu_svm *svm = to_svm(vcpu);
 996         u64 ratio;
 997         u64 khz;
 998
 999         /* Guest TSC same frequency as host TSC? */
1000         if (!scale) {
1001                 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1002                 return;
1003         }
1004
1005         /* TSC scaling supported? */
1006         if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1007                 if (user_tsc_khz > tsc_khz) {
1008                         vcpu->arch.tsc_catchup = 1;
1009                         vcpu->arch.tsc_always_catchup = 1;
1010                 } else
1011                         WARN(1, "user requested TSC rate below hardware speed\n");
1012                 return;
1013         }
1014
1015         khz = user_tsc_khz;
1016
1017         /* TSC scaling required  - calculate ratio */
1018         ratio = khz << 32;
1019         do_div(ratio, tsc_khz);
1020
1021         if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1022                 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1023                                 user_tsc_khz);
1024                 return;
1025         }
1026         svm->tsc_ratio             = ratio;
1027 }
1028
1029 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1030 {
1031         struct vcpu_svm *svm = to_svm(vcpu);
1032
1033         return svm->vmcb->control.tsc_offset;
1034 }
1035
1036 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1037 {
1038         struct vcpu_svm *svm = to_svm(vcpu);
1039         u64 g_tsc_offset = 0;
1040
1041         if (is_guest_mode(vcpu)) {
1042                 g_tsc_offset = svm->vmcb->control.tsc_offset -
1043                                svm->nested.hsave->control.tsc_offset;
1044                 svm->nested.hsave->control.tsc_offset = offset;
1045         } else
1046                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1047                                            svm->vmcb->control.tsc_offset,
1048                                            offset);
1049
1050         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1051
1052         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1053 }
1054
1055 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1056 {
1057         struct vcpu_svm *svm = to_svm(vcpu);
1058
1059         if (host) {
1060                 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1061                         WARN_ON(adjustment < 0);
1062                 adjustment = svm_scale_tsc(vcpu, (u64)adjustment);
1063         }
1064
1065         svm->vmcb->control.tsc_offset += adjustment;
1066         if (is_guest_mode(vcpu))
1067                 svm->nested.hsave->control.tsc_offset += adjustment;
1068         else
1069                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1070                                      svm->vmcb->control.tsc_offset - adjustment,
1071                                      svm->vmcb->control.tsc_offset);
1072
1073         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1074 }
1075
1076 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1077 {
1078         u64 tsc;
1079
1080         tsc = svm_scale_tsc(vcpu, native_read_tsc());
1081
1082         return target_tsc - tsc;
1083 }
1084
1085 static void init_vmcb(struct vcpu_svm *svm)
1086 {
1087         struct vmcb_control_area *control = &svm->vmcb->control;
1088         struct vmcb_save_area *save = &svm->vmcb->save;
1089
1090         svm->vcpu.fpu_active = 1;
1091         svm->vcpu.arch.hflags = 0;
1092
1093         set_cr_intercept(svm, INTERCEPT_CR0_READ);
1094         set_cr_intercept(svm, INTERCEPT_CR3_READ);
1095         set_cr_intercept(svm, INTERCEPT_CR4_READ);
1096         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1097         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1098         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1099         set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1100
1101         set_dr_intercepts(svm);
1102
1103         set_exception_intercept(svm, PF_VECTOR);
1104         set_exception_intercept(svm, UD_VECTOR);
1105         set_exception_intercept(svm, MC_VECTOR);
1106
1107         set_intercept(svm, INTERCEPT_INTR);
1108         set_intercept(svm, INTERCEPT_NMI);
1109         set_intercept(svm, INTERCEPT_SMI);
1110         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1111         set_intercept(svm, INTERCEPT_RDPMC);
1112         set_intercept(svm, INTERCEPT_CPUID);
1113         set_intercept(svm, INTERCEPT_INVD);
1114         set_intercept(svm, INTERCEPT_HLT);
1115         set_intercept(svm, INTERCEPT_INVLPG);
1116         set_intercept(svm, INTERCEPT_INVLPGA);
1117         set_intercept(svm, INTERCEPT_IOIO_PROT);
1118         set_intercept(svm, INTERCEPT_MSR_PROT);
1119         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1120         set_intercept(svm, INTERCEPT_SHUTDOWN);
1121         set_intercept(svm, INTERCEPT_VMRUN);
1122         set_intercept(svm, INTERCEPT_VMMCALL);
1123         set_intercept(svm, INTERCEPT_VMLOAD);
1124         set_intercept(svm, INTERCEPT_VMSAVE);
1125         set_intercept(svm, INTERCEPT_STGI);
1126         set_intercept(svm, INTERCEPT_CLGI);
1127         set_intercept(svm, INTERCEPT_SKINIT);
1128         set_intercept(svm, INTERCEPT_WBINVD);
1129         set_intercept(svm, INTERCEPT_MONITOR);
1130         set_intercept(svm, INTERCEPT_MWAIT);
1131         set_intercept(svm, INTERCEPT_XSETBV);
1132
1133         control->iopm_base_pa = iopm_base;
1134         control->msrpm_base_pa = __pa(svm->msrpm);
1135         control->int_ctl = V_INTR_MASKING_MASK;
1136
1137         init_seg(&save->es);
1138         init_seg(&save->ss);
1139         init_seg(&save->ds);
1140         init_seg(&save->fs);
1141         init_seg(&save->gs);
1142
1143         save->cs.selector = 0xf000;
1144         save->cs.base = 0xffff0000;
1145         /* Executable/Readable Code Segment */
1146         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1147                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1148         save->cs.limit = 0xffff;
1149
1150         save->gdtr.limit = 0xffff;
1151         save->idtr.limit = 0xffff;
1152
1153         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1154         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1155
1156         svm_set_efer(&svm->vcpu, 0);
1157         save->dr6 = 0xffff0ff0;
1158         kvm_set_rflags(&svm->vcpu, 2);
1159         save->rip = 0x0000fff0;
1160         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1161
1162         /*
1163          * This is the guest-visible cr0 value.
1164          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1165          */
1166         svm->vcpu.arch.cr0 = 0;
1167         (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1168
1169         save->cr4 = X86_CR4_PAE;
1170         /* rdx = ?? */
1171
1172         if (npt_enabled) {
1173                 /* Setup VMCB for Nested Paging */
1174                 control->nested_ctl = 1;
1175                 clr_intercept(svm, INTERCEPT_INVLPG);
1176                 clr_exception_intercept(svm, PF_VECTOR);
1177                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1178                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1179                 save->g_pat = 0x0007040600070406ULL;
1180                 save->cr3 = 0;
1181                 save->cr4 = 0;
1182         }
1183         svm->asid_generation = 0;
1184
1185         svm->nested.vmcb = 0;
1186         svm->vcpu.arch.hflags = 0;
1187
1188         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1189                 control->pause_filter_count = 3000;
1190                 set_intercept(svm, INTERCEPT_PAUSE);
1191         }
1192
1193         mark_all_dirty(svm->vmcb);
1194
1195         enable_gif(svm);
1196 }
1197
1198 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
1199 {
1200         struct vcpu_svm *svm = to_svm(vcpu);
1201         u32 dummy;
1202         u32 eax = 1;
1203
1204         init_vmcb(svm);
1205
1206         kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1207         kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1208 }
1209
1210 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1211 {
1212         struct vcpu_svm *svm;
1213         struct page *page;
1214         struct page *msrpm_pages;
1215         struct page *hsave_page;
1216         struct page *nested_msrpm_pages;
1217         int err;
1218
1219         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1220         if (!svm) {
1221                 err = -ENOMEM;
1222                 goto out;
1223         }
1224
1225         svm->tsc_ratio = TSC_RATIO_DEFAULT;
1226
1227         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1228         if (err)
1229                 goto free_svm;
1230
1231         err = -ENOMEM;
1232         page = alloc_page(GFP_KERNEL);
1233         if (!page)
1234                 goto uninit;
1235
1236         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1237         if (!msrpm_pages)
1238                 goto free_page1;
1239
1240         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1241         if (!nested_msrpm_pages)
1242                 goto free_page2;
1243
1244         hsave_page = alloc_page(GFP_KERNEL);
1245         if (!hsave_page)
1246                 goto free_page3;
1247
1248         svm->nested.hsave = page_address(hsave_page);
1249
1250         svm->msrpm = page_address(msrpm_pages);
1251         svm_vcpu_init_msrpm(svm->msrpm);
1252
1253         svm->nested.msrpm = page_address(nested_msrpm_pages);
1254         svm_vcpu_init_msrpm(svm->nested.msrpm);
1255
1256         svm->vmcb = page_address(page);
1257         clear_page(svm->vmcb);
1258         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1259         svm->asid_generation = 0;
1260         init_vmcb(svm);
1261
1262         svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1263                                    MSR_IA32_APICBASE_ENABLE;
1264         if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1265                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1266
1267         svm_init_osvw(&svm->vcpu);
1268
1269         return &svm->vcpu;
1270
1271 free_page3:
1272         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1273 free_page2:
1274         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1275 free_page1:
1276         __free_page(page);
1277 uninit:
1278         kvm_vcpu_uninit(&svm->vcpu);
1279 free_svm:
1280         kmem_cache_free(kvm_vcpu_cache, svm);
1281 out:
1282         return ERR_PTR(err);
1283 }
1284
1285 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1286 {
1287         struct vcpu_svm *svm = to_svm(vcpu);
1288
1289         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1290         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1291         __free_page(virt_to_page(svm->nested.hsave));
1292         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1293         kvm_vcpu_uninit(vcpu);
1294         kmem_cache_free(kvm_vcpu_cache, svm);
1295 }
1296
1297 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1298 {
1299         struct vcpu_svm *svm = to_svm(vcpu);
1300         int i;
1301
1302         if (unlikely(cpu != vcpu->cpu)) {
1303                 svm->asid_generation = 0;
1304                 mark_all_dirty(svm->vmcb);
1305         }
1306
1307 #ifdef CONFIG_X86_64
1308         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1309 #endif
1310         savesegment(fs, svm->host.fs);
1311         savesegment(gs, svm->host.gs);
1312         svm->host.ldt = kvm_read_ldt();
1313
1314         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1315                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1316
1317         if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1318             svm->tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1319                 __this_cpu_write(current_tsc_ratio, svm->tsc_ratio);
1320                 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1321         }
1322 }
1323
1324 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1325 {
1326         struct vcpu_svm *svm = to_svm(vcpu);
1327         int i;
1328
1329         ++vcpu->stat.host_state_reload;
1330         kvm_load_ldt(svm->host.ldt);
1331 #ifdef CONFIG_X86_64
1332         loadsegment(fs, svm->host.fs);
1333         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1334         load_gs_index(svm->host.gs);
1335 #else
1336 #ifdef CONFIG_X86_32_LAZY_GS
1337         loadsegment(gs, svm->host.gs);
1338 #endif
1339 #endif
1340         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1341                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1342 }
1343
1344 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1345 {
1346         return to_svm(vcpu)->vmcb->save.rflags;
1347 }
1348
1349 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1350 {
1351        /*
1352         * Any change of EFLAGS.VM is accompained by a reload of SS
1353         * (caused by either a task switch or an inter-privilege IRET),
1354         * so we do not need to update the CPL here.
1355         */
1356         to_svm(vcpu)->vmcb->save.rflags = rflags;
1357 }
1358
1359 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1360 {
1361         switch (reg) {
1362         case VCPU_EXREG_PDPTR:
1363                 BUG_ON(!npt_enabled);
1364                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1365                 break;
1366         default:
1367                 BUG();
1368         }
1369 }
1370
1371 static void svm_set_vintr(struct vcpu_svm *svm)
1372 {
1373         set_intercept(svm, INTERCEPT_VINTR);
1374 }
1375
1376 static void svm_clear_vintr(struct vcpu_svm *svm)
1377 {
1378         clr_intercept(svm, INTERCEPT_VINTR);
1379 }
1380
1381 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1382 {
1383         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1384
1385         switch (seg) {
1386         case VCPU_SREG_CS: return &save->cs;
1387         case VCPU_SREG_DS: return &save->ds;
1388         case VCPU_SREG_ES: return &save->es;
1389         case VCPU_SREG_FS: return &save->fs;
1390         case VCPU_SREG_GS: return &save->gs;
1391         case VCPU_SREG_SS: return &save->ss;
1392         case VCPU_SREG_TR: return &save->tr;
1393         case VCPU_SREG_LDTR: return &save->ldtr;
1394         }
1395         BUG();
1396         return NULL;
1397 }
1398
1399 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1400 {
1401         struct vmcb_seg *s = svm_seg(vcpu, seg);
1402
1403         return s->base;
1404 }
1405
1406 static void svm_get_segment(struct kvm_vcpu *vcpu,
1407                             struct kvm_segment *var, int seg)
1408 {
1409         struct vmcb_seg *s = svm_seg(vcpu, seg);
1410
1411         var->base = s->base;
1412         var->limit = s->limit;
1413         var->selector = s->selector;
1414         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1415         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1416         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1417         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1418         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1419         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1420         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1421
1422         /*
1423          * AMD CPUs circa 2014 track the G bit for all segments except CS.
1424          * However, the SVM spec states that the G bit is not observed by the
1425          * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1426          * So let's synthesize a legal G bit for all segments, this helps
1427          * running KVM nested. It also helps cross-vendor migration, because
1428          * Intel's vmentry has a check on the 'G' bit.
1429          */
1430         var->g = s->limit > 0xfffff;
1431
1432         /*
1433          * AMD's VMCB does not have an explicit unusable field, so emulate it
1434          * for cross vendor migration purposes by "not present"
1435          */
1436         var->unusable = !var->present || (var->type == 0);
1437
1438         switch (seg) {
1439         case VCPU_SREG_TR:
1440                 /*
1441                  * Work around a bug where the busy flag in the tr selector
1442                  * isn't exposed
1443                  */
1444                 var->type |= 0x2;
1445                 break;
1446         case VCPU_SREG_DS:
1447         case VCPU_SREG_ES:
1448         case VCPU_SREG_FS:
1449         case VCPU_SREG_GS:
1450                 /*
1451                  * The accessed bit must always be set in the segment
1452                  * descriptor cache, although it can be cleared in the
1453                  * descriptor, the cached bit always remains at 1. Since
1454                  * Intel has a check on this, set it here to support
1455                  * cross-vendor migration.
1456                  */
1457                 if (!var->unusable)
1458                         var->type |= 0x1;
1459                 break;
1460         case VCPU_SREG_SS:
1461                 /*
1462                  * On AMD CPUs sometimes the DB bit in the segment
1463                  * descriptor is left as 1, although the whole segment has
1464                  * been made unusable. Clear it here to pass an Intel VMX
1465                  * entry check when cross vendor migrating.
1466                  */
1467                 if (var->unusable)
1468                         var->db = 0;
1469                 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1470                 break;
1471         }
1472 }
1473
1474 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1475 {
1476         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1477
1478         return save->cpl;
1479 }
1480
1481 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1482 {
1483         struct vcpu_svm *svm = to_svm(vcpu);
1484
1485         dt->size = svm->vmcb->save.idtr.limit;
1486         dt->address = svm->vmcb->save.idtr.base;
1487 }
1488
1489 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1490 {
1491         struct vcpu_svm *svm = to_svm(vcpu);
1492
1493         svm->vmcb->save.idtr.limit = dt->size;
1494         svm->vmcb->save.idtr.base = dt->address ;
1495         mark_dirty(svm->vmcb, VMCB_DT);
1496 }
1497
1498 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1499 {
1500         struct vcpu_svm *svm = to_svm(vcpu);
1501
1502         dt->size = svm->vmcb->save.gdtr.limit;
1503         dt->address = svm->vmcb->save.gdtr.base;
1504 }
1505
1506 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1507 {
1508         struct vcpu_svm *svm = to_svm(vcpu);
1509
1510         svm->vmcb->save.gdtr.limit = dt->size;
1511         svm->vmcb->save.gdtr.base = dt->address ;
1512         mark_dirty(svm->vmcb, VMCB_DT);
1513 }
1514
1515 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1516 {
1517 }
1518
1519 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1520 {
1521 }
1522
1523 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1524 {
1525 }
1526
1527 static void update_cr0_intercept(struct vcpu_svm *svm)
1528 {
1529         ulong gcr0 = svm->vcpu.arch.cr0;
1530         u64 *hcr0 = &svm->vmcb->save.cr0;
1531
1532         if (!svm->vcpu.fpu_active)
1533                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1534         else
1535                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1536                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1537
1538         mark_dirty(svm->vmcb, VMCB_CR);
1539
1540         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1541                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1542                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1543         } else {
1544                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1545                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1546         }
1547 }
1548
1549 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1550 {
1551         struct vcpu_svm *svm = to_svm(vcpu);
1552
1553 #ifdef CONFIG_X86_64
1554         if (vcpu->arch.efer & EFER_LME) {
1555                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1556                         vcpu->arch.efer |= EFER_LMA;
1557                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1558                 }
1559
1560                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1561                         vcpu->arch.efer &= ~EFER_LMA;
1562                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1563                 }
1564         }
1565 #endif
1566         vcpu->arch.cr0 = cr0;
1567
1568         if (!npt_enabled)
1569                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1570
1571         if (!vcpu->fpu_active)
1572                 cr0 |= X86_CR0_TS;
1573         /*
1574          * re-enable caching here because the QEMU bios
1575          * does not do it - this results in some delay at
1576          * reboot
1577          */
1578         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1579         svm->vmcb->save.cr0 = cr0;
1580         mark_dirty(svm->vmcb, VMCB_CR);
1581         update_cr0_intercept(svm);
1582 }
1583
1584 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1585 {
1586         unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1587         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1588
1589         if (cr4 & X86_CR4_VMXE)
1590                 return 1;
1591
1592         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1593                 svm_flush_tlb(vcpu);
1594
1595         vcpu->arch.cr4 = cr4;
1596         if (!npt_enabled)
1597                 cr4 |= X86_CR4_PAE;
1598         cr4 |= host_cr4_mce;
1599         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1600         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1601         return 0;
1602 }
1603
1604 static void svm_set_segment(struct kvm_vcpu *vcpu,
1605                             struct kvm_segment *var, int seg)
1606 {
1607         struct vcpu_svm *svm = to_svm(vcpu);
1608         struct vmcb_seg *s = svm_seg(vcpu, seg);
1609
1610         s->base = var->base;
1611         s->limit = var->limit;
1612         s->selector = var->selector;
1613         if (var->unusable)
1614                 s->attrib = 0;
1615         else {
1616                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1617                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1618                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1619                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1620                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1621                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1622                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1623                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1624         }
1625
1626         /*
1627          * This is always accurate, except if SYSRET returned to a segment
1628          * with SS.DPL != 3.  Intel does not have this quirk, and always
1629          * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1630          * would entail passing the CPL to userspace and back.
1631          */
1632         if (seg == VCPU_SREG_SS)
1633                 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1634
1635         mark_dirty(svm->vmcb, VMCB_SEG);
1636 }
1637
1638 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1639 {
1640         struct vcpu_svm *svm = to_svm(vcpu);
1641
1642         clr_exception_intercept(svm, DB_VECTOR);
1643         clr_exception_intercept(svm, BP_VECTOR);
1644
1645         if (svm->nmi_singlestep)
1646                 set_exception_intercept(svm, DB_VECTOR);
1647
1648         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1649                 if (vcpu->guest_debug &
1650                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1651                         set_exception_intercept(svm, DB_VECTOR);
1652                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1653                         set_exception_intercept(svm, BP_VECTOR);
1654         } else
1655                 vcpu->guest_debug = 0;
1656 }
1657
1658 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1659 {
1660         if (sd->next_asid > sd->max_asid) {
1661                 ++sd->asid_generation;
1662                 sd->next_asid = 1;
1663                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1664         }
1665
1666         svm->asid_generation = sd->asid_generation;
1667         svm->vmcb->control.asid = sd->next_asid++;
1668
1669         mark_dirty(svm->vmcb, VMCB_ASID);
1670 }
1671
1672 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1673 {
1674         return to_svm(vcpu)->vmcb->save.dr6;
1675 }
1676
1677 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1678 {
1679         struct vcpu_svm *svm = to_svm(vcpu);
1680
1681         svm->vmcb->save.dr6 = value;
1682         mark_dirty(svm->vmcb, VMCB_DR);
1683 }
1684
1685 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1686 {
1687         struct vcpu_svm *svm = to_svm(vcpu);
1688
1689         get_debugreg(vcpu->arch.db[0], 0);
1690         get_debugreg(vcpu->arch.db[1], 1);
1691         get_debugreg(vcpu->arch.db[2], 2);
1692         get_debugreg(vcpu->arch.db[3], 3);
1693         vcpu->arch.dr6 = svm_get_dr6(vcpu);
1694         vcpu->arch.dr7 = svm->vmcb->save.dr7;
1695
1696         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1697         set_dr_intercepts(svm);
1698 }
1699
1700 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1701 {
1702         struct vcpu_svm *svm = to_svm(vcpu);
1703
1704         svm->vmcb->save.dr7 = value;
1705         mark_dirty(svm->vmcb, VMCB_DR);
1706 }
1707
1708 static int pf_interception(struct vcpu_svm *svm)
1709 {
1710         u64 fault_address = svm->vmcb->control.exit_info_2;
1711         u32 error_code;
1712         int r = 1;
1713
1714         switch (svm->apf_reason) {
1715         default:
1716                 error_code = svm->vmcb->control.exit_info_1;
1717
1718                 trace_kvm_page_fault(fault_address, error_code);
1719                 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1720                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1721                 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1722                         svm->vmcb->control.insn_bytes,
1723                         svm->vmcb->control.insn_len);
1724                 break;
1725         case KVM_PV_REASON_PAGE_NOT_PRESENT:
1726                 svm->apf_reason = 0;
1727                 local_irq_disable();
1728                 kvm_async_pf_task_wait(fault_address);
1729                 local_irq_enable();
1730                 break;
1731         case KVM_PV_REASON_PAGE_READY:
1732                 svm->apf_reason = 0;
1733                 local_irq_disable();
1734                 kvm_async_pf_task_wake(fault_address);
1735                 local_irq_enable();
1736                 break;
1737         }
1738         return r;
1739 }
1740
1741 static int db_interception(struct vcpu_svm *svm)
1742 {
1743         struct kvm_run *kvm_run = svm->vcpu.run;
1744
1745         if (!(svm->vcpu.guest_debug &
1746               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1747                 !svm->nmi_singlestep) {
1748                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1749                 return 1;
1750         }
1751
1752         if (svm->nmi_singlestep) {
1753                 svm->nmi_singlestep = false;
1754                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1755                         svm->vmcb->save.rflags &=
1756                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1757                 update_db_bp_intercept(&svm->vcpu);
1758         }
1759
1760         if (svm->vcpu.guest_debug &
1761             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1762                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1763                 kvm_run->debug.arch.pc =
1764                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1765                 kvm_run->debug.arch.exception = DB_VECTOR;
1766                 return 0;
1767         }
1768
1769         return 1;
1770 }
1771
1772 static int bp_interception(struct vcpu_svm *svm)
1773 {
1774         struct kvm_run *kvm_run = svm->vcpu.run;
1775
1776         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1777         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1778         kvm_run->debug.arch.exception = BP_VECTOR;
1779         return 0;
1780 }
1781
1782 static int ud_interception(struct vcpu_svm *svm)
1783 {
1784         int er;
1785
1786         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1787         if (er != EMULATE_DONE)
1788                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1789         return 1;
1790 }
1791
1792 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1793 {
1794         struct vcpu_svm *svm = to_svm(vcpu);
1795
1796         clr_exception_intercept(svm, NM_VECTOR);
1797
1798         svm->vcpu.fpu_active = 1;
1799         update_cr0_intercept(svm);
1800 }
1801
1802 static int nm_interception(struct vcpu_svm *svm)
1803 {
1804         svm_fpu_activate(&svm->vcpu);
1805         return 1;
1806 }
1807
1808 static bool is_erratum_383(void)
1809 {
1810         int err, i;
1811         u64 value;
1812
1813         if (!erratum_383_found)
1814                 return false;
1815
1816         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1817         if (err)
1818                 return false;
1819
1820         /* Bit 62 may or may not be set for this mce */
1821         value &= ~(1ULL << 62);
1822
1823         if (value != 0xb600000000010015ULL)
1824                 return false;
1825
1826         /* Clear MCi_STATUS registers */
1827         for (i = 0; i < 6; ++i)
1828                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1829
1830         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1831         if (!err) {
1832                 u32 low, high;
1833
1834                 value &= ~(1ULL << 2);
1835                 low    = lower_32_bits(value);
1836                 high   = upper_32_bits(value);
1837
1838                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1839         }
1840
1841         /* Flush tlb to evict multi-match entries */
1842         __flush_tlb_all();
1843
1844         return true;
1845 }
1846
1847 static void svm_handle_mce(struct vcpu_svm *svm)
1848 {
1849         if (is_erratum_383()) {
1850                 /*
1851                  * Erratum 383 triggered. Guest state is corrupt so kill the
1852                  * guest.
1853                  */
1854                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1855
1856                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1857
1858                 return;
1859         }
1860
1861         /*
1862          * On an #MC intercept the MCE handler is not called automatically in
1863          * the host. So do it by hand here.
1864          */
1865         asm volatile (
1866                 "int $0x12\n");
1867         /* not sure if we ever come back to this point */
1868
1869         return;
1870 }
1871
1872 static int mc_interception(struct vcpu_svm *svm)
1873 {
1874         return 1;
1875 }
1876
1877 static int shutdown_interception(struct vcpu_svm *svm)
1878 {
1879         struct kvm_run *kvm_run = svm->vcpu.run;
1880
1881         /*
1882          * VMCB is undefined after a SHUTDOWN intercept
1883          * so reinitialize it.
1884          */
1885         clear_page(svm->vmcb);
1886         init_vmcb(svm);
1887
1888         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1889         return 0;
1890 }
1891
1892 static int io_interception(struct vcpu_svm *svm)
1893 {
1894         struct kvm_vcpu *vcpu = &svm->vcpu;
1895         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1896         int size, in, string;
1897         unsigned port;
1898
1899         ++svm->vcpu.stat.io_exits;
1900         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1901         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1902         if (string || in)
1903                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1904
1905         port = io_info >> 16;
1906         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1907         svm->next_rip = svm->vmcb->control.exit_info_2;
1908         skip_emulated_instruction(&svm->vcpu);
1909
1910         return kvm_fast_pio_out(vcpu, size, port);
1911 }
1912
1913 static int nmi_interception(struct vcpu_svm *svm)
1914 {
1915         return 1;
1916 }
1917
1918 static int intr_interception(struct vcpu_svm *svm)
1919 {
1920         ++svm->vcpu.stat.irq_exits;
1921         return 1;
1922 }
1923
1924 static int nop_on_interception(struct vcpu_svm *svm)
1925 {
1926         return 1;
1927 }
1928
1929 static int halt_interception(struct vcpu_svm *svm)
1930 {
1931         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1932         return kvm_emulate_halt(&svm->vcpu);
1933 }
1934
1935 static int vmmcall_interception(struct vcpu_svm *svm)
1936 {
1937         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1938         kvm_emulate_hypercall(&svm->vcpu);
1939         return 1;
1940 }
1941
1942 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1943 {
1944         struct vcpu_svm *svm = to_svm(vcpu);
1945
1946         return svm->nested.nested_cr3;
1947 }
1948
1949 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1950 {
1951         struct vcpu_svm *svm = to_svm(vcpu);
1952         u64 cr3 = svm->nested.nested_cr3;
1953         u64 pdpte;
1954         int ret;
1955
1956         ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1957                                   offset_in_page(cr3) + index * 8, 8);
1958         if (ret)
1959                 return 0;
1960         return pdpte;
1961 }
1962
1963 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1964                                    unsigned long root)
1965 {
1966         struct vcpu_svm *svm = to_svm(vcpu);
1967
1968         svm->vmcb->control.nested_cr3 = root;
1969         mark_dirty(svm->vmcb, VMCB_NPT);
1970         svm_flush_tlb(vcpu);
1971 }
1972
1973 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1974                                        struct x86_exception *fault)
1975 {
1976         struct vcpu_svm *svm = to_svm(vcpu);
1977
1978         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
1979                 /*
1980                  * TODO: track the cause of the nested page fault, and
1981                  * correctly fill in the high bits of exit_info_1.
1982                  */
1983                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1984                 svm->vmcb->control.exit_code_hi = 0;
1985                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
1986                 svm->vmcb->control.exit_info_2 = fault->address;
1987         }
1988
1989         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
1990         svm->vmcb->control.exit_info_1 |= fault->error_code;
1991
1992         /*
1993          * The present bit is always zero for page structure faults on real
1994          * hardware.
1995          */
1996         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
1997                 svm->vmcb->control.exit_info_1 &= ~1;
1998
1999         nested_svm_vmexit(svm);
2000 }
2001
2002 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2003 {
2004         WARN_ON(mmu_is_nested(vcpu));
2005         kvm_init_shadow_mmu(vcpu);
2006         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
2007         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
2008         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
2009         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2010         vcpu->arch.mmu.shadow_root_level = get_npt_level();
2011         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
2012 }
2013
2014 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2015 {
2016         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2017 }
2018
2019 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2020 {
2021         if (!(svm->vcpu.arch.efer & EFER_SVME)
2022             || !is_paging(&svm->vcpu)) {
2023                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2024                 return 1;
2025         }
2026
2027         if (svm->vmcb->save.cpl) {
2028                 kvm_inject_gp(&svm->vcpu, 0);
2029                 return 1;
2030         }
2031
2032        return 0;
2033 }
2034
2035 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2036                                       bool has_error_code, u32 error_code)
2037 {
2038         int vmexit;
2039
2040         if (!is_guest_mode(&svm->vcpu))
2041                 return 0;
2042
2043         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2044         svm->vmcb->control.exit_code_hi = 0;
2045         svm->vmcb->control.exit_info_1 = error_code;
2046         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2047
2048         vmexit = nested_svm_intercept(svm);
2049         if (vmexit == NESTED_EXIT_DONE)
2050                 svm->nested.exit_required = true;
2051
2052         return vmexit;
2053 }
2054
2055 /* This function returns true if it is save to enable the irq window */
2056 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2057 {
2058         if (!is_guest_mode(&svm->vcpu))
2059                 return true;
2060
2061         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2062                 return true;
2063
2064         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2065                 return false;
2066
2067         /*
2068          * if vmexit was already requested (by intercepted exception
2069          * for instance) do not overwrite it with "external interrupt"
2070          * vmexit.
2071          */
2072         if (svm->nested.exit_required)
2073                 return false;
2074
2075         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
2076         svm->vmcb->control.exit_info_1 = 0;
2077         svm->vmcb->control.exit_info_2 = 0;
2078
2079         if (svm->nested.intercept & 1ULL) {
2080                 /*
2081                  * The #vmexit can't be emulated here directly because this
2082                  * code path runs with irqs and preemption disabled. A
2083                  * #vmexit emulation might sleep. Only signal request for
2084                  * the #vmexit here.
2085                  */
2086                 svm->nested.exit_required = true;
2087                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2088                 return false;
2089         }
2090
2091         return true;
2092 }
2093
2094 /* This function returns true if it is save to enable the nmi window */
2095 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2096 {
2097         if (!is_guest_mode(&svm->vcpu))
2098                 return true;
2099
2100         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2101                 return true;
2102
2103         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2104         svm->nested.exit_required = true;
2105
2106         return false;
2107 }
2108
2109 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2110 {
2111         struct page *page;
2112
2113         might_sleep();
2114
2115         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2116         if (is_error_page(page))
2117                 goto error;
2118
2119         *_page = page;
2120
2121         return kmap(page);
2122
2123 error:
2124         kvm_inject_gp(&svm->vcpu, 0);
2125
2126         return NULL;
2127 }
2128
2129 static void nested_svm_unmap(struct page *page)
2130 {
2131         kunmap(page);
2132         kvm_release_page_dirty(page);
2133 }
2134
2135 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2136 {
2137         unsigned port, size, iopm_len;
2138         u16 val, mask;
2139         u8 start_bit;
2140         u64 gpa;
2141
2142         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2143                 return NESTED_EXIT_HOST;
2144
2145         port = svm->vmcb->control.exit_info_1 >> 16;
2146         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2147                 SVM_IOIO_SIZE_SHIFT;
2148         gpa  = svm->nested.vmcb_iopm + (port / 8);
2149         start_bit = port % 8;
2150         iopm_len = (start_bit + size > 8) ? 2 : 1;
2151         mask = (0xf >> (4 - size)) << start_bit;
2152         val = 0;
2153
2154         if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, iopm_len))
2155                 return NESTED_EXIT_DONE;
2156
2157         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2158 }
2159
2160 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2161 {
2162         u32 offset, msr, value;
2163         int write, mask;
2164
2165         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2166                 return NESTED_EXIT_HOST;
2167
2168         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2169         offset = svm_msrpm_offset(msr);
2170         write  = svm->vmcb->control.exit_info_1 & 1;
2171         mask   = 1 << ((2 * (msr & 0xf)) + write);
2172
2173         if (offset == MSR_INVALID)
2174                 return NESTED_EXIT_DONE;
2175
2176         /* Offset is in 32 bit units but need in 8 bit units */
2177         offset *= 4;
2178
2179         if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2180                 return NESTED_EXIT_DONE;
2181
2182         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2183 }
2184
2185 static int nested_svm_exit_special(struct vcpu_svm *svm)
2186 {
2187         u32 exit_code = svm->vmcb->control.exit_code;
2188
2189         switch (exit_code) {
2190         case SVM_EXIT_INTR:
2191         case SVM_EXIT_NMI:
2192         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2193                 return NESTED_EXIT_HOST;
2194         case SVM_EXIT_NPF:
2195                 /* For now we are always handling NPFs when using them */
2196                 if (npt_enabled)
2197                         return NESTED_EXIT_HOST;
2198                 break;
2199         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2200                 /* When we're shadowing, trap PFs, but not async PF */
2201                 if (!npt_enabled && svm->apf_reason == 0)
2202                         return NESTED_EXIT_HOST;
2203                 break;
2204         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2205                 nm_interception(svm);
2206                 break;
2207         default:
2208                 break;
2209         }
2210
2211         return NESTED_EXIT_CONTINUE;
2212 }
2213
2214 /*
2215  * If this function returns true, this #vmexit was already handled
2216  */
2217 static int nested_svm_intercept(struct vcpu_svm *svm)
2218 {
2219         u32 exit_code = svm->vmcb->control.exit_code;
2220         int vmexit = NESTED_EXIT_HOST;
2221
2222         switch (exit_code) {
2223         case SVM_EXIT_MSR:
2224                 vmexit = nested_svm_exit_handled_msr(svm);
2225                 break;
2226         case SVM_EXIT_IOIO:
2227                 vmexit = nested_svm_intercept_ioio(svm);
2228                 break;
2229         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2230                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2231                 if (svm->nested.intercept_cr & bit)
2232                         vmexit = NESTED_EXIT_DONE;
2233                 break;
2234         }
2235         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2236                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2237                 if (svm->nested.intercept_dr & bit)
2238                         vmexit = NESTED_EXIT_DONE;
2239                 break;
2240         }
2241         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2242                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2243                 if (svm->nested.intercept_exceptions & excp_bits)
2244                         vmexit = NESTED_EXIT_DONE;
2245                 /* async page fault always cause vmexit */
2246                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2247                          svm->apf_reason != 0)
2248                         vmexit = NESTED_EXIT_DONE;
2249                 break;
2250         }
2251         case SVM_EXIT_ERR: {
2252                 vmexit = NESTED_EXIT_DONE;
2253                 break;
2254         }
2255         default: {
2256                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2257                 if (svm->nested.intercept & exit_bits)
2258                         vmexit = NESTED_EXIT_DONE;
2259         }
2260         }
2261
2262         return vmexit;
2263 }
2264
2265 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2266 {
2267         int vmexit;
2268
2269         vmexit = nested_svm_intercept(svm);
2270
2271         if (vmexit == NESTED_EXIT_DONE)
2272                 nested_svm_vmexit(svm);
2273
2274         return vmexit;
2275 }
2276
2277 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2278 {
2279         struct vmcb_control_area *dst  = &dst_vmcb->control;
2280         struct vmcb_control_area *from = &from_vmcb->control;
2281
2282         dst->intercept_cr         = from->intercept_cr;
2283         dst->intercept_dr         = from->intercept_dr;
2284         dst->intercept_exceptions = from->intercept_exceptions;
2285         dst->intercept            = from->intercept;
2286         dst->iopm_base_pa         = from->iopm_base_pa;
2287         dst->msrpm_base_pa        = from->msrpm_base_pa;
2288         dst->tsc_offset           = from->tsc_offset;
2289         dst->asid                 = from->asid;
2290         dst->tlb_ctl              = from->tlb_ctl;
2291         dst->int_ctl              = from->int_ctl;
2292         dst->int_vector           = from->int_vector;
2293         dst->int_state            = from->int_state;
2294         dst->exit_code            = from->exit_code;
2295         dst->exit_code_hi         = from->exit_code_hi;
2296         dst->exit_info_1          = from->exit_info_1;
2297         dst->exit_info_2          = from->exit_info_2;
2298         dst->exit_int_info        = from->exit_int_info;
2299         dst->exit_int_info_err    = from->exit_int_info_err;
2300         dst->nested_ctl           = from->nested_ctl;
2301         dst->event_inj            = from->event_inj;
2302         dst->event_inj_err        = from->event_inj_err;
2303         dst->nested_cr3           = from->nested_cr3;
2304         dst->lbr_ctl              = from->lbr_ctl;
2305 }
2306
2307 static int nested_svm_vmexit(struct vcpu_svm *svm)
2308 {
2309         struct vmcb *nested_vmcb;
2310         struct vmcb *hsave = svm->nested.hsave;
2311         struct vmcb *vmcb = svm->vmcb;
2312         struct page *page;
2313
2314         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2315                                        vmcb->control.exit_info_1,
2316                                        vmcb->control.exit_info_2,
2317                                        vmcb->control.exit_int_info,
2318                                        vmcb->control.exit_int_info_err,
2319                                        KVM_ISA_SVM);
2320
2321         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2322         if (!nested_vmcb)
2323                 return 1;
2324
2325         /* Exit Guest-Mode */
2326         leave_guest_mode(&svm->vcpu);
2327         svm->nested.vmcb = 0;
2328
2329         /* Give the current vmcb to the guest */
2330         disable_gif(svm);
2331
2332         nested_vmcb->save.es     = vmcb->save.es;
2333         nested_vmcb->save.cs     = vmcb->save.cs;
2334         nested_vmcb->save.ss     = vmcb->save.ss;
2335         nested_vmcb->save.ds     = vmcb->save.ds;
2336         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2337         nested_vmcb->save.idtr   = vmcb->save.idtr;
2338         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2339         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2340         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2341         nested_vmcb->save.cr2    = vmcb->save.cr2;
2342         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2343         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2344         nested_vmcb->save.rip    = vmcb->save.rip;
2345         nested_vmcb->save.rsp    = vmcb->save.rsp;
2346         nested_vmcb->save.rax    = vmcb->save.rax;
2347         nested_vmcb->save.dr7    = vmcb->save.dr7;
2348         nested_vmcb->save.dr6    = vmcb->save.dr6;
2349         nested_vmcb->save.cpl    = vmcb->save.cpl;
2350
2351         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2352         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2353         nested_vmcb->control.int_state         = vmcb->control.int_state;
2354         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2355         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2356         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2357         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2358         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2359         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2360         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2361
2362         /*
2363          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2364          * to make sure that we do not lose injected events. So check event_inj
2365          * here and copy it to exit_int_info if it is valid.
2366          * Exit_int_info and event_inj can't be both valid because the case
2367          * below only happens on a VMRUN instruction intercept which has
2368          * no valid exit_int_info set.
2369          */
2370         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2371                 struct vmcb_control_area *nc = &nested_vmcb->control;
2372
2373                 nc->exit_int_info     = vmcb->control.event_inj;
2374                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2375         }
2376
2377         nested_vmcb->control.tlb_ctl           = 0;
2378         nested_vmcb->control.event_inj         = 0;
2379         nested_vmcb->control.event_inj_err     = 0;
2380
2381         /* We always set V_INTR_MASKING and remember the old value in hflags */
2382         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2383                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2384
2385         /* Restore the original control entries */
2386         copy_vmcb_control_area(vmcb, hsave);
2387
2388         kvm_clear_exception_queue(&svm->vcpu);
2389         kvm_clear_interrupt_queue(&svm->vcpu);
2390
2391         svm->nested.nested_cr3 = 0;
2392
2393         /* Restore selected save entries */
2394         svm->vmcb->save.es = hsave->save.es;
2395         svm->vmcb->save.cs = hsave->save.cs;
2396         svm->vmcb->save.ss = hsave->save.ss;
2397         svm->vmcb->save.ds = hsave->save.ds;
2398         svm->vmcb->save.gdtr = hsave->save.gdtr;
2399         svm->vmcb->save.idtr = hsave->save.idtr;
2400         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2401         svm_set_efer(&svm->vcpu, hsave->save.efer);
2402         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2403         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2404         if (npt_enabled) {
2405                 svm->vmcb->save.cr3 = hsave->save.cr3;
2406                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2407         } else {
2408                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2409         }
2410         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2411         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2412         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2413         svm->vmcb->save.dr7 = 0;
2414         svm->vmcb->save.cpl = 0;
2415         svm->vmcb->control.exit_int_info = 0;
2416
2417         mark_all_dirty(svm->vmcb);
2418
2419         nested_svm_unmap(page);
2420
2421         nested_svm_uninit_mmu_context(&svm->vcpu);
2422         kvm_mmu_reset_context(&svm->vcpu);
2423         kvm_mmu_load(&svm->vcpu);
2424
2425         return 0;
2426 }
2427
2428 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2429 {
2430         /*
2431          * This function merges the msr permission bitmaps of kvm and the
2432          * nested vmcb. It is optimized in that it only merges the parts where
2433          * the kvm msr permission bitmap may contain zero bits
2434          */
2435         int i;
2436
2437         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2438                 return true;
2439
2440         for (i = 0; i < MSRPM_OFFSETS; i++) {
2441                 u32 value, p;
2442                 u64 offset;
2443
2444                 if (msrpm_offsets[i] == 0xffffffff)
2445                         break;
2446
2447                 p      = msrpm_offsets[i];
2448                 offset = svm->nested.vmcb_msrpm + (p * 4);
2449
2450                 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2451                         return false;
2452
2453                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2454         }
2455
2456         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2457
2458         return true;
2459 }
2460
2461 static bool nested_vmcb_checks(struct vmcb *vmcb)
2462 {
2463         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2464                 return false;
2465
2466         if (vmcb->control.asid == 0)
2467                 return false;
2468
2469         if (vmcb->control.nested_ctl && !npt_enabled)
2470                 return false;
2471
2472         return true;
2473 }
2474
2475 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2476 {
2477         struct vmcb *nested_vmcb;
2478         struct vmcb *hsave = svm->nested.hsave;
2479         struct vmcb *vmcb = svm->vmcb;
2480         struct page *page;
2481         u64 vmcb_gpa;
2482
2483         vmcb_gpa = svm->vmcb->save.rax;
2484
2485         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2486         if (!nested_vmcb)
2487                 return false;
2488
2489         if (!nested_vmcb_checks(nested_vmcb)) {
2490                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2491                 nested_vmcb->control.exit_code_hi = 0;
2492                 nested_vmcb->control.exit_info_1  = 0;
2493                 nested_vmcb->control.exit_info_2  = 0;
2494
2495                 nested_svm_unmap(page);
2496
2497                 return false;
2498         }
2499
2500         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2501                                nested_vmcb->save.rip,
2502                                nested_vmcb->control.int_ctl,
2503                                nested_vmcb->control.event_inj,
2504                                nested_vmcb->control.nested_ctl);
2505
2506         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2507                                     nested_vmcb->control.intercept_cr >> 16,
2508                                     nested_vmcb->control.intercept_exceptions,
2509                                     nested_vmcb->control.intercept);
2510
2511         /* Clear internal status */
2512         kvm_clear_exception_queue(&svm->vcpu);
2513         kvm_clear_interrupt_queue(&svm->vcpu);
2514
2515         /*
2516          * Save the old vmcb, so we don't need to pick what we save, but can
2517          * restore everything when a VMEXIT occurs
2518          */
2519         hsave->save.es     = vmcb->save.es;
2520         hsave->save.cs     = vmcb->save.cs;
2521         hsave->save.ss     = vmcb->save.ss;
2522         hsave->save.ds     = vmcb->save.ds;
2523         hsave->save.gdtr   = vmcb->save.gdtr;
2524         hsave->save.idtr   = vmcb->save.idtr;
2525         hsave->save.efer   = svm->vcpu.arch.efer;
2526         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2527         hsave->save.cr4    = svm->vcpu.arch.cr4;
2528         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2529         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2530         hsave->save.rsp    = vmcb->save.rsp;
2531         hsave->save.rax    = vmcb->save.rax;
2532         if (npt_enabled)
2533                 hsave->save.cr3    = vmcb->save.cr3;
2534         else
2535                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2536
2537         copy_vmcb_control_area(hsave, vmcb);
2538
2539         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2540                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2541         else
2542                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2543
2544         if (nested_vmcb->control.nested_ctl) {
2545                 kvm_mmu_unload(&svm->vcpu);
2546                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2547                 nested_svm_init_mmu_context(&svm->vcpu);
2548         }
2549
2550         /* Load the nested guest state */
2551         svm->vmcb->save.es = nested_vmcb->save.es;
2552         svm->vmcb->save.cs = nested_vmcb->save.cs;
2553         svm->vmcb->save.ss = nested_vmcb->save.ss;
2554         svm->vmcb->save.ds = nested_vmcb->save.ds;
2555         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2556         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2557         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2558         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2559         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2560         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2561         if (npt_enabled) {
2562                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2563                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2564         } else
2565                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2566
2567         /* Guest paging mode is active - reset mmu */
2568         kvm_mmu_reset_context(&svm->vcpu);
2569
2570         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2571         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2572         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2573         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2574
2575         /* In case we don't even reach vcpu_run, the fields are not updated */
2576         svm->vmcb->save.rax = nested_vmcb->save.rax;
2577         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2578         svm->vmcb->save.rip = nested_vmcb->save.rip;
2579         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2580         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2581         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2582
2583         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2584         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2585
2586         /* cache intercepts */
2587         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2588         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2589         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2590         svm->nested.intercept            = nested_vmcb->control.intercept;
2591
2592         svm_flush_tlb(&svm->vcpu);
2593         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2594         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2595                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2596         else
2597                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2598
2599         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2600                 /* We only want the cr8 intercept bits of the guest */
2601                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2602                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2603         }
2604
2605         /* We don't want to see VMMCALLs from a nested guest */
2606         clr_intercept(svm, INTERCEPT_VMMCALL);
2607
2608         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2609         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2610         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2611         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2612         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2613         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2614
2615         nested_svm_unmap(page);
2616
2617         /* Enter Guest-Mode */
2618         enter_guest_mode(&svm->vcpu);
2619
2620         /*
2621          * Merge guest and host intercepts - must be called  with vcpu in
2622          * guest-mode to take affect here
2623          */
2624         recalc_intercepts(svm);
2625
2626         svm->nested.vmcb = vmcb_gpa;
2627
2628         enable_gif(svm);
2629
2630         mark_all_dirty(svm->vmcb);
2631
2632         return true;
2633 }
2634
2635 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2636 {
2637         to_vmcb->save.fs = from_vmcb->save.fs;
2638         to_vmcb->save.gs = from_vmcb->save.gs;
2639         to_vmcb->save.tr = from_vmcb->save.tr;
2640         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2641         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2642         to_vmcb->save.star = from_vmcb->save.star;
2643         to_vmcb->save.lstar = from_vmcb->save.lstar;
2644         to_vmcb->save.cstar = from_vmcb->save.cstar;
2645         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2646         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2647         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2648         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2649 }
2650
2651 static int vmload_interception(struct vcpu_svm *svm)
2652 {
2653         struct vmcb *nested_vmcb;
2654         struct page *page;
2655
2656         if (nested_svm_check_permissions(svm))
2657                 return 1;
2658
2659         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2660         if (!nested_vmcb)
2661                 return 1;
2662
2663         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2664         skip_emulated_instruction(&svm->vcpu);
2665
2666         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2667         nested_svm_unmap(page);
2668
2669         return 1;
2670 }
2671
2672 static int vmsave_interception(struct vcpu_svm *svm)
2673 {
2674         struct vmcb *nested_vmcb;
2675         struct page *page;
2676
2677         if (nested_svm_check_permissions(svm))
2678                 return 1;
2679
2680         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2681         if (!nested_vmcb)
2682                 return 1;
2683
2684         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2685         skip_emulated_instruction(&svm->vcpu);
2686
2687         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2688         nested_svm_unmap(page);
2689
2690         return 1;
2691 }
2692
2693 static int vmrun_interception(struct vcpu_svm *svm)
2694 {
2695         if (nested_svm_check_permissions(svm))
2696                 return 1;
2697
2698         /* Save rip after vmrun instruction */
2699         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2700
2701         if (!nested_svm_vmrun(svm))
2702                 return 1;
2703
2704         if (!nested_svm_vmrun_msrpm(svm))
2705                 goto failed;
2706
2707         return 1;
2708
2709 failed:
2710
2711         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2712         svm->vmcb->control.exit_code_hi = 0;
2713         svm->vmcb->control.exit_info_1  = 0;
2714         svm->vmcb->control.exit_info_2  = 0;
2715
2716         nested_svm_vmexit(svm);
2717
2718         return 1;
2719 }
2720
2721 static int stgi_interception(struct vcpu_svm *svm)
2722 {
2723         if (nested_svm_check_permissions(svm))
2724                 return 1;
2725
2726         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2727         skip_emulated_instruction(&svm->vcpu);
2728         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2729
2730         enable_gif(svm);
2731
2732         return 1;
2733 }
2734
2735 static int clgi_interception(struct vcpu_svm *svm)
2736 {
2737         if (nested_svm_check_permissions(svm))
2738                 return 1;
2739
2740         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2741         skip_emulated_instruction(&svm->vcpu);
2742
2743         disable_gif(svm);
2744
2745         /* After a CLGI no interrupts should come */
2746         svm_clear_vintr(svm);
2747         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2748
2749         mark_dirty(svm->vmcb, VMCB_INTR);
2750
2751         return 1;
2752 }
2753
2754 static int invlpga_interception(struct vcpu_svm *svm)
2755 {
2756         struct kvm_vcpu *vcpu = &svm->vcpu;
2757
2758         trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
2759                           kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2760
2761         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2762         kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2763
2764         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2765         skip_emulated_instruction(&svm->vcpu);
2766         return 1;
2767 }
2768
2769 static int skinit_interception(struct vcpu_svm *svm)
2770 {
2771         trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2772
2773         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2774         return 1;
2775 }
2776
2777 static int wbinvd_interception(struct vcpu_svm *svm)
2778 {
2779         kvm_emulate_wbinvd(&svm->vcpu);
2780         return 1;
2781 }
2782
2783 static int xsetbv_interception(struct vcpu_svm *svm)
2784 {
2785         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2786         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2787
2788         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2789                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2790                 skip_emulated_instruction(&svm->vcpu);
2791         }
2792
2793         return 1;
2794 }
2795
2796 static int task_switch_interception(struct vcpu_svm *svm)
2797 {
2798         u16 tss_selector;
2799         int reason;
2800         int int_type = svm->vmcb->control.exit_int_info &
2801                 SVM_EXITINTINFO_TYPE_MASK;
2802         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2803         uint32_t type =
2804                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2805         uint32_t idt_v =
2806                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2807         bool has_error_code = false;
2808         u32 error_code = 0;
2809
2810         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2811
2812         if (svm->vmcb->control.exit_info_2 &
2813             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2814                 reason = TASK_SWITCH_IRET;
2815         else if (svm->vmcb->control.exit_info_2 &
2816                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2817                 reason = TASK_SWITCH_JMP;
2818         else if (idt_v)
2819                 reason = TASK_SWITCH_GATE;
2820         else
2821                 reason = TASK_SWITCH_CALL;
2822
2823         if (reason == TASK_SWITCH_GATE) {
2824                 switch (type) {
2825                 case SVM_EXITINTINFO_TYPE_NMI:
2826                         svm->vcpu.arch.nmi_injected = false;
2827                         break;
2828                 case SVM_EXITINTINFO_TYPE_EXEPT:
2829                         if (svm->vmcb->control.exit_info_2 &
2830                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2831                                 has_error_code = true;
2832                                 error_code =
2833                                         (u32)svm->vmcb->control.exit_info_2;
2834                         }
2835                         kvm_clear_exception_queue(&svm->vcpu);
2836                         break;
2837                 case SVM_EXITINTINFO_TYPE_INTR:
2838                         kvm_clear_interrupt_queue(&svm->vcpu);
2839                         break;
2840                 default:
2841                         break;
2842                 }
2843         }
2844
2845         if (reason != TASK_SWITCH_GATE ||
2846             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2847             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2848              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2849                 skip_emulated_instruction(&svm->vcpu);
2850
2851         if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2852                 int_vec = -1;
2853
2854         if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2855                                 has_error_code, error_code) == EMULATE_FAIL) {
2856                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2857                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2858                 svm->vcpu.run->internal.ndata = 0;
2859                 return 0;
2860         }
2861         return 1;
2862 }
2863
2864 static int cpuid_interception(struct vcpu_svm *svm)
2865 {
2866         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2867         kvm_emulate_cpuid(&svm->vcpu);
2868         return 1;
2869 }
2870
2871 static int iret_interception(struct vcpu_svm *svm)
2872 {
2873         ++svm->vcpu.stat.nmi_window_exits;
2874         clr_intercept(svm, INTERCEPT_IRET);
2875         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2876         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2877         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2878         return 1;
2879 }
2880
2881 static int invlpg_interception(struct vcpu_svm *svm)
2882 {
2883         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2884                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2885
2886         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2887         skip_emulated_instruction(&svm->vcpu);
2888         return 1;
2889 }
2890
2891 static int emulate_on_interception(struct vcpu_svm *svm)
2892 {
2893         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2894 }
2895
2896 static int rdpmc_interception(struct vcpu_svm *svm)
2897 {
2898         int err;
2899
2900         if (!static_cpu_has(X86_FEATURE_NRIPS))
2901                 return emulate_on_interception(svm);
2902
2903         err = kvm_rdpmc(&svm->vcpu);
2904         kvm_complete_insn_gp(&svm->vcpu, err);
2905
2906         return 1;
2907 }
2908
2909 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
2910                                             unsigned long val)
2911 {
2912         unsigned long cr0 = svm->vcpu.arch.cr0;
2913         bool ret = false;
2914         u64 intercept;
2915
2916         intercept = svm->nested.intercept;
2917
2918         if (!is_guest_mode(&svm->vcpu) ||
2919             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2920                 return false;
2921
2922         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2923         val &= ~SVM_CR0_SELECTIVE_MASK;
2924
2925         if (cr0 ^ val) {
2926                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2927                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2928         }
2929
2930         return ret;
2931 }
2932
2933 #define CR_VALID (1ULL << 63)
2934
2935 static int cr_interception(struct vcpu_svm *svm)
2936 {
2937         int reg, cr;
2938         unsigned long val;
2939         int err;
2940
2941         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2942                 return emulate_on_interception(svm);
2943
2944         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2945                 return emulate_on_interception(svm);
2946
2947         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2948         if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2949                 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2950         else
2951                 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2952
2953         err = 0;
2954         if (cr >= 16) { /* mov to cr */
2955                 cr -= 16;
2956                 val = kvm_register_read(&svm->vcpu, reg);
2957                 switch (cr) {
2958                 case 0:
2959                         if (!check_selective_cr0_intercepted(svm, val))
2960                                 err = kvm_set_cr0(&svm->vcpu, val);
2961                         else
2962                                 return 1;
2963
2964                         break;
2965                 case 3:
2966                         err = kvm_set_cr3(&svm->vcpu, val);
2967                         break;
2968                 case 4:
2969                         err = kvm_set_cr4(&svm->vcpu, val);
2970                         break;
2971                 case 8:
2972                         err = kvm_set_cr8(&svm->vcpu, val);
2973                         break;
2974                 default:
2975                         WARN(1, "unhandled write to CR%d", cr);
2976                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2977                         return 1;
2978                 }
2979         } else { /* mov from cr */
2980                 switch (cr) {
2981                 case 0:
2982                         val = kvm_read_cr0(&svm->vcpu);
2983                         break;
2984                 case 2:
2985                         val = svm->vcpu.arch.cr2;
2986                         break;
2987                 case 3:
2988                         val = kvm_read_cr3(&svm->vcpu);
2989                         break;
2990                 case 4:
2991                         val = kvm_read_cr4(&svm->vcpu);
2992                         break;
2993                 case 8:
2994                         val = kvm_get_cr8(&svm->vcpu);
2995                         break;
2996                 default:
2997                         WARN(1, "unhandled read from CR%d", cr);
2998                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2999                         return 1;
3000                 }
3001                 kvm_register_write(&svm->vcpu, reg, val);
3002         }
3003         kvm_complete_insn_gp(&svm->vcpu, err);
3004
3005         return 1;
3006 }
3007
3008 static int dr_interception(struct vcpu_svm *svm)
3009 {
3010         int reg, dr;
3011         unsigned long val;
3012
3013         if (svm->vcpu.guest_debug == 0) {
3014                 /*
3015                  * No more DR vmexits; force a reload of the debug registers
3016                  * and reenter on this instruction.  The next vmexit will
3017                  * retrieve the full state of the debug registers.
3018                  */
3019                 clr_dr_intercepts(svm);
3020                 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3021                 return 1;
3022         }
3023
3024         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3025                 return emulate_on_interception(svm);
3026
3027         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3028         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3029
3030         if (dr >= 16) { /* mov to DRn */
3031                 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3032                         return 1;
3033                 val = kvm_register_read(&svm->vcpu, reg);
3034                 kvm_set_dr(&svm->vcpu, dr - 16, val);
3035         } else {
3036                 if (!kvm_require_dr(&svm->vcpu, dr))
3037                         return 1;
3038                 kvm_get_dr(&svm->vcpu, dr, &val);
3039                 kvm_register_write(&svm->vcpu, reg, val);
3040         }
3041
3042         skip_emulated_instruction(&svm->vcpu);
3043
3044         return 1;
3045 }
3046
3047 static int cr8_write_interception(struct vcpu_svm *svm)
3048 {
3049         struct kvm_run *kvm_run = svm->vcpu.run;
3050         int r;
3051
3052         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3053         /* instruction emulation calls kvm_set_cr8() */
3054         r = cr_interception(svm);
3055         if (irqchip_in_kernel(svm->vcpu.kvm))
3056                 return r;
3057         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3058                 return r;
3059         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3060         return 0;
3061 }
3062
3063 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3064 {
3065         struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3066         return vmcb->control.tsc_offset +
3067                 svm_scale_tsc(vcpu, host_tsc);
3068 }
3069
3070 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
3071 {
3072         struct vcpu_svm *svm = to_svm(vcpu);
3073
3074         switch (ecx) {
3075         case MSR_IA32_TSC: {
3076                 *data = svm->vmcb->control.tsc_offset +
3077                         svm_scale_tsc(vcpu, native_read_tsc());
3078
3079                 break;
3080         }
3081         case MSR_STAR:
3082                 *data = svm->vmcb->save.star;
3083                 break;
3084 #ifdef CONFIG_X86_64
3085         case MSR_LSTAR:
3086                 *data = svm->vmcb->save.lstar;
3087                 break;
3088         case MSR_CSTAR:
3089                 *data = svm->vmcb->save.cstar;
3090                 break;
3091         case MSR_KERNEL_GS_BASE:
3092                 *data = svm->vmcb->save.kernel_gs_base;
3093                 break;
3094         case MSR_SYSCALL_MASK:
3095                 *data = svm->vmcb->save.sfmask;
3096                 break;
3097 #endif
3098         case MSR_IA32_SYSENTER_CS:
3099                 *data = svm->vmcb->save.sysenter_cs;
3100                 break;
3101         case MSR_IA32_SYSENTER_EIP:
3102                 *data = svm->sysenter_eip;
3103                 break;
3104         case MSR_IA32_SYSENTER_ESP:
3105                 *data = svm->sysenter_esp;
3106                 break;
3107         /*
3108          * Nobody will change the following 5 values in the VMCB so we can
3109          * safely return them on rdmsr. They will always be 0 until LBRV is
3110          * implemented.
3111          */
3112         case MSR_IA32_DEBUGCTLMSR:
3113                 *data = svm->vmcb->save.dbgctl;
3114                 break;
3115         case MSR_IA32_LASTBRANCHFROMIP:
3116                 *data = svm->vmcb->save.br_from;
3117                 break;
3118         case MSR_IA32_LASTBRANCHTOIP:
3119                 *data = svm->vmcb->save.br_to;
3120                 break;
3121         case MSR_IA32_LASTINTFROMIP:
3122                 *data = svm->vmcb->save.last_excp_from;
3123                 break;
3124         case MSR_IA32_LASTINTTOIP:
3125                 *data = svm->vmcb->save.last_excp_to;
3126                 break;
3127         case MSR_VM_HSAVE_PA:
3128                 *data = svm->nested.hsave_msr;
3129                 break;
3130         case MSR_VM_CR:
3131                 *data = svm->nested.vm_cr_msr;
3132                 break;
3133         case MSR_IA32_UCODE_REV:
3134                 *data = 0x01000065;
3135                 break;
3136         default:
3137                 return kvm_get_msr_common(vcpu, ecx, data);
3138         }
3139         return 0;
3140 }
3141
3142 static int rdmsr_interception(struct vcpu_svm *svm)
3143 {
3144         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3145         u64 data;
3146
3147         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
3148                 trace_kvm_msr_read_ex(ecx);
3149                 kvm_inject_gp(&svm->vcpu, 0);
3150         } else {
3151                 trace_kvm_msr_read(ecx, data);
3152
3153                 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, data & 0xffffffff);
3154                 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX, data >> 32);
3155                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3156                 skip_emulated_instruction(&svm->vcpu);
3157         }
3158         return 1;
3159 }
3160
3161 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3162 {
3163         struct vcpu_svm *svm = to_svm(vcpu);
3164         int svm_dis, chg_mask;
3165
3166         if (data & ~SVM_VM_CR_VALID_MASK)
3167                 return 1;
3168
3169         chg_mask = SVM_VM_CR_VALID_MASK;
3170
3171         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3172                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3173
3174         svm->nested.vm_cr_msr &= ~chg_mask;
3175         svm->nested.vm_cr_msr |= (data & chg_mask);
3176
3177         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3178
3179         /* check for svm_disable while efer.svme is set */
3180         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3181                 return 1;
3182
3183         return 0;
3184 }
3185
3186 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3187 {
3188         struct vcpu_svm *svm = to_svm(vcpu);
3189
3190         u32 ecx = msr->index;
3191         u64 data = msr->data;
3192         switch (ecx) {
3193         case MSR_IA32_TSC:
3194                 kvm_write_tsc(vcpu, msr);
3195                 break;
3196         case MSR_STAR:
3197                 svm->vmcb->save.star = data;
3198                 break;
3199 #ifdef CONFIG_X86_64
3200         case MSR_LSTAR:
3201                 svm->vmcb->save.lstar = data;
3202                 break;
3203         case MSR_CSTAR:
3204                 svm->vmcb->save.cstar = data;
3205                 break;
3206         case MSR_KERNEL_GS_BASE:
3207                 svm->vmcb->save.kernel_gs_base = data;
3208                 break;
3209         case MSR_SYSCALL_MASK:
3210                 svm->vmcb->save.sfmask = data;
3211                 break;
3212 #endif
3213         case MSR_IA32_SYSENTER_CS:
3214                 svm->vmcb->save.sysenter_cs = data;
3215                 break;
3216         case MSR_IA32_SYSENTER_EIP:
3217                 svm->sysenter_eip = data;
3218                 svm->vmcb->save.sysenter_eip = data;
3219                 break;
3220         case MSR_IA32_SYSENTER_ESP:
3221                 svm->sysenter_esp = data;
3222                 svm->vmcb->save.sysenter_esp = data;
3223                 break;
3224         case MSR_IA32_DEBUGCTLMSR:
3225                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3226                         vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3227                                     __func__, data);
3228                         break;
3229                 }
3230                 if (data & DEBUGCTL_RESERVED_BITS)
3231                         return 1;
3232
3233                 svm->vmcb->save.dbgctl = data;
3234                 mark_dirty(svm->vmcb, VMCB_LBR);
3235                 if (data & (1ULL<<0))
3236                         svm_enable_lbrv(svm);
3237                 else
3238                         svm_disable_lbrv(svm);
3239                 break;
3240         case MSR_VM_HSAVE_PA:
3241                 svm->nested.hsave_msr = data;
3242                 break;
3243         case MSR_VM_CR:
3244                 return svm_set_vm_cr(vcpu, data);
3245         case MSR_VM_IGNNE:
3246                 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3247                 break;
3248         default:
3249                 return kvm_set_msr_common(vcpu, msr);
3250         }
3251         return 0;
3252 }
3253
3254 static int wrmsr_interception(struct vcpu_svm *svm)
3255 {
3256         struct msr_data msr;
3257         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3258         u64 data = kvm_read_edx_eax(&svm->vcpu);
3259
3260         msr.data = data;
3261         msr.index = ecx;
3262         msr.host_initiated = false;
3263
3264         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3265         if (kvm_set_msr(&svm->vcpu, &msr)) {
3266                 trace_kvm_msr_write_ex(ecx, data);
3267                 kvm_inject_gp(&svm->vcpu, 0);
3268         } else {
3269                 trace_kvm_msr_write(ecx, data);
3270                 skip_emulated_instruction(&svm->vcpu);
3271         }
3272         return 1;
3273 }
3274
3275 static int msr_interception(struct vcpu_svm *svm)
3276 {
3277         if (svm->vmcb->control.exit_info_1)
3278                 return wrmsr_interception(svm);
3279         else
3280                 return rdmsr_interception(svm);
3281 }
3282
3283 static int interrupt_window_interception(struct vcpu_svm *svm)
3284 {
3285         struct kvm_run *kvm_run = svm->vcpu.run;
3286
3287         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3288         svm_clear_vintr(svm);
3289         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3290         mark_dirty(svm->vmcb, VMCB_INTR);
3291         ++svm->vcpu.stat.irq_window_exits;
3292         /*
3293          * If the user space waits to inject interrupts, exit as soon as
3294          * possible
3295          */
3296         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3297             kvm_run->request_interrupt_window &&
3298             !kvm_cpu_has_interrupt(&svm->vcpu)) {
3299                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3300                 return 0;
3301         }
3302
3303         return 1;
3304 }
3305
3306 static int pause_interception(struct vcpu_svm *svm)
3307 {
3308         kvm_vcpu_on_spin(&(svm->vcpu));
3309         return 1;
3310 }
3311
3312 static int nop_interception(struct vcpu_svm *svm)
3313 {
3314         skip_emulated_instruction(&(svm->vcpu));
3315         return 1;
3316 }
3317
3318 static int monitor_interception(struct vcpu_svm *svm)
3319 {
3320         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3321         return nop_interception(svm);
3322 }
3323
3324 static int mwait_interception(struct vcpu_svm *svm)
3325 {
3326         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3327         return nop_interception(svm);
3328 }
3329
3330 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3331         [SVM_EXIT_READ_CR0]                     = cr_interception,
3332         [SVM_EXIT_READ_CR3]                     = cr_interception,
3333         [SVM_EXIT_READ_CR4]                     = cr_interception,
3334         [SVM_EXIT_READ_CR8]                     = cr_interception,
3335         [SVM_EXIT_CR0_SEL_WRITE]                = cr_interception,
3336         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
3337         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
3338         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
3339         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
3340         [SVM_EXIT_READ_DR0]                     = dr_interception,
3341         [SVM_EXIT_READ_DR1]                     = dr_interception,
3342         [SVM_EXIT_READ_DR2]                     = dr_interception,
3343         [SVM_EXIT_READ_DR3]                     = dr_interception,
3344         [SVM_EXIT_READ_DR4]                     = dr_interception,
3345         [SVM_EXIT_READ_DR5]                     = dr_interception,
3346         [SVM_EXIT_READ_DR6]                     = dr_interception,
3347         [SVM_EXIT_READ_DR7]                     = dr_interception,
3348         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
3349         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
3350         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
3351         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
3352         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
3353         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
3354         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
3355         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
3356         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
3357         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
3358         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
3359         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
3360         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
3361         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
3362         [SVM_EXIT_INTR]                         = intr_interception,
3363         [SVM_EXIT_NMI]                          = nmi_interception,
3364         [SVM_EXIT_SMI]                          = nop_on_interception,
3365         [SVM_EXIT_INIT]                         = nop_on_interception,
3366         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
3367         [SVM_EXIT_RDPMC]                        = rdpmc_interception,
3368         [SVM_EXIT_CPUID]                        = cpuid_interception,
3369         [SVM_EXIT_IRET]                         = iret_interception,
3370         [SVM_EXIT_INVD]                         = emulate_on_interception,
3371         [SVM_EXIT_PAUSE]                        = pause_interception,
3372         [SVM_EXIT_HLT]                          = halt_interception,
3373         [SVM_EXIT_INVLPG]                       = invlpg_interception,
3374         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
3375         [SVM_EXIT_IOIO]                         = io_interception,
3376         [SVM_EXIT_MSR]                          = msr_interception,
3377         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
3378         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
3379         [SVM_EXIT_VMRUN]                        = vmrun_interception,
3380         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
3381         [SVM_EXIT_VMLOAD]                       = vmload_interception,
3382         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
3383         [SVM_EXIT_STGI]                         = stgi_interception,
3384         [SVM_EXIT_CLGI]                         = clgi_interception,
3385         [SVM_EXIT_SKINIT]                       = skinit_interception,
3386         [SVM_EXIT_WBINVD]                       = wbinvd_interception,
3387         [SVM_EXIT_MONITOR]                      = monitor_interception,
3388         [SVM_EXIT_MWAIT]                        = mwait_interception,
3389         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
3390         [SVM_EXIT_NPF]                          = pf_interception,
3391 };
3392
3393 static void dump_vmcb(struct kvm_vcpu *vcpu)
3394 {
3395         struct vcpu_svm *svm = to_svm(vcpu);
3396         struct vmcb_control_area *control = &svm->vmcb->control;
3397         struct vmcb_save_area *save = &svm->vmcb->save;
3398
3399         pr_err("VMCB Control Area:\n");
3400         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3401         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3402         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3403         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3404         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3405         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3406         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3407         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3408         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3409         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3410         pr_err("%-20s%d\n", "asid:", control->asid);
3411         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3412         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3413         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3414         pr_err("%-20s%08x\n", "int_state:", control->int_state);
3415         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3416         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3417         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3418         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3419         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3420         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3421         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3422         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3423         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3424         pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3425         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3426         pr_err("VMCB State Save Area:\n");
3427         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3428                "es:",
3429                save->es.selector, save->es.attrib,
3430                save->es.limit, save->es.base);
3431         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3432                "cs:",
3433                save->cs.selector, save->cs.attrib,
3434                save->cs.limit, save->cs.base);
3435         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3436                "ss:",
3437                save->ss.selector, save->ss.attrib,
3438                save->ss.limit, save->ss.base);
3439         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3440                "ds:",
3441                save->ds.selector, save->ds.attrib,
3442                save->ds.limit, save->ds.base);
3443         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3444                "fs:",
3445                save->fs.selector, save->fs.attrib,
3446                save->fs.limit, save->fs.base);
3447         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3448                "gs:",
3449                save->gs.selector, save->gs.attrib,
3450                save->gs.limit, save->gs.base);
3451         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3452                "gdtr:",
3453                save->gdtr.selector, save->gdtr.attrib,
3454                save->gdtr.limit, save->gdtr.base);
3455         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3456                "ldtr:",
3457                save->ldtr.selector, save->ldtr.attrib,
3458                save->ldtr.limit, save->ldtr.base);
3459         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3460                "idtr:",
3461                save->idtr.selector, save->idtr.attrib,
3462                save->idtr.limit, save->idtr.base);
3463         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3464                "tr:",
3465                save->tr.selector, save->tr.attrib,
3466                save->tr.limit, save->tr.base);
3467         pr_err("cpl:            %d                efer:         %016llx\n",
3468                 save->cpl, save->efer);
3469         pr_err("%-15s %016llx %-13s %016llx\n",
3470                "cr0:", save->cr0, "cr2:", save->cr2);
3471         pr_err("%-15s %016llx %-13s %016llx\n",
3472                "cr3:", save->cr3, "cr4:", save->cr4);
3473         pr_err("%-15s %016llx %-13s %016llx\n",
3474                "dr6:", save->dr6, "dr7:", save->dr7);
3475         pr_err("%-15s %016llx %-13s %016llx\n",
3476                "rip:", save->rip, "rflags:", save->rflags);
3477         pr_err("%-15s %016llx %-13s %016llx\n",
3478                "rsp:", save->rsp, "rax:", save->rax);
3479         pr_err("%-15s %016llx %-13s %016llx\n",
3480                "star:", save->star, "lstar:", save->lstar);
3481         pr_err("%-15s %016llx %-13s %016llx\n",
3482                "cstar:", save->cstar, "sfmask:", save->sfmask);
3483         pr_err("%-15s %016llx %-13s %016llx\n",
3484                "kernel_gs_base:", save->kernel_gs_base,
3485                "sysenter_cs:", save->sysenter_cs);
3486         pr_err("%-15s %016llx %-13s %016llx\n",
3487                "sysenter_esp:", save->sysenter_esp,
3488                "sysenter_eip:", save->sysenter_eip);
3489         pr_err("%-15s %016llx %-13s %016llx\n",
3490                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3491         pr_err("%-15s %016llx %-13s %016llx\n",
3492                "br_from:", save->br_from, "br_to:", save->br_to);
3493         pr_err("%-15s %016llx %-13s %016llx\n",
3494                "excp_from:", save->last_excp_from,
3495                "excp_to:", save->last_excp_to);
3496 }
3497
3498 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3499 {
3500         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3501
3502         *info1 = control->exit_info_1;
3503         *info2 = control->exit_info_2;
3504 }
3505
3506 static int handle_exit(struct kvm_vcpu *vcpu)
3507 {
3508         struct vcpu_svm *svm = to_svm(vcpu);
3509         struct kvm_run *kvm_run = vcpu->run;
3510         u32 exit_code = svm->vmcb->control.exit_code;
3511
3512         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3513                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3514         if (npt_enabled)
3515                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3516
3517         if (unlikely(svm->nested.exit_required)) {
3518                 nested_svm_vmexit(svm);
3519                 svm->nested.exit_required = false;
3520
3521                 return 1;
3522         }
3523
3524         if (is_guest_mode(vcpu)) {
3525                 int vmexit;
3526
3527                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3528                                         svm->vmcb->control.exit_info_1,
3529                                         svm->vmcb->control.exit_info_2,
3530                                         svm->vmcb->control.exit_int_info,
3531                                         svm->vmcb->control.exit_int_info_err,
3532                                         KVM_ISA_SVM);
3533
3534                 vmexit = nested_svm_exit_special(svm);
3535
3536                 if (vmexit == NESTED_EXIT_CONTINUE)
3537                         vmexit = nested_svm_exit_handled(svm);
3538
3539                 if (vmexit == NESTED_EXIT_DONE)
3540                         return 1;
3541         }
3542
3543         svm_complete_interrupts(svm);
3544
3545         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3546                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3547                 kvm_run->fail_entry.hardware_entry_failure_reason
3548                         = svm->vmcb->control.exit_code;
3549                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3550                 dump_vmcb(vcpu);
3551                 return 0;
3552         }
3553
3554         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3555             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3556             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3557             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3558                 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3559                        "exit_code 0x%x\n",
3560                        __func__, svm->vmcb->control.exit_int_info,
3561                        exit_code);
3562
3563         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3564             || !svm_exit_handlers[exit_code]) {
3565                 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
3566                 kvm_queue_exception(vcpu, UD_VECTOR);
3567                 return 1;
3568         }
3569
3570         return svm_exit_handlers[exit_code](svm);
3571 }
3572
3573 static void reload_tss(struct kvm_vcpu *vcpu)
3574 {
3575         int cpu = raw_smp_processor_id();
3576
3577         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3578         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3579         load_TR_desc();
3580 }
3581
3582 static void pre_svm_run(struct vcpu_svm *svm)
3583 {
3584         int cpu = raw_smp_processor_id();
3585
3586         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3587
3588         /* FIXME: handle wraparound of asid_generation */
3589         if (svm->asid_generation != sd->asid_generation)
3590                 new_asid(svm, sd);
3591 }
3592
3593 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3594 {
3595         struct vcpu_svm *svm = to_svm(vcpu);
3596
3597         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3598         vcpu->arch.hflags |= HF_NMI_MASK;
3599         set_intercept(svm, INTERCEPT_IRET);
3600         ++vcpu->stat.nmi_injections;
3601 }
3602
3603 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3604 {
3605         struct vmcb_control_area *control;
3606
3607         control = &svm->vmcb->control;
3608         control->int_vector = irq;
3609         control->int_ctl &= ~V_INTR_PRIO_MASK;
3610         control->int_ctl |= V_IRQ_MASK |
3611                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3612         mark_dirty(svm->vmcb, VMCB_INTR);
3613 }
3614
3615 static void svm_set_irq(struct kvm_vcpu *vcpu)
3616 {
3617         struct vcpu_svm *svm = to_svm(vcpu);
3618
3619         BUG_ON(!(gif_set(svm)));
3620
3621         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3622         ++vcpu->stat.irq_injections;
3623
3624         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3625                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3626 }
3627
3628 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3629 {
3630         struct vcpu_svm *svm = to_svm(vcpu);
3631
3632         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3633                 return;
3634
3635         clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3636
3637         if (irr == -1)
3638                 return;
3639
3640         if (tpr >= irr)
3641                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3642 }
3643
3644 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3645 {
3646         return;
3647 }
3648
3649 static int svm_vm_has_apicv(struct kvm *kvm)
3650 {
3651         return 0;
3652 }
3653
3654 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3655 {
3656         return;
3657 }
3658
3659 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3660 {
3661         return;
3662 }
3663
3664 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3665 {
3666         struct vcpu_svm *svm = to_svm(vcpu);
3667         struct vmcb *vmcb = svm->vmcb;
3668         int ret;
3669         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3670               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3671         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3672
3673         return ret;
3674 }
3675
3676 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3677 {
3678         struct vcpu_svm *svm = to_svm(vcpu);
3679
3680         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3681 }
3682
3683 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3684 {
3685         struct vcpu_svm *svm = to_svm(vcpu);
3686
3687         if (masked) {
3688                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3689                 set_intercept(svm, INTERCEPT_IRET);
3690         } else {
3691                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3692                 clr_intercept(svm, INTERCEPT_IRET);
3693         }
3694 }
3695
3696 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3697 {
3698         struct vcpu_svm *svm = to_svm(vcpu);
3699         struct vmcb *vmcb = svm->vmcb;
3700         int ret;
3701
3702         if (!gif_set(svm) ||
3703              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3704                 return 0;
3705
3706         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3707
3708         if (is_guest_mode(vcpu))
3709                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3710
3711         return ret;
3712 }
3713
3714 static void enable_irq_window(struct kvm_vcpu *vcpu)
3715 {
3716         struct vcpu_svm *svm = to_svm(vcpu);
3717
3718         /*
3719          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3720          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3721          * get that intercept, this function will be called again though and
3722          * we'll get the vintr intercept.
3723          */
3724         if (gif_set(svm) && nested_svm_intr(svm)) {
3725                 svm_set_vintr(svm);
3726                 svm_inject_irq(svm, 0x0);
3727         }
3728 }
3729
3730 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3731 {
3732         struct vcpu_svm *svm = to_svm(vcpu);
3733
3734         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3735             == HF_NMI_MASK)
3736                 return; /* IRET will cause a vm exit */
3737
3738         /*
3739          * Something prevents NMI from been injected. Single step over possible
3740          * problem (IRET or exception injection or interrupt shadow)
3741          */
3742         svm->nmi_singlestep = true;
3743         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3744         update_db_bp_intercept(vcpu);
3745 }
3746
3747 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3748 {
3749         return 0;
3750 }
3751
3752 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3753 {
3754         struct vcpu_svm *svm = to_svm(vcpu);
3755
3756         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3757                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3758         else
3759                 svm->asid_generation--;
3760 }
3761
3762 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3763 {
3764 }
3765
3766 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3767 {
3768         struct vcpu_svm *svm = to_svm(vcpu);
3769
3770         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3771                 return;
3772
3773         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3774                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3775                 kvm_set_cr8(vcpu, cr8);
3776         }
3777 }
3778
3779 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3780 {
3781         struct vcpu_svm *svm = to_svm(vcpu);
3782         u64 cr8;
3783
3784         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3785                 return;
3786
3787         cr8 = kvm_get_cr8(vcpu);
3788         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3789         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3790 }
3791
3792 static void svm_complete_interrupts(struct vcpu_svm *svm)
3793 {
3794         u8 vector;
3795         int type;
3796         u32 exitintinfo = svm->vmcb->control.exit_int_info;
3797         unsigned int3_injected = svm->int3_injected;
3798
3799         svm->int3_injected = 0;
3800
3801         /*
3802          * If we've made progress since setting HF_IRET_MASK, we've
3803          * executed an IRET and can allow NMI injection.
3804          */
3805         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3806             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3807                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3808                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3809         }
3810
3811         svm->vcpu.arch.nmi_injected = false;
3812         kvm_clear_exception_queue(&svm->vcpu);
3813         kvm_clear_interrupt_queue(&svm->vcpu);
3814
3815         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3816                 return;
3817
3818         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3819
3820         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3821         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3822
3823         switch (type) {
3824         case SVM_EXITINTINFO_TYPE_NMI:
3825                 svm->vcpu.arch.nmi_injected = true;
3826                 break;
3827         case SVM_EXITINTINFO_TYPE_EXEPT:
3828                 /*
3829                  * In case of software exceptions, do not reinject the vector,
3830                  * but re-execute the instruction instead. Rewind RIP first
3831                  * if we emulated INT3 before.
3832                  */
3833                 if (kvm_exception_is_soft(vector)) {
3834                         if (vector == BP_VECTOR && int3_injected &&
3835                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3836                                 kvm_rip_write(&svm->vcpu,
3837                                               kvm_rip_read(&svm->vcpu) -
3838                                               int3_injected);
3839                         break;
3840                 }
3841                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3842                         u32 err = svm->vmcb->control.exit_int_info_err;
3843                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
3844
3845                 } else
3846                         kvm_requeue_exception(&svm->vcpu, vector);
3847                 break;
3848         case SVM_EXITINTINFO_TYPE_INTR:
3849                 kvm_queue_interrupt(&svm->vcpu, vector, false);
3850                 break;
3851         default:
3852                 break;
3853         }
3854 }
3855
3856 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3857 {
3858         struct vcpu_svm *svm = to_svm(vcpu);
3859         struct vmcb_control_area *control = &svm->vmcb->control;
3860
3861         control->exit_int_info = control->event_inj;
3862         control->exit_int_info_err = control->event_inj_err;
3863         control->event_inj = 0;
3864         svm_complete_interrupts(svm);
3865 }
3866
3867 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3868 {
3869         struct vcpu_svm *svm = to_svm(vcpu);
3870
3871         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3872         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3873         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3874
3875         /*
3876          * A vmexit emulation is required before the vcpu can be executed
3877          * again.
3878          */
3879         if (unlikely(svm->nested.exit_required))
3880                 return;
3881
3882         pre_svm_run(svm);
3883
3884         sync_lapic_to_cr8(vcpu);
3885
3886         svm->vmcb->save.cr2 = vcpu->arch.cr2;
3887
3888         clgi();
3889
3890         local_irq_enable();
3891
3892         asm volatile (
3893                 "push %%" _ASM_BP "; \n\t"
3894                 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
3895                 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
3896                 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
3897                 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
3898                 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
3899                 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
3900 #ifdef CONFIG_X86_64
3901                 "mov %c[r8](%[svm]),  %%r8  \n\t"
3902                 "mov %c[r9](%[svm]),  %%r9  \n\t"
3903                 "mov %c[r10](%[svm]), %%r10 \n\t"
3904                 "mov %c[r11](%[svm]), %%r11 \n\t"
3905                 "mov %c[r12](%[svm]), %%r12 \n\t"
3906                 "mov %c[r13](%[svm]), %%r13 \n\t"
3907                 "mov %c[r14](%[svm]), %%r14 \n\t"
3908                 "mov %c[r15](%[svm]), %%r15 \n\t"
3909 #endif
3910
3911                 /* Enter guest mode */
3912                 "push %%" _ASM_AX " \n\t"
3913                 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
3914                 __ex(SVM_VMLOAD) "\n\t"
3915                 __ex(SVM_VMRUN) "\n\t"
3916                 __ex(SVM_VMSAVE) "\n\t"
3917                 "pop %%" _ASM_AX " \n\t"
3918
3919                 /* Save guest registers, load host registers */
3920                 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
3921                 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
3922                 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
3923                 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
3924                 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
3925                 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
3926 #ifdef CONFIG_X86_64
3927                 "mov %%r8,  %c[r8](%[svm]) \n\t"
3928                 "mov %%r9,  %c[r9](%[svm]) \n\t"
3929                 "mov %%r10, %c[r10](%[svm]) \n\t"
3930                 "mov %%r11, %c[r11](%[svm]) \n\t"
3931                 "mov %%r12, %c[r12](%[svm]) \n\t"
3932                 "mov %%r13, %c[r13](%[svm]) \n\t"
3933                 "mov %%r14, %c[r14](%[svm]) \n\t"
3934                 "mov %%r15, %c[r15](%[svm]) \n\t"
3935 #endif
3936                 "pop %%" _ASM_BP
3937                 :
3938                 : [svm]"a"(svm),
3939                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3940                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3941                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3942                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3943                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3944                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3945                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3946 #ifdef CONFIG_X86_64
3947                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3948                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3949                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3950                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3951                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3952                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3953                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3954                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3955 #endif
3956                 : "cc", "memory"
3957 #ifdef CONFIG_X86_64
3958                 , "rbx", "rcx", "rdx", "rsi", "rdi"
3959                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3960 #else
3961                 , "ebx", "ecx", "edx", "esi", "edi"
3962 #endif
3963                 );
3964
3965 #ifdef CONFIG_X86_64
3966         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3967 #else
3968         loadsegment(fs, svm->host.fs);
3969 #ifndef CONFIG_X86_32_LAZY_GS
3970         loadsegment(gs, svm->host.gs);
3971 #endif
3972 #endif
3973
3974         reload_tss(vcpu);
3975
3976         local_irq_disable();
3977
3978         vcpu->arch.cr2 = svm->vmcb->save.cr2;
3979         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3980         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3981         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3982
3983         trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
3984
3985         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3986                 kvm_before_handle_nmi(&svm->vcpu);
3987
3988         stgi();
3989
3990         /* Any pending NMI will happen here */
3991
3992         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3993                 kvm_after_handle_nmi(&svm->vcpu);
3994
3995         sync_cr8_to_lapic(vcpu);
3996
3997         svm->next_rip = 0;
3998
3999         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4000
4001         /* if exit due to PF check for async PF */
4002         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4003                 svm->apf_reason = kvm_read_and_reset_pf_reason();
4004
4005         if (npt_enabled) {
4006                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4007                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4008         }
4009
4010         /*
4011          * We need to handle MC intercepts here before the vcpu has a chance to
4012          * change the physical cpu
4013          */
4014         if (unlikely(svm->vmcb->control.exit_code ==
4015                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
4016                 svm_handle_mce(svm);
4017
4018         mark_all_clean(svm->vmcb);
4019 }
4020
4021 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4022 {
4023         struct vcpu_svm *svm = to_svm(vcpu);
4024
4025         svm->vmcb->save.cr3 = root;
4026         mark_dirty(svm->vmcb, VMCB_CR);
4027         svm_flush_tlb(vcpu);
4028 }
4029
4030 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4031 {
4032         struct vcpu_svm *svm = to_svm(vcpu);
4033
4034         svm->vmcb->control.nested_cr3 = root;
4035         mark_dirty(svm->vmcb, VMCB_NPT);
4036
4037         /* Also sync guest cr3 here in case we live migrate */
4038         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4039         mark_dirty(svm->vmcb, VMCB_CR);
4040
4041         svm_flush_tlb(vcpu);
4042 }
4043
4044 static int is_disabled(void)
4045 {
4046         u64 vm_cr;
4047
4048         rdmsrl(MSR_VM_CR, vm_cr);
4049         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4050                 return 1;
4051
4052         return 0;
4053 }
4054
4055 static void
4056 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4057 {
4058         /*
4059          * Patch in the VMMCALL instruction:
4060          */
4061         hypercall[0] = 0x0f;
4062         hypercall[1] = 0x01;
4063         hypercall[2] = 0xd9;
4064 }
4065
4066 static void svm_check_processor_compat(void *rtn)
4067 {
4068         *(int *)rtn = 0;
4069 }
4070
4071 static bool svm_cpu_has_accelerated_tpr(void)
4072 {
4073         return false;
4074 }
4075
4076 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4077 {
4078         return 0;
4079 }
4080
4081 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4082 {
4083 }
4084
4085 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4086 {
4087         switch (func) {
4088         case 0x80000001:
4089                 if (nested)
4090                         entry->ecx |= (1 << 2); /* Set SVM bit */
4091                 break;
4092         case 0x8000000A:
4093                 entry->eax = 1; /* SVM revision 1 */
4094                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4095                                    ASID emulation to nested SVM */
4096                 entry->ecx = 0; /* Reserved */
4097                 entry->edx = 0; /* Per default do not support any
4098                                    additional features */
4099
4100                 /* Support next_rip if host supports it */
4101                 if (boot_cpu_has(X86_FEATURE_NRIPS))
4102                         entry->edx |= SVM_FEATURE_NRIP;
4103
4104                 /* Support NPT for the guest if enabled */
4105                 if (npt_enabled)
4106                         entry->edx |= SVM_FEATURE_NPT;
4107
4108                 break;
4109         }
4110 }
4111
4112 static int svm_get_lpage_level(void)
4113 {
4114         return PT_PDPE_LEVEL;
4115 }
4116
4117 static bool svm_rdtscp_supported(void)
4118 {
4119         return false;
4120 }
4121
4122 static bool svm_invpcid_supported(void)
4123 {
4124         return false;
4125 }
4126
4127 static bool svm_mpx_supported(void)
4128 {
4129         return false;
4130 }
4131
4132 static bool svm_xsaves_supported(void)
4133 {
4134         return false;
4135 }
4136
4137 static bool svm_has_wbinvd_exit(void)
4138 {
4139         return true;
4140 }
4141
4142 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4143 {
4144         struct vcpu_svm *svm = to_svm(vcpu);
4145
4146         set_exception_intercept(svm, NM_VECTOR);
4147         update_cr0_intercept(svm);
4148 }
4149
4150 #define PRE_EX(exit)  { .exit_code = (exit), \
4151                         .stage = X86_ICPT_PRE_EXCEPT, }
4152 #define POST_EX(exit) { .exit_code = (exit), \
4153                         .stage = X86_ICPT_POST_EXCEPT, }
4154 #define POST_MEM(exit) { .exit_code = (exit), \
4155                         .stage = X86_ICPT_POST_MEMACCESS, }
4156
4157 static const struct __x86_intercept {
4158         u32 exit_code;
4159         enum x86_intercept_stage stage;
4160 } x86_intercept_map[] = {
4161         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
4162         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
4163         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
4164         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
4165         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
4166         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
4167         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
4168         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
4169         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
4170         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
4171         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
4172         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
4173         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
4174         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
4175         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
4176         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
4177         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
4178         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
4179         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
4180         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
4181         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
4182         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
4183         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
4184         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
4185         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
4186         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
4187         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
4188         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
4189         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
4190         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
4191         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
4192         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
4193         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
4194         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
4195         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
4196         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
4197         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
4198         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
4199         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
4200         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
4201         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
4202         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
4203         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
4204         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
4205         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
4206         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
4207 };
4208
4209 #undef PRE_EX
4210 #undef POST_EX
4211 #undef POST_MEM
4212
4213 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4214                                struct x86_instruction_info *info,
4215                                enum x86_intercept_stage stage)
4216 {
4217         struct vcpu_svm *svm = to_svm(vcpu);
4218         int vmexit, ret = X86EMUL_CONTINUE;
4219         struct __x86_intercept icpt_info;
4220         struct vmcb *vmcb = svm->vmcb;
4221
4222         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4223                 goto out;
4224
4225         icpt_info = x86_intercept_map[info->intercept];
4226
4227         if (stage != icpt_info.stage)
4228                 goto out;
4229
4230         switch (icpt_info.exit_code) {
4231         case SVM_EXIT_READ_CR0:
4232                 if (info->intercept == x86_intercept_cr_read)
4233                         icpt_info.exit_code += info->modrm_reg;
4234                 break;
4235         case SVM_EXIT_WRITE_CR0: {
4236                 unsigned long cr0, val;
4237                 u64 intercept;
4238
4239                 if (info->intercept == x86_intercept_cr_write)
4240                         icpt_info.exit_code += info->modrm_reg;
4241
4242                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4243                     info->intercept == x86_intercept_clts)
4244                         break;
4245
4246                 intercept = svm->nested.intercept;
4247
4248                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4249                         break;
4250
4251                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4252                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4253
4254                 if (info->intercept == x86_intercept_lmsw) {
4255                         cr0 &= 0xfUL;
4256                         val &= 0xfUL;
4257                         /* lmsw can't clear PE - catch this here */
4258                         if (cr0 & X86_CR0_PE)
4259                                 val |= X86_CR0_PE;
4260                 }
4261
4262                 if (cr0 ^ val)
4263                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4264
4265                 break;
4266         }
4267         case SVM_EXIT_READ_DR0:
4268         case SVM_EXIT_WRITE_DR0:
4269                 icpt_info.exit_code += info->modrm_reg;
4270                 break;
4271         case SVM_EXIT_MSR:
4272                 if (info->intercept == x86_intercept_wrmsr)
4273                         vmcb->control.exit_info_1 = 1;
4274                 else
4275                         vmcb->control.exit_info_1 = 0;
4276                 break;
4277         case SVM_EXIT_PAUSE:
4278                 /*
4279                  * We get this for NOP only, but pause
4280                  * is rep not, check this here
4281                  */
4282                 if (info->rep_prefix != REPE_PREFIX)
4283                         goto out;
4284         case SVM_EXIT_IOIO: {
4285                 u64 exit_info;
4286                 u32 bytes;
4287
4288                 if (info->intercept == x86_intercept_in ||
4289                     info->intercept == x86_intercept_ins) {
4290                         exit_info = ((info->src_val & 0xffff) << 16) |
4291                                 SVM_IOIO_TYPE_MASK;
4292                         bytes = info->dst_bytes;
4293                 } else {
4294                         exit_info = (info->dst_val & 0xffff) << 16;
4295                         bytes = info->src_bytes;
4296                 }
4297
4298                 if (info->intercept == x86_intercept_outs ||
4299                     info->intercept == x86_intercept_ins)
4300                         exit_info |= SVM_IOIO_STR_MASK;
4301
4302                 if (info->rep_prefix)
4303                         exit_info |= SVM_IOIO_REP_MASK;
4304
4305                 bytes = min(bytes, 4u);
4306
4307                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4308
4309                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4310
4311                 vmcb->control.exit_info_1 = exit_info;
4312                 vmcb->control.exit_info_2 = info->next_rip;
4313
4314                 break;
4315         }
4316         default:
4317                 break;
4318         }
4319
4320         vmcb->control.next_rip  = info->next_rip;
4321         vmcb->control.exit_code = icpt_info.exit_code;
4322         vmexit = nested_svm_exit_handled(svm);
4323
4324         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4325                                            : X86EMUL_CONTINUE;
4326
4327 out:
4328         return ret;
4329 }
4330
4331 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4332 {
4333         local_irq_enable();
4334 }
4335
4336 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4337 {
4338 }
4339
4340 static struct kvm_x86_ops svm_x86_ops = {
4341         .cpu_has_kvm_support = has_svm,
4342         .disabled_by_bios = is_disabled,
4343         .hardware_setup = svm_hardware_setup,
4344         .hardware_unsetup = svm_hardware_unsetup,
4345         .check_processor_compatibility = svm_check_processor_compat,
4346         .hardware_enable = svm_hardware_enable,
4347         .hardware_disable = svm_hardware_disable,
4348         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4349
4350         .vcpu_create = svm_create_vcpu,
4351         .vcpu_free = svm_free_vcpu,
4352         .vcpu_reset = svm_vcpu_reset,
4353
4354         .prepare_guest_switch = svm_prepare_guest_switch,
4355         .vcpu_load = svm_vcpu_load,
4356         .vcpu_put = svm_vcpu_put,
4357
4358         .update_db_bp_intercept = update_db_bp_intercept,
4359         .get_msr = svm_get_msr,
4360         .set_msr = svm_set_msr,
4361         .get_segment_base = svm_get_segment_base,
4362         .get_segment = svm_get_segment,
4363         .set_segment = svm_set_segment,
4364         .get_cpl = svm_get_cpl,
4365         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4366         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4367         .decache_cr3 = svm_decache_cr3,
4368         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4369         .set_cr0 = svm_set_cr0,
4370         .set_cr3 = svm_set_cr3,
4371         .set_cr4 = svm_set_cr4,
4372         .set_efer = svm_set_efer,
4373         .get_idt = svm_get_idt,
4374         .set_idt = svm_set_idt,
4375         .get_gdt = svm_get_gdt,
4376         .set_gdt = svm_set_gdt,
4377         .get_dr6 = svm_get_dr6,
4378         .set_dr6 = svm_set_dr6,
4379         .set_dr7 = svm_set_dr7,
4380         .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4381         .cache_reg = svm_cache_reg,
4382         .get_rflags = svm_get_rflags,
4383         .set_rflags = svm_set_rflags,
4384         .fpu_activate = svm_fpu_activate,
4385         .fpu_deactivate = svm_fpu_deactivate,
4386
4387         .tlb_flush = svm_flush_tlb,
4388
4389         .run = svm_vcpu_run,
4390         .handle_exit = handle_exit,
4391         .skip_emulated_instruction = skip_emulated_instruction,
4392         .set_interrupt_shadow = svm_set_interrupt_shadow,
4393         .get_interrupt_shadow = svm_get_interrupt_shadow,
4394         .patch_hypercall = svm_patch_hypercall,
4395         .set_irq = svm_set_irq,
4396         .set_nmi = svm_inject_nmi,
4397         .queue_exception = svm_queue_exception,
4398         .cancel_injection = svm_cancel_injection,
4399         .interrupt_allowed = svm_interrupt_allowed,
4400         .nmi_allowed = svm_nmi_allowed,
4401         .get_nmi_mask = svm_get_nmi_mask,
4402         .set_nmi_mask = svm_set_nmi_mask,
4403         .enable_nmi_window = enable_nmi_window,
4404         .enable_irq_window = enable_irq_window,
4405         .update_cr8_intercept = update_cr8_intercept,
4406         .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4407         .vm_has_apicv = svm_vm_has_apicv,
4408         .load_eoi_exitmap = svm_load_eoi_exitmap,
4409         .sync_pir_to_irr = svm_sync_pir_to_irr,
4410
4411         .set_tss_addr = svm_set_tss_addr,
4412         .get_tdp_level = get_npt_level,
4413         .get_mt_mask = svm_get_mt_mask,
4414
4415         .get_exit_info = svm_get_exit_info,
4416
4417         .get_lpage_level = svm_get_lpage_level,
4418
4419         .cpuid_update = svm_cpuid_update,
4420
4421         .rdtscp_supported = svm_rdtscp_supported,
4422         .invpcid_supported = svm_invpcid_supported,
4423         .mpx_supported = svm_mpx_supported,
4424         .xsaves_supported = svm_xsaves_supported,
4425
4426         .set_supported_cpuid = svm_set_supported_cpuid,
4427
4428         .has_wbinvd_exit = svm_has_wbinvd_exit,
4429
4430         .set_tsc_khz = svm_set_tsc_khz,
4431         .read_tsc_offset = svm_read_tsc_offset,
4432         .write_tsc_offset = svm_write_tsc_offset,
4433         .adjust_tsc_offset = svm_adjust_tsc_offset,
4434         .compute_tsc_offset = svm_compute_tsc_offset,
4435         .read_l1_tsc = svm_read_l1_tsc,
4436
4437         .set_tdp_cr3 = set_tdp_cr3,
4438
4439         .check_intercept = svm_check_intercept,
4440         .handle_external_intr = svm_handle_external_intr,
4441
4442         .sched_in = svm_sched_in,
4443 };
4444
4445 static int __init svm_init(void)
4446 {
4447         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4448                         __alignof__(struct vcpu_svm), THIS_MODULE);
4449 }
4450
4451 static void __exit svm_exit(void)
4452 {
4453         kvm_exit();
4454 }
4455
4456 module_init(svm_init)
4457 module_exit(svm_exit)